NKT Cell Ligand Recognition Logic: Molecular Basis for a Synaptic Duet and Transmission of Inflammatory Effectors

The immune system evolved with the descent of multicellular metazoans as a means to recognize and respond to an altered internal milieu (homeostasis). Both internal and external stressors, such as toxic substances or microbial and parasitic infections, are known to incite tissue injury. Containment and removal of the stressor, which are essential for initiating tissue repair, are accomplished initially by the archaic, multimodular innate immune system. The innate-like lymphocyte module consists of NK, B1 B, γδ T, NKT, and other cells. NKT cells have evolved to jump-start and fine-tune the innate and adaptive immune responses. The adaptive immune system consists of B and T lymphocytes, which are recruited to assist in the healing process should the innate mechanisms fail to contain and clear the inciter. The quick-acting innate system senses an altered homeostatic state with pattern recognition receptors. In contrast, the slow-responding adaptive immune system uses Ag-specific receptors that are expressed clonally by B and T lymphocytes, BCRs (and Abs) and TCRs, respectively, to sense alterations in the internal milieu. Although each module plays a specific role, multiple modules act in concert resulting in an inflammatory response that is essential in maintaining homeostasis (reviewed in Ref. 1). In this review, we discuss the current knowledge of a duet between NKT cells and APCs, pivotal to which is the understanding of the TCR–ligand recognition logic, and its impact on inflammation.

NKT cells, which express both NK and T cell phenotypic and functional features, are thymus-derived, innate-like lymphocytes whose functions are regulated by self and nonself lipid ligands presented by CD1d molecules. CD1d molecules are expressed by APCs, such as dendritic cells (DCs), macrophages, and B cells, as well as CD4⁺8⁺ thymocytes, hepatocytes, and intestinal epithelial cells. Hence, under different experimental and pathologic conditions, each of these CD1d⁺ cell types can present self and microbial lipids and activate NKT cells (2–12).

The majority of NKT cells express an invariant TCR α-chain generated by TRAV11*02 (mouse Vα14i) or TRAV10 (human Vα24i) to TRAJ18 (Jα18) rearrangement. The invariant α-chain pairs predominantly with mouse TRBV13-2*01 (Vβ8.2), TRBV29*02 (Vβ7), TRBV1 (Vβ2), or human TRBV25-1 (Vβ11) β-chain to form a functional semi-invariant TCR. A small subset, referred to as type II NKT cells, expresses a more diverse TCR repertoire, but little is known regarding its properties and hence is not discussed here. NKT cells regulate microbial and tumor immunity as well as autoimmune diseases by their ability to secrete rapidly large amounts of immunoregulatory cytokines and to upregulate costimulatory molecules to alert and modulate the effector functions of myeloid and lymphoid cells (13, 14).

CD1d is a member of the CD1 family of Ag-presenting molecules. The original report by Brenner and colleagues demonstrating CD1 restriction of Mycobacterium tuberculosis-reactive T cells and the recognition of M. tuberculosis lipids indicated that CD1 molecules present lipid ligands (15–18). Consistent with this is the finding that CD1d assembles with cellular phospholipids and sphingolipids (19–23). CD1d then acquires self and microbial NKT cell agonists in the endosomes/lysosomes (24). Much of our understanding of NKT cell biology, however, has been gleaned from numerous in vitro and in vivo studies that use the marine sponge-derived, synthetic (KRN7000) α-galactosylceramide (αGalCer) and its analogues as the probe (Fig. 1) (13, 14, 24–29). Sphingomonas spp., which is a Gram-negative α-Proteobacteria that lacks LPS, synthesize α-glucuronosylceramide and α-galacturonosylceramide (αGalACer) that resemble αGalCer (Fig. 1). αGalACer directly activates NKT cells in a CD1d-restricted manner (30–32). NKT cells activated by αGalACer appear to be important in Sphingomonas-specific immunity because high-dose infection of wild-type mice results in septic shock caused by rapid release of inflammatory cytokines, whereas low-dose infection of NKT cell-deficient mice delays bacterial clearance (30, 31).

