Attenuating Lymphocyte Activity

Five CD28-like proteins exert positive or negative effects on immune cells. Only four of these five receptors interact with members of the B7 family. The exception is BTLA (B and T lymphocyte attenuator), which instead interacts with the tumor necrosis factor receptor superfamily member HVEM (herpes virus entry mediator). To better understand this interaction, we determined the 2.8-Å crystal structure of the BTLA-HVEM complex. This structure shows that BTLA binds the N-terminal cysteine-rich domain of HVEM and employs a unique binding surface compared with other CD28-like receptors. Moreover, the structure shows that BTLA recognizes the same surface on HVEM as gD (herpes virus glycoprotein D) and utilizes a similar binding motif. Light scattering analysis demonstrates that the extracellular domain of BTLA is monomeric and that BTLA and HVEM form a 1:1 complex. Alanine-scanning mutagenesis of HVEM was used to further define critical binding residues. Finally, BTLA adopts an immunoglobulin I-set fold. Despite structural similarities to other CD28-like members, BTLA represents a unique co-receptor.

Recently the CD28 family member BTLA was unexpectedly shown to bind and be activated by the TNFRSF member herpes virus entry mediator (HVEM, also known as TNFRSF14, HveA, ATAR, TR2, or LIGHTR) (4,5). This is the first example of cross-talk between the CD28 family and the TNFRSF. Whereas HVEM has been previously described as a co-stimulator triggered by the TNF-like ligands lymphotoxin ␣ (LT␣) and LIGHT (6), recent results from HVEM knock-out mice as well as the interaction between BTLA and HVEM are consistent with HVEM playing a co-inhibitory role (7). In addition to binding BTLA, LIGHT, and LT␣, human HVEM is also a host cell receptor for herpes simplex virus 1 by binding to herpes simplex virus 1 glycoprotein D (gD) (8).
Structurally, the connection between the IgSF family represented by BTLA and the TNFRSF proteins such as HVEM is unexpected. Crystal structures of CD28, CTLA-4, and programmed death-1 have revealed that these co-stimulatory and co-inhibitor receptors are all members of the immunoglobulin superfamily with each protein containing an extracellular IgV domain (9 -11). Based on sequence analysis, BTLA was also expected to contain an extracellular IgV domain. Similarly, the extracellular domains of B7-like proteins are comprised of Ig domains. Cocrystal structures of B7-1 and B7-2 bound to CTLA4 show that Ig domains from the receptor and ligand pack against each other forming a compact interface (12,13). In contrast, the TNFSF and TNFRSF members are formed by very different structural elements and interact in a distinctive manner determined by the quaternary structure of TNF-like ligands. These proteins are homotrimeric or occasionally heterotrimeric proteins comprised of jelly-roll monomers. Multidomain TNFRSF family members are comprised of multiple pseudo-repeats of a cysteinerich motif. Structures of signaling complexes formed by TNF-like ligands and receptors show that the elongated receptors bind at monomer-monomer interfaces on the ligands in a manner much different from the compact B7-CD28-type interaction (14).
The BTLA-HVEM interaction is also unusual in that it represents the first example of a TNFRSF functioning as a "ligand" and is one of a handful of examples of a TNFRSF interacting with a non-TNF-like ligand. In addition to the BTLA-HVEM and gD-HVEM interactions, other TNFSFR, which bind unusual ligands, include the low affinity neurotrophin receptor p75, which binds the cystine knot growth factor NGF, and feline OX40, which acts as a co-receptor for the feline immunodeficiency virus (15,16). Crystal structures of the relevant complexes show that gD protein interacts primarily with the N-terminal cysteinerich domain (CRD1) of HVEM on the surface opposite the TNFSF binding site (17). Similarly, p75 uses the same respective surface on CRD1 and a part of CRD2 to bind NGF (18). Previous biochemical characterization suggests that BTLA also binds to HVEM on the CRD1 distal to the TNFSF binding site (4,5).
