Natural killer T (NKT) cells have been found to influence diverse immune responses, especially in mice, including the maintenance of self-tolerance and the surveillance for tumours1. Homologous populations of NKT cells have been found in rodents and primates2, and although this evolutionary conservation indicates that NKT cells have a distinct role in immunity, their precise physiological function has eluded definition. Several recent studies, however, indicate that NKT cells could have been selected in evolution primarily for their role in antimicrobial defence.

NKT cells were originally named because of the co-expression of a T-cell receptor (TCR) along with typical surface receptors for NK cells2,3. This co-expression perfectly illustrates the hybrid nature of these cells. NK cells are part of the innate immune system, as they effect rapid killing and cytokine responses without the need for extensive cell division or differentiation. By contrast, conventional T cells are, together with B cells, the prototypic cell types of adaptive immunity. Adaptive immune cells express highly diverse antigen receptors formed by somatic recombination of the variable (V) and joining (J) gene segments encoding them. Although one lymphocyte expresses only a single rearranged antigen receptor gene, the diversity in the population confers on lymphocytes the capacity to recognize an almost infinite number of different antigens. Responses mediated by B and T cells, therefore, require several days to clonally expand those rare cells that express a particular antigen-specific receptor. Following a second exposure to the antigen, these antigen-specific clones exhibit a more rapid and vigorous response — the hallmarks of immune memory.

Conventional T cells recognize peptides presented by or bound to cell-surface proteins encoded in the major histocompatibility complex (MHC). T cells were initially divided into two categories (Fig. 1a,b): those that express CD4 on their cell surface (CD4+ T cells) recognize peptides presented by MHC class II molecules (Fig. 1a), secrete cytokines and are known as 'helpers' because they regulate or 'help' the responses of other cell types; CD8+ T cells recognize peptides that are presented by MHC class I molecules (Fig. 1b), and although they can secrete cytokines, they are also potent killers of cells that present antigenic peptides.

Figure 1: Antigen recognition by T cells.
figure 1

The three main types of antigen recognition by T cells that express T-cell receptors (TCRs) composed of α- and β-chains. a | Recognition of peptides that are presented by major histocompatibility complex (MHC) class II molecules by CD4+ T cells. b | A CD8+ T cell recognizing its antigenic peptide presented by MHC class I molecules. c | Invariant natural killer T (iNKT) cells are a third category of T cell, and can be CD4+, CD8+ or double negative for these surface markers and recognize lipid antigens presented by CD1 molecules. The 'X' indicates variable antigen-receptor chains. CD8 interacts with a non-polymorphic portion of the MHC class I molecule, whereas CD4 interacts with a non-polymorphic part of the MHC class II molecule.

We now recognize a third category of T cell, which recognizes antigens presented by CD1 molecules (Fig. 1c). These cells can, but do not necessarily, express CD4 or CD8. In humans and most other mammals, four CD1 genes (CD1A–D) encode antigen-presenting proteins4; mice and rats have Cd1d only. The analyses of crystal structures show that CD1 molecules have an MHC-type antigen-binding groove that is bordered by α-helical domains5 (Fig. 2). MHC class I and class II molecules are encoded by highly polymorphic genes that are closely linked. By contrast, CD1 genes are encoded outside the MHC, and with few alleles, they are nearly invariant. Moreover, CD1 molecules have a deeper, narrower and more hydrophobic antigen-binding groove than their MHC-encoded counterparts4,5. The CD1 groove is adapted for the presentation of lipid antigens, which are mostly glycolipids, instead of peptide presentation. CD1 molecules traffic through vesicles of the endolysosomal system4, where they bind the lipid antigens that will be presented on the cell surface. Once bound to CD1, polar head groups of the lipid antigen point upwards from the middle of the CD1 groove (Fig. 2), and the TCR probably recognizes these polar head groups together with α-helical regions of the CD1 molecule.

Figure 2: Structure of a glycolipid bound to, or presented by, CD1d.
figure 2

A ribbon diagram of a side view of the mouse CD1d molecule (grey colour). The bound lipid is a synthetic version of the glycosphingolipid from Sphingomonas yanoikuyae (yellow backbone). Bound palmitic acid (green) completely fills the CD1d groove. Note how the hexose sugar ring in the horizontal orientation protrudes from the CD1d antigen-binding groove for possible T-cell-receptor recognition. This figure was modified with permission from Ref. 101 © (2006) National Academy of Sciences.

