Structural basis for the recognition of nectin-like protein-5 by the human-activating immune receptor, DNAM-1

Nectin and nectin-like (Necl) adhesion molecules are broadly overexpressed in a wide range of cancers. By binding to these adhesion molecules, the immunoreceptors DNAX accessory molecule-1 (DNAM-1), CD96 molecule (CD96), and T-cell immunoreceptor with Ig and ITIM domains (TIGIT) play a crucial role in regulating the anticancer activities of immune effector cells. However, within this axis, it remains unclear how DNAM-1 recognizes its cognate ligands. Here, we determined the structure of human DNAM-1 in complex with nectin-like protein-5 (Necl-5) at 2.8 Å resolution. Unexpectedly, we found that the two extracellular domains (D1–D2) of DNAM-1 adopt an unconventional “collapsed” arrangement that is markedly distinct from those in other immunoglobulin-based immunoreceptors. The DNAM-1/Necl-5 interaction was underpinned by conserved lock–and–key motifs located within their respective D1 domains, but also included a distinct interface derived from DNAM-1 D2. Mutation of the signature DNAM-1 “key” motif within the D1 domain attenuated Necl-5 binding and natural killer cell–mediated cytotoxicity. Altogether, our results have implications for understanding the binding mode of an immune receptor family that is emerging as a viable candidate for cancer immunotherapy.

The activity of immune effector cells is tightly regulated to enable the identification and elimination of infected, transformed, or stressed cells while preventing autoimmunity toward the surrounding healthy tissue. This specificity is achieved by the coordinated activity of a variety of cell surface-bound receptors that either stimulate or dampen immune cell activation. For example, natural killer (NK) 3 cell function is regulated by the integration of signals derived from a variety of paired receptor families that include the Ly49 receptors in rodents (1), the killer cell and leukocyte immunoglobulin receptors (KIRs and LIRs, respectively) in humans (2,3), and the CD94-NKG2 (4) and NKR-P1 (5) families in both species. Within these systems, the inhibitory receptors typically recognize "self"-molecules that are broadly expressed on the surface of healthy cells (6 -10), while their stimulatory counterparts predominantly bind to nonself or stress-induced ligands (11)(12)(13).
Alternatively, some paired receptor families comprise members that exhibit overlapping ligand specificities, and as such they compete with each other to fine-tune the immune response. For example, the nectin receptors DNAM-1 (CD226), CD96 (TACTILE), and T-cell immunoglobulin and immunoreceptor tyrosine-based inhibitory motif (TIGIT) are expressed on T and NK cells, where they mediate effector/target cell adhesion and immune synapse formation and regulate effector function (14). Within this context, DNAM-1 is a stimulatory receptor that was first identified by its ability to activate T cells independently of T-cell receptor engagement (15)(16)(17)(18)(19). DNAM-1 has since been shown to be critical for tumor immunosurveillance, as demonstrated in mouse models of chemically-induced tumors, and by DNAM-1 Ϫ/Ϫ mice, which exhibit accelerated tumor growth (17). In contrast, TIGIT is an inhibitory receptor (20 -22), and CD96 is able to counter-balance DNAM-1 function in mice (16), although human CD96 has been reported to stimulate NK cell cytotoxicity (23).
Nectin receptor function depends on their capacity to recognize certain nectin and nectin-like (Necl) adhesion molecules, namely nectin-2 (CD112) and Necl-5 (CD155), which are broadly overexpressed in a wide range of cancers (24 -27). During homeostasis, nectin (like) proteins tether adjacent cells together via the formation of homo-and heterodimers. These interactions rely on signature "lock" (AX 6 G) and "key" (T(F/ Y)P) motifs located within their most membrane distal (D1) domains (28). The nectin receptors also contain similar lock and key motifs within their D1 domains, although they differ in the complexity of their ectodomains. For example, TIGIT, DNAM-1, and CD96 possess one, two, and three extracellular Ig domains, respectively. Structural studies of TIGIT and CD96 have indicated that their D1 domains, which are sufficient for ligand binding, engage their cognate ligands by a docking mode that mimics nectin homo/heterodimerization (29 -31). A recent structure of mouse DNAM-1 bound to human Necl-5 supports this proposition, but it also indicated that the second DNAM-1 domain (D2) can modulate the interaction (32). However, the molecular basis underpinning recognition of Necl-5 by human DNAM-1 remains unclear.
Here, we determined the structure of human DNAM-1 in complex with one of its cognate ligands, Necl-5. DNAM-1 bound Necl-5 in a 1:1 stoichiometry via a docking mode that enabled the interaction between lock and key motifs within their respective D1 domains. The unconventional "collapsed" architecture of the two extracellular domains of DNAM-1 positioned the D2 domain such that it made direct contacts with Necl-5. Binding and functional studies indicated that residues within DNAM-1 D2 did not modulate the interaction with ligands or functional response of DNAM-1 to Necl-5/nectin-2expressing target cells. In contrast, the D1-localized key residue was critical for DNAM-1-dependent ligand binding and NK cell cytotoxicity. Altogether, these data expand our understanding of the molecular features that govern nectin receptorligand interactions.

