Structure of human cytomegalovirus UL144, an HVEM orthologue, bound to the B and T cell lymphocyte attenuator

Human cytomegalovirus (HCMV) is a β-herpesvirus that has co-evolved with the host immune system to establish lifelong persistence. HCMV encodes many immunomodulatory molecules, including the glycoprotein UL144. UL144 is a structural mimic of the tumor necrosis factor receptor superfamily member HVEM (herpesvirus entry mediator), which binds to the various ligands LIGHT, LTα, BTLA, CD160, and gD. However, in contrast to HVEM, UL144 only binds BTLA, inhibiting T-cell activation. Here, we report the crystal structure of the UL144–BTLA complex, revealing that UL144 utilizes residues from its N-terminal cysteine-rich domain 1 (CRD1) to interact uniquely with BTLA. The shorter CRD2 loop of UL144 also alters the relative orientation of BTLA binding with both N-terminal CRDs. By employing structure-guided mutagenesis, we have identified a mutant of BTLA (L123A) that interferes with HVEM binding but preserves UL144 interactions. Furthermore, our results illuminate structural differences between UL144 and HVEM that explain its binding selectivity and highlight it as a suitable scaffold for designing superior, immune inhibitory BTLA agonists.

Although the immune system has evolved to provide protection against various pathogens, many have developed strategies to thwart host immunity, including large dsDNA viruses such as ␤-herpesviruses and poxviruses (1,2). Pathogen strategies include both preventing immune-activating ligands from reaching the infected cell surface and encoding mimics of host cell proteins to avoid missing self-recognition (3)(4)(5)(6). The molecular mimicry strategy is often used by human cytomega-lovirus (HCMV). 3 HCMV is a ␤-herpesvirus that is largely asymptomatic in the immunocompetent host, but can mediate severe disease in the immunocompromised or naïve, and it is the number one infectious cause of congenital infection (7). In addition, HCMV has an enormous impact on shaping the circulating immune repertoire over a lifetime of infection (8), and therefore vaccine development is a high priority (7,9). Members of the tumor necrosis factor (TNF) receptor superfamily (TNFRSF) play a crucial role in maintaining immune homeostasis by both mediating antiviral defenses and restricting tissue pathogenesis during infection (10). In turn, HCMV encodes multiple immunomodulatory proteins that restrict expression of TNF receptors (UL144, UL141, etc.) to dampen immune activation (11).
UL144 encoded by clinical/low-passage isolates of HCMV is the only primary sequence orthologue of TNFRSFs encoded by the herpesvirus family (12). Unlike other TNFRSF orthologues secreted by poxviruses, UL144 is a membrane-anchored glycoprotein that appears to be expressed largely intracellularly in infected cells, although it can localize to the cell surface when overexpressed (13,14). UL144 is also expressed in latently infected myeloid cells, suggesting a potentially important role in the immune regulation of HCMV reactivation (15). UL144 is composed of an N-terminal signal peptide, an extracellular region containing two cysteine-rich domains (CRD) followed by a transmembrane domain and a short cytoplasmic tail. UL144 sequence varies significantly between viral clinical isolates, and three major sequence groups exist (14,16,17).
UL144 is a viral mimic of the TNFRSF member HVEM (herpesvirus entry mediator), originally identified as an entry receptor for herpes simplex virus that binds the major viral gD protein (18). HVEM contains four CRDs (18) and acts as a checkpoint regulator for directing T-cell activation (19,20). HVEM can both activate immune cells by interacting with the TNF family ligands LIGHT and LT␣ and be immune inhibitory by binding the immunoglobulin superfamily members BTLA and CD160 (19,21,22). Earlier structural and functional studies showed that the CRD1 region of HVEM is responsible for binding to BTLA, CD160, and gD, whereas the CRD2/CRD3 regions contain the LIGHT-or LT␣-binding site (16). In contrast to promiscuous binding of HVEM, UL144 has evolved to exclusively bind to BTLA (B and T lymphocyte attenuator), an inhibitory co-receptor of the CD28-like immunoglobulin superfamily member (23)(24)(25)(26).
UL144 binds to BTLA via its CRD1 region, and all UL144 sequence groups retain strong BTLA binding despite significant sequence divergence among clinical viral isolates (16). UL144 has been reported to have several functional consequences on antiviral immunity, one being to induce production of CCL22 (a macrophage-derived chemokine) via TRAF6 (TNF receptor-associated factor 6)-mediated NF-B signaling to enhance Th2 responses (12). In contrast, UL144 can be antiinflammatory by evading CD160-mediated activation of NK cells (22). Together, these data indicate that UL144 likely plays multiple roles in regulating immunity to CMV infection.
To elucidate the structural basis of how UL144 has evolved to exclusively target BTLA, we have determined the crystal structure of the UL144 -BTLA complex at a resolution of 2.7 Å. Subsequent structure-guided mutagenesis has revealed specific hot spots within the UL144 -BTLA interaction that could be optimized further in an attempt to design superior BTLA agonists that may have indications as anti-inflammatory therapeutics.

