The Molecular Basis for Recognition of CD1d/α-Galactosylceramide by a Human Non-Vα24 T Cell Receptor

Human Vα24− CD1d-restricted T cells use variation in their CDR1α loop to respond to lipid antigens presented by CD1d, altering their specificities from that of invariant natural killer T cells.


Introduction
Natural killer T (NKT) cells are a highly conserved lineage of T lymphocytes found in both human and mice that are involved in the modulation of the immune response in autoimmunity, infection, and tumor development [1]. Unlike conventional CD4 + and CD8 + ab T cells that recognize peptides presented by MHC molecules, NKT cells are reactive to a broad range of self and foreign lipids displayed by the MHC class I-like molecule CD1d [2,3]. This reactivity is initiated by the recognition of the CD1d-lipid complex via the NKT T cell receptor (NKT-TCR) followed by Th1 and/or Th2 biased cytokine secretion that can regulate the activity of other immune cells such as conventional ab T cells, B cells, and Natural Killer (NK) cells [4].
The most extensively studied NKT cells in humans and mice are invariant (iNKT) or type I NKT cells that express TCRs composed of a highly conserved a chain encoded by a Va24-Ja18 rearranged gene segment in humans and Va14-Ja18 in mice. This invariant a chain is covalently paired with a b chain in which the variable region is encoded in humans by the Vb11 gene and can be Vb8, Vb7, or Vb2 in mice [1]. NKT cells expressing these TCRs have a pre-activated phenotype that is due to the expression of the transcription factor pro-myelocytic leukemia zinc finger (PLZF) [5,6] and are also characterized by high reactivity towards the potent stimulatory lipid antigen a-galactosylceramide (aGal-Cer) [7]. In both humans and mice there are additional classes of T cells that respond to CD1d, one that expresses diverse TCRs but do not respond to aGalCer; these are generally called Type II or non-invariant NKT cells [8]. These NKT cells are typically reactive to lipid antigens such as sulfatide and use an entirely different molecular strategy for recognizing the CD1d/lipid complex [9,10]. A third group of T cells exist that do respond to CD1d presenting aGalCer and also express TCRs different from that of the iNKT-TCR. In mice these NKT cells express a TCR comprised of a Va10-Ja50/Vb8 pair [11]. These cells are called Va10 NKT cells and show a preference for a-glucosylceramide (aGlcCer) over aGalCer; indeed, Va10 NKT cells can produce a several magnitudes greater cytokine response relative to iNKT cells when stimulated by the related a-glucuronosyldiacylglycerl (a-GlcA-DAG) [11].
In humans this third group of CD1d reactive T cells express TCRs with many different Va domains joined with Ja18, paired with the Vb11 domain [12,13]. In contrast to both Type I and Type II NKT cells, these T cells do not typically express CD161, a Natural Killer cell marker found on NKT cells [13]. They have been called Va242 NKT cells or CD1d-restricted, Va242 T cells due to their use of alternative Va domains rearranged to Ja18, paired with the Vb11 domain in their TCRs. These cells are found in all individuals sampled [13] at appreciable frequency (,10 25 ) [14] and express either the CD8ab or CD4 co-receptors, can be cytotoxic, and can secrete IL-2, IFN-c, and IL-13 (and in some cases IL-4) [13]. In contrast to human iNKT cells, they express low to intermediate levels of PLZF and have a naïve phenotype [14]. Importantly, these NKT cells have shifted lipid specificities from that of iNKT cells with an inability to recognize and respond to aGlcCer [12]. The distinctive difference in reactivity between aGalCer and aGlcCer suggests that this population of NKT cells focuses on a different repertoire of lipid antigens than those of iNKT cells.
Despite the variability that exists in NKT cell populations, most of our current knowledge of NKT cell recognition of antigen derives from structural studies that have focused on self and foreign lipid antigen recognition by Type I iNKT TCRs [15]. iNKT-TCRs recognize, through their complementary determining regions (CDR) loops, a composite surface composed of the ahelices of CD1d and the solvent exposed head group of the CD1dpresented lipid antigens. The CDR3a loop plays a prominent, conserved role in CD1d-lipid recognition, predominantly via residues encoded by the Ja18 segment, which is found in all iNKT TCRs. There are also important contributions from the CDR1a and CDR2b loops, which explain the restricted use of specific Va and Vb domains (which encode the CDR1 and CDR2 loops) [16,17]. For each Vb chain used in mouse, the docking of iNKT-TCRs on the CD1d/lipid antigen surface is remarkably conserved [18,19], indeed variation of the lipid antigen is accommodated mainly through structural modifications of the lipid antigen as opposed to changes in the iNKT TCR footprint [20][21][22][23][24]. The number of human iNKT TCR complex structures are fewer yet reflect some flexibility in docking of the iNKT TCR depending on the lipid antigen [16,19,23,25], yet appear to be similarly anchored via conserved positioning of the CDR3a loop.
The crystal structure of a murine Va10 NKT TCR in complex with murine CD1d-aGlcCer [11] has shed light onto the molecular mechanisms that murine non-canonical NKT TCRs use to recognize CD1d. Despite significant sequence divergence in the a chain amino acid sequence (40% sequence identity), the Va10 NKT TCR assumes a very similar docking mode to that of the iNKT TCR on CD1d. However, unlike the iNKT TCR, all CDR loops of the Va10 NKT TCR contribute to CD1d/aGlcCer recognition, with seemingly important contacts being contributed by the CDR2b and CDR3b loops. Thus the two Va chains of these divergent murine NKT cell populations (iNKT and Va10) have convergently evolved a similar molecular strategy for recognizing CD1d. Recently, crystal structures of the Type II NKT TCR recognition of CD1d presenting sulfatide [9] and lysosulfatide [10] provided an interesting contrast to the conserved recognition of CD1d by the iNKT and murine Va10 TCRs. The Type II TCRs use all six CDR loops in CD1d/ligand engagement and dock on a separate site on CD1d, concentrating on residues surrounding the A9 pocket. Thus, NKT cells have a range of docking modes used in CD1d/ligand engagement.
Structural data on NKT cell recognition in humans remains limited, and information of how Va242 T cells recognize CD1d/ lipid is, to our knowledge, absent. To better understand how this functionally distinct human T cell population recognizes CD1d/ lipid, we have co-crystallized a Va242 TCR with CD1d/aGalCer and present here the structure of this complex resolved to 2.5 Å resolution. This structure provides an excellent model by which to understand how functionally distinct human T cells, via their TCR, can recognize CD1d with a shifted specificity from that found in the iNKT cell population.

