Another View of T Cell Antigen Recognition: Cooperative Engagement of Glycolipid Antigens by Va14Ja18 Natural TCR 1

Va14Ja18 natural T (iNKT) cells rapidly elicit a robust effector response to different glycolipid Ags, with distinct functional outcomes. Biochemical parameters controlling iNKT cell function are partly defined. However, the impact of iNKT cell receptor β-chain repertoire and how α-galactosylceramide (α-GalCer) analogues induce distinct functional responses have remained elusive. Using altered glycolipid ligands, we discovered that the Vb repertoire of iNKT cells impacts recognition and Ag avidity, and that stimulation with suboptimal avidity Ag results in preferential expansion of high-affinity iNKT cells. iNKT cell proliferation and cytokine secretion, which correlate with iNKT cell receptor down-regulation, are induced within narrow biochemical thresholds. Multimers of CD1d1-αGalCer- and αGalCer analogue-loaded complexes demonstrate cooperative engagement of the Va14Ja18 iNKT cell receptor whose structure and/or organization appear distinct from conventional αβ TCR. Our findings demonstrate that iNKT cell functions are controlled by affinity thresholds for glycolipid Ags and reveal a novel property of their Ag receptor apparatus that may have an important role in iNKT cell activation.

F undamental to the initiation of a cellular immune response is cell-to-cell communication through receptor triggering and the dynamic formation of an immunological synapse. Central to this process is the interaction between the Ag and its cognate receptor, which relays the specificity of recognition. Although much has been learned regarding the interactions between peptide Ags and their cognate TCR, comparatively little is known about the recognition of CD1-restricted glycolipid Ags by specific T cells.
Va14Ja18 natural T (iNKT) 3 cells are a unique subset of CD1d1-restricted T lymphocytes whose invariant ␣-chain preferentially pairs with Vb8.2 ␤-chain and less commonly with Vb7. Remarkably, in vivo iNKT cell activation through the TCR results in rapid (i.e., within 60 -90 min) and robust IL-4 response and a spectrum of Th1 and Th2 cytokines (reviewed in Ref. 1). In striking contrast, conventional T lymphocytes require up to a day to produce significant amounts of cytokines in response to Ag.
The natural Ag recognized by iNKT cells remains unknown. A variety of CD1d-positive cells activate freshly isolated thymic iNKT cells and derived hybridomas without the addition of any exogenous Ag (2)(3)(4)(5)(6), which suggests the recognition of self-Ags. Moreover, presentation of self-Ags requires CD1d trafficking through the late endosomes/lysosomes (3,4,(7)(8)(9). The recognition of ␣-galactosylceramide (␣GalCer) and ␣-glucosylceramide by iNKT cells (10 -14) suggests a glycosphingolipid nature of the elusive Ags. Although ␣GalCer is a nonphysiological Ag, our recent studies indicate that it may be a very close mimic of at least one natural iNKT cell ligand (15). Consistent with this conclusion is the fact that ␣GalCer and/or its close analogue OCH, with a shortened long-chain sphingosine base and acyl chain, exhibit immunopharmacological effects in vivo. Thus, ␣GalCer acts like an adjuvant enhancing immunity to malaria and other infectious pathogens (16). Furthermore, ␣GalCer and/or OCH can prevent autoimmune diseases in mouse models of type I diabetes and multiple sclerosis (17)(18)(19)(20)(21). Interestingly, the ability of OCH to induce IL-4 alone and no IFN-␥ appears to underlie its pharmacological action (19). Thus, delineating the biochemical parameters of Va14Ja18 TCR/Ag interactions is of paramount pharmacological significance.
Interactions of soluble Ag receptors of conventional T cells with cognate Ags are of low affinity (0.1-50 M) and relatively fast dissociation half-life (t 1/2 ϭ 10 -50 s) (22)(23)(24)(25). Va14Ja18 TCR of an iNKT cell hybridoma has been demonstrated to interact with ␣GalCer-loaded CD1d1 with relatively high affinity (0.2 M) and very long half-life (t 1/2 ϭ 175 s) (26). The high-avidity interaction of Va14Ja18 TCR with CD1d1-␣GalCer dimer appears to be influenced by TCR ␤-chain repertoire (27). Recent studies have implicated both optimal dwell time (28) and affinity (29) of TCR-Ag interaction as critical determinants of T cell sensitivity and activation. Furthermore, interactions of conventional TCR with Ag are thought to be stabilized by CD4 and CD8 coreceptors (30 -32). Long dwell time of Va14Ja18 TCR/CD1d1-␣GalCer interaction (26) appears counterintuitive to the optimal dwell-time requirements for T cell activation. Because iNKT cells might not use coreceptors during Ag engagement, this interaction might require

Generation of multimers
Preparation of CD1d1-glycolipid (15) and H2K b -peptide tetramers (27,28) has been described. Dimers of CD1d1 (custom order; BD PharMingen, San Diego, CA) and H2K b (DimerX; BD PharMingen) are dimeric owing to their fusion to IgG1 H chains. To obviate the potential for artifacts induced by detection mediated via a fluorochrome-conjugated secondary Ab, the dimers were Alexa Fluor 647-or PE-conjugated via Fab specific for the Fc portion of IgG1 (anti-mouse IgG1 Alexa Fluor 647 and PE Zenon kits; Molecular Probes, Eugene, OR). Every batch of tetramer generated was tested for complete loading of ␣GalCer and its analogues by glycolipid titration loading and testing by reaction with the best characterized iNKT hybridoma N38-2C12.