NKT cells also recognize diacylglycerol-based microbial lipids; for example, α-galactosyldiacylglycerol (αGalDAG) and phosphatidylinositol tetramannoside (PtdInoMan₄) (Fig. 1) (33, 34). These glycolipids are cell wall components or their precursors synthesized by Borrelia burgdorferi (35, 36)—the agent of Lyme disease—and M. tuberculosis (34), respectively. Additionally, NKT cells are activated by Helicobacter pylori-derived cholesteryl-6-O-acyl α-glucoside (37). Hence, NKT cells have broad ligand specificity. How the structurally distinct ceramide- and diacylglycerol-based glycolipids are recognized by the semi-invariant NKTCR is discussed in this review, yet how the cholesterylglucoside is recognized remains unknown.

NKT cells are also autoreactive; that is, they react to self-lipids presented by the host APCs (38, 39). An initial search for the endogenous NKT cell agonist revealed that neither cells deficient in β-glucosylceramide (βGlcCer) synthase and transiently expressing CD1d nor cell-free CD1d–βGlcCer complexes activate mouse NKT cell-derived hybridomas (40). This finding suggested that one endogenous mouse NKT cell agonist is a cellular, βGlcCer-derived glycosphingolipid (GSL) (Fig. 2). Notwithstanding, current evidence suggests that both β-linked GSLs [e.g., cellular βGlcCer, isoglobotrihexosylceramide (iGb3; Fig. 1), GD3, and an analogue, β-mannosylceramide (41–44)] as well as glycerophospholipids [e.g., Ptd-inositol, Ptd-ethanolamine, and lyso-Ptd-choline (45–47)] are agonists for a subset or all mouse and/or human NKT cells. The identity of other self-ligands awaits identification and characterization.

The importance of self-lipid recognition was realized in studies demonstrating human and mouse NKT cell activation by DCs cocultivated with either the Gram-positive Staphylococcus aureus or the Gram-negative Salmonella typhimurium. In the case of S. typhimurium, NKT cell activation resulted from bacterial LPS-mediated stimulation of DCs through TLR4 and the secretion of IL-12 (48). This response requires β-hexosaminidase B (HexB), which converts, among other GSLs, the precursor isoglobotetrahexosylceramide (iGb4) to agonistic iGb3 (Fig. 2). These data were interpreted to mean that Salmonella activates NKT cells indirectly through the recognition of self-iGb3 in the presence of IL-12 (31).

Whether iGb3 is the sole endogenous NKT cell agonist has been a contentious issue as NKT cells from iGb3 synthase-deficient mice are fully functional (49). Additionally, only mouse dorsal root ganglion but neither human thymocytes nor DCs appear to synthesize iGb3 (49). Nonetheless, human thymocytes synthesize iGb4 (50) and HexB (44), which can convert iGb4 to iGb3. Moreover, iGb3 is detectable in the absence of the regioisomer Gb3 (51) suggesting that iGb3 is ephemeral and that its levels are regulated by either rapid anabolism to iGb4 or catabolism by the lysosomal α-galactosidase A (GalA; Fig. 2). Indeed, APCs deficient in GalA, which cleaves reducing α-linked Gal residues, cause overt activation of wild-type NKT cells, suggesting the accumulation of an agonist (52). GalA also converts iGb3 (and Gb3) to lactosylceramide (Fig. 2), and therefore its deficiency increases iGb3 levels in cells up to 5-fold (51). A caveat with experiments that use cells or cell lines deficient in lipid metabolic enzymes is that they are known to cause lipid storage disease, which can in turn alter lipid and protein trafficking within cells (53–57). Nonetheless, these findings have implications for the role of NKT cell responses to microbial infections because Salmonella-infected cells or cells stimulated by certain microbial products that downregulate GalA expression (52) increase the levels of iGb3.

NKT cells also respond to a sialylated endogenous lipid when DCs are activated by CpG, a TLR7 ligand, and produce IFN-α (58). They also respond to a combination of inflammatory cytokines such as IL-12 and IL-18 in the absence of a CD1d-restricted agonist (59–61). This latter mechanism is important for immunity to CMV (61). Hence, NKT cells have evolved multiple ways to sense microbial stressors including direct recognition of CD1d-restricted exogenous glycolipids. Alternatively, they sense stressors indirectly, either through the recognition of CD1d–self-lipid complex or in a CD1d-independent manner, in the presence of inflammatory cytokines.