Protein expression and characterization of the UL144 -BTLA complex
The design of expression constructs for both BTLA and UL144 is based on the domain architecture shown in Fig. 1, A and B. Full-length BTLA was expressed on the surface of HEK293T cells for subsequent binding analysis of UL144-Fc fusion proteins from the three different groups using a FACSbased binding assay (Fig. 1C). UL144-Fc fusion proteins were generated in HEK293T cells. Our studies confirmed that the Crystal structure of the UL144 -BTLA complex group 3 protein (UL144 -Fiala) shows modestly enhanced BTLA binding compared with representative groups 1 and 2 proteins (Fig. 1C). Therefore, we chose to express the UL144 protein from the Fiala strain (UL144-F) (residues 21-132) (14) for our structural and biochemical studies. Because the ectodomain of UL144 contains a total of eight putative N-linked glycosylation sites that were spread across CRD2 and the membrane extension region, expression of UL144 in mammalian HEK293T cells yielded a heavily-glycosylated protein that we assumed would impede subsequent crystallization. Therefore, for structural studies, we have expressed the UL144 protein in Sf9 insect cells. The insect cell-expressed WT UL144 protein exhibited considerably reduced glycosylation compared with that produced in the mammalian expression system. However, attempts to obtain crystals of WT UL144 were unsuccessful, likely due to the still high amounts of flexible N-linked glycans. To overcome this, we generated random combinations of different N-linked glycosylation site mutants of UL144 and purified the individual mutants using ion metal-affinity chromatography using the C-terminal hexahistidine tag, followed by sizeexclusion chromatography (Fig. 1D). The SDS-PAGE analysis of these variants revealed disparity in the extent of glycosylation modification and the relative expression levels (Fig. 1E). UL144 variants were subjected to crystallization trials, either by themselves or in complex with BTLA. For structural studies, the immunoglobulin-like domain (residues 31-137) of human BTLA was expressed in Escherichia coli and refolded as reported previously (24). Several diffracting crystals of the UL144 -BTLA complex containing construct no. 11 of UL144 were obtained. Construct no. 11 lacked all N-glycans except at Asn-86 and migrated as a homogeneous band, yet resulted in lower expression levels (Fig. 1E).

Structure of the UL144 -BTLA complex
The UL144 -BTLA crystals belonged to space group P2 1 , and the structure was determined by molecular replacement using the available BTLA structure, combined with experimental phases obtained by sulfur-single anomalous dispersion. The structure was refined to a resolution of 2.7 Å ( Table 1). The asymmetric unit of the crystal contained four copies of the complex. In the final structure, with the exception of some flexible loops, the BTLA structure was well-ordered in all four protomers (amino acids 34 -135). However, although the CRD1 and the majority of the CRD2 region of UL144 are ordered, we have not observed obvious electron density corresponding to the membrane extension region (residues 96 -132) in any of the four copies. This may reflect a greater flexibility in this region that links the UL144 ectodomain to the membrane. Superposition of all four copies of the UL144 -BTLA complex indicated high-similarity in the structure of BTLA and the CRD1 region of UL144 (root mean square deviation value (RMSD) of 0.124 Å) (Fig. S1A). In contrast, the structure of the CRD2 region of UL144 exhibited slight differences between the four copies within the asymmetric unit, likely resulting from crystal packing. However, superposition of all four individual UL144 molecules yielded a similar overall structure with an RMSD value of less than 1 Å across all C␣ atoms (Fig. S1B).
The HCMV UL144 ectodomain folds into an elongated molecule composed of two CRDs that form a contiguous structure (Fig. 2). A total of 12 cysteine residues that were distributed equally in both the CRD regions are involved in intramolecular disulfide bridges and maintain the structural integrity of UL144. The structure also showed ordered electron density for the N-linked glycan at Asn-86 in two out of four UL144 molecules. The ectodomain of UL144 shares around 31% sequence identity with its human orthologue HVEM because all six disulfide linkages (12 cysteines) are conserved between the two molecules (Fig. 2, A and B). Considering that these cysteines are important for the overall fold and classification as a TNFRSF member, the actual sequence similarity between UL144 and HVEM is much lower. Nonetheless, the global structure of the UL144 in complex with BTLA is analogous to HVEM (PDB 2AW2) with considerable structural adaptations in its CRD1 and CRD2 regions (Fig. 2C). The N-terminal CRD1 region of UL144 possesses an extended loop, whereas HVEM contains two canonical anti-parallel ␤-strands (Fig. S2A). The N-terminal CRD2 region of UL144 contains two extended anti-parallel ␤-strands connected by a short loop that together replace the long C-shaped loop found in HVEM (Fig. S2B). These structural differences between UL144 and HVEM appear to be driven by their differing lengths of CRD2. Although the topological fold is comparable between UL144 and HVEM, superposition of monomeric UL144 with monomeric HVEM results in large RMSD values of 2.19 Å between 64 C␣ atoms from both molecules (Fig. 2C). As the CRD1 of both these proteins superpose well, the relative orientation of CRD2 with respect to CRD1 changes in UL144 compared with that in HVEM. Sequence alignment shows that the N-terminal loop of the CRD2 region in UL144 is three amino acids shorter compared Crystal structure of the UL144 -BTLA complex with the corresponding region in HVEM, which allow it to orient differently with respect to CRD1. Strikingly, when we superpose only the CRD2 region of UL144 with that of HVEM, it exhibits higher similarity with RMSD values of 0.7 Å between 24 C␣ atoms (Fig. S2B). The Ig-like domain of human BTLA is composed of two ␤-sheets wherein strands B, D, and E form the outer face, and AЈ, C, CЈ, G, G 0 , and F strands form the inner face of the Ig fold, as reported previously (Fig. S2C) (24). Structural superposition of human BTLA complexed with UL144 over unbound mouse BTLA (25) results in an RMSD value of 0.7 Å between 82 CA atoms. The binding of UL144 did not induce any major conformational changes in the overall architecture of BTLA.

