A molecular switch in mouse CD1d modulates natural killer T cell activation by α-galactosylsphingamides

Type I natural killer T (NKT) cells are a population of innate like T lymphocytes that rapidly respond to α-GalCer presented by CD1d via the production of both pro- and anti-inflammatory cytokines. While developing novel α-GalCer analogs that were meant to be utilized as potential adjuvants because of their production of pro-inflammatory cytokines (Th1 skewers), we generated α-galactosylsphingamides (αGSA). Surprisingly, αGSAs are not potent antigens in vivo despite their strong T-cell receptor (TCR)–binding affinities. Here, using surface plasmon resonance (SPR), antigen presentation assays, and X-ray crystallography (yielding crystal structures of 19 different binary (CD1d-glycolipid) or ternary (CD1d-glycolipid-TCR) complexes at resolutions between 1.67 and 2.85 Å), we characterized the biochemical and structural details of αGSA recognition by murine NKT cells. We identified a molecular switch within murine (m)CD1d that modulates NKT cell activation by αGSAs. We found that the molecular switch involves a hydrogen bond interaction between Tyr-73 of mCD1d and the amide group oxygen of αGSAs. We further established that the length of the acyl chain controls the positioning of the amide group with respect to the molecular switch and works synergistically with Tyr-73 to control NKT cell activity. In conclusion, our findings reveal important mechanistic insights into the presentation and recognition of glycolipids with polar moieties in an otherwise apolar milieu. These observations may inform the development αGSAs as specific NKT cell antagonists to modulate immune responses.

Type I or semi-invariant natural killer T (NKT) 4 cells are a specialized population of T lymphocytes that express a semi-conserved TCR ␣-chain rearrangement (TRAV11/TRAJ18 (V␣14J␣18) in mice, TRAV10/TRAJ18 (V␣24J␣18) in humans) (1). They are restricted by the nonpolymorphic and nonclassical major histocompatibility class (MHC) I analog CD1d and recognize synthetic, self, and microbial glycolipid antigens (2)(3)(4)(5)(6)(7)(8)(9)(10)(11)(12)(13)(14). NKT cells are considered innate like T lymphocytes and respond to antigen challenge within 2-4 h, producing copious amounts of both pro-and anti-inflammatory cytokines. Many ␣-GalCer analogs have been studied that induce a superior production of IFN-␥, often combined with a reduced amount of IL-4 compared with ␣-GalCer itself (also known as Th1 skewing) (15)(16)(17). It has been observed that Th1 skewing antigens also lead to increased production of IL-12 by the presenting DC, which together with IFN-␥ from the NKT cell transactivate NK cells (18 -20). NK cells are the major producers of IFN-␥ at later time points. As such NKT cells are of therapeutic interest because they exert a central role in bridging the innate and adaptive immune system. The mechanism of Th1 cytokine skewing involves TCR ligation with CD1d-antigen complexes on DCs, which induces up-regulation of costimulatory molecules, such as CD40L on the NKT cell and CD40 on the DC, ultimately leading to DC maturation and IL-12 production. Several non-mutually exclusive mechanisms have been proposed for the Th1 skewing activity of various ␣-GalCer analogs. These involve the loading of glycolipids to CD1d located in lipid rafts or an increased in vivo half-life of CD1d-glycolipid complexes often via enhanced CD1d-glycolipid interactions (16,17,21).
We have previously hypothesized that ␣-GalCer analogs that contain aromatic groups to enhance binding to CD1d while not altering TCR recognition are superior Th1 skewing antigens (16). In one of these glycolipids, NU-␣-GalCer, the C6Љ-OH of the galactose was replaced by a naphthyl urea group that induced a structural change in the AЈ roof of CD1d, leading to a cavity in which the aromatic group is bound (16). In addition to the two lipid tail "anchors," we proposed this novel interaction as the "third anchor" hypothesis (16). Several other ␣-GalCer analogs that had aromatic groups at the 6Љ-OH of the galactose also caused an increased serum IFN-␥ and IL-12 production, albeit not inducing a structural change within the AЈ roof, but rather mediating hydrophobic contacts with CD1d or the TCR (17). Other strong Th1 skewing ligands include 7DW8-5, containing a terminal aromatic para-fluoro-phenyl group in the acyl chain, and other ligands with terminal phenyl anchors in the acyl chain that bound within the AЈ pocket of CD1d are currently being tested as vaccine adjuvants (22)(23)(24)(25)(26). The sphingosine chain of ␣-GalCer is the only moiety that had not been systematically substituted with aromatic anchors. We have recently synthesized a panel of ␣-GalCer analogs in which terminal or in-chain phenyl groups are incorporated in the phytosphingosine moiety of short chain ␣-GalCer (PBS-25) (27) via an amide linker (28). These short-chain ␣-galactosylsphingamides (␣GSA) were bound by the TCR with high affinity, yet did not induce a robust cytokine production when administered to mice by intraperitoneal injection (28). However, when increasing the length of the acyl chain to 26 carbons, ␣GSAs induced the production of both the pro-and anti-inflammatory cytokines IL-4 and IFN-␥, respectively, albeit at markedly reduced levels compared with ␣-GalCer (28). In the current study, we have systematically analyzed these ␣GSAs with regard to the TCR-binding kinetics and their ability to activate a murine iNKT cell hybridoma. We also determined the crystal structures of 19 different binary (CD1d-glycolipid) or ternary (CD1d-glycolipid-TCR) complexes in an effort to unravel the molecular basis of the reduced antigenicity of the ␣GSAs. We have discovered that the activity of ␣GSAs is regulated by an interplay of acyl chain length and a molecular switch within CD1d that forms a novel interaction with the amide group in the phytosphingosine of ␣GSAs.