The concept that CD1 molecules present microbial lipids is not a new one. It has been established for some years that the CD1a, CD1b and CD1c molecules present various antigens from the mycobacterial cell wall4,5,6. CD1d, the most distantly related member of the CD1 family of proteins, has been known primarily for its ability to present antigens to a unique population of cells known as NKT cells2,4. The precise definition of an NKT cell is not simple, but in this Review we will consider only those NKT cells that recognize CD1d molecules and that have an invariant or canonical TCR α-chain (Box 1) — here called iNKT cells. In mice, the invariant α-chain is encoded by Vα14 (Vα14i), whereas in humans it is encoded by the orthologous Vα24 (Vα24i). For nearly a decade, the only known antigen presented by CD1d and recognized by iNKT cells was α-galactosylceramide (α-GalCer) (Fig. 3a). α-GalCer is a synthetic glycosphingolipid (GSL) variant of a closely related molecule originally extracted from a marine sponge in a screen for compounds that could prevent tumour metastases to the liver of mice7. It was later found to be a potent agonist for iNKT cells8. However, α-GalCer was not thought to be a good candidate for the natural antigen because of its unusual α-linkage of the sugar to the ceramide lipid, but it has proved to be a useful tool for dissecting the role of iNKT cells in immunity. For example, by using α-GalCer and mouse models deficient in Vα14i NKT cells, numerous studies have shown the pivotal roles that iNKT cells have in cancer, autoimmunity and inflammatory disease, and in host defence against pathogens1.

Figure 3: Structure of some microbial glycolipid antigens recognized by iNKT cells.
figure 3

a | Structure of α-galactosyl ceramide (α-GalCer), the first known antigen for invariant natural killer T (iNKT) cells, originally extracted from a marine sponge. b | Structure of phosphatidylinositol tetramannoside (PIM4), a weak iNKT-cell antigen originally extracted from Mycobacteria spp. c | Structure of GalAGSL (glycosphingolipid containing galacturonic acid) and GlcAGSL (glycosphingolipid containing glucuronic acid) originally extracted from Sphingomonasspp. The only difference between the two compounds being the noted 4′ hydroxyl on the sugar in the equatorial rather than the axial position. The asterisk indicates that the 2-hydroxyl on the acyl chain of the Sphingomonas spp. glycolipids is not always present. d | Structure of BbGL-IIc and BbGL-IIf, monogalactosyl diacylglycerol lipids, originally extracted from Borrelia burgdorferi. BbGL-IIc is the most potent BbGL-II antigen in a mouse model, whereas BbGL-IIf is the most potent in humans.

In this Review, we describe the diverse infectious agents, including bacteria, viruses and protozoan parasites, in which iNKT cells have been reported to have an important role in the host response; this role is usually a beneficial one that aids clearance of the infectious agent, although in a few cases iNKT cells are detrimental. We summarize data indicating that iNKT cells respond rapidly to cytokines from dendritic cells (DCs) that have been activated by microorganisms, even when their invariant TCR is not engaged by a foreign lipid antigen, although recognition of a self antigen presented by CD1d might be involved in modulating or initiating this response. Finally, we review the bacterial sources of the glycolipid antigens that engage the invariant TCR expressed by iNKT cells, as well as the unique structures of these antigens.

i NKT cells in host defence against pathogens

Many studies have been carried out to define the role of Vα14i NKT cells in the response to microbial pathogens9,10,11,12,13,14,15,16,17,18,19,20,21,22,23,24,25,26,27. Most of these used strains that are deficient for Vα14i NKT cells. Jα18−/− mice28 specifically lack Vα14i NKT cells because they do not have the Jα gene segment required to form the invariant α-chain by rearrangement. The CD1d protein must be engaged by the TCR, otherwise Vα14i NKT cells fail to differentiate in the thymus29,30,31; Cd1d−/−mice therefore lack Vα14i NKT cells, as well as some other CD1d-reactive T cells with more diverse TCRs32,33. In some studies, blocking anti-CD1d antibodies were used to assess Vα14i NKT-cell function. By these methods, it was shown that Vα14i NKT cells participate in the response to various microbial pathogens including bacteria, fungi, parasites and viruses4,34,35,36 (Tables 1,2).

Table 1 The role of Vα14i NKT cells in parasite, fungal or viral infection
Table 2 The role of Vα14i NKT cells in parasite, fungal or viral infection

Responses to bacteria. The protective role of Vα14i NKT cells was shown in an acute pneumonia model following Streptococcus pneumoniae infection37. Almost all Jα18−/− mice died within several days after intratracheal infection with S. pneumoniae, whereas most wild-type mice survived. Jα18−/− mice had significantly more bacteria, decreased numbers of neutrophils and lower levels of several cytokines in the lung. Similarly, in pulmonary infection with Pseudomonas aeruginosa , it was shown that Cd1d−/− mice or mice treated with anti-CD1d antibodies had decreased clearance of bacteria from the lung38. The number of neutrophils was decreased in the bronchoalveolar lavage fluid of Cd1d−/− mice, and this correlated with decreased production of a neutrophil chemotactic protein. However, in another study, it was recently reported that Jα18−/− mice were not more susceptible to P. aeruginosa infection following intratracheal infection39. These discrepant observations regarding P. aeruginosa might be due to the different strains of bacteria used or the different routes of infection, but they also illustrate the complexity of analysing the role of the relatively small iNKT-cell subpopulation, and they reflect controversies that are endemic to the field.

Another example of the importance of Vα14i NKT cells is provided by a study of a lyme disease model system. After intradermal infection with Borrelia burgdorferi , Cd1d−/− mice developed a greater thickening of the knee and tibiotarsal joints, indicative of arthritis40. Bacterial dissemination from skin to urinary bladder was observed in all Cd1d−/− mice, but not in most control mice. It has been reported that T cells are not crucial for resistance to B. burdorferi41; this work, however, was performed on C57BL/6 mice, a background known to be particularly resistant to the development of Lyme disease, and our work in progress (E.T., unpublished observations) indicates an important role for iNKT cells in vivo when using mice on the BALB/c background.