Interaction interface of the UL144 -BTLA complex
The BTLA monomer binds to one monomeric UL144, forming a 1:1 arrangement as the minimal biological unit, similar to what has been observed in the HVEM-BTLA complex (Fig.  3A). The binding interface is almost equally shared between UL144 and BTLA, in which ϳ881 Å 2 area is buried on the BTLA and 832 Å 2 is buried on UL144. In the binding interface, UL144 employs residues exclusively from its CRD1 region to contact BTLA (Fig. 3A). Detailed analysis of the interface revealed that predominantly hydrophilic and polar contacts are formed between UL144 and BTLA with a few additionally charged salt bridges and hydrophobic interactions. Throughout the interface, both UL144 and BTLA residues mostly employ their side-chain atoms to interact with each other. The interface can be divided into three major binding sites. Site 1 contains an extended interface wherein BTLA recruits residues from its N-terminal long loop to stabilize the CRD1 of UL144 (Fig. 3B). Site 2 contains a small patch of interaction interface with residues contributed by the C strand and the CCЈ loop of BTLA with the N-terminal CRD1 region of UL144 (Fig. 3C).

Figure 2. Comparison between UL144 and HVEM.
A, sequence alignment of the extracellular regions of UL144 and HVEM. The conserved cysteine residues in both UL144 and HVEM are colored blue. The six N-linked glycosylation sites present in the CRD2 region of UL144 are colored red. In the crystallized construct, all N-glycosylation sites, except Asn-86, are mutated and are colored. B, cartoon representation of the crystal structure of a UL144 monomer overlaying a transparent surface. All cysteine residues that form disulfide linkages and N-linked glycosylation sites (green) are shown as sticks. The N-glycans at Asn-86 are denoted as sticks. C, structural alignment of UL144 with HVEM (yellow cartoon) shows its significant structural alteration within the CRD2 region. B and C, CRD1 region of UL144 is colored cyan, and the CRD2 region is colored pink. All figures were generated with PyMOL.

Crystal structure of the UL144 -BTLA complex
Site 3 is the major interaction site where the G 0 strand of BTLA forms an extended anti-parallel ␤-sheet with UL144. This shared ␤-sheet between both proteins is formed by main-chain hydrogen-bonding interactions and supported by few sidechain interactions (Fig. 3D). At this region, Tyr-42 of UL144 protrudes into the interior of the BTLA-binding site and facilitates hydrophobic interactions between Leu-58 of UL144 and Leu-123 of BTLA, stabilizing the anti-parallel intermolecular ␤-sheet. This is a key interaction forming a structural "hot spot" for the UL144 -BTLA complex, as the mutation of Tyr-42 to Ala completely abolished the binding to BTLA (Fig. 3E). Furthermore, close to this site the previously described mutation, Gly-46 of UL144 to lysine (G46K) (22), markedly increases the binding affinity between UL144 and BTLA (Fig. 3E). Structural analysis suggests that the increased binding affinity is due to the formation of a new salt bridge between Lys-46 of UL144 with Asp-35 of BTLA present at the extreme N-terminal end (Fig.  S3).