␣-Galactosyl sphingamide overview
We have reported previously a novel class of ␣-GalCer derivatives termed ␣-galactosylsphingamides (␣GSAs) or in short sphingamides (28). The sphingamides are similar to ␣-GalCer but feature an amide group within the sphingosine moiety at a fixed distance from the anomeric carbon ( Fig. 1). This amide was used to introduce either a terminal or in-chain phenyl group in the sphingosine. The acyl chain length was varied from 8 to 26 carbons. We have synthesized various control compounds that mimic the overall structure of the sphingamides but lack the central amide group to assess the influence of this polar moiety in NKT cell activation.

SPR studies using the V␣14V␤8.2 TCR (clone 2C12)
In analogy to our previous study (28), all sphingamides with a terminal phenyl group were bound by the V␣14V␤8.2 TCR with similar high affinities in the range from 16 to 37 nM (Fig. 2). There were no significant differences between the TCR association and dissociation rates. However, the sphingamide ␣GSA [26,P5p], which has a C26 fatty acid and an in-chain parasubstituted phenyl moiety was bound by the TCR with considerably weaker affinity, suggesting that the phenyl moiety impairs TCR binding to CD1d. We have not measured the TCR-binding affinities for all sphingamides or control compounds, because our selected lipids establish a similar TCRbinding recognition. However, we failed to detect any TCR binding to ␣GSA [8,2P] and ␣GSA [8,4P], suggesting that these relatively short sphingamides do not bind to CD1d. This is also corroborated by our inability to obtain crystal structures of these two lipids in complex with CD1d (see below).

iNKT cell activation by sphingamides
Previously, we reported that sphingamides are not potent agonists for murine iNKT cells in vivo (28). Therefore, we have assessed in detail the antigenicity of sphingamides using a cellfree and cell-based antigen presentation assay. Surprisingly, despite their high TCR-binding affinity, none of the short acyl chain sphingamides were able to robustly activate the iNKT cell hybridoma 1.2 when presented by recombinant mouse CD1d, suggesting that solely varying the position of the phenyl anchor does not affect antigenicity (Fig. 3A). In contrast, short acyl chain ␣-GalCer (␣GC [8,18]) strongly activated iNKT cells. We next investigated whether the sphingamide modification influenced loading in cellular compartments and presentation by CD1d. Therefore, we used a cell-based antigen-presentation assay in which A20 cells were transfected with either WT or tail-deleted mCD1d (Fig. 3, B and C). The tail-deleted mCD1d lacks the tyrosine containing trafficking motif within the cytoplasmic tail and prevents its recycling through lysosomal compartments after mCD1d has been expressed on the cell surface; the WT mCD1d can recycle through lysosomal compartments allowing for lipid loading and/or processing. Again, none of the short-chain sphingamides activated iNKT cells, regardless of whether mCD1d had access to lysosomes for potential sphingamide loading using lipid transfer proteins. To increase the binding affinity of the sphingamides to mCD1d, we then synthesized and analyzed the two full-length sphingamides ␣GSA [26,18P] and ␣GSA [26,2P5p]. Surprisingly, both lipids potently activated iNKT cells and the response was optimal already at the lowest lipid concentration (0.5 g/ml), despite ␣GSA [26,2P5p] having a 10-fold reduced TCR-binding affinity (Figs. 2 and 3D). The antigen-presentation data led us to hypothesize that the acyl chain length influences iNKT cell activation by sphingamides and, therefore, we have also generated and included intermediate acyl chain ␣GSAs (C12, C16, and C20) in our subsequent studies. These ␣GSAs were based on the sphingamide 6P backbone.