To obtain a more detailed understanding of the novel interaction between the CD28-like protein BTLA and the TNFRSF member HVEM, we have determined the 2.8-Å crystal structure of the BTLA-HVEM complex. This structure shows that despite major structural differences between BTLA and gD, they bind to an overlapping site on HVEM using a similar ␤-sheet binding motif. We have used alaninescanning mutagenesis and Scatchard assays to identify critical BTLAbinding residues on HVEM. Mutations that significantly reduced binding affinities were in close agreement with the crystal structure. Light scattering demonstrates that the recombinant extracellular domain of BTLA is monomeric and that BTLA and HVEM form a 1:1 complex in solution. Despite the homology and functional similarities between BTLA and the CD28 family, BTLA contains structurally unique features. Moreover, compared with the CD28-B7 binding site, BTLA uses a distinct surface to interact with HVEM. Finally, using the BTLA-HVEM structure, we propose a hypothetical model for a BTLA-HVEM-TNF ternary complex.

MATERIALS AND METHODS
Recombinant Protein Expression and Purification-DNA encoding human BTLA residues 26 -137 (the initial methionine is residue 1) with the addition of a C-terminal His 8 tag was expressed in Escherichia coli. Inclusion bodies from BTLA expressing E. coli were extracted under denaturing conditions, and the protein was purified on a Ni-NTA metal chelate column as described (19). Fractions were pooled and diluted to 50 g/ml with buffer containing 0.1 M Tris (pH 8.6), 0.3 M NaCl, 20 mM glycine, 1 mM EDTA, 1 mM glutathione (oxidized), and 1 mM glutathione (reduced). The refolding mixture was incubated overnight at 2-8°C, and the pH adjusted to pH 3.0 with trifluoroacetic acid. The acidified refolding mixture was loaded onto an RP-HPLC Vydac C4 column (1.0 ϫ 25 cm) equilibrated with 0.1% (w/v) trifluoroacetic acid in water and eluted with a linear gradient of acetonitrile (from 15 to 55%) in 0.1% trifluoroacetic acid at 3 ml/min for a total of 35 min. Protein fractions were pooled, and the acetonitrile was removed by evaporation assisted by a gentle stream of N 2 . The RP-HPLC pool was buffer exchanged using a HiTrap Desalting column (Amersham Biosciences) equilibrated with buffer containing 10 mM HEPES (pH 6.8), 0.15 M NaCl. BTLA activity was evaluated using SPR (Biacore).
BTLA-Fc fusion protein was expressed in Chinese hamster ovary cells as previously described (4). The BTLA-Fc construct contains a Genenase site (New England Biolabs) between the BTLA extracellular domain (ECD) and the Fc fusion. BTLA ECD was cleaved from the Fc domain by the addition of a 1:100 ratio of Genenase to BTLA-Fc in phosphate-buffered saline. After 2 h at room temperature, the reaction was quenched with a protease inhibitor mixture and loaded onto a protein A-Sepharose column. Cleaved BTLA was collected from the flow through.
DNA encoding a truncated human HVEM ECD, residues 1-105 (residue numbered as in Protein Data Bank code 1JMA; residue 1 is the first residue in the mature protein), corresponding to the ordered portion of HVEM in the gD-HVEM complex, including CRD1, CRD2, and the N-terminal 22 residues of CRD3, was subcloned into the pET15b vector and subsequently cloned into the baculovirus secretion vector pAcGP67-B. Virus was made by co-transfection and three rounds of viral amplification in Sf9 cells. High titer virus was used to infect Hi5 insect cells, and protein was expressed for 3 days and the media treated as described (20) prior to being loaded onto a Ni-NTA-agarose column and eluted with 250 mM imidazole buffer. Thrombin was added (1 unit/mg of HVEM) to remove the N-terminal His tag, and the solution was dialyzed overnight at 4°C against 150 mM NaCl, 20 mM Tris (pH 8.0). The dialysate was concentrated and loaded onto an S-75 sizing column equilibrated in 150 mM NaCl, 20 mM Tris (pH 8). Fractions of purified HVEM were pooled and yielded ϳ0.5 mg of protein/liter of Hi5 cells. As a control, a construct of full-length human HVEM ECD (here-after referred to as HVEM L ) with a C-terminal His tag was also expressed and purified. This construct contains the same HVEM residues as that used by Carfi et al. (17) and Whitbeck et al. (21). For crystallography, the BTLA-HVEM complex was made by adding excess HVEM residues 1-105 to E. coli-expressed BTLA and incubating at room temperature for 1 h. The complex was concentrated and purified from excess HVEM (residues 1-105) with an S-75 sizing column equilibrated in 150 mM NaCl, 20 mM Tris (pH 8.0). Fractions containing purified complex were pooled and concentrated to 9 mg/ml. DNA encoding LIGHT extracellular domain residues 91-240 was expressed and purified in the same manner as HVEM with a final purification on Superdex S-200 sizing equilibrated in 150 mM NaCl, 20 mM Tris (pH 8.0). Expression yielded greater than 1 mg of purified protein/ liter of Hi5 cells.