Comparison with HVEM-BTLA
We then compared the structure of the UL144 -BTLA complex with the previously solved HVEM-BTLA complex (PDB 2AW2) by aligning both BTLA molecules from these complexes. Structural superposition revealed that both complexes align with an RMSD value of ϳ0.8 Å between 127 C␣ atoms in which both BTLA and the CRD1 region of UL144 and HVEM superpose perfectly, whereas their CRD2 regions deviate to some degree (Fig. 4A). The structural alignment reveals that despite the relatively low sequence identity between UL144 and HVEM, UL144 is a clear structural mimic of HVEM in the way it engages BTLA (Fig. 4B). Analysis of the binding interface revealed that BTLA utilizes similar residues to bind HVEM and UL144 (Fig. 4C); correspondingly, UL144 and HVEM also employ residues from their CRD1s to engage BTLA. Although the overall binding interface seems to be conserved in both complexes, considerable variations exist in the specific amino Crystal structure of the UL144 -BTLA complex acids used to form the individual contacts. At the interface, more than 50% of CRD1 residues from UL144 and HVEM participate in the binding event, and the majority of these interactions are not conserved between them (Fig. 4B). UL144 shows unique interactions with BTLA residues Gln-43, Lys-41, Ser-121, Leu-74, and Thr-77 that were not observed in the HVEM-BTLA complex (Fig. 4C). Because the two N-terminal ␤-strands found in HVEM are replaced by a long loop in UL144 ( Fig. S2A and Fig. 5A), different interactions are formed in that region. In addition, to avoid steric clashes with this long loop of UL144, certain interface residues of BTLA adopted alternative conformations in the UL144 -BTLA complex. The key specific contacts that were reformed in the UL144 -BTLA complex compared with the HVEM-BTLA complex because of these structural differences are listed. First, Lys-24 of UL144 interacts with Gln-43 (at the N-terminal loop) of BTLA, whereas the conserved residue of HVEM (Lys-43) forms a hydrogen-bond interaction with Ser-112 (G 0 strand) of BTLA (Fig. 5B). Second, Gln-37 of BTLA adopts a different side-chain conformation, because it lacks any interaction with UL144, although it interacts with Gly-72 of HVEM. Similarly, to avoid a steric clash with Gln-33 of UL144, Arg-42 of BTLA also adopts a different sidechain conformation to which it forms salt bridge with Glu-27 at the N-terminal end of UL144, whereas the same residue forms a salt bridge with Glu-69 at the C-terminal end of the CRD1 region of HVEM (Fig. 5B). Another substantial difference appears around the G 0 strand, where Asn-122 of BTLA interacts with Thr-57 of UL144, while in HVEM a critical contact is formed with Lys-64, since the K64A mutant of HVEM lacks BTLA binding (16). Furthermore, Arg-114 of BTLA makes a salt bridge contact with Asp-45 of HVEM, whereas it lacks the partner residue in the UL144 -BTLA complex. Although UL144 possess an acidic Glu-26 in this region, the conformational variation within its N-terminal loop moves the side chain of Glu-26 far away from Arg-114.
Although unique interactions are present in both the UL144 -BTLA complex and the HVEM-BTLA complex, the G 0 strand of BTLA in both complexes forms anti-parallel intermolecular ␤-sheet that is further stabilized by hydrophobic interactions around this region (Fig. 5A). Of note, we have observed that Tyr-42 of UL144 and Tyr-61 of HVEM form a hydrophobic triad between Leu-123 of BTLA and Leu-58 of UL144 or Pro-77 of HVEM, respectively (Fig. 5C). As the Y42A mutant of UL144 and the Y61A mutant of HVEM (25,27) lose BTLA binding, we assessed the impact of Leu-123 on binding by transfecting 293T cells with plasmid encoding WT and an L123A mutant of BTLA. Although the L123A mutant of BTLA showed decreased binding toward HVEM, interestingly, UL144 binding was not impacted (Fig. 5D). This suggests that the tyrosine residues in both UL144 and HVEM are involved in structural stabilization, and the mutation of the Tyr-42 or Tyr-61 to alanine abolishes the binding toward BTLA by an allosteric mechanism, wherein its hydrophobic side chain stabilizes the overall structural fold in that region rather than forming a direct contact with BTLA, and removal of that side chain destabilizes the complex formation resulting in loss of binding.
Overall, our studies strongly demonstrate that although the BTLA contact residues of UL144 and HVEM differ significantly, both molecules directly compete for the same binding site on BTLA. To test this, we employed a competition binding assay testing whether a pre-existing HVEM-BTLA interaction could inhibit UL144 binding in trans. HEK293T cells were cotransfected with a plasmid encoding BTLA and either HVEM or control vector (Fig. 5E). These co-transfections resulted in a strong cell-surface expression of both BTLA and HVEM, as detected with specific monoclonal antibodies. Although we observed a dose-dependent binding of soluble UL144-Fc to BTLA expressed in the absence of HVEM, the co-expression of the BTLA-HVEM complex formation completely prevented the binding of UL144-Fc to BTLA, demonstrating competition for the same BTLA-binding site, as predicted by the crystal structure (Fig. 5E).