Sphingamide presentation by CD1d
To rationalize why all sphingamides are bound by the iNKT TCR with high affinity, but only long acyl chain sphingamides potently activate iNKT cells, we examined the structural basis of sphingamide presentation by mCD1d. The individual crystal structures of mCD1d-sphingamide complexes were determined to resolutions between 1.67 and 2.45 Å (Table S1). The two short-chain sphingamides ␣GSA [8,6P] and ␣GSA [8,8P] had been crystallized previously (termed as 5d and 5e in Ref. 28). Most sphingamides show well-defined electron density for the hydroxyls of both the sugar moiety and the lipid backbone, indicative of an ordered presentation by CD1d (Fig. 4).
The exception was ␣GSA [12,6P], which showed well-defined electron density for the galactose but less-ordered electron density around the amide group, suggesting a more flexible binding of that portion. As expected for ␣-GalCer analogs, the acyl chain is always inserted into the AЈ pocket, regardless of its length, whereas the sphingosine moiety containing the amide group is bound in the FЈ pocket. Up to an acyl chain length of C12, all sphingamides also recruited a spacer lipid into the AЈ pocket of CD1d. Interestingly the helical orientation of the acyl chain circling in the AЈ pocket differs between the longer sphingamides ␣GSA [20,6P] and ␣GSA [26,6P]. Whereas ␣GSA[20,6P] circles clockwise when looking down at the pocket, ␣GSA [26,6P] circles in a counterclockwise fashion. Whether there is a functional consequence of the acyl chainbinding orientation is not known.
The sphingamide ␣GSA[8,6P] was structure-based designed and predicted to form an additional hydrogen bond between its amide oxygen and the hydroxyl of mCD1d residue Tyr-73. Surprisingly, only the long-chain sphingamide ␣GSA [26,6P] exhibited this H-bond, whereas the amide group of the shorter homologues and other sphingamides exhibit an altered orientation (Fig. 5). It appears there is a high degree of rotational freedom for this group, with the amide oxygen either pointing toward or away from Tyr-73. However, because the electron density in that region is well-defined for most sphingamides, we propose that the acyl chain length influences the precise positioning of the amide group in the FЈ pocket, which in turn influences iNKT cell activation. Except for the amide group, the sphingamides bound in a conserved orientation with the galactose being stabilized by a core hydrogen-bond network involving CD1d residues Asp-80, Asp-153, and Thr-156 and the 2Љ-OH, 3Љ-OH of the galactose as well as the 3Ј and/or 4Ј hydroxyls of the phytosphingosine chain. The only exception is ␣GSA [20,6P], which showed a slight rotation in the galactose positioning. This binding orientation resulted in the loss of the H-bond interaction between Asp-153 of CD1d and 2Љ-OH and 3Љ-OH of the galactose, whereas a new H-bond is formed with the 4Љ-OH. However, the electron density for the galactose of this ligand is less contoured compared with other sphingamides, suggesting some flexibility in the binding orientation, especially because there is no well-defined electron density for the galactose hydroxyls (Fig. 4).  [8,18]) with increasing alkyl chain length between the terminal phenyl anchor and the amide group. The ␣GSA was then selected and variants synthesized with acyl chain length from C12 to C26 (left panel). Control ␣-GalCer analogs with terminal phenyl anchors lacking the amide group were also prepared (right column). Bottom row, ␣GSAs that have a short or long acyl chain and a central phenyl anchor.