Crystallographic Data Collection and Structure Determination-Crystals of the BTLA-HVEM complex were grown by vapor diffusion at 19°C using the sitting drop method. Crystals formed in drops containing protein solution were mixed with an equal volume of reservoir solution containing 2.0 M sodium formate, 0.1 M sodium acetate (pH 4.6). The resulting small, clustered crystals were used to seed new drops yielding larger, single crystals. The crystals were transferred briefly to a droplet containing reservoir solution with 30% glycerol before flashcooling in liquid nitrogen. The crystals belonged to space group C222 1 , and the asymmetric unit contained two copies of the BTLA-HVEM complex. A data set to 2.8-Å resolution was measured from a single crystal at beam line 5.0.1 of the Berkeley Center for Structural Biology at the Advanced Light Source. The data were processed using the HKL package (22).
Structures of HVEM (chain B in PDB code 1JMA) and murine BTLA (PDB 1XAU; structure determined and deposited by D. Fremont and co-workers in 2004) were used as search models to determine the structure of the BTLA-HVEM complex by molecular replacement. Side chains, which differed between murine and human BTLA, were manually trimmed to C-␤. The program Phaser (23) gave a clear solution. A 2-fold noncrystallographic symmetry-averaged and solvent-flattened map using program dm (24) revealed clear density for the missing hBTLA side chains. The model was refined with REFMAC5 (25) with tight noncrystallographic symmetry restraints on residues 34 -144 of BTLA and residues 4 -83 of HVEM. The last 10 residues of BTLA were not included in the noncrystallographic symmetry restrains as they differed significantly because of the involvement of the C-terminal His tag in crystal packing contacts. Additional density was observed for a metal ion interacting with the His tag of symmetry related copies of chain A. This ion was modeled as a Ni(II) and is likely an artifact because of purification of BTLA by Ni-NTA affinity chromatography. Refinement and model statistics are shown in TABLE ONE. The coordinates for the BTLA-HVEM complex have been deposited in the Protein Data Bank and assigned access code 2AW2.
Alanine Scanning Mutagenesis and Cell Binding Affinity Assays-The QuikChange site-directed mutagenesis kit (Stratagene) was used as recommended by the manufacturer to generate single alanine mutations in HVEM. Mutant HVEM sequences were confirmed by DNA sequencing. Recombinant BTLA-Fc was iodinated by the lactoperoxidase (Biotrend) method and LIGHT by the IODO-GEN (PerkinElmer Life Sciences) method. Displacement binding studies were done as previously described (4) with 0.5 nM labeled BTLA and varying concentrations of unlabeled protein. LIGHT binding was confirmed with 0.5 nM labeled LIGHT with or without 1000-fold excess of unlabeled protein.
LIGHT binding to alanine mutants was normalized as a percentage of wild type binding. AD-293 cells were transiently transfected with either wild type or alanine mutant HVEM cDNA as previously described (4) using pRK mock-transfected cells as a negative control. Expression of wild type and alanine mutant HVEM was confirmed by flow cytometry as previously described using fluorescein isothiocyanate-conjugated mouse anti-human HVEM (clone 122, MBL) in comparison to an isotype control (BD Biosciences). The HVEM antibody epitope is contained in all HVEM point mutants. Expression levels of point mutants are normalized to that of wild type HVEM as denoted by the percentage of cells above a given mean fluorescence threshold determined by isotype control antibody staining.