CD1d is a heterodimer consisting of an H chain that is noncovalently associated with the L chain β₂-microglobulin. The H chain folds into five domains: the extracellular α1, α2, and α3 domains (Fig. 4A), which are membrane-anchored by the transmembrane region, ending in a short cytoplasmic tail. Solution of the three-dimensional structures of mouse and human CD1d molecules, which differ subtly from each other, in complex with several lipid ligands—αGalCer, αGalACer, OCH, αGalDAG, sulfatide, PtdInoMan₂, PtdCho, iGb3 (8, 62–69)—revealed that the α1 and α2 domains of the H chain fold into a superdomain to form the Ag-binding groove (ABG; Fig. 3). The ABG is laterally confined by two antiparallel α-helices that are supported at the bottom by an eight-stranded antiparallel β-sheet platform. The membrane-proximal Ig-like α3 domain and the noncovalently associated L chain support the superdomain (Fig. 4A). Therefore, the topology of CD1d resembles peptide–Ag-presenting MHC class I molecules.

The arrangement of the amino acids that make up the ABG is such that the narrow apical entrance leads into two deep-seated pockets (A′ and F′; Figs. 1, 3). The two pockets are lined predominantly by hydrophobic amino acid residues and hence permit the binding of hydrocarbon tails of lipid molecules of varying lengths. The N-acyl chain of αGalCer and related compounds, OCH and αGalACer, tucks into the large A′ pocket, and the long-chain base of GSLs fits into the F′ pocket (Figs. 1, 3). This binding mode exposes the polar head group out from the ABG (Figs. 3, 4). Moreover, the charged amino acids at the entrance of the ABG form a conserved hydrogen-bond network with polar atoms of the head groups of these α-anomeric GSLs (8, 62, 63, 66–68). The same CD1d residues also form hydrogen bonds with β-anomeric GSLs, such as sulfatide, a type II NKT cell ligand, and iGb3 (68, 69). This hydrogen-bond network provides stability to the CD1d–lipid interaction. Thus, the physicochemical architecture of the ABG dictates how the polar epitope is disposed for recognition by the Vα14i/Vα24i TCR or the TCR of more diverse type II NKT cells in the case of sulfatide.

Microbial αGalDAG Ags are structurally similar to αGalCer in that they also have an α-anomeric galactose attached to a lipid backbone (Fig. 1). However, in contrast to αGalCer, the DAG backbone is characterized by two fatty acids esterified to both the sn-1 and sn-2 position of a glycerol moiety, whereas the α-anomeric galactose is attached to the sn-3 position. Borrelial αGalDAG lipids bind to CD1d in two different orientations, depending on the nature of the acyl chains linked to sn-1 and sn-2 positions of glycerol. As a result, the lipid backbone is important in the formation of a TCR epitope, as certain αGalDAGs are NKT cell agonists and others not, because they bind in the opposite orientation (65). B. burgdorferi glycolipid 2c, which is bound with the sn-1 oleic acid (C_18:1) in the A′ pocket and the sn-2 palmitic acid (C_16:0) in the F′ pocket, is a mouse NKT cell agonist (Fig. 4D), whereas B. burgdorferi glycolipid 2f that binds in a reversed orientation with the sn-2 linked oleic acid (C_18:1) in the A’ pocket and the sn-1 linked linoleic acid (C_18:2) in the F′ pocket does not activate mouse NKT cells (33). In contrast, B. burgdorferi glycolipid 2f is a human NKT cell agonist (33), but how αGalDAGs are presented by human CD1d is currently unknown. Even though chemical modifications, such as unsaturations, do not directly make contact with the TCR, by virtue of affecting the orientation of the hexose sugar they contribute to the formation of the NKT cell epitope. Similar changes in the ceramide backbone of αGalCer analogues do not lead to alternative GSL binding orientation, as the ceramide backbone is bound in a conserved orientation through the conserved hydrogen-bond network. In sum, αGalDAG presentation reveals for the first time, to our knowledge, striking differences between mouse and human glycolipid Ag recognition that could not have been appreciated by using strong agonists, such as αGalCer.