Structural determination of the DNAM-1/Necl-5 complex
Although the mode of interaction of TIGIT and CD96 with their cognate ligands has been established (29 -31), the structural basis for human DNAM-1 ligand recognition is unknown. To address this gap in our knowledge, we expressed the extracellular Ig domains of human DNAM-1 (D1-D2) and Necl-5 (D1-D3) in human embryonic kidney (HEK)-293S and Hi5 insect cells, respectively. DNAM-1 behaved as a heavily glycosylated monomeric species, as judged by its apparent migration during SDS-PAGE analysis and size-exclusion chromatography (Fig. S1). Necl-5 also appeared to be a glycosylated monomer, consistent with previous studies (30). To aid in crystallization, we treated DNAM-1 with endoglycosidase H (endo H), which trims high-mannose N-glycans that are the only glycoforms produced by HEK-293S cells to a single N-acetylglucosamine (GlcNAc) residue, prior to mixing with a 1:1 molar ratio of Necl-5. Using this approach, we generated crystals of the DNAM-1/Necl-5 co-complex and determined the structure to 2.8 Å resolution (Table S1). The crystallographic asymmetric unit comprised two molecules of DNAM-1 and two molecules of Necl-5 that formed two similar DNAM-1/Necl-5 heterodimeric complexes (root mean square deviation (r.m.s.d.) of 1.78 over 430 C␣ atoms). Although one of the DNAM-1/Necl-5 complexes (chains B and D) was relatively poorly resolved, the electron density for the other heterodimer (chains A and C) was clear, allowing the building of the entire polypeptide chains encompassing the D1 and D2 domains of DNAM-1 (Glu-19 -Ala-241) and the D1-D3 domains of Necl-5 (Glu-27-Val-327), with the exception of a single loop in DNAM-1 D1 (Ser-84 -Ser-89) and two loops within Necl-5 D3 (Asp-252-Gln-259 and Val-302-Thr-308), none of which were located at the DNAM-1/Necl-5 interface. Indeed, the structure refined well (Fig. S2) to an R factor and R free of 20.8 and 23.0%, respectively, thereby permitting a detailed understanding of the interactions that governed DNAM-1/Necl-5 recognition.

DNAM-1 comprises an atypical configuration of two tandem Ig domains
DNAM-1 was composed of two Ig-variable domains that were tethered together by a relatively long 12-amino acid linker region (Gln-126 -Asn-137) (Fig. 1A). Six individual GlcNAc residues were visible attached to Asn residues 32, 83, 90, and 97 within D1 and Asn-147 and -186 within D2. The DNAM-1 D1 domain was composed of one small (sheet I: BED) and one large (sheet II: AGFCCЈCЉ) ␤-sheet that were bridged by a single disulfide bond located between the B-and F-strands (Cys-37-Cys-108) (Fig. 1B). DNAM-1 D2 was organized similarly to the D1 domain (r.m.s.d. 2.6 Å over 63 C␣ atoms), whereby it possessed a canonical disulfide bond, linking the B-and F-strands (Cys-152-Cys-222). However, within the D2 domain, the regions equivalent to the CЉ-and D-strand were no longer H-bonded to sheets I and II, respectively, due in part to an additional disulfide bond that attached the CЈ-strand to the CЈ-D linker (Cys-179 -Cys-199).
Within the DNAM-1 structure, the D1 and D2 domains were collapsed on each other in an unusual side-by-side arrangement, such that the ␤-strands of D1 and D2 formed a single extended ␤-sheet (Fig. 1, A and B). This unexpected architecture, where D2 was rotated relative to D1 by ϳ100°, was facilitated by the extended nature of the inter-domain linker that connected the G-strand of D1 to the A-strand of D2. The DNAM-1 D1/D2 interface occupied a moderate solvent-inaccessible surface area (Ϸ900 Å 2 ) and was primarily hydrophobic in nature (Fig. 1C). The core of the interaction zone was derived from the A-strand of D1 and the CЈ-strand of D2, where four contiguous H-bonds between the main chain atoms of Trp-23-Val-27 (D1) and Ile-173-Leu-175 (D2) formed an anti-parallel ␤-sheet that stitched the D1 and D2 domains together ( Fig. 1D and Table S2). The interface was further supported by the DNAM-1 N terminus, where Leu-22-His-24 interacted with several residues that spanned the length of the C and CЈ strands of D2 (Fig. 1E).
Notably, the relative juxtapositioning of DNAM-1 domains is completely distinct to the "beads-on-a-string" arrangement evident within the nectin (-like) ligands (Fig. 2), to which the nectin receptors were previously considered to be closely structurally related. Indeed, although many immune receptors are Immune recognition of nectin adhesion molecules composed of multiple tandem Ig domains, the inter-domain angles are typically large (Ͼ75°), resulting in either linear (PD1-L2) (33), perpendicular (LIR-B1) (34), or open V-shaped (NKp46 and KIR3DL1) (9, 35) configurations (Fig. 2). Among the immune receptors, the DNAM-1 architecture is most similar to that found in FcyRIIA (36), although the rotational angle between D1 and D2 domains within these two receptors is quite distinct. Notably, the overall architecture of DNAM-1 reported here is similar to that of (unliganded) human and mouse DNAM-1 recently reported by Wang et al. (32) (r.m.s.d. 0.79 and 1.37 Å over 170 and 185 aligned C␣ atoms, respectively), indicating that this unusual domain configuration represents the bona fide native state of DNAM-1. Accordingly, DNAM-1 possessed a collapsed side-by-side architecture that is atypical among immune receptors whose structures have been solved to date.