Determination of relative avidity (K av )
Equilibrium (Ͼ2 h) binding experiments were performed using increasing tetramer concentrations in 100 l of PBS containing 2% FCS (Invitrogen, Carlsbad, CA) and 0.05% NaN 3 at 4°C, to prevent capping and internalization of the TCR. K av was calculated from specific mean fluorescence intensity (MFI; difference between total MFI at a defined tetramer concentration and background MFI derived from ligand-free tetramer binding to the same cells) using nonlinear regression analysis fitted to classical Michaelis-Menten kinetics (Prism 3.02; GraphPad Software, San Diego, CA). MFI (% maximum) shown in the relevant figures is based on V max calculated from nonlinear regression analysis of the data for adequate graphical representation. This permits easy and reliable comparison of data generated in different experiments. Nonlinear Michaelis-Menten regression analysis was preferred, because Scatchard transformation, which uses linear regression, amplifies any variation of the data from the linear curve. That notwithstanding, the results from the Michaelis-Menten kinetics were confirmed by using classical Scatchard transformations to derive the K av (42).

Determination of off-rates
T cells were labeled with 50 g/ml H2K b -peptide or 10 g/ml CD1dglycolipid tetramers, respectively, incubated at 4°C for 3 h, and washed extensively. Cells were also stained with 10 g/ml anti-TCR C␤-FITC to monitor TCR levels. Following initial tetramer binding, 10 6 cells were chased in 3 ml of buffer with rocking at 4 or 37°C for the indicated time periods and analyzed by flow cytometry.

Measurement of in vivo and in vitro cytokine response
Mice were injected i.v. with the indicated concentrations of glycolipids diluted in PBS from a 220 g/ml stock solution in vehicle (0.5% v/v polysorbate and 0.9% w/v NaCl). Controls were injected with corresponding dose of vehicle. After 90 min, IL-2, IL-4, IL-13, CSF-2, IFN-␥, and TNF-␣ in control and immune sera were measured by ELISA using Abs and methods that we have described previously (43).

TCR down-regulation and in vitro expansion
Bulk C57BL/6 splenocytes were incubated for the indicated amounts of time with increasing concentrations of glycolipid Ags. Following stimulation, iNKT cell receptor level was determined by flow-cytometric analysis following staining with CD1d1-␣GalCer tetramer, anti-TCR␤ Ab, within electronically gated B220 and CD8-negative lymphocytes. In other experiments, Ags were first equilibrium-loaded overnight onto B6.129-Tcra 0/0 splenocytes, and then mixed with C57BL/6 splenocytes magnetically depleted of MHC class II-positive cells. To directly evaluate iNKT cell division during culture, splenocytes were labeled with 2 M CFSE (Molecular Probes) in PBS for 8 min at room temperature, followed by quenching with cold FCS and washing with ice-cold RPMI 1640 supplemented with 10% FCS before culture. Evaluation of iNKT cell proliferation was performed by multiplying the percentage of iNKT cells determined by flow cytometry with the total cell number.

Cell-free Ag dissociation assay
Soluble mouse CD1d1 was Ni-affinity purified, as described (44), and bound to ELISA plates at a concentration of 10 g/ml. Following binding at 4 o C for 18 h and blocking of unbound sites with 2% FCS, plate-bound soluble mouse CD1d1 was loaded with 0.1 M lipids for 12 h at 37 o C. After removing excess lipids, the Ag was allowed to dissociate for the indicated times at 37°C. The wells were washed again, and ϳ5 ϫ 10 4 hybridoma cells were added to each well. Controls included wells bound with 2 g/ml anti-CD3⑀ (positive) or with 5 g/ml BSA (negative) loaded with 1 g/ml ␣GalCer. IL-2 secreted upon activation was monitored by ELISA. Data are presented as the percentage of maximum activation.

Determination of Hill coefficient
Hill coefficient was determined from epitope titration experiments. Briefly, CD1d1 and H2K b -H28 tetramers or dimers were loaded with increasing amounts of glycolipid ligand or peptide epitopes, respectively. Note that H2K b tetramers were initially folded with H28-derived epitope, a peptide with low affinity for H2K b , allowing rapid and efficient ligand exchange (Y. Yoshimura and S. Joyce, unpublished data). Glycolipid and peptide loading occurred at 37 o C and room temperature, respectively, for 16 -18 h.
Hill curve was derived from data transformation; fractional saturation (Ys) of the receptor was determined as the ratio of specific MFI to maximum MFI (V max ) at a defined ligand concentration and plotted against the concentration of added ligand (glycolipid or peptide). Linear graph of logarithmic roots of the values for the x-and y-axes were used to determine the slope of the Hill curve revealing the Hill coefficient (34).