Sulfatide binds CD1d in a distinct manner such that the 3′-sulfated galactose is solvent exposed and projects up and away from the ABG as a result of its β-linkage (Figs. 1, 3) (68). This contrasts the more intimate binding of the galactosyl headgroup of αGalCer to CD1d (8, 63). However, despite differences in the binding mode, sulfatide engages CD1d via a hydrogen-bond network mediated by the same residues involved in stabilizing αGalCer (8, 63, 68).

The first hexose of iGb3 (i.e., glucose), akin to sulfatide, is β-linked to ceramide and hence would be predicted to be solvent exposed in a manner similar to the sulfated galactose. This disposition of the first glucose of iGb3 results in an almost perpendicular exposition of the two terminal galactoses [Glc–β(1-4)Gal–α(1-3)Gal] of iGb3 out of the ABG as revealed by the structure of the mouse CD1d (mCD1d)–iGb3 complex (Figs. 1, 3) (69). Nevertheless, the β-linked glucose, which unlike sulfatide lacks a 3′-sulfate and whose 4′-hydroxyl is equatorially disposed, perhaps results in poor binding to CD1d because the 3′-sulfate and the axial 4′-hydroxyl are involved in hydrogen bonding of sulfatide with CD1d.

Taken together, the presentation principles for α- and β-linked glycolipids are distinct. How the same Vα14i/Vα24i TCRs recognize these structurally distinct agonists remains to be elucidated. Finally, it will be interesting to see why sulfatide and iGb3, which share the same core structure, are recognized by different NKT cells: sulfatide being a type II NKT cell agonist and the latter a semi-invariant NKT cell agonist.

Lyso-Ptd-choline, but not Ptd-choline, is a human NKT cell agonist (45). As it consists of only one acyl–sn-1–glycerol, it will be interesting to see into which pocket this single-chain lipid binds or whether two different binding orientations exist of which only one orientation results in an agonist, similar to what has been observed in the case of the borellial DAG ligands (65). As most agonists that are structurally characterized in complex with the NKTCR contain an α-linked galactose that shows a conserved TCR binding footprint, it is difficult to predict how the more complex glycolipids, such as PIM₄ or iGb3, are recognized and engaged by the same NKTCRs.

By contrast to TCR–peptide MHC complexes, wherein the receptor docks diagonal on the Ag (70), the NKTCR docks parallel onto the extreme C-terminal end of the CD1d ABG above the F′ pocket by using three of the six CDRs (CDR1α, CDR3α, and CDR2β) while almost excluding CDR2α, CDR1β, and CDR3β from the interface (Fig. 4) (71–73). This docking mode enables a lock-and-key interaction with the α-linked galactose epitope that was predicted from biophysical studies of Vα14i TCR–ligand binding (71–74). Furthermore, alanine-scanning mutagenesis of the mouse Vα14i TCR as well as the crystal structures of Vα14i-Vβ8.2 and Vα14i-Vβ7 co-complexed with mCD1d–αGalCer revealed that the mouse NKTCR interfaces its ligand in a manner similar to the Vα24i TCR–ligand interaction (72, 73, 75–77).

The above germline-encoded recognition logic raises the question of how the mouse Vα14i and human Vα24i TCRs recognize structurally distinct ligands such as iGb3, GD3, PtdInoMan₄, PtdIno, PtdEtN, and lysoPtdCho. Alanine-scan mutants of Vα14i TCR revealed that the NKTCR recognizes many α-linked GSLs (αGalCer, OCH, αGalACer, αGalDAG, and iGb3, which contains an α-linked terminal galactose) by means of a “hot spot” of germline-encoded amino acids within CDR1α, CDR3α, and CDR2β loops (77). The recent structure of mCD1d–PtdIno bound to an autoreactive Vα14i TCR, in which the β-chain has been mutated to increase affinity toward self-antigens, surprisingly revealed that CDR3α residues do not directly contact the glycolipid, although the conserved TCR footprint on CD1d is maintained, while additional residues in CDR2α contact the phosphoinositol headgroup (Fig. 4) (78). Those interactions are novel and have not been reported for any other Vα14i TCR. Whether recognition of glycolipids by CDR2α residues is unique to self-antigens, however, is currently unknown.