Overview of the DNAM-1/Necl-5 heterodimer
DNAM-1 bound exclusively to the D1 domain of Necl-5 via an extensive interface that buried a total solvent-inaccessible surface area of ϳ1880 Å 2 and was characterized by high shape complementarity (Sc ϭ 0.79). Within the complex, Necl-5 adopted a linear configuration of its three Ig domains, such that the D2 and D3 domains projected away from DNAM-1 and did

Immune recognition of nectin adhesion molecules
not participate in the interaction (Fig. 3A). Although some subtle differences in the inter-domain angle were evident, the overall architecture of Necl-5 was similar to that previously described for the unbound and CD96-bound forms ( Fig. S3) (28,30). In this arrangement, the C termini of both receptor and ligand are positioned at opposing ends of the complex (Fig. 3A). Accordingly, the DNAM-1/Necl-5 interaction could conceivably occur either in cis or in trans, i.e. between molecules located on single or adjacent cellular membranes, respectively (Fig. 3B). Although a higher-order heterotetrameric arrangement was visible within the crystal lattice (Fig. S4), we do not consider this to be a physiologically relevant assembly because the interface was mediated solely by interactions between two DNAM-1 monomers, and DNAM-1 was judged to be monomeric in solution (Fig. S1A).

DNAM-1 D1 binds to Necl-5 via a conserved docking mode
The DNAM-1/Necl-5 interface was primarily centered around DNAM-1 D1, which engaged Necl-5 via a head-tohead docking mode reminiscent of that utilized by TIGIT and CD96 ( Fig. S5) (29,31). Here, DNAM-1 D1 interacted with necl-5 in an orthogonal orientation across the breadth of their respective AGFCCЈCЉ sheets, giving this region of the interface a pseudo 2-fold symmetrical appearance (Fig. 3A). The defining feature of this region of the interface was derived from the lock and key motifs, whereby aromatic key residues within the F-G loop of both DNAM-1 (Tyr-113) and Necl-5 (Phe-128) were inserted into hydrophobic pockets formed by the CЈ-CЉ loops of their respective binding partner (Fig. 3, C and D, and Table  S3). These interactions were further stabilized by the C-CЈ loop of each molecule, which flexed inward to enable a single residue (Gln-57 in DNAM-1 and Ser-74 in Necl-5) to make interactions with the G-strand of the binding partner (Fig. 3E). Together, these two major points of contact served to latch the molecules together at the periphery of the interface. However, residues located within the central region of this interface also played a prominent role and accounted for 8 of the 11 total H-bonds (Fig. 3F). Here, several residues within the central C-, F-, and G-strands of DNAM-1 and Necl-5 projected directly toward the opposing sheet where they primarily made H-bond interactions with main-chain atoms of their binding partner. Thus, the D1 domain of DNAM-1 binds to Necl-5 via a conserved docking topology that is characteristic of the nectin receptor family.

DNAM-1 D2 makes direct interactions with Necl-5
In addition to the conventional interface derived from their respective D1 domains, we observed an interaction site located between DNAM-1 D2 and Necl-5 D1. This interface contributed a relatively small fraction (13%) of the total buried surface area (

Comparison with the mouse DNAM-1/Necl-5 structure
Overall, the docking mode of DNAM-1 onto Necl-5 is conserved between human and mouse DNAM-1 (r.m.s.d. 1.44 Å over 216 aligned C␣ atoms) (Fig. S7A). However, closer inspection of the interface reveals several subtle differences. In particular, of the 22 human DNAM-1 residues that formed interactions with Necl-5, only seven are conserved in mouse DNAM-1 (Fig. S7B). This includes the signature aromatic key (Tyr-113), which is a phenylalanine in mouse DNAM-1. Within the interface, the differences between mouse and human DNAM-1 are most obvious in the D2 domain, where the CЈ-D loop, which lacks a glycan consensus motif in mouse DNAM-1, adopts distinct conformations within the two structures resulting in altered interactions with the C-CЈ loop of Necl-5 (Fig. S7A).