Sphingamide recognition by NKT cells TCR recognition of sphingamides
Because we had previously established that the TCR of iNKT cells can induce a structural change in both glycolipid as well as mCD1d upon binding (29), we also determined the crystal structures of a panel of sphingamides presented by mCD1d and in complex with the V␣14V␤8.2 TCR (clone 2C12) at resolutions between 2.0 and 2.85 Å (Table S1). The TCR binds above mCD1d in the canonical parallel-binding orientation, with the complementarity-determining region (CDR) 1␣ sitting above the galactose of the glycolipids and CDR3␣ binding centered above the FЈ pocket of CD1d. CDR3␣ ]contacted both the gly-colipid ligand as well as the FЈ roof using the hydrophobic finger Leu-99␣ (Fig. 6).
The electron density of each glycolipid is unambiguous, even in the lower-resolution structures and is a consequence of the TCR stabilizing and locking the binding orientation of the glycolipids inside the CD1d binding groove. A close examination of the CD1d-glycolipid, as well as TCR-glycolipid interactions reveals no differences between the different sphingamides, in contrast to the binary complexes, in which the sphingamides bind differently to mCD1d with respect to the positioning of the amide group in the phytosphingosine-like moiety. The TCR Figure 2. TCR-binding kinetics. SPR sensorgrams are shown as red curves, with fitted data (1:1 Langmuir interaction) shown as black curves. TCR concentrations are indicated and inset shows the kinetic data including TCR association rate (k a ), TCR dissociation rate (k d ) and equilibrium binding constant (K D ). Each binding experiment was performed at least twice with two different TCR preparations. One representative sensorgram is shown.

Figure 3. NKT cell activation by sphingamides.
A-C, short-chain ␣GSAs were assessed for their ability to induce IL-2 production by the murine NKT cell hybridoma 1.2 when added to recombinant mCD1d (A), or when added to A20 cells that were transfected with either tail-deleted CD1d (B), or WT CD1d (C). IL-2 was measured in the supernatant using ELISA. GalGalCer was used as a control for CD1d trafficking through the lysosome. D, same as A but including long chain ␣GSAs and a higher antigen concentration dose. Experiments were performed at least three times with one representative experiments shown.

Sphingamide recognition by NKT cells
forms the canonical hydrogen-bond interactions, especially Gly-96␣ with the 2Љ-OH and Asn-31␣ with the 3Љ-and 4Љ-OH groups of the galactose, whereas Arg-95␣ contacts Asp-80 of CD1d (Fig. 7). In the ternary complexes, all sphingamides, regardless of whether they contain a central or terminal phenyl attached to the amide group within the phytosphingosine-like moiety, now form an additional H-bond between the hydroxyl of Tyr-73 of CD1d and the amide oxygen of the sphingamides. This suggests that upon TCR binding, the TCR forces each sphingamide into a conserved binding orientation, the hallmark of which is the formation of this novel H-bond interaction.

Identification and characterization of a molecular switch in CD1d
Given the fact that the long-chain sphingamide agonist ␣GSA [26,6P] has the identical binding orientation in the ternary complex compared with the inactive short-chain sphingamide ␣GSA [8,6P], we compared the binding orientation before and after TCR binding for all crystallized sph-ingamides (Fig. 8, A and B). Surprisingly, only the active sphingamide ␣GSA [26,6P] shows the additional H-bond with Tyr-73 of CD1d in the absence of TCR binding (Fig.  8A), whereas all sphingamides form this H-bond upon TCR binding (Fig. 8B). This led us to speculate that the length of the acyl chain determines the binding orientation of the amide group in the FЈ pocket of CD1d and whether or not it is further stabilized by the additional H-bond with Tyr-73. We hypothesize that this interaction forms a molecular switch that affects sphingamide activity. A preformed H-bond (in the absence of TCR) resembles an active state (switch ON), whereas lack of this H-bond will lead to an inability of activating iNKT cells (switch OFF). A systematic analysis of a panel of sphingamides with increasing acyl chain length revealed a direct correlation between acyl chain length and antigenicity. The short C8 acyl chain failed to robustly activate an iNKT cell hybridoma in a cell-free antigen presentation assay using recombinant mCD1d, whereas the antigenicity successively increases with increasing acyl