Results
The CD1d1-␣GalCer/Va14Ja18 TCR interaction has high relative avidity There are a number of different methods available to assess the kinetics and extent of ligand-receptor interactions. Biophysical methods using purified recombinant molecules have been extremely useful in the study of a variety of immunological receptors (45)(46)(47). That notwithstanding, methods that examine molecules on living cells are particularly powerful (24,26,28,48,49). To gain insight into the parameters that govern the binding interactions of the CD1d1 ligand to the specialized TCR of iNKT cells required for their activation, we first determined the K av (measured affinity of tetrameric Ags for the cognate TCR) between tetrameric CD1d1-glycolipid Ag and its receptor on live cells. For comparison, the K av of peptide Ags for the TCR expressed by recently activated CD8 ϩ T lymphocyte (CTL) clones specific for two H2K b -restricted mH (H60 and H28 (27)) Ags and a viral (SV-40 T Ag-derived epitope IV (39)) Ag was measured. The comparison with class I-restricted Ags was used, because iNKT cells reflect memory/activated T lymphocyte phenotype similar to CTL clones.
CD1d1-␣GalCer tetramer binds Va14Ja18 ϩ but not Va14-negative NKT hybridomas (Fig. 1A). Similarly, H2K b -peptide tetramers specifically bind their cognate, but not irrelevant, CTL clones (data not shown and Refs. 38 and 39). Nonspecific binding was Ͻ5% in all cases ( Fig. 1A and data not shown). From the binding isotherms, the K av of Ag-TCR interaction was calculated (see Materials and Methods).
Va14Ja18 TCR binds CD1d1-␣GalCer with a K av ranging from 7 to 17 nM ( Fig. 1B and Table I). TCR of conventional T cells bind H2K b -peptide tetramers with a wide range of K av , ranging from ϳ20 to 220 nM ( Fig. 1C and Table I). Note that, in this study, the saturation binding isotherms at equilibrium were derived at 4°C. Whereas the K av values obtained at 4°C may not be the same as those at 37°C, the relationship of the K av between different Ag-TCR interactions remains unaltered (50). Consistent with that reported (50), we also found that the K av determined for two iNKT hybridomas at 4 o C (N37-1H5a, 10.5 nM; N38-2C12, 18.8 nM) maintained their avidity relationship at 37 o C (N37-1H5a, 39.3 nM; N38-2C12, 65.2 nM).
To obtain a more physiological estimate of the K av between CD1d1-␣GalCer and Va14Ja18 TCR, the above binding analysis was extended to NKT cell-enriched thymocytes and splenocytes. The K av of CD1d1-␣GalCer for Va14Ja18 TCR of live NKT cells ( Fig. 1D and Table I) is similar to that observed with iNKT hybridomas (B and Table I). Taken together, the K av of the CD1d1-␣GalCer-Va14Ja18 TCR engagement is similar to or higher than that of immunodominant peptide Ag-TCR interaction.

Glycolipid Ag-Va14Ja18 TCR interaction is long-lived
Functional T cell responses following Ag recognition have been correlated with the dwell time, measured as the t 1/2 of ligand engagement by its receptor (28,49,51,52). We found that the t 1/2 of glycolipid Ag/Va14Ja18 TCR is long, lasting between 10 and 40 min (Fig. 1E, top two panels; Table I), which is longer than the t 1/2 observed for peptide Ag/TCR interactions investigated (Fig. 1E, bottom panel; Table I). Hence, the off-rate of CD1d1-␣GalCer/ TCR interaction on the surface of intact NKT cells appears quantitatively distinct from that of conventional T lymphocytes.
Our data are consistent with previously published reports (28,53) for studies of peptide-Ag-specific T cells. Due to the manner in which our experiments were performed and analyzed, the data may appear inconsistent with recent dwell-time measurements between CD1d1-␣GalCer and iNKT cell receptors (26). Specifically, we did not use anti-CD1d1 or anti-MHC Abs in our experiments, and CD1d1-lipid or MHC-peptide levels detected postchase were not normalized to prechase reacted anti-TCR␤ levels. Abs to MHC molecules added during the chase period prevent the dissociating monomers from reassociating with their receptor (49). In our experiments, t 1/2 of H60 tetramer for cognate TCR determined in the absence (9 Ϯ 0.01 min; see Table I) or presence of an H2K breactive Ab, EH144 (2-10 min; n ϭ 5), was similar. Likewise, t 1/2 of CD1d1-␣GalCer tetramers for cognate iNKT cell receptor determined in the absence (9.8 Ϯ 0.9 min; n ϭ 5) or presence of 100 g/ml CD1d1-reactive Ab 1B1 (17.6 Ϯ 7.4 min; n ϭ 2) was similar. It should also be noted that, unlike MHC class II, which is not expressed by mouse T cells, MHC class I and CD1d1 are expressed by T lymphocytes. The use of anti-class I or anti-CD1d1 has the potential to cross-link the dissociating tetramer to the T cells, thereby skewing the data toward increased dwell time. Hence, the comparative off-rate measurements were performed in the absence of Abs.
We also found that the TCR levels on iNKT cells and CTL at time zero and at 120 min of chase were similar when they were stained with anti-TCR␤ Ab postchase (data not shown). In contrast, prestaining with anti-TCR␤ resulted in a significant loss of TCR␤ staining during chase, most likely due to the t 1/2 of anti-TCR␤ Ab and cell surface TCR interaction. Thus, we chose not to normalize the remaining CD1d1-␣GalCer tetramer bound postchase to prechase reacted anti-TCR␤ staining, as described in published reports (28,53). Nevertheless, our results are consistent with the conclusion that iNKT cell receptor interaction with CD1d1presenting glycolipid Ag exhibits longer dwell time than that of CTL receptor interaction with peptidic Ags (26).
TCR ␤-chain repertoire of iNKT cells impacts Ag specificity and the K av of their interaction Altered glycolipid ligands derived from ␣GalCer elicit distinct functional responses from iNKT cells in vivo and in vitro (19). Recently, the TCR ␤-chain repertoire of iNKT cells was implicated in high-affinity dimeric CD1d1-␣GalCer binding; the Vb8.2 ϩ iNKT cells have higher affinity for Ag than those that express Vb7 (27). Differences in TCR ␤-chain repertoire and/or the affinity for altered glycolipid ligands could explain the differential Ag specificity and functional outcomes. Tetramers of CD1d1-␣GalCer and its analogues OCH, 3,4D, and NH were generated concurrently under saturating conditions. Tetramers of Table I. Kinetic parameters of Ag-TCR interactions CD1d1-␣GalCer, -OCH, and -3,4D have exquisite specificity for iNKT cells ( Fig. 2A). However, CD1d1-3,4D (an analogue lacking the two hydroxyl groups at C atoms 3 and 4 of the long-chain base) and especially CD1d1-NH (C atom 2Ј amine-modified ␣GalCer) bind poorly or not at all, respectively, to iNKT hybridomas ( Fig. 2A).
Considering that the TCR ␤-chain repertoire of cells recognizing OCH was Vb8.1,8.2 skewed, we hypothesized that TCR ␤-chain of the iNKT cell receptor impacts K av for Ag. We found that Vb7 ϩ iNKT cells have 50% lower K av for both CD1d1-␣Gal-Cer and -OCH compared with Vb8.1,8.2 ϩ iNKT cells (Fig. 2E and Table II). Note that K av determination was performed with Vb7 ϩ cells that detectably bound CD1d1-OCH tetramer, which represented only ϳ50% of total CD1d1-␣GalCer tetramer-reactive Vb7 ϩ iNKT cells. Therefore, the results potentially represent a higher K av than that of the entire Vb7 ϩ iNKT population. Because the dwell time of TCR and Ag interaction correlates with the capacity for T cell activation, the t 1/2 of CD1d1-␣GalCer, and CD1d1-OCH from iNKT cell receptor was determined as described above. The results indicate that both glycolipid Ags have similar dwell times for their cognate receptors (Fig. 2F and Table  II). Taken together, the data suggest that the TCR ␤-chain repertoire and the K av of Ag-receptor interaction, but not the dwell time, might govern distinct functional outcomes from iNKT cells.