Mutations within DNAM-1 D1 but not D2 impact binding to nectin-2 and Necl-5
Next, to investigate the energetic contribution of DNAM-1 D2 to ligand binding, we first attempted to express the DNAM-1 D1 domain in isolation. However, we were unable to produce this truncated protein using mammalian cells, most probably due to instability associated with the exposure of a hydrophobic surface that would normally reside at the D1/D2 interface (Fig. 1C). Instead, we generated a panel of DNAM-1 point mutants and assessed their interaction with nectin-2 and Necl-5 using surface plasmon resonance (SPR). Here, both

Immune recognition of nectin adhesion molecules
Necl-5 and nectin-2 bound to WT HEK-293S produced DNAM-1 with comparable affinity (K D ϭ 1.8 Ϯ 0.6 and 2.9 Ϯ 0.8 M, respectively) (Fig. 4A). Although we were only able to couple lower amounts on to the SPR chip, mutation of the DNAM-1 key residue (Y113A) within the D1 domain drastically reduced nectin-2 and Necl-5 binding when compared with a control mutation (R96A) in a surface-exposed DNAM-1 residue that was distal to the Necl-5-binding site (Fig. 4A). However, mutations within DNAM-1 D2 that disrupted Asn-186 and its associated N-glycan (N186A) or the glycosylation consensus motif, defined as NX(S/T), where X P (T188A), did not markedly impact the binding affinity of DNAM-1 for either nectin-2 or Necl-5 (K D ϭ 2.0 -2.5 M) (Fig. 4A). To further confirm that the nature of the glycan chain attached to Asn-186 did not impact the interaction with Necl-5/nectin-2, we also tested the binding of recombinant DNAM-1 produced in HEK-293T cells, which was decorated with a variety of complex N-glycans similar to those that would be expected to occur in vivo, as opposed to the more homogeneous high-mannose N-glycans of HEK-293S-produced DNAM-1 (Fig. S6B). However, the binding of HEK-293T-produced DNAM-1 to Necl-5 and nectin-2 was of similar affinity to DNAM-1 produced in HEK-293S cells (Fig. 4A).
To further probe the impact of DNAM-1 mutants in a cellular system, we also tetramerized the biotinylated DNAM-1 mutants and assessed their ability to bind human myelogenous leukemia (K562) cells, which express Necl-5 and nectin-2 ( Fig.  4B) by flow cytometry. Although no binding was observed using a control murine cell line lacking human DNAM-1 ligands, WT DNAM-1 bound well to the K562 cells (Fig. 4, C and D). With the exception of the T188A mutant, which exhibited a 68% increase in binding, the overall pattern of binding of the mutant or variant forms of DNAM-1 closely matched that obtained by SPR. The R96A and N186A mutations as well as HEK-293T-produced DNAM-1 bound equivalently to HEK-293S DNAM-1, whereas the Y113A mutation reduced binding to the K562 cells. Thus, the aromatic key residue of DNAM-1 is critical for the interaction with Necl-5 and nectin-2, whereas Asn-186 and its associated N-glycan appear to be dispensable for ligand binding under these experimental conditions.

Mutations within DNAM-1 D1, but not D2, impact NK cell killing
Finally, to investigate the functional relevance of DNAM-1/Necl-5 interactions, we transfected WT DNAM-1 (and mutants thereof) into an NK cell line (NK92) that lacked endogenous DNAM-1 expression (Fig. 5A) and co-incubated these cells with K562 cells at a range of NK/target cell ratios (Fig. 5B). Analysis of the expression of DNAM-1 in these cell lines demonstrated that DNAM-1 N186A could not be detected using the anti-DNAM-1 antibody 11A8. Thus, we confirmed the expression of DNAM-1 N186A in these cells using nectin-2 tetramers.

Immune recognition of nectin adhesion molecules
Similar to the observations with DNAM-1 tetramer staining of K562 cells, (Fig. 4C), nectin-2 tetramers bound to each DNAMexpressing cell line with the exception of NK92 DNAM-1 Y113A cells (Fig. 5A). Given that NK92 cells preferentially kill via NKG2D, we performed the killing assays in the presence of the NKG2D-blocking antibody 1D11. Here, expression of WT DNAM-1 markedly improved NK cell cytotoxicity compared with the parental line, indicating that within this experimental system, killing was DNAM-1-dependent (Fig. 5B). Although the functional capacity of NK92 cells transduced with the control R96A DNAM-1 mutant was identical to that of WT DNAM-1, introduction of the Y113A mutation significantly reduced killing across all NK/target cell ratios, highlighting the importance of the conserved aromatic key residue (Fig. 5B). In contrast, mutations within DNAM-1 D2 (N186A and T188A) had no effect on NK cell killing at any effector/target cell ratio (Fig. 5B). Thus, in this NK cell assay, interactions derived from Asn-186 or its associated N-glycan chain did not influence DNAM-1-mediated cytotoxicity of nectin-2/ Necl-5-expressing target cells.