Sphingamide recognition by NKT cells
chain length (Fig. 8C). This suggests that the presence or absence of the preformed H-bond involving Tyr-73 are the two extreme states (ON versus OFF) and that the acyl chain length modulates the overall intrinsic activity of sphingamides. We have further tested the influence of the H-bond by site-directed mutagenesis of the molecular switch (Tyr-73). Y73F-modified CD1d retains its overall structure and interaction with sphingamide, while excluding H-bond bond formation with the amide. With the exception of the longchain sphingamide ␣GSA [26,6P], the activity of all sphingamides is markedly reduced, suggesting that the H-bond stabilizes the binding orientation of the amide group within CD1d after TCR binding to allow for potent iNKT cell activation. However, the H-bond itself is not the only driver of activity. The acyl chain length itself dictates the binding orientation of the amide group even in the absence of this stabilizing H-bond. The Y73F mutant also showed slightly decreased cytokine production by the control antigen ␣GC [8,18] but not to the extent observed for the sphingamides. We next speculated that modifying the molecular switch to stabilize the orientation of the amide group in the ON position would increase antigenicity. Hereto, we generated the mutant mCD1d Y73H. Indeed, all sphingamides now show hyperactivity, even the otherwise inactive short chain ␣GSA [8,6P], whereas in contrast the activity of the control antigen was greatly reduced (ϳ4-fold) (Fig. 8C). This suggests that introducing a polar group into an apolar environment (the hydrophobic FЈ pocket of CD1d) reduces the activity of traditional hydrophobic glycolipid antigens, whereas it is beneficial for engineered glycolipids that contain polar groups within the sphingosine chain.
The polarity likely stabilizes the amide group and positions it away from the FЈ roof, which is the major binding site of the TCR␣ chain, especially the hydrophobic finger Leu-

Sphingamide recognition by NKT cells
99a, which was shown to be crucial for iNKT cell activation (30). If our hypothesis holds true, we would expect that the inactive sphingamide ␣GSA [8,6P] now forms the additional H-bond with His-73 in the mCD1d Y73H mutant before TCR binding, leading to an active complex. We have, therefore, determined the crystal structure of the mCD1d Y73H/ ␣GSA[8,6P] complex (Table S1 and Fig. 8D). As expected, ␣GSA [8,6P], which had the amide oxygen pointing away from the hydroxyl of Tyr-73 in WT CD1d (switch OFF) (Fig.  8A, yellow lipid), now features a rotated amide group to form an H-bond with His-73 (switch ON). Electron density in that region is unambiguous; however, the introduction of polar- Figure 6. Overview of ternary complex structures with electron density for the glycolipids. Similar view and coloring as Fig. 4 but with TCR ␣-chain shown as cyan cartoon and TCR ␤-chain shown as orange loops (not contacting the antigen). CDR1␣ and 3␣ sit above the glycolipid.

Sphingamide recognition by NKT cells
ity in the FЈ pocket slightly affects binding of the terminal acyl chain carbons in the AЈ pocket, for which electron density becomes disordered.
In summary, two factors appear to drive sphingamide activity. The acyl chain length, which dictates amide positioning within CD1d, and the ability to form a stabilizing interaction

Sphingamide recognition by NKT cells
with Tyr-73 to lock the binding orientation of the amide group in an active state.

Discussion
In this study, we have characterized a panel of ␣-GalCer analogs termed ␣-galactosylsphingamides that contain a phenyl group connected to the phytosphingosine moiety via an amide linker. The rationale was that the phenyl groups would act as an anchor to increase the binding interaction with CD1d, while not affecting the interaction with the TCR of iNKT cells. Unexpectedly, although the TCR bound the sphingamides with high affinity, short acyl chain (C8 -C12) ␣GSAs failed to activate iNKT cells in vitro and in vivo, whereas long acyl chain ␣GSAs conferred antigenicity. The obvious conundrum was that both short-chain (inactive) and long-chain (active) ␣GSAs showed no difference in the TCR-binding kinetics, as determined by surface plasmon resonance studies using recombinant molecules in solution. Even though this assay does not account for other interactions (e.g. co-receptor binding, formation of the immunological synapse) that would occur in vivo upon antigen recognition, previous studies on NKT cells observed a correlation between the 2D TCR-binding affinity and the cytokine production (5,7,(31)(32)(33). By determining the crystal structures of various ␣GSAs and control compounds before and after TCR binding we identified a molecular switch that controls the anti-genicity of ␣GSAs, yet does not affect the TCR-binding kinetics. The molecular switch is formed between the amide oxygen of the ␣GSAs and Tyr-73 of CD1d and controlled by the acyl chain length of the ligands. This molecular switch has an ON and OFF position before TCR binding, which correlates with the subsequent ability to activate NKT cells. Considering that there is no difference in the TCR-binding kinetics between inactive and active ␣GSAs, we speculate that a compensatory mechanism is masking the biophysical properties leading to the observed biological differences. The active sphingamide ␣GSA [26,P5p] with an in-chain phenyl group is an outlier, because it lacks the preformed H-bond with Tyr-73 but also has reduced TCR-binding affinity. Previously, we had reported that the TCR association rate toward CD1d-glycolipid complexes is affected by the positioning of the galactose moiety (29). Because ␣-GalCer is already presented in the ideal orientation, a fast TCR association rate is observed, whereas ligands that need to undergo a re-orientation of the sugar upon TCR binding have a slower association rate (29). TCR dissociation is affected by whether the FЈ roof is open or closed before TCR binding (29). For ␣GSAs, TCR binding positioned the amide group in a way that the amide oxygen formed the novel H-bond interaction with Tyr-73 of CD1d. We believe that this interaction stabilized TCR binding (and reduced dissociation), because CD1d itself