Structural differences between UL144 and HVEM limit LIGHT binding
HVEM is a multifunctional molecule that can engage the TNF ligand LIGHT, LT␣, and CD160, in addition to BTLA (19). However, UL144 has evolved to bind only BTLA of these multiple ligands. Although the UL144 -BTLA complex presented here does not definitively explain the absence of LIGHT binding by UL144, the structure of UL144 raises several possibilities. The analysis of the HVEM-LIGHT interaction interface compared with the UL144 structure reveals several key features that might account for the lack of interaction between UL144 and LIGHT. First, LIGHT is a homotrimeric TNF ligand, and the crystal structure of the HVEM-LIGHT complex (PDB 4RSU) Crystal structure of the UL144 -BTLA complex revealed that each HVEM binds at the interface formed by two adjacent LIGHT protomers via its CRD2 and CRD3 regions (Fig. 6B). As UL144 lacks the CRD3, the absence of interactions from this region mainly interferes with the complex formation with LIGHT. Second, although UL144 possesses a CRD2 that is topologically similar to HVEM, very little sequence identity is shown, whereas the majority of the LIGHT-interacting residues of HVEM is not conserved between them (Fig. 6A). In addition, structural superimposition of UL144 CRD2 region with that of HVEM in the HVEM-LIGHT complex revealed that the shorter N-terminal CRD2 loop of UL144 acquired a different orientation compared with the corresponding region of HVEM, due to which it sterically clashes with the DE loop of one of the LIGHT protomers if UL144 has to bind (Fig. 6C). It was reported earlier that in many of the TNF-TNFR complexes, the conserved hydrophobic residue of this DE loop energetically favors the complex formation, and any obstruction in this interaction occludes the complex formation. Hence, the absence of interactions from the DE loop residues (Tyr-173) might prevent the binding of UL144 to LIGHT. Third, WT UL144 has six glycosylation sites located in the CRD2 region, and three of these N-glycan sites are present toward the side of the LIGHT-binding interface. Hence, the heavily glycosylated WT UL144 gets obstructed by N-glycans at positions Asn-70, Asn-73, and Asn-78 toward binding to LIGHT (Fig. 6D).