iNKT cells recognize OCH and ␣GalCer in vivo with similar sensitivity
A number of in vitro studies have indicated that iNKT cells recognize CD1d1-␣GalCer with nanomolar sensitivity (10 -12, 14, 35). Ags with different binding affinity for their TCR activate T cells with distinct activation thresholds (54 -58). To determine the sensitivity of effector responses by iNKT cells in vivo, C57BL/6 mice were injected i.v. with ␣GalCer and OCH, and serum cytokine response was measured after 90 min. CD1d1-restricted NKT cell (B6.129-CD1d1 0/0 )-and iNKT cell (B6-Ja18 0/0 )-deficient mice do not respond to these glycolipids (Fig. 3A), nor do C57BL/6 mice injected with the vehicle used to dissolve the glycolipid Ags (data not shown).
Mice administered 0.5 g of OCH elicited substantial amounts of IL-2 and IL-4; TNF-␣, IL-13, and CSF-2 (GM-CSF) were also detectable within 90 min (Fig. 3A). Administration of 1.0 g of ␣GalCer or OCH elicited a robust cytokine response including TNF-␣, IL-13, and CSF-2 (Fig. 3A). Note that the observed IFN-␥ response is at the very low end of maximum at this early time point. Furthermore, the previously reported differential IFN-␥ response to OCH and ␣GalCer are strikingly apparent only at or after 6 h (19), because that is the time point at which IFN-␥ peaks (59). Thus, at early time, ␣GalCer and OCH are recognized with similar sensitivity in vivo.

Kinetics of CD1d1 loading with ␣GalCer and OCH explain similar early iNKT cell response in vivo
Activation of T cells is an effect of Ag-TCR engagement and consequent intracellular signaling. T cell activation correlates with the extent of receptor down-regulation due to signal-dependent altered intracellular TCR trafficking (60 -62). Surprisingly, iNKT cells respond to ␣GalCer and OCH with similar early sensitivity (Fig.  3A), despite different equilibrium binding properties of TCR and specific Ag (Fig. 2, D-F, and Table II). To determine the cellular basis of ␣GalCer and OCH sensitivity, the kinetics and extent of TCR down-regulation following addition of increasing concentrations of ␣GalCer and OCH to splenocytes in vitro were evaluated. Both ␣GalCer and OCH down-regulated similar levels of surface TCR within 4 -12 h of Ag stimulation (Fig. 3B, top three panels). However, ␣GalCer was ϳ10-fold more potent at inducing surface TCR down-regulation after 24 h of stimulation. Thus, the kinetics of TCR down-regulation reflected the early induced iNKT cell response in vivo.
Two plausible mechanisms can explain the difference observed in early and late iNKT cell responses to ␣GalCer and OCH. ␣Gal-Cer, because of its higher K av for Va14Ja18 TCR compared with OCH, is a more potent iNKT cell ligand resulting in more sustained TCR down-regulation and activation. Alternatively, OCH, because of its shortened sphingosine and acyl chains, binds CD1d1 faster than ␣GalCer, and hence compensates for its low K av and elicits an early iNKT cell response. To distinguish between the two possibilities, B6.129-Tcra 0/0 splenocytes, which lack T and iNKT cells, were incubated with increasing quantities of ␣GalCer and OCH for 24 h. They were then used to stimulate C57BL/6 splenocytes depleted of MHC class II-positive cells, after which iNKT cell receptor down-regulation was evaluated. OCH was 10-to 20fold less efficient in TCR down-regulation compared with ␣GalCer at all time points tested (Fig. 3C). This result is consistent with the hypothesis that ␣GalCer and OCH have different kinetics of CD1d1 loading, and that the similar early iNKT cell response to the two Ags in vivo reflects rapid on-rate of OCH compared with ␣GalCer.
To determine the concentration threshold required for the elicitation of distinct cytokines from iNKT cells by ␣GalCer and its analogue OCH, C57BL/6 splenocytes were stimulated with increasing concentrations of these glycolipids. The results revealed that IL-2 and IFN-␥ response after 48 h (Fig. 3D) required at least 50% iNKT cell receptor down-regulation measured at 24 h (B, bottom panel) and medium Ag concentration threshold of ␣GalCer and OCH (D). In contrast, secretion of CSF-2 and IL-4 was more sensitive to low concentrations of glycolipid Ags and, hence, responded to low levels of TCR down-modulation (Fig. 3, B and D). In support of our previous report (19), OCH preferentially induced an IL-4 response, whereas 50-fold higher concentration of OCH was required to produce an IFN-␥ response similar to that induced by ␣GalCer (Fig. 3D). Thus, the secretion of cytokines by iNKT cells follows a hierarchical Ag response pattern, with higher avidity and higher concentrations required for secretion of IFN-␥ and IL-2 compared with both low avidity and low concentration for CSF-2 and IL-4.
To fully understand the properties of CD1d1-OCH interaction, we used a cell-free Ag presentation assay to determine its dissociation kinetics. Plate-bound soluble CD1d1 was loaded with equimolar quantities of ␣GalCer or OCH. After removing unbound lipid, the complexes were allowed to dissociate for varying time periods at 37°C. The t 1/2 of Ag-CD1d1 complex was monitored by its ability to activate iNKT cell hybridomas. OCH interaction with CD1d1 was more labile, because it dissociated faster than ␣GalCer from CD1 (Fig. 3E). Thus, similar early sensitivity of iNKT cells to ␣GalCer and OCH in vivo reflects the differences in the kinetics of their interaction with CD1d1 and also the differences in their equilibrium parameters of TCR engagement.