The recent solution of the Vα14i-Vβ8.2/mCD1d–αGalDAG and CD1d–αGalACer crystal structures revealed that the NKTCR has the capacity to induce structural changes in both CD1d and the ligand orientation to maintain the conserved TCR footprint (Fig. 4B–E) (72). Similar to αGalCer and αGalACer, the NKTCR contacts αGalDAG exclusively through CDR1α and CDR3α (72). In each of these ternary structures, CDR1α Asn30 hydrogen bonds with the 2′ and 3′ hydroxyls of the galactose or galacturonic acid of the GSL (72). However, for αGalDAG, this conserved interaction with the NKTCR required a reorientation of the galactose moiety (Fig. 4E) (72). CDR3α residue Gly96 contacts the 2′-OH through a main chain carbonyl, while Arg95 contacts the 3″-OH of the ceramide backbone of both αGalCer and αGalACer (72). However, this hydrogen bond interaction is lost in the αGalDAG structure due to the different lipid backbone structure (72). These findings suggest that the interaction of NKTCR with structurally distinct α-linked ligands is accomplished by similar recognition logic, which involves the germline-encoded hot spot composed of amino acids within CDR1α, CDR3α, and CDR2β loops.

The recently determined crystal structures of nine ternary complexes with bound αGalCer analogues, such as OCH, C20:2 αGalCer, C20:2 αGluCer, 3′,4″-deoxy αGalCer, 4′,4″-deoxy αGalCer (79), as well as C-glycoside, BnNH–GSL-1', and naphthylurea (NU)–αGalCer (80), provide insights into the mechanisms of NKTCR binding and illustrate novel and unexpected findings about the flexibility of the Ag-presenting molecule CD1d. The successive elimination of individual hydroxyl groups of αGalCer analogues at either the galactose moiety or the phytosphingosine chain disrupts individual H-bond interaction between the glycolipid and the NKTCR (or CD1d, in the case of ceramide modifications) and as such affects their recognition and biological outcome. However, certain 6′-galactose modifications can furthermore induce structural changes in CD1d itself, as demonstrated for the ligand NU–αGalCer (80). The aromatic NU modification is not contacted by the NKTCR but instead is inserted into the A′ roof, inducing the formation of a small pocket within that roof. It was proposed that the NU group serves as a third anchor in addition to the two alkyl chains that are bound in the A′ and F′ pockets and as such increases the stability of the CD1d–glycolipid complex, possibly affecting its in vivo activity (80).

The kinetic parameters of NKTCR–ligand interaction have been extensively studied. Surface plasmon resonance and tetramer-binding experiments have revealed high-affinity interaction between Vα14i or Vα24i TCR/CD1d–αGalCer (or derived analogues): the relative avidity of this interaction is similar to that of high-affinity interactions between the TCR–peptide MHC complexes (74, 81–85). Notably, the half-life of mouse NKTCR/CD1d–αGalCer interaction was unusually long (Supplemental Table I and references therein). How these kinetic parameters relate to the rapid and robust NKT cell response remains to be elucidated. In this regard, it is interesting to note that an αGalCer analogue, OCH, which has a shortened long-chain sphingosine base (C₉ versus C₁₈) and acyl chain (C₂₄ versus C₂₆; Fig. 1) and interacts with the Vα14i and Vα24i TCR with lower relative affinity/avidity compared with αGalCer (Supplemental Table I) (64, 82, 83), specifically elicits sustained IL-4 with very little IFN-γ response (86). A similar IL-4–biased response is elicited by a diunsaturated (C20-diene) N-acyl analogue of αGalCer (Fig. 1) (87) whose binding constant is similar to that of αGalCer but whose dissociation rate is similar to that of OCH (Supplemental Table I) (64, 83), the structural basis for which is described later. Hence, the relative TCR binding affinities do not seem to be responsible for the observed Th1/Th2 predisposition of structurally related glycolipids. Rather, the ability of αGalCer-loaded CD1d molecules to accumulate in lipid rafts in vivo, in contrast to CD1d molecules that contain OCH or C20-diene, appears to influence the cytokine profile (88, 89).