Discussion
In this study, we have reported the crystal structure of the HCMV immunomodulatory protein UL144 bound to BTLA. The general architecture of the UL144 -BTLA complex shows significant conservation with the previously published HVEM-BTLA structure, with BTLA using a largely similar binding site to interact with both UL144 and HVEM. The central footprint of BTLA-HVEM binding is largely conserved, but different residues contribute to the fine binding specificity of the two receptors. In contrast, the BTLA contact residues employed by UL144 or HVEM differ significantly, supporting a model where UL144 has evolved uniquely to interact with this immune Crystal structure of the UL144 -BTLA complex inhibitory molecule. Recently published results identified seven critical UL144 residues (Glu-27, Gln-33, Pro-36, Gly-41, Tyr-42, Thr-52, and Leu-68) that contribute to BTLA binding (27). Our structural data show that not all of these UL144 residues directly contact BTLA. Specifically, the salt-bridge contact mediated by Glu-27 of UL144 with Arg-42 of BTLA is critical as the E27A mutant of UL144 does not bind to BTLA. In contrast, although a UL144 Q33A mutation severely impeded BTLA binding, this residue makes no direct contacts with BTLA, nor does it appear to stabilize the UL144 molecule. Nonetheless, Gln-33 is part of the long loop that replaces the antiparallel ␤-strands of HVEM in CRD1, and any structural change in this region might indirectly impact BTLA binding by affecting the overall loop structure. Other critical residues Pro 36, Tyr-42, and Thr 52 of UL144 mediate van der Waals contacts with His-127, Leu-123, and Tyr-39 of BTLA via their hydrophobic side chains. Both proline and tyrosine residues favor strong hydrophobic interactions in this region of the anti-parallel intermolecular ␤-sheet formed between BTLA and UL144. Both our crystal structure and previous mutational data support a hypothesis that whereas the energetics of the HVEM-BTLA complex are centered at the anti-parallel intermolecular ␤-sheet region, the binding energy in the UL144 -BTLA complex is distributed at both its N-terminal and intermolecular anti-parallel ␤-sheet regions. Our data imply that UL144 directly recruits Glu-27 to engage Arg-42 of BTLA and in addition promotes strong hydrophobic interactions through Pro-36 and Tyr-42, and thereby forms more stable interactions than between BTLA and HVEM (Fig. 3E). Distinct HCMV UL144 group proteins contain hypervariable CRD1 and CRD2 primary sequences; however, all these variants bind BTLA within a 5-fold affinity range (16). Primary CRD1 sequences among the UL144 variants show that although many of the BTLA-interacting residues are not conserved, all three groups have main-tained the critical hot spots (Glu-27 and Pro 36) required for BTLA binding. Surprisingly, Tyr-42 of UL144-F (a group 3 UL144 variant) is critical for BTLA binding, but this Tyr is not conserved in any other UL144 group proteins, suggesting this may explain why UL144 -group 3 does bind BTLA with modestly higher affinity (Fig. S4).
HCMV UL144 is incapable of binding to both LIGHT and CD160 (22). BTLA and CD160 directly compete for binding to the CRD1 region of HVEM, but the structural basis for this competition remains unclear. Our results suggest that alterations in the CRD1 region of UL144 might account for its inability to bind CD160, perhaps contributing to evasion of NK cell responses. Notably, a UL144-Fc protein inhibits T-cell activation to a much greater extent than HVEM-Fc, despite binding BTLA ϳ5 times worse than HVEM (28), perhaps because evolving to bind only one of the four known HVEM ligands has selectively enhanced its immune inhibitory activity. In summary, the crystal structure presented here explains the selective BTLA binding by UL144. In addition, these results form a platform that will help facilitate the generation of a potent T-cell inhibitor molecule.

Generation of UL144, HVEM, and BTLA constructs
For structural studies, the ectodomain of HCMV UL144 (amino acids 21-132) carrying a C-terminal hexa-histidine tag was cloned downstream of the gp67 secretion signal sequence into the baculovirus transfer vector pAcGP67A. Human BTLA (residues 31-137) was cloned into the pET22bϩ vector without any purification tag. For binding studies, UL144 from three different HCMV clades, UL144-1A (aa 21-138), UL144-2 (aa 21-138), and UL144 -Fiala (aa 19 -137), as well as human HVEM (aa 39 -99), were cloned into a modified mammalian Crystal structure of the UL144 -BTLA complex expression vector pCR 3.1 downstream of the HA signal sequence and upstream of the Fc domain of human IgG1 and expressed in mammalian HEK293T cells. Similarly, full-length BTLA was cloned into the vector pcDNA 3.1(ϩ) containing a C-terminal glycosylphosphatidylinositol anchor to assist its expression on the cell surface of HEK293T cells. The correct sequence for all the clones was confirmed by DNA sequencing.

Generation of UL144 and BTLA mutants
Various N-linked glycosylation site mutants of UL144 were generated by site-directed mutagenesis using the QuikChange II site-directed mutagenesis kit (Stratagene, La Jolla, CA). The Asn residues carrying the N-linked glycan were mutated individually or in combination at residues Asn-61, Asn-70, Asn-73, Asn-78, Asn-86, Asn-91, Asn-99, and Asn-116 (starting from initiating methionine) to either Ser, Asp, or Gln. Similarly, a BTLA L123A mutant was prepared for binding studies. Variants of UL144 -Fiala-Fc fusion constructs (UL144 (Y42A) and UL144 (G46K)) were also generated for binding studies. All mutants of UL144-Fc and BTLA were expressed in mammalian HEK293T cells, and the mutants of His-tagged UL144 were expressed in Sf9 insect cells.