Structure of a Va242 TCR in Complex with CD1d/ aGalCer
In order to understand the molecular basis of Va242 TCR recognition of CD1d, we expressed a soluble, heterodimeric version of the extracellular domains of the J24.N22 TCR [12], which uses the Va3.1 (TRAV17) gene segment rearranged with Ja18 complexed with Vb11, in insect cells. The purified TCR was co-crystalized with recombinant, soluble CD1d loaded with aGalCer; X-ray data were collected to 2.5 Å , and the structure was solved via molecular replacement. Data collection and refinement statistics are listed in Table 1. One TCR/CD1d/ aGalCer ternary complex was identified in the asymmetric unit. All components of this complex were well resolved in the electrondensity, enabling unambiguous assignment of TCR-CD1d/lipid antigen contacts. Table 2 presents a comparison between the amino acid sequences of the a and b CDR loops of the Va242 (Va3.1+) TCR studied here and an iNKT Va24+ TCR studied previously [25]. Va3.1 and Va24 share 46% amino acid identity overall, with only 33% (2/6) identity at the CDR1a and 15% (1/7) at the CDR2a loop. However, the shared usage between these TCRs of the Ja18 segment and the canonical DRGSTLGR motif that it encodes gives high sequence identity to the CDR3a loops of these TCRs with different residues encoded only at the Va-Ja junction, with ATY and VVS motifs in the Va242 and Va24+ TCRs,