Activation of iNKT cells by 3,4D in vitro causes selective expansion of high-avidity clones
The altered lipid ligand, 3,4D, engages the iNKT cell receptor, albeit with low K av compared with ␣GalCer and OCH ( Fig. 2B and Table II), and elicits a weak cytokine response in vivo (19). To elucidate the biochemical basis of this weak response, the proliferative capacity of iNKT cells to Ag engagement was determined in vitro by CFSE dye dilution assay. After stimulation of splenocytes with Ags for 96 h, iNKT cells were costained with CD1d1-␣GalCer tetramer and TCR␤-specific Ab. At high concentration (575 nM), ␣GalCer, OCH, and 3,4D induced extensive iNKT cell proliferation (Fig. 4A). In contrast, at a lower concentration (2.9 nM) of these same Ags, ␣GalCer induced a strong proliferative response; OCH induced a partial proliferative response, whereas 3,4D and NH elicited a very weak or no response, respectively (Fig. 4A). Furthermore, quantitation of the proliferative response revealed that ␣GalCer induced maximum proliferation at 5.75 nM, and OCH at 57.5 nM, whereas maximum expansion was not reached even with 575 nM of 3,4D (Fig. 4B). We also noted that stimulation with supraoptimal Ag concentrations does not result in increased proliferation, but actually reduces total iNKT cell expansion (Fig. 4B).
Together, the data reveal that, despite differences in K av and the TCR ␤-chain repertoire, the altered lipid ligands induce proliferative response (Figs. 2, B-F, and 4, A and B). Therefore, the ␤-chain repertoire and the K av of Ag-expanded iNKT cells were determined. TCR ␤-chain repertoire of iNKT cells following ␣GalCer, OCH, and 3,4D stimulation remains largely unaltered at Ag concentrations inducing a maximum proliferative response, although a slight decrease in the percentage of Vb8-negative iNKT cells was noted (ϳ35% of expanded iNKT (Fig. 4C) compared with ϳ45% for naive iNKT cells (Fig. 2B)). Additionally, very little if any difference was observed in the Vb repertoire of iNKT cells expanded with different suboptimal doses of ␣GalCer and OCH (data not shown). Interestingly, iNKT cell activation by 3,4D, but not ␣GalCer or OCH, resulted in the expansion of iNKT cells responding to Ag with higher K av for ␣GalCer and OCH ( Fig.  4D and Table II). Thus, high-avidity iNKT cells preferentially expand to suboptimal TCR engagement.