Discussion
The nectin receptors are emerging as a critical regulatory checkpoint governing immune effector function. For example, overexpression of nectin-2 and Necl-5 on tumors has been reported to dampen T and NK cell responses via TIGIT and CD96 (16 -19), whereas recognition of these same ligands by DNAM-1 promotes tumor immunosurveillance and clearance (15,(37)(38)(39)(40)(41). Moreover, DNAM-1 has been associated with a number of adverse autoimmune pathologies (42)(43)(44) and directly drives the allo-immune response in graft versus host disease (45). Thus, understanding the factors governing recognition within this paired receptor family has the potential to

Immune recognition of nectin adhesion molecules
inform future strategies for therapeutic intervention in a range of human diseases.
Within paired receptor families whose members compete with each other for ligand binding, the affinity of each receptor toward any given ligand constitutes a key factor governing the overall outcome of effector/target cell engagement. Here, inhibitory receptors are generally considered to bind to ligands with higher affinity than their activating counterparts, as evidenced within the KIRs (46) and the CTLA-4/CD28 (47) receptor systems. This bias favoring inhibition is likely important for the maintenance of immune tolerance toward healthy tissues under steady-state conditions. However, our binding data indicate that DNAM-1 binds to its ligands with an affinity in the same range as that of TIGIT and CD96 (2-10 M) (29,30). Thus, the nectin receptor axis likely represents a highly-sensitive system where relatively minor alterations in receptor/ligand expression levels could markedly impact immune cell activation.
The last few years have seen considerable progress in our understanding of the molecular basis underpinning nectin receptor recognition, in part due to the determination of multiple receptor-ligand complexes, including TIGIT/Necl-5 (31), TIGIT/nectin-2 (29), and CD96-D1/Necl-5 (30). Altogether, these studies have highlighted the important role that the D1 domain of each receptor, and in particular the lock and key motifs, plays in ligand binding. However, whereas TIGIT is composed of only a single Ig domain, DNAM-1 and CD96 possess two and three Ig domains, respectively, raising questions as to the evolutionary driving force behind this increased complexity, particularly because the isolated CD96 D1 domain is sufficient for ligand binding (30).
Here, we extend these findings by reporting the structure of the entire ectodomain of human DNAM-1 bound to Necl-5. DNAM-1 adopted an unusual stacked configuration of its two Ig domains that differed markedly from the linear arrangement observed within the nectin adhesion molecules. This collapsed architecture placed the D2 domain of DNAM-1 adjacent to Necl-5, such that Asn-186 and its associated N-glycan made direct interactions with the C-CЈ loop of the ligand. Although we were only able to visualize a single GlcNAc moiety at the interface due to our inability to obtain crystals of natively glycosylated DNAM-1, MS-based glycopeptide and glycan analyses established that Asn-186 of recombinant DNAM-1 was extensively decorated with a heterogeneous mixture of large, complex N-glycans. Although our SPR and tetramer staining experiments demonstrated robust binding of fully glycosylated DNAM-1 to both nectin-2 and Necl-5, precisely how extended glycan chains attached to Asn-186 are accommodated at the relatively tight juncture between DNAM-1 D1, D2, and necl-5 D1 is at this stage unclear. Indeed, mutation of Asn-186 or removal of its associated N-glycan had no impact on DNAM-1 ligand-binding affinity or DNAM-1-mediated NK cell cytotoxicity. In this regard, it is important to note that Arg-185 of DNAM-1 D2 also makes interactions with Gly-70 of Necl-5. However, because these contacts were restricted to the mainchain atoms of the receptor and ligand, we were unable to test their functional importance using mutagenesis. Thus, although at this stage we have no evidence that DNAM-1 D2 contributes to ligand recognition, we cannot exclude this possibility entirely.
Our data raise new questions regarding the function of the D2 domain within the nectin receptor family. Because TIGIT comprises just a single extracellular Ig domain, it seems unlikely that the D2 domain is required as a structural scaffold to place the D1 domain in the appropriate position/orientation for ligand binding. Whether the D2 domain can impact recognition of other, as of yet unidentified, nectin receptor ligands or ancillary molecules that are directly associated with DNAM-1, such as leukocyte function-associated antigen-1 (LFA-1), will be an interesting avenue for future investigation. Moreover, it will also be insightful to identify whether the CD96 D2 domain adopts a configuration similar to that observed in DNAM-1, particularly because human CD96 exists as two isoforms that differ by a 16-residue insertion within the D2 domain, and this alteration has been reported to modulate Necl-5 binding (48).
Overall, the structure described herein is similar to that of mouse DNAM-1 bound to human Necl-5 recently reported by Wang et al. (32). The close overall architecture and binding affinity of human and mouse DNAM-1 indicate the evolutionarily conserved nature of this receptor system. However, mouse and human DNAM-1 have diverged somewhat in regard to primary amino acid sequence (53% identity within the ectodomain) and indeed the nature of many of the interactions were distinct. Although the D2 domain of both human and mouse DNAM-1 made interactions with Necl-5, mouse DNAM-1 did so via a solitary interaction derived from Glu-185 (equivalent to Gly-184 in human DNAM-1), whereas human DNAM-1 formed a more extensive interface that included Arg-185, Asn-186, and its associated N-glycan. Although the importance of Glu-185 in mouse DNAM-1/Necl-5 recognition was not formally tested, a truncated DNAM-1 construct comprising only the D1 domain exhibited reduced binding to Necl-5expressing cells, leading the authors to conclude that the D2 domain modulated the interaction with Necl-5. This was in direct contrast to our binding and functional assays that indicated that mutation of contact residues within DNAM-1 D2 did not impact Necl-5 binding or DNAM-1 function. These results can be reconciled by our observation that we were unable to express properly folded DNAM-1 D1 in mammalian cells, indicating that the D2 domain might be important for stability of the receptor. Nevertheless, knowledge of these differences between human and mouse DNAM-1 will be important when interpreting data from mouse studies.