It is surprising that despite conserved NKTCR–ligand binding, the equilibrium binding affinity toward microbial glycolipids can vary up to 600-fold compared with αGalCer (Supplemental Table I) (72). Essentially, two factors have been identified that affect both the association rate of the TCR as well as the dissociation rate. First, the need to reorient the galactose of borrelial αGalDAG results in a reduced TCR association (72). Second, upon TCR binding onto mCD1d–αGalDAG or CD1d–αGalACer, the NKTCR induces a structural change in mCD1d above the F′ pocket, namely the formation of the F′ roof (72). The F′ roof is already preformed upon αGalCer binding to mouse and human CD1d, but not when other known NKT cell agonists bind (Fig. 2) (8, 63, 72, 78, 79, 83), and, as such, the TCR does not invest energy into keeping the roof closed upon binding. That results in a more stable complex, indicated by a reduced TCR dissociation rate. The F′ roof is also closed in the previously mentioned ternary complexes of the various αGalCer analogues, as well as in the PtdIno structure (78–80). However, in light of the lack of structures without bound NKTCR, it is not clear whether the F′ roof is already preformed in those CD1d molecules before NKTCR engagement.

In summary, the agonistic potency of αGalCer and related compounds appears to correlate with the extent to which the F′ roof is preformed as well as the ability of the glycolipids to induce further structural changes within CD1d that could enhance CD1d–ligand stability or CD1d–TCR binding stability. Those factors could in turn dictate the biologic outcome upon engaging different ligands, in addition to the pharmacological differences of the glycolipids.

Conventional T cells and APCs as well as NK cells and target cells form immune synapses in preparation for eliciting an appropriate effector response (90–92), and so do NKT cells and APCs/CD1d-containing planar membrane (93), the specificity of which lies within NKTCR–ligand interactions. Consistent with the kinetic parameters (Supplemental Table I), αGalCer and C20-diene efficiently elicit classic immune synapses between NKT cells and the planar membrane by engaging ∼10 molecules of CD1d–ligand μm⁻². By contrast, OCH forms immune synapses at a 10-fold higher concentration (83). Thus, at equal ligand concentration, αGalCer and C20-diene induce very quick and sustained iCa²⁺ flux (a measure of very early T cell activation) compared with OCH (83).

Conventional T cells polarize certain cytokine receptors (IFN-γR), cytokines [e.g., IFN-γ and IL-2 but not TNF-α and CCL3 (94–96)], and lytic granules to the immune synapse (91). αGalCer–pulsed DCs also form immune synapse with freshly isolated NKT cells within 30 min and polarize IFN-γ to the synapse within 50–60 min (93). Similarly, αGalCer and C20-diene rapidly polarized lytic granules to the immune synapse compared with OCH (83). In this way, they engage in synaptic transmission of effector molecules to modulate inflammatory responses to changes in cellular lipid content.

NKT cells localize to portals on microbial entry around cells that express the lipid-presenting molecule, CD1d (7, 97, 98). The stability of CD1d–lipid complexes depends on whether the hydrocarbon chain occupying the F′ pocket permits the formation of a roof. NKTCR interfaces its cognate ligand—CD1d–lipid complexes—in a unique mode, which involves germline-encoded hot spots that lie within CDR1α, CDR3α, and CDR2β loops. The NKT cell–APC synaptic duet is driven by the binding kinetics of NKTCR–ligand interactions. Synapse formation prepares for effector functions and permits synaptic transmission of certain effector cytokines and lytic granules.

Cellular lipid gradients are tightly regulated. Internal and external stressors are known to alter this gradient (52, 99–101). Because CD1d molecules evolved to present lipids, any alterations in the gradient is displayed at the cell surface for an appraisal by NKT cells. By virtue of sensitive ligand recognition [perhaps based on cooperativity (82)], NKT cells can respond quickly to changes in ligand concentration and/or structure. As such, they are known to regulate autoimmune diseases and microbial immunity and hence inflammation arising from stressors from within as occurs in autoimmunity or from the outside as in an infection. In this manner, NKT cells can regulate homeostasis.

We thank Dr. L. Van Kaer, Vanderbilt University, and Dr. J.S. Bezbradica, Yale University, for critical reading of the manuscript and helpful suggestions.

The authors have no financial conflicts of interest.