Light Scattering-Molar Mass determination was carried out using a Agilent 1100 series (Agilent, Palo Alto, CA) HPLC system in line with a Wyatt MiniDawn MALS (multiangle light scattering) detector (Wyatt Technology, Santa Barbara, CA). Concentration measurements were made using an online Wyatt OPTILAB DSP interferometric refractometer (Wyatt Technology). Astra software (Wyatt Technology) was used for light scattering data acquisition and processing. Either a Shodex 803 or a S75 10/300 column (Amersham Biosciences) equilibrated with filtered phosphate-buffered saline (pH 7.2) was used with a flow rate of 1 ml/min. Both the light scattering unit and the refractometer were calibrated as per the manufacturer's instructions. A value of 0.180 ml/g was assumed for the dn/dc ratio of the protein. Measuring the signal from monomeric bovine serum albumin normalized the detector responses. The temperature of the light scattering unit was maintained at 25°C, and the temperature of the refractometer was kept at 35°C. The column and all external connections were at ambient temperature (20 -25°C). Recombinant BTLA produced in Chinese hamster ovary cells, the purified HVEM-BTLA complex used for crystallization, and HVEM L was loaded at 1.0 mg/ml.

RESULTS
The structure of the human BTLA-HVEM complex was solved by molecular replacement using the structures of the HVEM ECD (PDB code 1JMA, chain B) and the 1.8-Å structure of murine BTLA ECD (PDB code 1XAU) as search models with the program Phaser (23). A solution was found with two copies of the BTLA-HVEM complex forming the asymmetric unit. The structure was manually rebuilt and refined to an R/R free of 23.1 and 27.8%, respectively (TABLE ONE, Fig. 1). The final model consists of BTLA residues 34 -137 and HVEM residues 5-92 and 95-102. In both copies of BTLA, an additional 4 -6 residues from the C-terminal His tag were well ordered in the electron density and are included in the model. Noncrystallographic symmetry restraints were used throughout the refinement and the two copies of the BTLA-HVEM complex are very similar.
This structure reveals that human BTLA, like murine BTLA (PDB code 1XAU), is a compact IgG domain composed of two flat ␤-sheets, which are formed by strands B, E, and D in one sheet and strands AЈ, G, F, C, and CЈ in the other (Fig. 1B). The sheets are buttressed by three disulfides (residues 72-79, 34 -63, and 58 -115). The Cys 72 -Cys 79 disulfide connects the C and CЈ strands; the Cys 34 -Cys 63 disulfide joins the N-terminal region preceding strand AЈ to the CD loop, and the Cys 58 -Cys 115 disulfide connects the B and F strands. This disulfide is completely buried in the hydrophobic core of BTLA and is part of the "Y-corner" motif, DX(G/A)DXYXC. The B-F strand disulfide and the Y-corner motif are both highly conserved features in IgSF domains. In addition to the Cys 58 -Cys 115 disulfide, the Cys 34 -Cys 63 disulfide is also conserved in murine BTLA. The Cys 72 -Cys 79 disulfide is not conserved in all murine BTLA alleles (see below). This disulfide is not present in the variant of murine BTLA, which was crystallized (PDB code 1XAU) in which the cysteine corresponding to Cys 79 is replaced by a tryptophan.