Protein expression and purification
The construct encoding human DNAM-1 (amino acids 19 -250) was codon-optimized for human expression (Integrated DNA Technologies) and cloned between AgeI and KpnI restriction sites into the pHLSec vector to include either a C-terminal hexa-histidine tag (for structural studies) or a consensus site for the biotin protein ligase enzyme (BirA) followed by a hexa-histidine tag (for SPR and tetramerization). The human nectin-2 ectodomain (residues 32-350) was cloned in the same manner. DNAM-1 (and mutants thereof) and nectin-2 were expressed via transient transfection in HEK-293S cells. For some SPR, tetramer staining, and MS experiments, WT DNAM-1 was also expressed in HEK-293T cells, where indicated. The human Necl-5 ectodomain (residues 32-350 and 28 -334) was codon-optimized for expression in Trichoplusia ni (Integrated DNA Technologies) and cloned into the pFASTBac vector (Invitrogen) between XhoI and SpeI restriction sites to include a C-terminal hexa-histidine tag. The plasmid was incorporated into a recombinant baculovirus, and the viral titer was expanded in SF9 cells as described in the Bac-to-Bac manual (Invitrogen). Soluble Necl-5 was obtain by infecting Hi5 cells with 2% P3 virus. Secreted protein from mammalian and baculoviral systems was buffer-exchanged into 10 mM Tris-HCl (pH 8.0), 500 mM NaCl and purified via nickel-affinity and size-exclusion chromatography using a Superdex S200 16/600 column (GE Healthcare) in a buffer comprising 10 mM Tris-HCl (pH 8.0), 150 mM NaCl. For biotinylation, purified BirAtagged DNAM-1 was buffer-exchanged into 10 mM Tris-HCl (pH 8.0) and biotinylated overnight at 4°C.

Crystallization and data collection
For crystallization, purified DNAM-1 was enzymatically digested with endo H in the presence of 10 mM sodium citrate (pH 5.0) at 4°C overnight, prior to being mixed with Necl-5 in a 1:1 molar ratio at a total concentration of 10 mg/ml. Crystals were obtained using the hanging-drop vapor diffusion method using a reservoir solution comprising 20% (v/v) PEG3350, 0.18 mM K 2 SO 4 , 10 mM EDTA and were cryo-protected in in the presence of 10% (v/v) glycerol. X-ray diffraction data were recorded on a Quantum-315 CCD detector at the MX2 beamline of the Australian Synchrotron. Data were integrated by MOSFLM and scaled using SCALA within the CCP4 suite of programs. Details of the data-processing statistics are given in Table S1.

Structure determination and refinement
The structure was determined by molecular replacement using Phaser. Isolated models for each of the Necl-5 D1, D2, and D3 domains were generated from the unliganded structure (PDB code 4FQP) (28) using PyMOL (Schrödinger, Inc.) and used as search models for the D1, D2, and D3 domains of Necl-5, respectively. Molecular replacement solutions for the DNAM-1 D1 and D2 domains were obtained using a model consisting of the isolated nectin-2 D1 (PDB code 5V52) as a search model. The structure was refined via iterative cycles of model building in Coot and refinement using Buster (http:// globalphasing.com/buster/). 4 N-Linked glycans were manually incorporated into regions of positive density that correlated to the requisite sequence motif, NX(S/T), where X is any amino acid except proline. The final structure comprised two molecules of DNAM-1 and two molecules of Necl-5 within the asymmetric unit that were arranged into two heterodimers (heterodimer 1, chains A and C; heterodimer 2, chains B and D). Structural analysis was restricted to heterodimer 1 as the electron density surrounding chains A and C was better defined than that of chains B and D. The final structure refined to final R factor /R free values of 20.8 and 23.0. Details of the refinement statistics are given in Table S1. The structure factor file and associated atomic coordinates for the DNAM-1/Necl-5 structure have been deposited within the Protein Data Bank with the accession code 6O3O.