Sphingamide recognition by NKT cells
contributed binding energy to keep the amide group in place. As such, we did not see reduced TCR dissociation. A similar scenario was reported for ␣-glycosyl diacylglycerolipids from Streptococcus pneumoniae compared with ␣-galactosyl diacylglycerolipids from Borrelia burgdorferi. Here, the TCR dissociated much slower from the glucose-containing ligand compared with the galactose-containing ligand (34). This was likely because of a novel contact formed between the glucose-specific axial 4Љ-OH and Thr-156 of CD1d, which cannot be formed with the galactose version (34). Overall, the binding affinity between both galactose and glucose antigens was similar.
Interestingly, a similar phenomenon was recently described for peptide-(p)MHC restricted TCRs. Here, TCRs were also found to bind with high affinity to pMHC complexes, yet this binding event did not lead to T-cell activation (35). Using single-molecule force measurements, the authors demonstrated the emergence of catch bonds in the activating pMHC-TCR interface, whereas nonagonist peptides formed slip bonds with the same TCR under force, which accounted for the differences in T-cell activation (35).
Recently p-MeOBz amide-containing ␣-GalCer analogs have been reported that contain aromatic groups within the acyl chain but also carry an amide linker, as well as an additional oxygen following the aromatic group. The polar amide was designed to interact with a region in the AЈ pocket of CD1d that for a slightly polar side pocket (Ser-28 and Gln-14) (36). Surprisingly, the incorporation of polar groups within the acyl chain results in a Th2-polarized NKT cell response characterized by a reduced production of IFN-␥ and increased levels of IL-4. Therefore, the incorporation of polar moieties in an otherwise apolar glycolipid antigen is a novel strategy for modulation NKT cell responses, but the outcome of the associated immune response is difficult to predict at this point.
The sphingamides are also of interest for the development of NKT cell antagonists, if their binding affinity toward CD1d can be increased without leading to antigenicity. Such a sphingamide would contain the short acyl chain (for inactivity) but introduce a third anchor, such as we have identified in the glycolipid NU-␣-GalCer (16). Development of NKT cell antagonists using short-chain sphingamides as a scaffold are currently ongoing.

Protein expression and purification
Mouse CD1d/b2M was expressed in Spodoptera frugiperda insect cells using the baculovirus expression system and purified as reported previously (5). The V␣14V␤8.2 TCR from the iNKT cell hybridoma 2C12 was generated by expressing the separate TCR ␣ and ␤ chains in Escherichia coli and refolding from inclusion bodies as reported (5).

Lipid synthesis
␣-Galactosylceramides containing either a C8 or C26 fatty acyl chain and sphingosine chains of varying length with terminal phenyl moieties were synthesized as reported (37). ␣-Galactosylsphingamides were synthesized as reported previously (28). Structural representations of all lipids are found in Fig. 1.