Author Summary
Certain lineages of T cells can recognize lipids as stimulatory antigens when presented in the context of CD1 molecules. We know how most Natural Killer T (NKT) cells react with this unusual ligand because they use a single invariant T cell receptor (TCR) alpha chain to do the job. NKT cells place particular emphasis on their CDR3a and CDR2b loops in recognition of antigen-these complementarity determining regions (CDRs) are the hypervariable parts of the TCR that ''complement'' an antigen's shape. How do these other T cells recognize closely related yet distinct lipid antigens? Here we show that human CD1d-restricted T cells, typically called Va242 T cells due to their use of diverse Va domains in their TCRs, use similar molecular strategies to respond to lipid antigens presented by CD1d. To this end we present a 2.5 Å complex structure of a Va242 TCR complexed with CD1d presenting the protypical lipid, a-galactosylceramide (aGalCer). The TCR examined in this study notably shifts its binding slightly, placing more emphasis on the interaction with the CDR1a loop as revealed through alanine scanning mutagenesis. This shift explains the inability of these T cells to respond to lipids that vary at this site of contact (the 49OH), like the related a-linked glucosylceramide. These results provide a molecular basis for the finespecificity of different CD1d-restricted T cell lineages.
respectively. The Vb11 domain is also shared between these TCRs; therefore, the CDR1b and CDR2b sequences are identical. However, the rearranged CDR3b loops differ due to differences introduced during the rearrangement process.

Recognition of CD1d/aGalCer by the Va242 TCR
Overall, the Va242 TCR recognizes CD1d/aGalCer with the a and b chains oriented on CD1d in a parallel fashion unlike the typical diagonal mode of MHC-I peptide-TCR complexes and similar to that of iNKT-TCR and Va10 NKT-TCR in complex with CD1d/aGalCer ( Figure 1A and 1B) [11,16,19]. However, the binding angle of the Va242 TCR in relation to the CD1d/ aGalCer surface is more acute than the almost perpendicular orientation observed with the Va24+ iNKT TCR-CD1d/ aGalCer structure ( Figure 1A) [16,19]. The CDRa loops adopt a similar yet slightly shifted footprint for the a-chain, yet the bchain CDR loop positioning is counter-clockwise rotated compared with the Va24+ TCR complexed with aGalCer [16,19], which is even more extreme than rotations observed in structures of human NKT-TCRs complexed with CD1d presenting LPC or bGalCer ( Figure 1B) [23,25]. The TCR-CD1d-lipid contacts mostly fall in the F9 pocket area of the CD1d molecule ( Figure 1C), where there are slight differences in TCR contact surface between the Va242 and Va24+. The total buried surface area (BSA) between the Va242 TCR and the CD1d-aGalCer complex was 747 A 2 , which is slightly smaller than the previously reported interface area for the Va24+ TCR, ,910 A 2 . This difference is more pronounced in the b-chain loops with ,37% less contribution in the Va242 complex (205.7 A 2 versus 325.3 A 2 for the Va242 and Va24+, respectively).
aGalCer Positioning in the Complex with the Va242 TCR The conformation and positioning of aGalCer presented by CD1d is almost identical in both complexes with the Va24+ and the Va242 TCRs. The sphinosine base and acyl chain of aGalCer fall in the F9 and A9 pockets, respectively ( Figure 1D). The aGalCer headgroup also adopts a very similar conformation, with solvent exposed with the sugar oxygens displayed for recognition by the TCR. The conformation of the a helical side chains of CD1d were also highly conserved between the Va24+ and Va242 complex structures, with only a few exceptions that are noted later in the text.

Convergent Recognition Strategy of a Va242 TCR
In all three human iNKT TCR-CD1d/lipid complexes that have been resolved to date, the CDR1a loop makes important contacts with the lipid headgroup [16,19,23,25]. In recognition of aGalCer and bGalCer the O c of Ser30 and the mainchain carbonyl oxygen of Phe29 make hydrogen-bonds (some watermediated) with the 39OH of aGalCer and bGalCer, and in the case of LPC, the O c Ser27 and the mainchain carbonyl oxygen of Phe29 establish hydrogen bonds with the phosphate oxygens of the phosphorylcholine headgroup. Pro28 establishes van der Waals (VDW) contacts with the galactose headgroup; mutagenesis of this residue has a marked effect on recognition but is likely due to global structural changes in the conformation of the TCR as this mutation also disrupted binding of a conformational-specific antibody [17]. In our structure the Va242 CDR1a loop is slightly shifted from the Va24+ CDR1a loop ( Figure 1B); therefore, the equivalent structural positions to the Va24+ S 27 P 28 F 29 S 30 motif are T 26 S 27 I 28 N 29 in Va242. Despite the Table 1. Data collection and refinement statistics (molecular replacement).