SPR
SPR experiments were performed using a BIAcore 3000 system (GE Healthcare) at 25°C with a buffer comprising 10 mM Tris-HCl (pH 8.0), 150 mM NaCl, and 0.005% (v/v) surfactant P20. Varying response units of WT biotinylated DNAM-1 or mutants thereof (see Fig. 4 for details) were immobilized onto streptavidin-coated sensor chips (GE Healthcare), and flow cells were quenched with 10 l of free D-biotin (1 mg/ml) at a flow rate of 5 l/min prior to injection of analyte. Varying concentrations of soluble nectin-2 and Necl-5 were passed over the flow cells for 30 -60 s, in duplicate, at a flow rate of 10 l/min. The final responses were double-referenced by subtracting responses from an "empty" flow cell containing biotin-blocked streptavidin as well as from buffer-only injections. The responses at equilibrium were used to construct equilibrium binding curves that fit by a single-site binding model. The calculated equilibrium dissociation constants represent the mean Ϯ S.E. of the mean from n ϭ 3 (for WT DNAM-1 produced in HEK-293S and HEK-293T) or n ϭ 2 (for all others) independent experiments. Data were analyzed with Scrub-ber2.0 (BioLogic Software, Campbell, ACT, Australia) and Prism 7.0 (GraphPad Software).

FACS analysis
K562 and NK92 cells were harvested and washed in PBS/FCS (1%), and Fc receptors were blocked using 10% normal goat serum. K562 cells were then washed and stained with DNAM-1 tetramers, anti-CD155 (TX24; Biolegend, San Diego), or anti-CD112 (TX31; Biolegend) for 30 min at 4°C. After staining with DNAM-1 tetramers, cells were directly fixed by the addition of paraformaldehyde (10% in PBS) for 20 min at 4°C. K562 cells stained with antibodies were not fixed. NK92 cells were stained with anti-DNAM antibody (11A8; Biolegend) or nectin-2 tetramers for 30 min at 4°C. After staining with nectin-2 tetramers, cells were directly fixed by the addition of paraformaldehyde (10% in PBS) for 20 min at 4°C. After staining/fixation, K562 and NK92 cells were washed twice in PBS/FCS and prepared for FACS. For acquisition, events were electronically gated on FSC-A versus FSC-H (singlets), followed by FSC-A and SSC-A (to exclude doublets and debris). Among the remaining populations, at least 5000 electronic events of interest were collected using an LSR-II or X-20 Fortessa (BD Biosciences).

Protein denaturation
The cysteine residues of recombinant human DNAM-1 expressed in HEK293T or HEK-293S cells were reduced using 10 mM aqueous DTT, 45 min at 56°C, and then carbamidomethylated using 25 mM aqueous iodoacetamide, 30 min in the dark at 20°C. The alkylation reaction was quenched using 30 mM aqueous DTT (final concentrations stated).

N-Glycan preparation
All reagents were obtained from Sigma unless otherwise stated. The water used for all experimental procedures was of ultra-high purity (MilliQ filtration system, Millipore). The DNAM-1 N-glycans were released and prepared for glycomics as described previously (49). In brief, DNAM-1 (20 g) was immobilized in triplicate spots on a primed 0.45 m polyvinylidene fluoride (PVDF) membrane (Merck-Millipore). The dried PVDF spots were stained with Direct Blue, excised, transferred to separate wells in a flat bottom polypropylene 96-well plate (Corning Life Sciences), blocked with 1% (w/v) polyvinylpyrrolidone in 50% (v/v) aqueous methanol, and washed with water. De-N-glycosylation was performed using 2 units of Elizabethkingia miricola N-glycosidase F (Promega) per 20 g of DNAM-1 in 10 l of water/well, 16 h at 37°C. The released N-glycans were transferred into fresh tubes and hydroxylated by the addition of 100 mM aqueous ammonium acetate (pH 5), 1 h at 20°C. The glycans were reduced using 1 M sodium borohydride in 50 mM aqueous potassium hydroxide, 3 h at 50°C. The reaction was quenched using glacial acetic acid. Dual desalting of the reduced N-glycans was performed using first strong cation-exchange resin (where the N-glycans are not retained) and then porous graphitized carbon (PGC) (where N-glycans are retained) packed as micro-columns on top of C 18 discs in P10 solid-phase extraction (SPE) formats. The N-glycans were eluted from the PGC micro-columns using 0.05% trifluoroacetic acid (TFA): 40% acetonitrile (ACN): 59.95% water (v/v/v), dried, and redissolved in 10 l of water for LC-MS/MS analysis. Bovine fetuin was included as a sample handling and LC-MS/MS control.