Surface plasmon resonance studies
For kinetic experiments, enzymatically biotinylated CD1d (C-terminal birA tag) was used for lipid loading (see Lipid loading, complex formation, and crystallization). Individual CD1dglycolipid complexes were immobilized on a CAP sensor chip (GE Healthcare) at levels between 100 and 400 response units using a Biacore T200. Increasing concentrations of TCR (15 nM to 1 M) in HBS-EP buffer (10 mM HEPES, 150 mM NaCl, 3 mM EDTA, and 0.05% surfactant Tween 20) were passed over the sensor chip at a flow rate of 30 l/min, without regenerating the sensor chip surface in between cycles. Association was observed for 4 min; dissociation was continued for up to 15 min. All binding curves were corrected for background and bulk refractive index contribution by subtracting the response of the reference flow cells from the active surface. Kinetic data were analyzed using the Biacore T200 Evaluation software 2.0 (GE Healthcare) using a 1:1 Langmuir interaction kinetic model to calculate the association rate (k a ) and the dissociation rate (k d ) and the equilibrium dissociation constant (K D ϭ k d /k a ) by nonlinear fitting. Kinetic experiments were performed at least twice with two different TCR preparations.

Antigen presentation assay
The cell-free antigen-presentation assay has been performed as reported (38). 96-well plates were coated with 1 g of CD1d and incubated overnight at room temperature with three different concentrations of glycolipids as triplicates. Buffer was removed and hybridoma cells (5 ϫ 10 4 ) in culture media were added to each well and co-incubated with CD1d-glicolipid complexes overnight at 37°C in a CO 2 incubator. The cellbased antigen-presentation assay also has been described (39). Briefly, antigen-presenting cells (1 ϫ 10 5 per well) were pulsed with indicated amounts of glycolipid and were incubated overnight. The cells were then combined with 5 ϫ 10 4 cells of the V␣14V␤8.2 NKT cell hybridoma DN3A4 -1.2 (1.2) for 24 h. The 1.2 type I NKT hybridoma cell line has been described previously (40). TCR stimulation and T cell activation was measured using a sandwich ELISA for IL-2 cytokines in the supernatant of hybridoma cultures.

Lipid loading, complex formation, and crystallization
Mouse CD1d was incubated with a 6-to 10-fold molar excess of glycolipids (dissolved in DMSO at 10 mg/ml) in 50 mM Hepes, pH 7.5, 150 mM NaCl and incubated overnight at room temperature under slight agitation. For structural analysis of CD1d-glycolipid complexes, the CD1d/lipid mixture was spun down (14,000 ϫ g at 4°C for 10 min), concentrated using 30 kDa molecular filtration devices, and purified further by size exclusion chromatography using a Superdex S-200 GL10/300 column. For structural analysis of CD1d-glycolipid-TCR complexes, the CD1d-lipid mixtures were incubated for 2 h at room temperature with purified TCR at 2-fold molar excess of CD1d, concentrated to ϳ250 l, and subjected to size exclusion chromatography to separate ternary complexes from unbound CD1d and TCR using a Superdex S-200 GL10/300. Fractions containing CD1d-glycolipid or CD1d-glycolipid-TCR complexes were pooled, concentrated to ϳ5 mg/ml in 10 mM HEPES, pH 7.5, 30 mM NaCl and subjected to crystallization Sphingamide recognition by NKT cells trials using the sitting drop vapor diffusion method. Typically, 1 l of protein solution was mixed with 1 l precipitant using selected conditions of the PEG/ION screen that previously yielded crystals of related CD1d-glycolipid or CD1d-glycolipid-TCR complexes. Plates were incubated at 22.3°C for several days (ternary complexes) or 1-2 weeks (CD1d-glycolipid) before harvesting individual crystals for diffraction analysis. Crystallization conditions for each individual complex are listed in Table S1.

Data collection and structure determination
Single crystals were cryo-protected by immersion in crystallization solution containing 20 -25% glycerol and flash cooled in liquid nitrogen. Diffraction data were collected remotely at beam line 7-1, 14-1, and 9-2 at the Stanford Synchrotron Radiation Light source (Menlo Park, CA). Diffraction data for the individual crystals were processed and scaled using HKL2000. All structures were determined by molecular replacement method using the protein coordinates from the mCD1d-iGb3 structure (PDB ID 2Q7Y) and in case of ternary complexes also the TCR coordinates (PDB ID 3QUZ) using PHASER (41) as part of the CCP4 suite (42). The models were built in COOT (43) and iteratively refined using REFMAC5 (44). Refinement progress was monitored by a continuous drop in R free values and improvement in electron density. Data collection and refinement statistics are presented in Table S1. The quality of the models was examined using Molprobity (45). Crystal structure illustrations were prepared using MacPyMOL (Schroedinger, New York City, NY).

Accession numbers
The coordinates and structure factors of the various complexes have been deposited in the Protein Data Bank, under the accession codes listed in Table S1.