Protein expression and purification of UL144 from insect cells
The His-tagged fusion proteins of WT and various N-glycan site mutants of UL144 were transfected individually into BacPAK6DNA under aseptic conditions according to the manufacturer's protocol. For transfection, we have incubated 1 g of pAcGp67A transfer vector containing the UL144 gene mixed with 5 l of BacPAK6DNA and 5 l of Bacfectin reagent in a total volume to 100 l of serum-free media at room temperature in a dark environment for 15 min. Simultaneously, a control transfection mixture that lacks the BacPAK6DNA was also used. 2 ϫ 10 6 healthy dividing Spodoptera frugiperda (Sf) 9 cells were seeded, and then both the control and the transfection mixture were added to these cells and grown at 27°C in serumfree medium containing the following antibiotics: 100 units/ml penicillin and 100 g/ml streptomycin. After 5 days, the supernatant was collected by centrifugation at 1000 ϫ g for 10 min, which then used for first round of virus amplification. Consequently, after 5 days, the second round of virus amplification was performed to achieve the virus titer with (multiplicity of infection) m.o.i. ϭ 1. For protein production, high-titer virus was prepared from low-titer virus m.o.i. ϭ 1 of the second virus amplification, which then added to several individual 2-liter Erlenmeyer flasks seeded with 2 ϫ 10 6 Sf9 cells/ml. Protein expression was continued for 72-84 h as a suspension culture (135 rpm) at 27°C, and the cell supernatant was collected by centrifugation. The supernatant containing the protein of interest was concentrated and then buffer-exchanged against 1ϫ PBS using a Millipore filtration device using a 10-kDa molecular mass cutoff membrane. The supernatant was loaded onto a nickel-nitrilotriacetic acid column, and the His-tagged UL144 fusion protein was eluted with 250 mM imidazole. Final purification of UL144 was carried out by size-exclusion chromatography using a Superdex S200 column, and the protein was concentrated to ϳ3 mg/ml and used for crystallization.

Protein expression and purification of BTLA from E. coli
The ectodomain of human BTLA was prepared in E. coli. BL21 DE3 cells were grown in LB medium at 37°C until an A 600 of 0.6. Protein expression was then induced by the addition of 1 mM isopropyl 1-thio-␤-D-galactopyranoside at 37°C and continued for 4 h. The cell pellet was collected, and the cells were disrupted by sonication in lysis buffer (100 mM Tris-HCl, pH 7.0, 5 mM EDTA, 5 mM DTT, and 0.5 mM phenylmethylsulfonyl fluoride) and then centrifuged at high speed to collect the lysed pellet. The pellet was further washed, and the inclusion bodies were extracted from BTLA expressing E. coli cells in extraction buffer (50 mM Tris-HCl, pH 7.0, 5 mM EDTA, 2 mM DTT, 6 M guanidine HCl). For protein purification, ϳ15 mg of inclusion bodies were dissolved in refolding buffer (100 mM Tris, pH 8.5, 20 mM glycine, 300 mM NaCl, 1 mM EDTA, 146.8 mg of oxidized GSH, and 73.6 mg of reduced GSH) in a total volume of 250 ml and incubated overnight at 4°C. The refolding mixture was concentrated using 10-kDa molecular mass cutoff ultrafiltration devices (Millipore) and then further purified by size-exclusion chromatography using an S200 column. The desired protein fractions were pooled and concentrated to ϳ2 mg/ml for subsequent crystallization trials.

Crystallization of UL144 -BTLA complex
For crystallization studies, the purified UL144 mutant devoid of N-linked glycans at positions 61, 70, 73, 78, 91, 99, and 116 and a slight molar excess of BTLA were mixed and incubated at room temperature for 1 h. The complex was then concentrated and isolated from unbound proteins using a Superdex S-200 size-exclusion column in 50 mM HEPES and 150 mM NaCl, pH 7.5, running buffer. The peak fractions containing the UL144 -BTLA complex were then pooled, concentrated to 3 mg/ml, and subjected to crystallization. Initial crystallization trials were performed at both 22 and 4°C and tested over 800 different crystallization conditions (JCSG coreϩ, 1-4, Wizard, MB suite, and PEG ion screens) by a sitting-drop vapor diffusion method in a 96-well format using a nanoliter dispensing liquid handling robot (Art Robbins Phenix). Optimization of crystallization conditions was performed manually by both hangingdrop and sitting-drop methods by equilibrating 1.2 l of protein (3.2 mg/ml UL144 -BTLA complex in 50 mM HEPES, pH 7.0, and 150 mM NaCl) and 0.8 l of reservoir solution at 4°C. The crystals were grown over 15 days at 4°C by the hangingdrop method using the precipitant 0.1 M CHES, pH 9.5, 20% w/v PEG 8000 and generated high-quality diffraction. All crystals were flash-cooled in liquid nitrogen in their crystallization buffer containing 20% glycerol for subsequent data collection.