Data Collection
Va242 TCR-CD1d-aGalCer chemical and structural differences of the CDR1a loops between these TCRs, specific side-chain-mediated hydrogen bonds are still formed in the Va242 CDR1a loop, both with the galactose headgroup of aGalCer and through VDW contacts with CD1d's Val72 ( Figure 2A and Table 3). The shifted position of Ser27 in this complex enables a hydrogen bond between its O c with the 69OH of aGalCer, whereas the N d2 of Asn29 hydrogen bonds with the 39OH and 49OH of aGalCer and Asn29 also forms VDW contacts with the galactose headgroup. Therefore, alternative residues in the CDR1a loop are effectively used in recognition of aGalCer with a focus on the 49OH of the galactose ring, with a novel contact with CD1d also noted.
We have also noted residues in the CDR2a loop that make water-mediated contacts with the aGalCer galactose headgroup: Ser50 and Asn51 both establish water-mediated hydrogen bonds with the 49OH of aGalCer ( Figure 2B). In the other human complexes, Phe51 of the Va24+ CDR2a loop makes VDW contacts with both bGalCer and LPC, however hydrogen bonds have not been noted for the CDR2a loop of Va24+ TCRs. In contrast to the sequence and contact differences at the CDR1a and CDR2a loops, the residues of the CDR3a loop in the Va242 TCRs adopt a similar conformation to that of the Va24+ iNKT TCRs ( Figure 2C). Yet despite the similarity in footprint, the Va242 CDR3a loop establishes fewer contacts with CD1d and aGalCer than does the CDR3a loop of the iNKT TCR (Table 3) (25 instead of 32, respectively, for CD1d and eight instead of 19, respectively, for aGalCer). There are fewer hydrogen bonds (two versus eight with CD1d and one versus four with aGalCer) and, in the case of aGalCer, fewer than half (seven versus 15) VDW contacts of those observed in the Va24+ complex. The residues of (slate, a chain; orange, b chain) in complex with human CD1d-b 2 m (ribbon, white) and aGalCer (sticks, yellow). Right panel, the Va242 TCR-CD1d-aGalCer complex (orange) is shown superimposed with a Va24+ NKT TCR-CD1d-aGalCer complex (PDB ID: 3HUJ; TCR in green) [19]. Complexes were aligned via the main-chain CA carbons of the CD1d heavy chain. (B) Positioning of the four different human NKT TCR loops on the CD1d-ligand surface: purple, Va242 TCR; orange, Va24+ NKT TCR-CD1d-bGalCer (PDB ID: 3SDX) [23]; berry, Va24+ NKT TCR-CD1d-aGalCer complex; and green, the iNKT TCR-CD1d-LPC complex (PDB ID: 3TZV) [25]. Shown is CD1d (white)-aGalCer (yellow) from the Va242 TCR-CD1-aGalCer complex. (C) Upper panel, footprint of the Va242 TCR on the surface of CD1d-aGalCer. Residues that are contacted by the TCR a chain, b chain, or both are colored in blue, orange, and green, respectively. Lower panel, footprint of the Va24+ NKT TCR on the surface of CD1d-aGalCer; colors of CD1d are as for the Va242 TCR. (D) Electron density of the aGalCer ligand in the Va242 TCR-CD1d-aGalCer complex. Electron density, shown as a blue mesh, corresponds to a composite omit map (2Fo-Fc) contoured at 1s around the aGalCer ligand (yellow). CD1d is shown in grey ribbons, and the a2 helix has been omitted to facilitate the visualization of the ligand. The TCR a and b chains in light blue and yellow-orange, respectively. doi:10.1371/journal.pbio.1001412.g001 Va242 TCR Recognition of CD1d/aGalCer PLOS Biology | www.plosbiology.org the Va24+ CDR3a were previously shown to be energetically critical for CD1d/aGalCer recognition [17], a finding recapitulated in our data (discussed further below) despite the lower contact number.