The sequence of human BTLA contains three putative N-linked glycosylation sites. Expression of recombinant BTLA in eukaryotic cells results in a protein with significant glycosylation that is unsuitable for crystallization. For the structural studies, recombinant BTLA was expressed in E. coli cells to produce a protein without glycosylation. Examination of the BTLA-HVEM complex (below) shows that the putative N-linked glycosylation sites are away from the binding site and would not be expected to affect the interaction between BTLA and HVEM ( Fig. 1). Analysis of the interaction between BTLA expressed in E. coli and HVEM by surface plasmon resonance shows that glycosylation is not required for HVEM binding. 3 The BTLA-HVEM Complex-The BTLA-HVEM complex consists of a single globular BTLA interacting with the membrane distal region of rod-shaped HVEM (Fig. 1). BTLA binds HVEM using two short segments: an N-terminal extension preceding strand AЈ (residues 35-43) and the short G°strand (residues 118 -128). The HVEM binding site for BTLA consists almost exclusively of residues from CRD1. BTLA residues 35-43 interact with HVEM residues 26 -33, which form the "tip" of HVEM CRD1 distal to the C terminus. This loop in HVEM leads to a strand (residues 33-38) that, together with the G°strand from BTLA, makes the heart of the binding interface. These two strands form a small anti-parallel intermolecular ␤-sheet. This interaction is primarily mediated by main chain hydrogen bonds and includes relatively few side chain contacts. These two interactions, in conjunction with small contribution from the BTLA CCЈ loop, generate an interface that buries ϳ1800 Å 2 of solvent accessible surface area, which is contributed equally by both binding partners. This complex positions the C termini of the two proteins in opposite directions consistent with the BTLA-HVEM complex forming between proteins resident on different cells. 3 L. Gonzalez, unpublished data. Moreover, as full-length HVEM includes an additional ϳ60 residues prior to the transmembrane helix, it is also possible that the binary BTLA-HVEM interaction could occur between proteins residing on the same cell. Sequence Polymorphism in Murine BTLA-Three different BTLA alleles have been isolated from 23 mice strains. These alleles have been labeled BALB/c-like, MRL/lpr-like, and C57BL/6-like according to the strains from which they were derived (26). The crystal structure of murine BTLA is of the BALB/c-like variant. The BALB/c and MLR/lpr alleles differ at only one amino acid, whereas the C57BL/6-like allele differs from the other two by 10 or 11 amino acids, respectively, in the extracellular Ig domain (26). The structural and functional consequences of these differences can be predicted based on the structure of the human BTLA-HVEM complex. The most interesting difference is the polymorphism between Trp and Cys at position 85 corresponding to the human residue Cys 79 (hereafter human BTLA residue numbers will follow in parentheses). In the C57BL/6 allele, the presence of a Cys at this position suggests that a disulfide analogous to the Cys 72 -Cys 79 disulfide in hBTLA will be formed. The disulfide containing variant may be more stable as solvent-exposed unpaired cysteine residues can lead to inappropriate oxidation, oligomerization, or misfolding. Another interesting difference between the alleles is that in the C57BL/6 variant, glutamines replace a glutamate and an arginine at positions 72 (h66) and 102 (h93), respectively, resulting in exchange of a solvent-exposed salt bridge with a potential hydrogen bonding interaction. Three of the remaining differences, E41P(h35), N45T(h39), and K47T(h41), affect residues in the vicinity of the expected receptor binding site but are not predicted to make significant contacts to murine HVEM. The remaining polymorphisms in murine BTLA occur outside the HVEM binding site and are not expected to affect either BTLA structure or HVEM affinity. In summary, because none of the side chains of any of the variant residues are predicted to be in intimate contact with murine HVEM, all three alleles are expected to code for proteins with comparable affinity for HVEM.
Functional Characterization of the BTLA-HVEM Interface-Alanine scanning mutagenesis (27) was used to identify the functional epitope on HVEM for BTLA. Based on prior characterization of this complex (4), 15 residues were selected for mutagenesis based on three criteria: conservation in murine and human BTLA, location on the side of HVEM away from the expected TNFSF binding site, and solvent exposure (TABLE TWO). Single alanine-substituted variants of HVEM were expressed transiently in AD-293 cells, and BTLA binding data were measured for Scatchard analysis. Six residues (Glu 8, Pro 17 , Tyr 23 , Lys 26 , Glu 31 , and Val 36 ) showed greater than 2-fold reduction in affinity when mutated to alanine (Fig. 2). These six residues can be divided into two groups based on the severity of the effect of the alanine substitution. Substitution at positions Pro 17 , Tyr 23 , and Val 36 had a more pronounced effect than substitutions at positions Glu 8 , Lys 26 , and Val 31 . Variant proteins with alanine at either Pro 17 or Tyr 23 had no detectable affinity for BTLA, although LIGHT binding was largely intact indicating that the HVEM fold was not seriously compromised. The side chains of these two residues pack against either end of the intermolecular ␤-sheet likely providing crucial van der Waals contacts. Mutation of HVEM residue Val 36 to alanine resulted in a 10-fold reduction in BTLA affinity. The Val 36 side chain packs against BTLA residue Ile 124 , whereas the backbones of these two residues form reciprocal anti-parallel hydrogen   bonds at the center of the intermolecular anti-parallel ␤-sheet. Significantly, the three most disruptive alanine substitutions were located at residues involved in the formation of the HVEM-BTLA intermolecular ␤-sheet. Alanine substitutions at Glu 8 , Lys 26 , and Glu 31 all caused an ϳ3-fold decrease in HVEM affinity for BTLA. These residues are located at the periphery of the binding interface (Fig. 2). Alanine mutation at nine other residues had no effect on BTLA binding. Of these residues, only Thr 33 and Pro 39 bury more than 50% of their accessible surface area in the BTLA-HVEM interface. Thus the alanine scanning and structural data both indicate that the anti-parallel strand and immediately surrounding interactions are the energetic and structural core of the BTLA-HVEM interface.