Cooperative glycolipid Ag recognition by iNKT cells
Self-Ag recognition must be finely tuned to prevent iNKT cell activation during physiological conditions, but respond rapidly to disturbances in cellular physiology. In other words, iNKT cells need to be very sensitive to modest changes in Ag concentration. Kinetics of CD1d1 loading with ␣GalCer and OCH explain similar early iNKT cell response in vivo to the two glycolipids. A, C57BL/6 mice or control B6.129-CD1d1 0/0 and B6-Ja18 0/0 mice were injected i.v. with the indicated concentrations of ␣GalCer, OCH, or vehicle. After 90 min, serum cytokines were monitored. Background cytokine level (Ͻ3%) elicited by vehicletreated mice was subtracted from the Ag-treated response. The data represent cytokine responses (ϮSE) elicited by four individual mice in two identical experiments. B, Va14Ja18 TCR down-regulation was monitored at the indicated time points following addition of ␣GalCer or OCH to C57BL/6 splenocytes. iNKT cell receptor level was determined by flow-cytometric analysis following reaction with CD1d1-␣GalCer tetramer and anti-TCR␤ Ab, within electronically gated B220 and CD8-negative lymphocytes. C, Va14Ja18 TCR down-regulation was monitored following reaction of C57BL/6 splenocytes magnetically depleted of MHC class IIpositive cells with B6.129-Tcra 0/0 splenocytes equilibrium loaded with Ag overnight. D, Cytokines elicited by C57BL/6 splenocytes were monitored 48 h following addition of indicated quantities of ␣GalCer or OCH in vitro by sandwich ELISA. E, The dissociation of ␣GalCer and OCH from plate-bound soluble CD1d1 was monitored after removing excess glycolipids, and chasing the Ag for 4, 12, 24, and 36 h at 37°C, using an iNKT cell hybridoma, N38-2C12, as a probe. Activation-induced IL-2 was determined and plotted as percentage of maximum, a value obtained at start of chase.
In biological systems, this kind of fine-tuning is often achieved by using cooperative ligand-receptor interactions (33,63). To determine whether cooperativity participates in sensitive glycolipid Ag recognition, this mode of interaction was determined by calculating the Hill coefficient (see Materials and Methods). The Hill coefficient of the interaction between the tetrameric Ag and the iNKT cell receptor was Ͼ2 (Fig. 5A and Table I). In stark contrast, all MHC class I-restricted TCR had a calculated Hill coefficient of ϳ1 ( Fig. 5B and Table I), indicating a lack of cooperativity. Peptide binding to each H2K b monomer of the tetrameric molecule is an independent event. Saturation binding of the tetramer to the TCR with increasing concentration of added peptide indicates occupancy of all four sites (Fig. 5B). Furthermore, an analysis of the stoichiometry of class I H chain, ␤ 2 -microglobulin, and peptide following ligand exchange by Edman sequence determination (64) revealed a 1:1:1 ratio of the three components (data not shown). Thus, a Hill coefficient of 1 is not due to incomplete loading of the class I tetramer.
OCH is a structurally different Ag, particularly in the hydrophobic component thought to interact with CD1d1. Also, OCH interaction with CD1d1 has distinct kinetic parameters compared with ␣GalCer (Figs. 2F and 3E). Thus, to exclude the possibility that the biochemical or structural properties of ␣GalCer loading onto CD1d1 account for the observed cooperative response, Hill coefficient was measured for the binding of CD1d1-OCH to the Va14Ja18 TCR. As expected, we found that the Hill coefficient for CD1d1-OCH and CD1d1-␣GalCer for the same Va14Ja18 TCR are very similar (Fig. 5C and Table II). Thus, Hill coefficient measurement does not reflect the loading properties of glycolipid Ags, but rather, it is the property of the Ag receptors with which it interacts.
To independently demonstrate cooperative Ag engagement by iNKT cells with multimeric Ags other than soluble, biotinylated monomers of CD1d1 and H2K b prepared in-house, we determined the Hill coefficients with commercially obtained dimeric IgG1-CD1d1 and IgG1-H2K b fusion molecules loaded with ␣GalCer and H60 peptide, respectively, for iNKT cells and H60-specific SPH60 CTL clone. iNKT cells demonstrated cooperative engagement of both dimeric and tetrameric Ag by the Va14Ja18 TCR ( Fig. 5D and Table I). As expected, neither dimeric nor tetrameric H2K b cooperatively engaged their cognate TCR ( Fig. 5D and Table I). Thus, we conclude that, in contrast to conventional T lymphocytes, glycolipid Ag recognition by iNKT cells involves cooperativity.
iNKT cell receptor appears to have distinct structure and/or organization A plausible model for cooperative tetrameric Ag engagement by Va14Ja18 TCR is receptor partitioning and oligomerization within lipid rafts (50). To test this model, Hill coefficients for Ag-receptor interactions were determined for two representative iNKT hybridomas (N38-2C12 and N37-1H5a), NKT cell-enriched thymocytes, and two CTL clones (SPH60 and BH60), following disruption of their lipid rafts. Lipid rafts were disrupted by cholesterol depletion with methyl-␤-cyclodextrin (65) or alternatively by filipin-mediated intercalation of this membrane microdomain (66). Disruption of lipid rafts did not alter the Hill coefficient for any of the interactions tested (data not shown), suggesting that these membrane microdomains are not critical for cooperative Ag engagement by iNKT cell receptor.
To further examine the structural properties of iNKT cell Ag receptor, we used fluorescence resonance energy transfer (FRET) D, C57BL/6 splenocytes were stimulated with the indicated glycolipid. The resulting iNKT cell population was reacted with the indicated concentration of CD1d1-␣GalCer or CD1d1-OCH tetramers. From the binding isotherms, K av was determined as described in Fig. 1. Binding reactions were performed at 4 o C in the presence of sodium azide to prevent capping and internalization. measurements between CD1d1 and H2K b multimers and Abs specific for components of the TCR complex. In the course of our studies, we observed that costaining of iNKT cells ex vivo by allophycocyanin-or PE-conjugated CD1d1 tetramers and PE-or allophycocyanin-TCR␤ (clone H57-597) or anti-CD3⑀ (clone 145-2C11) Abs resulted in large and repeatable increase in FL3 channel fluorescence in a properly compensated flow-cytometric experiment (Fig. 6, A (iNKT cell hybridoma N38-2C12) and B (thymic iNKT cells)). Such large FRET shift was not observed with highintensity staining Abs specific for cell surface molecules not within the TCR complex (e.g., anti-CD44, clone IM7; Fig. 6). As PE and allophycocyanin have overlapping fluorescence emission and absorption spectra, respectively, it was likely that this result was a consequence of nonradiative FRET. This hypothesis was tested by running samples on the flow cytometer with the red diode laser (emission, 635 nm) and its FL4 filter switched off. Indeed, we still observed FL3 fluorescence only when costaining with CD1d1 multimers and TCR complex-specific Abs. The large FRET observed upon tetramer-anti-TCR␤/anti-CD3⑀ binding to iNKT cells presented an opportunity to test the hypothesis that the structural orientation and/or organization of Va14Ja18 TCR are distinct from ␣␤ TCR of conventional CTL. When H60-specific CTL were costained in a manner identical with that of iNKT cells, and analysis was restricted to equivalent MFI of anti-TCR␤ or anti-CD3⑀ and H2K b multimer, very little FRET was detected (Fig. 6A). Similarly, we observed FRET using PE-conjugated CD1d1 tetramers and allophycocyanin-conjugated anti-TCR␤ or anti-CD3⑀ Abs (data not shown). FRET between CD1d1-␣GalCer tetramers and TCR␤ on iNKT cells directly correlated with the staining intensity, even at relatively low concentrations (ϳ2.5 nM; Fig. 6C). In contrast, no FRET between H2K b tetramers and TCR␤ or CD3⑀ was observed, even with saturating concentrations of H2K b tetramers (ϳ250 nM; Fig. 6C). FRET is exquisitely sensitive to small changes in donor and acceptor fluorochrome distances (FRET, ϳr 6 ). Thus, these results strongly suggest that iNKT cell receptor has a distinct structure and/or organization, resulting in shorter distance between donor and acceptor fluorochromes used.