N-Glycan analysis
The N-glycans were separated on a Hypercarb PGC column (particle size 3 m, column length 100 mm, inner diameter 0.18 mm, Thermo Fisher Scientific) heated to 27°C over an 83-min gradient of 0 -45% (v/v) ACN (solvent B) in 10 mM aqueous NH 4 HCO 3 (solvent A) at a flow of 2 l/min delivered by a Dionex Ultimate-3000 HPLC (Dionex, CA). The glycans were detected using an LTQ Velos Pro ion trap mass spectrometer (Thermo Fisher Scientific) in negative ion polarity mode with a source voltage of 3.2 kV. The MS1 scan range was m/z 550 -2000 with a zoom scan resolution of m/z 0.25 (FWHM). The automatic gain control (AGC) for the MS1 scan was 5 ϫ 10 4 with a maximum accumulation time of 50 ms. Data-dependent tandem mass spectrometry (MS/MS) was performed using collision-induced dissociation at a normalized collision energy (NCE) of 33% of the 10 most abundant precursors in each MS1 scan. For the MS2 events, the resolution was set to m/z 0.35 (FWHM); the AGC was 2 ϫ 10 4 , and the maximum accumulation time was 300 ms. Dynamic exclusion was disabled. All MS and MS/MS spectra were acquired in profile mode. The LC-MS/MS instrument was tuned and calibrated, and its performance was bench-marked using bovine fetuin N-glycans prior to use. The raw LC-MS/MS data were browsed and annotated using Xcalibur version 2.2 (Thermo Fisher Scientific) using assisting software, i.e. GlycoMod and GlycoWorkBench, and manual de novo glycan sequencing as published previously (50). The relative abundances of the individual N-glycans were determined from relative area-under-the-curve (AUC) measurements based on extracted ion chromatograms (EICs) performed for the monoisotopic m/z values using Skyline (51).

N-Glycopeptide analysis
The enriched N-glycopeptides were separated on a nanoflow reversed-phase Helo C18 column (particle size 2.7 m, column length 100 mm, inner diameter 75 m, Thermo Fisher Scientific) heated to 50°C. A gradient over 70 min of 0 -60% (v/v) solvent B (0.1% (v/v) formic acid (FA) in ACN) was applied (solvent A was 0.1% (v/v) aqueous FA). A constant flow rate of 450 nl/min was delivered by an Eksigent nano-400 LC system (SCIEX). The glycopeptides were detected using an Orbitrap Elite mass spectrometer (Thermo Fisher Scientific) in positive polarity mode with a source voltage of 3 kV. The MS1 scan range was m/z 500 -200; AGC was 1 ϫ 10 6 , and the maximum Immune recognition of nectin adhesion molecules accumulation time was 50 ms. The 10 most abundant precursors in each MS1 scan were selected for data-dependent MS/MS acquisition (m/z 110 -2000) using higher-energy C-trap dissociation (HCD) at an NCE of 35%. The AGC was 5 ϫ 10 4 , and the maximum accumulation time was 200 ms for the HCD-MS/MS events. Dynamic exclusion was enabled with a maximum repeat count of three for every 30 s. All MS and MS/MS spectra were acquired in profile mode. The LC-MS/MS instrument was tuned and calibrated, and its performance bench-marked using BSA peptides prior to use. The raw LC-MS/MS data were browsed and annotated using Xcalibur version 2.2 (Thermo Fisher Scientific) using assisting software, i.e. GPMAW version 10 (Lighthouse Data, Denmark) and Byonic version 2.3 (Protein Metrics) and manual de novo glycopeptide sequencing (53). The relative abundances of the individual N-glycopeptides were determined from relative AUC measurements based on monoisotopic m/z EICs performed manually using Xcalibur version 2.2 (Thermo Fisher Scientific).

Generation NK92-DNAM-1 cell lines
The full-length sequences of WT and mutated DNAM-1 were cloned into the MSCV plasmid and retroviral supernatants generated using HEK-293T cells. For NK92 transduction, 50 g/ml retronectin was coated onto a nontreated plate for 2 h at room temperature prior to washing (2 times) and blocking with 2% BSA in PBS for 30 min. Plates were then washed with PBS (2 times), and viral supernatant was added to the wells after filtration through 0.45-m filters. Virus was adhered to the plates by centrifugation at 1500 ϫ g (2 h at room temperature). Supernatant was then removed, and the wells were washed with PBS, 0.1% BSA prior to addition of NK92 cells at 1 ϫ 10 6 cells/ml. Cells and virus were integrated by centrifugation at 200 ϫ g for 5 min at room temperature. NK92 cells infected with retrovirus were sorted on the basis of GFP expression. Populations were sorted until stable with insertion of DNAM-1 determined by FACS.

NK cell-killing assays
K562 cells were labeled with Cell Trace Violet according to the manufacturer's instructions (Thermo Fisher Scientific). NK92 cells were co-cultured with K562 cells in the presence of a saturating amount (10 g/ml) of anti-NKG2D antibody (1D11; Biolegend). Different ratios of cells were added to U-bottom plates and centrifuged at 1400 rpm for 1 min. The co-cultures were then incubated for 4 h at 37°C at which time they were washed and labeled with Zombie NR. For acquisition, events were electronically gated on FSC-A and SSC-A to detect live and dead cells. Effectors were excluded in the GFP channel, and targets were identified on the V450/50 channel. Among the remaining population in this channel, at least 200 viable electronic events of interest (CTV ϩ /Zombie NR Ϫ ) were collected using an LSR-II or X-20 Fortessa (BD Biosciences).