BTLA-HVEM Complex Forms a Stable Heterodimer in Solution-
The crystallographic asymmetric unit contains two BTLA-HVEM complexes forming a dimer of dimers (Fig. 3). To ascertain the relevance of this interaction, the stoichiometry of the BTLA-HVEM complex in solution was characterized by multiangle light scattering. This analysis showed that the complex formed by E. coli-derived BTLA and truncated HVEM ECD, which was used for crystallography, as well as the complex of Chinese hamster ovary-derived glycosylated BTLA with the entire HVEM ECD both form stable heterodimers with no indication of higher order assemblies in solution at a concentration of ϳ40 M (Fig. 3). In addition, both glycosylated and E. coli-derived BTLA were found to be monomeric in solution. This data, in combination with the relatively small area of contact between the adjacent heterodimers in the crystal-  Fig. 1a. b, representative light scattering data showing glycosylated BTLA ECD is monomeric in solution with apparent molecular mass of 22,000 g/mol. C, representative light scattering data showing that the ECD of BTLA and HVEM form a heterodimeric complex in solution at a concentration of 40 M apparent molar mass of 24,600 g/mol. HVEM L (the entire HVEM ECD with a C-terminal His tag) and the soluble complex between HVEM L and glycosylated BTLA were also assessed using this procedure. Soluble HVEM L was found to be monomeric with an apparent molecular mass of 12,400 g/mol, whereas the HVEM L -glycosylated BTLA complex appears heterodimeric with an approximate molecule mass of 32,400 g/mol. The apparent molecular mass of the HVEM L and the HVEM L -glycosylated BTLA complexes are slightly lower than predicted likely because of proteolytic degradation of the C-terminal portion of HVEM L . lographic asymmetric unit, suggests that the interaction between the two heterodimers is weak.

DISCUSSION
Comparison of the gD and BTLA Binding Sites on HVEM: gD Mimics BTLA-Comparison of the BTLA-HVEM structure to the previously determined gD-HVEM structure (17) shows that the BTLA and gD binding sites on HVEM are largely overlapping and involve similar structural motifs as well as similar energetics (Fig. 4). In particular, the BTLA binding site on HVEM is almost entirely contained within the gD binding site; of the 24 HVEM residues that lose solvent-exposed surface area upon binding with BTLA, 20 are also involved in binding gD. Like BTLA, gD presents a short ␤-strand that makes an anti-parallel interaction with HVEM strand 34 -39. Peripheral interactions between gD and HVEM are formed by a loop in gD contacting the surface of HVEM at the junction of CRD1 and CRD2. The interaction between the short ␤-strands presented by BTLA or gD and HVEM are very similar and are at the center of both the BTLA-HVEM and gD-HVEM interfaces. Despite the similarity of the structure and function of this strand in BTLA and gD, a comparison of the sequences of the two strands shows that they have no sequence identity and only modest sequence similarity, indicating the importance of the backbone hydrogen bonds in forming the inter-molecular ␤-sheet (Fig. 4). Of the similar residues, Ile 124 in BTLA and Leu 28 in gD make the most comparable side chain interactions. Both of these residues pack against hydrophobic residue Val 36 in HVEM, which was identified by alanine scanning mutagenesis as important for BTLA binding and is energetically important for gD binding as well (28).