Discussion
In summary, our findings demonstrate that iNKT cell receptors recognize glycolipid Ags with avidities similar to, if not higher than, those of immunodominant, high-affinity ␣␤ TCR of conventional T cells. In contrast to CTL, which recognize Ag over a large avidity range (20 -220 nM), iNKT cells efficiently recognize Ag within a narrow window of avidity (10 -40 nM). Interestingly, although the TCR-Ag dwell time for ␣GalCer and OCH are very similar, TCR down-regulation as well as the proliferative and cytokine response of iNKT cells to these Ags directly correlated with avidity for Ag. Strikingly, both ␣GalCerand OCH-bound CD1d1 tetramers and dimers display cooperative engagement of the iNKT cell receptor, a property that CTL clones tested in this study lack. Additional data revealed FRET between specific combinations of fluorochromes conjugated to CD1d1 tetramers or dimers (data not shown) and TCR ␤-chain or CD3⑀-specific Abs. These findings suggest that the iNKT cell receptor structure and/or organization may be distinct from conventional ␣␤ TCR.
Conventional T cells recognize peptide Ags with a wide range of avidities and dwell times (23-25, 28, 29). In contrast, strong recognition of ␣GalCer and OCH, poor recognition of 3,4D, and no recognition of NH (19), by the Va14Ja18 TCR with distinct K av points toward a narrow kinetic window for iNKT cell activation. We demonstrate that both optimal Ag concentration and relative avidity are essential to elicit a strong proliferative response by iNKT cells. Interestingly, as observed with conventional T cell effector functions (56), iNKT cells exhibit hierarchical functional consequences to Ag quality and concentration. In support of our previous study (19), we also find a dissociation from a clear avidity-concentration dependence in IL-4 secretion following OCH compared with ␣GalCer stimulation. Both low Ag concentration and low K av are sufficient for selective IL-4 secretion and iNKT cell proliferation. In contrast, higher K av and Ag concentration are required for IFN-␥ response. Consistent with this conclusion is the finding that dendritic cells presenting a high concentration of the high K av Ag ␣GalCer induce sustained IFN-␥ response from iNKT cells (67). In this regard, iNKT cell response closely follows the principle of Ag concentration threshold set for IFN-␥ and IL-4 responses elicited by conventional T cells (68).
Due to their potent immunoregulatory properties, therapeutic modulation of iNKT cell number and functional responses has been proposed for prevention of autoimmunity as well as for the enhancement of immune responses to tumors and vaccines. In the nonobese diabetic mouse model of autoimmune type I diabetes, iNKT number and function are low (43,69,70). Increasing the iNKT cell number (71)(72)(73) or the ␣GalCer treatment-induced Th2 bias (17,18,74) effectively reduces the incidence of type I diabetes in nonobese diabetic mice. Our data demonstrate that distinct glycolipid administration regimens may be required to induce tolerizing activity compared with IFN-␥-dependent antitumor and adjuvant properties of iNKT cells.
The natural self-Ag recognized by iNKT cells and its structural relationship to ␣GalCer remain unknown. However, we recently discovered that a cell line deficient in ␤-glucosylceramide (␤Gl-cCer) is defective in the presentation of a self-Ag to iNKT cell hybridomas (15). Together with the evidence that the defect was not due to altered folding, intracellular traffic of CD1d1, or recognition of ␤GlcCer itself, these results suggest that ␤GlcCer is either a precursor or an essential factor in the synthesis and/or loading of a natural Ag. Thus, it is possible that the elusive self-Ag may be ␣GlcCer or a similar compound. Further support for the hypothesis that a self-Ag similar to ␣GalCer is recognized is the finding that transgenic overexpression of CD1d1 results in preferential deletion of Vb8.1,8.2 ϩ iNKT cells (75). This result is fully consistent with our finding that Vb8.1,8.2 ϩ iNKT cells have a higher K av for ␣GalCer and OCH than Vb8.1,8.2-negative iNKT cells. Furthermore, the high K av binding of CD1d1-␣GalCer with Vb8.1,8.2 ϩ Va14Ja18 TCR is consistent with the high K av binding of dimeric Ag to similar TCR (27). Importantly, for the first time, we demonstrate that the repertoire and Ag K av of iNKT cell receptors are regulated during proliferation and result in selection of high-avidity iNKT cells under conditions of suboptimal stimulation. These data, taken together, suggest that the narrow kinetic window of recognition of ␣GalCer and its analogues is reflective of the parameters of natural self-Ag recognition.