Data collection and refinement
Native diffraction data for UL144 -BTLA complex crystals were collected remotely at Stanford Synchrotron Radiation Light Source (SSRL) beamline 9-2 using a PILATUS 6M PAD detector at a wavelength of 0.97 Å and a temperature of 100 K. Each image was collected at 0.25°oscillation and a 5-s exposure time. The data were processed and scaled using HKL2000. Attempts to determine the structure of UL144 by the molecular replacement method using the structure of HVEM were unsuccessful. Therefore, we have experimentally determined the Crystal structure of the UL144 -BTLA complex phase information by molecular replacement combined with single anomalous dispersion using the BTLA model and sulfur-SAD phasing by collecting diffracting data on several crystals that belong to P2 1 space group (unit cell dimensions: a ϭ 66.9 Å, b ϭ 77.2 Å, c ϭ 101.7 Å, ␣ ϭ ␥ ϭ 90°, and ␤ ϭ 91.6°), while enhancing the sulfur anomalous signal at 2.07 Å wavelength and taking advantage of the shutter-less and thin-slicing (0.1°/ image) capabilities of the PAD detector. The phasing data were collected with over 50-fold redundancy and averaged over different pixels (four different detector distances) and a much lower absorbed dose (208 gray) per image due to the attenuated 1.8e10 photon flux used at 2.07 Å wavelength.

Molecular replacement combined with single anomalous dispersion phasing
The XSCALE output was used to identify the anomalous sulfur sites in the hybrid substructure search program (HySS) as a part of Phenix graphical interface (29 -31). Within HySS, the search for 15 sulfur sites (out of 96 sulfur atoms possible; some disordered; some in disulfides and seen as a single peak) in the 40 to 2.7 Å range was successful. The phase information for the BTLA molecule in the UL144 -BTLA complex was obtained by molecular replacement method using previously solved human BTLA (PDB 1XAU) as a search model (32). Experimental phasing combined with molecular replacement partial structure was executed in the PHASER EP program as part of CCP4 (33)(34)(35). Thirty-two sulfur sites were found, and the improved combined phases were used to build the secondary structural elements for UL144 and BTLA by performing (PARROT) density modification and BUCCANEER (31) model building. The model and improved phases from PHASER EP were refined in PHENIX/REFMAC (36), and the missing loops were built manually, and the improvements in the model were performed in COOT (37,38). The final model was refined in PHENIX/REFMAC to 2.7 Å resolution with residual factors R/R free ϭ 22.9/27.6%. The model had excellent stereochemistry with only two residues as Ramachandran outliers, which are confirmed by the electron density. All figures were made in PyMOL (39).

Expression and purification of UL144-Fc and HVEM-Fc from mammalian HEK293T cells
The mammalian expression vector pCR3 containing the Fc protein constructs (HVEM, UL144-1A, UL144-2, UL144 -Fiala, UL144 -Fiala-G46K, and UL144 -Y42A) was transiently transfected into mammalian HEK293T cells cultured in complete Dulbecco's modified Eagle's medium (DMEM supplemented with 100 units/ml penicillin, 100 g/ml streptomycin, 2 mM L-glutamine, and 10% v/v fetal bovine serum (FBS)) using standard calcium phosphate transfection. 12-16 h after transfection, the culture media were changed to fresh DMEM supplemented with 5% FBS, and cells were maintained at 37°C under 5% CO 2 for an additional 72 h. Supernatant containing secreted Fc protein was collected and buffer-exchanged against 1ϫ PBS by tangential flow-through filtration using 10-kDa MWCO membranes. The supernatant was loaded onto a HiTrap TM protein A HP column, and the Fc protein was eluted with 100 mM Na citrate, pH 3.0, buffer. The protein was further purified by size-exclusion chromatography using a Superdex S-200 column in 1ϫ PBS buffer, and the peak fractions were concentrated and stored at Ϫ80°C.

Flow cytometric-binding assays
HEK293T cells were transfected with mammalian expression vector PCR3 containing either hBTLA or hHVEM using standard calcium phosphate transfection. Twelve hours after transfection, cells were washed with 1ϫ PBS, and media were changed to fresh complete media and incubated 24 h at 37°C under 5% CO 2 . Cells were collected and washed one time with FACS buffer (1ϫ PBS (GIBCO), 2% FBS (Sigma), 0.1% sodium azide (Fisher)) and then incubated for 30 min at 4°C with the different Fc proteins (0.001 to 200 g/ml). Cells were then washed two times with FACS buffer and incubated for 30 min at 4°C with Alexa Fluor 647-conjugated (Jackson Immuno-Research) and then washed two additional times. In parallel, the membrane expressions of BTLA and HVEM were assessed using, respectively, an allophycoeryhtrin-conjugated anti-BTLA and a phycoerythrin-conjugated anti-HVEM (Biolegend). Cells were analyzed on an LSRII (BD Biosciences), and data were analyzed using FlowJo softwareX (TreeStar).