The energetics of these two binding sites are also similar, reflecting the structural mimicry of the BTLA binding site by gD. Connolly et al. (28,29) used alanine scanning to dissect the HVEM-gD binding site and identified 11 positions in HVEM for which alanine substitutions resulted in impaired gD binding. Six of these residues were also analyzed in this work and we find that alanine substitutions at four of them, Pro 17 , Lys 26 , Val 36 , and Tyr 23 , also affect BTLA binding. The gD functional epitope also includes Pro 39 and Ser 74 , which are neutral for BTLA binding. Remarkably, Tyr 23 appears to be the energetic heart of both the gD and BTLA binding sites as replacement of this residue by alanine abolishes both BTLA and gD binding.
Whereas gD and BTLA both bind HVEM and compete for the same site on HVEM, their structures are strikingly different (Fig. 4). gD is formed by a core Ig domain that is decorated with a N-terminal extension (ϳ50 residues) lacking regular secondary structure and a longer partially helical C-terminal extension (ϳ100 residues). The gD Ig domain does not contact HVEM directly. Instead, the N-terminal extension forms most of the contacts to HVEM and is supported by an ␣-helix from the C-terminal extension (17). In contrast, BTLA consists solely of an Ig domain and uses this domain to bind HVEM. Despite these differences, BTLA and gD have converged on a very similar structural and energetic solution for specifically binding HVEM.
This convergence suggests that formation of this anti-parallel intermolecular ␤-sheet is a favorable interaction. Intriguingly, the HVEM strand ( 35 TVCEP 39 ) at the heart of the BTLA and gD binding sites is one of the relatively few conserved elements of secondary structure in the TNFRSF-fold. This strand is present in both death receptor-5 (30,31) and TNFR1 (32), the only other multidomain TNFRSF family members that have been structurally characterized. In TNFR1, the strand is present but shorter than in HVEM. The structure of death receptor-5, which has a truncated CRD1, retains a diminished strand at this position although it is occluded by the death receptor-5 N terminus. At the sequence level, Thr 35 -Val 36 is present at this position in ϳ30% of multidomain TNFRSF members and Cys 37 is invariant. Position 38 is solvent exposed and is not conserved, whereas Pro 39 is present in ϳ60% of multidomain TNFRSF sequences. This conservation of sequence and structure suggests that other TNFRSF could interact with non-TNFSF binding partners by forming a similar anti-parallel ␤-sheet binding motif.
BTLA Is a Member of the I-set of IgSF Domains-In contrast to previous suggestions based on sequence analysis (33), examination of the BTLA ECD structure indicates that it belongs to a different subset of the IgSF than the CD28-like family. IgSF domains can be divided into 4 main classes, the Variable (V), Constant 1 (C1), Constant 2 (C2), and Intermediate (I) sets, based on their structures and sequences (34). One of the distinguishing features of I-set domains is the lack of a CЉ strand. The present structure shows that the BTLA ECD domain lacks this strand and thus is better described as a member of the I-set of Ig domains. Examination of 20 "fingerprint " residues, which differentiate I-set and V-set Ig domains from other Ig domains (34), confirms this analysis (Fig.  5). The structures of CTLA4 (10), programmed death-1 (11), and CD28 (9) as well as the sequence of inducible T cell costimulator indicate that these proteins all possess a CЉ strand and are better described as members of the V-set of Ig domains. Furthermore, these four known CD28like proteins are all located in close proximity on chromosome 2, whereas BTLA is located separately on chromosome 3. This analysis, as well as the fact that BTLA binds a structurally different ligand, strongly suggests that BTLA is distinct from the CD28 family.