A Shifted Va242 TCR b Chain Maintains Conserved Contacts through the CDR2b
While the CDR3a loop serves to anchor human iNKT TCRs on the CD1d/lipid platforms with highly similar conformations [16,19,23,25], the remaining loops have demonstrated rotational flexibility in how they are positioned over the CD1d/lipid surface, in particular at the CDR2b, which establishes energetically critical contacts with CD1d [17]. A similar rotation is seen in the Va242 TCR docking on the CD1d/aGalCer platform in the complex structure presented here ( Figure 1B and Figure 3A). As in the Va24+ complexes, the involvement of the CDR2b loop in CD1d binding is predominantly mediated by Tyr48 and Tyr50. Despite an average shift of 4.6 Å between the Va242 and Va24+ CDR2b CA backbones, the rotationally flexible tyrosine side chains maintain highly similar contacts between the two complexes ( Figure 3A). Glu83 on CD1d takes a central role in contact with the CDR2b in both complexes, establishing a hydrogen-bonded network with both Tyr48 and Tyr50 hydroxyls. Met87 also contributes VDW contacts with Tyr50 in both complexes. However, in contrast to the Va24+ complex, where Glu56 of the CDR2b establishes a robust salt-bridge with Lys86 of CD1d (3.7 Å distance), in the Va242 complex Lys86 has shifted such that is it 4.6 Å from Glu56 ( Figure 3A). Thus, the critical contacts of the CDR2b loop are maintained in the Va242 complex despite large main chain shifts of the CDR2b backbone.
The highly variable CDR3b loop has been demonstrated to confer reactivity to specific lipids presented by CD1d by both human [26] and mouse [27] iNKT cells. In the Va242 complex, the CDR3b loop is well resolved in the electron density and establishes only one weak hydrogen bond and a VDW contact with Gln150 on CD1d's a2-helix via Ser97 ( Figure 3B). Thus, unlike the murine Va10 NKT TCRs, which have CDR3b sequence specificity and use this loop in CD1d binding, this Va242 TCR does not appear to rely heavily on its CDR3b loop for binding.

Conformational Flexibility of Va242 CDR3a Loops
The availability of a Va242 TCR also expressing a Va3.1 domain (named 5B) [28] in the unliganded state allows a direct comparison between the loop structures between the TCR examined here (bound to CD1d) and a Va242, Va3.1+, TCR in its unbound state. Due to the use of different Jb gene segments that results in global domain orientation shifts, the TCRs are not perfectly superimposable ( Figure 4A) and there are two amino acid differences in the CDR3a sequences of these TCRs due to junctional diversity ( Figure 4B). Alignment of the two Va3.1 domains shows the CDR1 and CDR2 loops are essentially identical structurally ( Figure 4B), yet examination of the CDR3a loops ( Figure 4B) shows significant structural differences. While the unliganded structure of J24.N22 is not known, modeling of the 5B TCR onto our complex structure suggests a large shift in loop conformation would need to occur in the CDR3a loop for it to dock onto CD1d/aGalCer in a similar fashion. Because of the similarities between these TCRs in all other loops save the CDR3b, it is very likely that the 5B TCR would dock in a similar fashion as seen here. Thus in contrast to the Va24+ NKT TCRs' recognition of CD1/aGalCer, where loop conformation was highly conserved in the liganded and unliganded state, we suggest that the CDR3a loop can be flexibile in Va3.1+, Va242 TCRs,

Residues Contributing to Va242 TCR Binding of CD1d/ aGalCer
To evaluate the kinetics involved in binding of our Va242 TCR with CD1d/aGalCer, we used surface plasmon resonance to measure the association (k on ) and dissociation rates (k off ) of this interaction and determine the dissociation constant (K D ) ( Figure 5A). We also used this to calculate K D by equilibrium analysis (Figure 5A, insets). We included an iNKT (Va24+) TCR in our kinetic measurements such that we could compare these values to a representative of the iNKT population. The affinity of the Va242 TCR used in this study for CD1d/aGalCer (2.1 mM kinetic, 2.5 mM equilibrium) was similar to the affinity we measured for the iNKT TCR (2.1 mM kinetic, 1.9 mM equilibrium) as well as affinities from previous measurements with Va242 TCRs (using Va3.1 and Va10.3 domains) [28]. Stronger affinities (0.5 mM) have been noted for other human iNKT TCRs [17].
We sought to further evaluate the residues contributing most to Va242 TCR binding to CD1d/aGalCer. We chose key TCR residues identified as interacting with CD1d/aGalCer in our complex and evaluated their contribution to binding via alaninescanning mutagenesis and SPR. We first evaluated the CDR1a loop residues Ser27 and Asn29, as these appeared to mediate the side-chain-specific contacts that differed most from the Va24+ TCRs. While mutation of Ser27 to Ala (S27A) did not drastically change Va242 TCR binding kinetics, mutating Asn29 to Ala (N29A) resulted in a significant disruption to binding with changes in both the association and dissociation rates and an increase in the K D by an order of magnitude ( Figure 5B). Thus the CDR1a loop provides a clear contribution to Va242 TCR binding to CD1d/aGalCer. Previous mutational analysis of the CDR1a loop of a Va24+ TCR [17] of Pro28 to Alanine disrupted binding, however this was assumed to be due to changes in the TCR architecture as conformational-specific antibodies failed to bind this mutant.
Mutation of the CDR2a side chains Ser50 and Asn51 had subtle effects on k on and k off ( Figure 5B) yet did not appear to have a substantial effect on the overall affinity of CD1d/aGalCer binding, similar to what we observed with mutation of Ser97 in the CDR3b loop sequence. Because of the similarities in CDR3a loop contacts between Va242 and Va24+ TCRs, we included a mutation of Arg95 of the CDR3a as a positive control; this side chain has been shown to be central to iNKT TCR binding to CD1d/aGalCer [17]. We also observed that mutation of this side chain to Ala (R95A) abrogated binding of the Va242 TCR and thus supports the importance of the CDR3a loop to Va242 TCR docking.