The 2C transgenic TCR exhibits differing peptide Ag binding modes on naive and effector cells, suggesting cooperativity (50). The existence of two TCR ␣␤ molecules within a single CD3 complex was evoked to explain this result (50). However, recent data suggest that the stoichiometry of TCR ␣␤ assembly with CD3 complex is 1:1 (76). Whether this stoichiometry changes during CTL activation remains to be established. Using tetramers of CD1d1 and H2K b , we demonstrate cooperative Ag engagement of glycolipid Ags by iNKT cell receptors but not that of peptidic Ags FIGURE 6. iNKT cell receptor has distinct structure and/or organization. CTL clone SPH60 (A), iNKT hybridoma N38-2C12 (A), and thymic iNKT cells (B) were reacted with allophycocyanin-conjugated H2K b -H60 and CD1d1-␣GalCer tetramers, respectively. They were also reacted with PE-conjugated Abs against TCR␤ (H57-597; specific for TCR␤ chain FG loop), CD3⑀, or CD44. FRET was measured as the fluorescence in the FL3 channel, with the red diode laser off. CD1d1-and H2K b -tetramer concentrations were adjusted to obtain equal MFI of tetramer staining. Considering that the 488-nm emission of the argon-ion laser cannot excite allophycocyanin, fluorescence detected in the FL3 channel is due to FRETmediated allophycocyanin excitation. FRET-induced fluorescence intensity is indicated as a function of CD1d1-␣GalCer tetramer concentration (C). All binding reactions were performed at 4 o C in the presence of sodium azide to prevent capping and internalization. by conventional ␣␤ TCR. The binding mode of 2C transgenic TCR was investigated using an IgG1-H2K b dimer, and evidence for TCR ␣␤ dimerization was obtained by data deconvolution. To confirm the cooperative engagement of glycolipid Ag, we also used IgG1-CD1d1 and IgG1-H2K b dimer similar to that used to investigate the 2C TCR (50). The results supported cooperative Ag engagement by iNKT cell, but not CTL receptors. Thus, the relationship of our findings with those previously reported is unclear.
Because H2K b and CD1d1 tetramers were built upon the same batches of streptavidin-PE/allophycocyanin, cooperativity in one and not the other precludes conformational change in streptavidin or the fluorochrome. Furthermore, because of the wide separation between monomeric subunits of tetrameric CD1d1, it is extremely unlikely that a conformational change within CD1d1 itself is responsible for the observed Hill coefficient. This is further emphasized by the fact that CD1d1 dimers made in a manner distinct from tetramers also show cooperative binding. Cooperativity is independent of the parameters of glycolipid binding to CD1d1, because OCH, which interacts with CD1d1 with differing properties than ␣GalCer, had essentially the same Hill coefficient as ␣GalCer. Thus, the change in Hill coefficient does not reflect a change in the structure of CD1d1 tetramer, but rather a different organization and/or orientation of the TCR engaging such Ags. How iNKT cells respond to self-Ag and yet remain quiescent in physiological situations remains unclear. In this study, we demonstrate that iNKT cell receptors exhibit cooperative engagement of glycolipid Ag. Cooperativity in biological systems is a common mechanism for achieving sensitivity to relatively modest changes in the strength of the signal (33,63). In other words, a relatively small change in ligand concentration will result in full binding/ activation of an enzyme/receptor. It is possible that iNKT cells use cooperativity to induce sensitive response to a small change in the concentration of self-Ag. In support of this hypothesis, self-Ag recognition of ex vivo-isolated iNKT cells is dependent on high levels of CD1d1 expression by target cells (2), and conversely, iNKT cell hybridomas recognizing physiologic levels of CD1d1 on target thymocytes or dendritic cells have high levels of Va14Ja18 TCR expression (6). Thus, the finding of cooperativity in iNKT cell Ag engagement, but not among CTL recognizing peptidic Ags may be one mechanism by which iNKT cells recognize self-Ag(s).
Our data indicate that the structure and/or organization of the iNKT cell receptor may be distinct from ␣␤ TCR of CTL. FG loop within the Cb domain is a large, evolutionarily conserved structure, which forms a wall at the region where Cb and Vb domains of the TCR ␤-chain join to form a cavity (77). Ab mapping studies revealed that the FG loop is in close proximity to one of the CD3⑀ subunits (78). Transgenic mice expressing the TCR ␤-chain mutant lacking the FG loop have no gross deficiencies in the development and function of conventional CD4 and CD8 T cells (79), implying that ␣␤ TCR pairing and surface expression are not grossly impaired. However, a careful analysis in a single specificity TCR transgenic system revealed that thymocytes lacking the FG loop had impaired negative selection (80), but TCR ␣␤ pairing and expression were unhindered. In contrast, however, Va14Ja18 TCR ␣-chain was found not to pair at all with a Vb8.2-FG loop mutant, and hence, the mutant mice were impaired in iNKT cell development (81). Interestingly, the anti-TCR␤ Ab H57-597, which exhibits strong FRET in conjunction with the CD1d1-␣Gal-Cer tetramer specifically binds the FG loop (77,79). However, FRET was not observed in conjunction with H2K b tetramers or dimers. FRET is observed between CD8a of 2C-transgenic CTL and H2K b , suggesting the engagement of CD8a by monomeric H2K b (82). Taken together, the data strongly suggest that the structure and/or the organization of the Va14Ja18 TCR complex are distinct from ␣␤ TCR of conventional T cells, which might potentially account for the cooperative engagement of glycolipid Ags.
In conclusion, our findings demonstrate that iNKT cell functions are controlled by narrow avidity thresholds for glycolipid Ags and demonstrate novel properties of their Ag receptor that may have an important role in iNKT cell activation. These findings have important implications for the therapeutic use of iNKT cells.