BTLA Utilizes a Distinct Binding Surface Compared with CD28 Family Members-The structures of CTLA4 bound to B7-1 and B7-2, which is expected to be representative of CD28 family interactions, shows that the conserved MYPPPYYL sequence in the CTLA4 GF-loop, in conjunction with residues on the C and F strand, form a relatively flat HVEM residues crucial for gD affinity (C37 and Y23 (28)) are colored red and labeled. c, sequence of the structurally similar short ␤-strands in BTLA and gD that bind HVEM.
binding surface (12,13). In contrast, the BTLA binding surface is located along the edge of the I-set Ig domain almost orthogonal to the CTLA4 binding surface (Fig. 5). Even more strikingly, the CTLA4-B7 complexes suggests that at least some of these proteins are capable of forming a periodic repeating array at the immunological synapse. For instance, a covalent "outward" facing CTLA4 dimer can interact with two separate B7-like dimers (Fig. 6). This type of interaction may lead to the clustering of a large number of CTLA4 molecules resulting in high local intracellular concentrations of ITIM or ITAM domains (12,13). In contrast, multiangle light scattering experiments on purified BTLA-HVEM extracellular domain complexes indicate that the only species present in solution, even at relatively high concentrations, is a single BTLA-HVEM complex (Fig. 3). Thus BTLA signaling must be activated in a different manner than that of CTLA4 or CD28. Instead, altered recruitment or increased lifetime at the immunological synapse may facilitate BTLA attenuation of T cell activation.
Implications for BTLA-HVEM-Ligand Ternary Complexes and HVEM Signaling-In addition to binding BTLA, HVEM is also known to bind TNFSF family members LIGHT and LT␣. BTLA binding does not occlude the TNFSF binding site on the opposite face of HVEM and thus HVEM should be able to simultaneously bind BTLA and either LIGHT or LT␣ to form a stable ternary complex. We have been able to purify a complex containing recombinant LIGHT, HVEM, and BTLA by size exclusion chromatography although the stoichiometry of this complex has not been determined. 4 As LIGHT is not required for BTLA activation, the physiological relevance of a ternary complex is unclear at this time (7,33). A model of a possible BTLA-HVEM-LT␣ ternary complex was generated by assuming that the LT␣-HVEM complex will be very similar in geometry to the LT␣-TNFR1 complex (32). The geometry of this complex model predicts that LIGHT and BTLA would be expressed on the same cell, whereas HVEM would be expressed on a different cell (Fig. 6b). Finally, because BTLA does not promote trimeric clustering of HVEM, it is unlikely to have the same effect on HVEM activation as LIGHT binding, which causes three copies of HVEM to cluster and trigger cytosolic signal transduction.
Conclusions-The BTLA-HVEM complex simultaneously provides examples of both convergent and divergent evolution and demonstrates the plasticity of evolutionary relationships determining protein-protein interactions. Unexpectedly, our data showed that two structurally very different proteins, BTLA and gD, have converged not just on the same binding site but also use similar elements of secondary structure to bind HVEM. In addition, the BTLA-HVEM structure further differentiates BTLA from the CD28 family of proteins. BTLA and CTLA4 differ more, in both structure and binding mode, than had been expected from amino acid sequence. BTLA may have diverged from the CD28-like  (white, PDB code 1I8L), and CD28 (green, PDB code 1YJD). Loop regions have been smoothed for clarity. Strands are labeled. The CЈ and CЉ region of CD28 and CTLA4, which is missing in BTLA, is boxed. b, molecular surface of BTLA in the same orientation as in a. Residues forming the BTLA-HVEM interface are colored by % accessible surface area buried upon complex formation (25-50%, yellow; 50 -75%, orange; 75-100%, red). c, molecular surface of CTLA4 in the same orientation as in a. Residues forming the CTLA4-B7-1 interface are colored as described for BTLA. d, sequence alignment of Ig domains of human BTLA, murine BTLA, human CD28, and human CTLA4. Secondary structure of hBTLA is shown above the alignment. The CЈ and CЉ strands in CTLA4 and CD28 are labeled with green italics. A blue asterisk appears above every fifth BTLA residue. The 20 V-frame fingerprint residues are indicated with black dots. Residues that fulfill this criterion are in bold. Orange shading, BTLA residues that bury at least 25% of accessible surface area upon binding HVEM. Green shading, residues in CTLA4 that bury at least 25% of accessible surface area upon binding B7-1. Predicted glycosylation sites in hBTLA are boxed in purple. Murine BTLA residues that differ in BALB/c-like, MLP/lpr-like, and C57BL/6-like alleles (26) are underlined.
family and evolved to recognize the TNF receptor HVEM as imparting an evolutionarily favorable and distinct function.