Discussion
Our complex structure of a Va242 TCR with CD1d/aGalCer provides a model by which to understand how this diverse population of CD1d-restricted human T cells recognize antigen. These cells differ from iNKT cells in their specificity, effector function, and the markers expressed on their cell surface; these factors combined argue that these cells provide another arm of Tcell-mediated lipid recognition in humans. Here we provide a structural and biophysical foundation upon which to understand the molecular basis of differential reactivity observed at the cellular level in this NKT cell population.
Despite the divergent amino acid sequences encoded by the Va3.1 domain for the CDR1a and CDR2a loops, the Va242 TCR adopts a similar footprint to that of Va24+ iNKT TCRs. This docking orientation is primarily dictated by the conserved docking of the CDR3a loop, containing the highly similar sequence encoded by the Ja18 segment of iNKT TCRs. The contacts mediated by the other loops, while not identical to those of iNKT TCRs, were very similar, suggesting that despite sequence differences in the Va loops they could establish contacts with similar regions of the CD1d/aGalCer surface. The aGalCer headgroup position was almost identical to that observed in the iNKT complex structures [16,19]. This docking mode, also shared with that of the murine Va10 NKT TCR [11], is strikingly different from that of the recently resolved type II NKT TCR structures [9,10], where the TCRs dock on an entirely different surface of CD1d (the A9 pocket) and use all six of the TCR's CDR loops in recognition (similar to what is observed in conventional ab TCR recognition of MHC/peptide). These structures demonstrate that CD1d-restricted T cells can use at least two divergent ways to recognize their antigens [29].
Our complex structure provides a useful model to compare other Va242 TCRs' structures, notably the structure of a highly related unliganded TCR called 5B [28]. If we assume the 5B TCR would dock similarly to the Va242 TCR examined in our study, a significant conformational change would have to occur in 5B's CDR3a loop. This conformational flexibility was a feature we also observed in human iNKT TCR binding to CD1d/LPC [25]. In contrast to what was observed with the iNKT TCR complex structure with CD1d/aGalCer [16,19], this suggests that not all CD1d-TCR interactions are ''lock and key'' and that changes to CDR3a loop conformation may contribute to differences in binding kinetics and thermodynamics. A similar phenomenon of loop movement was observed in the murine Va10 NKT TCR upon binding [11].
The CDR3a loop footprint on CD1d/aGalCer is conserved in all the iNKT-TCR/CD1d structures noted to date as it is here. However, the number of contacts in this complex structure were less than that observed in the iNKT-TCR CD1d/aGalCer complex structure, yet the binding affinities measured for the Va24+ and Va242 TCRs in this study did not differ substantially (,2 mM for both TCRs). The alanine-scanning mutagenesis revealed important contributions from the CDR1a loop (in particular, residue N29) in the Va242 TCR binding that were not noted in Va24+ TCR binding (mutation of the equivalent position, S30 in the Va24+ TCR, showed little effect [17]). This shift of importance toward the CDR1a likely compensates for fewer CDR3a loop contacts and would explain the altered reactivity patterns of Va242 TCRs for lipids that are recognized similarly by Va24+ TCRs (such as aGlcCer and aGalCer, discussed more below). We cannot rule out that contributions from other loops, such as the CDR2a and CDR3b, contribute as well; while individual mutagenesis of these residues had small effects upon TCR binding, in combination they may have a cumulative effect in binding CD1d/lipid, evident only when they are mutated in concert.
Extensive studies in the mouse iNKT cell system have revealed how lipid ligands are structurally modified during recognition by the iNKT TCR. Even though extensive structural variability exists in the glycolipid headgroups, each carbohydrate structure adopts a similar orientation when bound by the TCR [20][21][22][23][24]. Therefore, contributions of the CDR1a in recognition of alternative lipids, both aand b-linked glycolipids, could be an important factor in Va242 T cell reactivity towards different lipids. Directly relevant to this point is the clear distinction between Va242 T cells and Va24+ iNKT cells in their differential reactivity to the a-linked glycolipids aGlcCer and aGalCer. Va24+ iNKT cells respond well to both lipids, whereas Va242 T cells do not respond to aGlcCer. The only difference present between these two lipids is the orientation of the 49OH group on the sugar ring (glucose versus galactose). Our structural and biophysical data provide an explanation for this difference in reactivity. Asn29, a residue in the Va242 CDR1a, establishes both VDW and hydrogen bonds with Va242 TCR Recognition of CD1d/aGalCer the 39OH and 49OH. Mutation of this residue to alanine results in an order of magnitude decrease in binding of the Va242 TCR, presumably due to disruption of these contacts. Furthermore, the CDR2a loop residues Ser50 and Asn51 establish water-mediated hydrogen bonds with the 49OH that may help to stabilize the interaction despite lacking clear energetic contributions (as Va242 TCR Recognition of CD1d/aGalCer PLOS Biology | www.plosbiology.org assessed in our alanine-mutagenesis studies). We therefore propose that modification to the 49OH between the galactose (aGalCer) and glucose (aGlcCer) structure is the primary molecular factor mediating the differences in reactivity of the Va242 population of CD1d-restricted T cells. The alternative contacts with the carbohydrate headgroup in the iNKT TCR/CD1d/aGalCer structure may explain why iNKT cells can respond to both lipids; the main contacts with the 49OH are mediated by Ser30, which when mutated to alanine only had a minimal effect on binding [17]. The greater number of contacts and BSA of the Va24+ TCR CDR3a loop on CD1d/aGalCer may make these T cells relatively insensitive to variation in the glycolipid headgroup at other positions. The difference in 49OH recognition may translate to alternative reactivity to other glycolipid and non-glycolipid lipid structures both in development of these T cells in the thymus and their effector functions in the periphery. Despite their shared use of Ja18 and Vb11, the Va242 T cells are differentiated from iNKT cells in their development and activation state; presumably altered TCR recognition of a selecting antigen during thymic development plays a role in these differences. Our structure provides a model by which to understand the molecular basis of this altered reactivity.
Our results, which focus much of the differences in reactivity to aGlcCer on the CDR1a loop and its interaction with the 49OH, contrast with the murine Va10 NKT cell preferred reactivity to aGlcCer [11], where preference in binding appears due to many factors. The highly convergent recognition of aGlcCer by these TCRs distributes the binding contacts over much of the CDR loop surfaces [11]. While mutagenesis data for these residues are not available, it is clear there are differences in the nature of the contacts between the Va10 and iNKT TCRs with CD1d (VDW versus hydrogen bonds), that many new contacts are established with CD1d, and therefore modification to the sugar ring may have more of a distributed effect over the Va10 NKT interaction than what we observe in our Va242 TCR complex structure. Both structures, however, provide molecular models for the observed differences in lipid reactivity and demonstrate how divergent NKT TCR structures can convergently recognize similar CD1d/lipid antigen structures. The molecular basis of the differences in recognition we have described here are the first clues into understanding why Va242 cells are developmentally and functionally distinct from the iNKT population.

Human Wild-Type CD1d2 b 2 m Expression and Purification
The ectodomain region of human CD1d and human b 2 microglobulin (b 2 m) were co-expressed in insect cells and purified as described [25].

Va24 + and Va24 2 TCR Expression and Purification
The cDNAs corresponding to the a and b chains of the Va24 + NKT TCR clone J24L.17 and the a and b chains of Va242 TCR clone J24N.22 were separately cloned into different versions of the pAcGP67A vector each containing a 3C protease site followed by either acidic or basic zippers and a 6xHis tag. Both chains were coexpressed in Hi5 cells via baculovirus transduction. The heterodimeric TCRs was captured with Nickel NTA Agarose (Qiagen) and further purified by anion exchange and size-exclusion chromatography.

Generation of Va242 TCR Mutants
Mutants of the Va242 TCR (S27A, N29A, S50A, N51A, R95A for the alpha chain, and S97A for the beta chain) were generated through overlapping PCR with specific primers containing the desired mutation. Mutant heterodimeric TCR was expressed in insect cells as described above.

CD1d Loading with aGalCer
Purified human CD1d was used for loading with aGalCer at room temperature with a three molar excess of lipid for 16 h. The excess of lipid was then removed with a Superdex 200 (10/30) column (GE Healthcare).

Surface Plasmon Resonance Measurements
A human CD1d construct bearing a 3C protease site + 6X-Histidine tag at the C-terminus was expressed in Hi5 cells and purified as described [25]. All interaction experiments were performed in a BIAcore 3000 Instrument (GE Healthcare). Three hundred RUs of wild-type Va242 NKT TCR or a mutant version of it were captured in a flow channel of an Ni-NTA sensor chip (GE Healthcare) previously treated with NiCl 2 . Insect-cellderived recombinant IgFc was used to block unbound sensor chip surface to minimize nonspecific binding events. Increasing concentrations (0, 0.037, 0.111, 0.333, 1, 3, 9, and 27 mM) of CD1d-aGalCer were injected at a flow rate of 30 ml/min in 10 mM Hepes pH 7.4, 150 mM NaCl, and 0.005% Tween-20. Both kinetic and equilibrium parameters were calculated off of these curves using BIAevaluation software 3.2RC1 (GE Healthcare) and GraphPad Prism.

Ternary Complex Formation and Crystallization
Nickel agarose-purified Va242 TCR was digested with 3C protease for 16 h at 4uC to remove the zippers and His tags and purified by anion exchange chromatography in a MonoQ column (GE Healthcare). Endoglycosidase F3 (EndoF3) was used next at a 1:10 enzyme-to-protein ratio for 2 h at 37uC in order to minimize the sugar content present in the protein. The digested protein was purified by a new round of anion exchange followed by sizeexclusion chromatography. Both aGalCer-loaded CD1d and EndoF3-treated Va242 TCR protein samples were mixed in HBS at 1:1 molar ratio and concentrated in Nanosep Centrifugal Devices (Pall Life Sciences) to 10 mg/ml. Initial hits were found in 0.1 M sodium acetate, 20% PEG 4000, and were optimized to birefringent crystals that grew in 0.1 M sodium acetate pH 5.0, 17% PEG 4000, and 0.1 M ammonium acetate.

Crystallographic Data Collection, Structure Determination, and Refinement
Crystals were cryo-cooled in mother liquor supplemented with 20% glycerol prior to data collection. All data sets were collected on a MarMosaic 300 CCD at the LS-CAT Beamline 21-ID-G at the Advanced Photon Source (APS) at Argonne National Laboratory and processed with HKL2000 [30].
The structure of the ternary complex was solved by molecular replacement with the program Phaser [31] using the human CD1d-b2m (Protein Data Bank (PDB) accession number 1ZT4) and an iNKT Va24+ TCR (2EYS) as search models. Refinement with Phenix software suite [32] was initiated through rigid body and followed with XYZ coordinates and individual B-factor refinement. These first steps of refinement yielded clear unbiased and continuous density for aGalCer. Next, extensive cycles of manual building in Coot [33] and refinement were carried out and ligands such as aGalCer or covalently bound sugars were introduced guided by Fo2Fc positive electron density. Ligands structures and chemical parameters were defined with C.C.P.4.'s Sketcher [34] and included in subsequent refinement and manual building steps. Translation/libration/screw (TLS) partitions were calculated and incorporated at later refinement stages. All the refinement procedures were performed taking a random 5% of reflections and excluding them for statistical validation purposes (Rfree).

Structure Analysis
Intermolecular contacts and distances were calculated using the program Contacts from the CCP4 software package [34], interface surface areas were calculated using the PISA server (http://www. ebi.ac.uk/msd-srv/prot_int/pistart.html), and all structural figures were generated using the program Pymol (Schrödinger, LLC).

Accession Numbers
Coordinates and structure factors for the J24.N22 Va242 TCR/CD1d/aGalCer complex have been deposited in the Protein Data Bank under the accession code 4EN3.