Responses to parasites. In visceral Leishmania donovani infection, it was reported that the parasite burden was significantly higher in the liver and spleen of Cd1d−/− mice compared with wild-type mice42. Cd1d−/− mice showed a defective granulomatous response, which is important for resistance. By using Jα18−/− mice, the protective role of Vα14i NKT cells also was shown following cutaneous infection with Leishmania major 43. More recently, it was reported that these cells are more important in the response to L. major in visceral infection than in cutaneous infection44, and perhaps this is because Vα14i NKT cells are abundant in the liver and spleen.

Following intraperitoneal infection with T. cruzi, most of the Jα18−/− mice died within a few weeks, and they alldeveloped severe inflammation in the liver, spleen and muscle45. The production of cytokines such as interferon (IFN)-γ, tumour necrosis factor (TNF) and nitric oxide (NO) was significantly higher in cultures of spleen cells from Jα18−/− mice. These data indicate that mice without iNKT cells died from immunopathology as opposed to rampant parasite infection, and that in this experimental system Vα14i NKT cells have a beneficial, anti-inflammatory role.

These examples from different parasite infections illustrate the perplexing dual nature of Vα14i NKT-cell function. Although in many cases Vα14i NKT cells promote microbial clearance, in other cases they might inhibit the synthesis of proinflammatory cytokines, thereby preventing inflammation and decreasing immunopathology.

Responses to viruses. Vα14i NKT cells also are involved in the responses to viruses, although unlike bacteria and parasites, viruses contain only host lipids. In a skin infection (zosteriform) model with a virulent strain of herpes simplex virus type 1 (HSV-1), it was shown that Cd1d−/− mice are more susceptible to infection, with a more rapid spread of virus to spinal ganglia and delayed virus clearance46. Similarly, the virus titres in spinal ganglia were higher in Jα18−/− mice than in wild-type mice. In another study, however, Cd1d−/− mice were not susceptible to HSV-1 infection47, although the less virulent HSV-1 strain might account for the different outcome. It was also reported that Cd1d−/− mice are more susceptible to genital HSV-2 infection, and that iNKT cells are an early source of IFN-γ production following this infection48.

Several mouse viruses, including vaccinia virus and lymphocytic choriomeningitis virus, downregulate Cd1d expression49,50, suggesting a viral immune evasion mechanism that prevents CD1d-mediated antigen presentation to Vα14i NKT cells. This is also true in humans, as Kaposi's sarcoma-associated herpesvirus51, HSV52 and HIV53,54 each use different mechanisms to decrease CD1D expression. Additionally, there is a single case report of a girl who died after receiving the varicella vaccine, and the only immune defect characterized was a lack of Vα24i NKT cells55. Furthermore, boys deficient for the lymphocyte adaptor protein SAP (signalling lymphocyte activation molecule-associated protein), encoded by the SH2D1A gene locus on the X chromosome, lack Vα24i NKT cells, and die from uncontrolled Epstein–Barr virus infections56,57. Similar findings have recently been reported for individuals defective for the X-linked inhibitor of apoptosis protein (XIAP)58. However, the data linking Vα24i NKT cells to protection from viral infections in humans is still fragmentary, and SAP-deficient individuals have a number of immune defects in addition to Vα24i NKT-cell deficiency.

Detrimental role for iNKT cells. iNKT cells are not always protective for the host, as illustrated by studies of chlamydia infection59. Following intranasal infection with Chlamydia trachomatis mouse pneumonitis (C. muridarum), Cd1d−/− mice showed less body-weight loss and decreased bacterial numbers in the lung compared with control mice. Furthermore, Cd1d−/− mice showed decreased production of interleukin (IL)-4 and IL-5, and lower levels of the immunoglobulin isotypes IgE and IgG1. These are the hallmarks of what immunologists call a T helper 2 (TH2) response (Box 2). TH2 cytokines such as IL-4 can inhibit the other main type of TH-cell response, the TH1 response, which is important for microbial clearance because TH1 immunity is characterized by the production of cytokines that activate macrophages. These results indicate that iNKT cells enhanced chlamydial infection through the augmentation of the TH2 response, although for reasons that are not known, in other infections iNKT cells can promote a TH1 response and microbial clearance. There are also data indicating that Vα14i NKT cells are detrimental in the cases of Toxoplasma gondii 60,61 and Listeria monocytogenes infection62, but other laboratories have obtained different results63,64,65 (Tables 1,2), perhaps reflecting microbial strain differences or differences in complex experimental systems.

In summary, despite being a minority T-cell population, there is abundant evidence that iNKT cells are important in mice for host defence against various bacteria, viruses and parasites. There are also data, especially for viral infections, that indicate that this population could be important in humans as well. Work carried out in the last few years, described below, has begun to elucidate the two main mechanisms whereby iNKT cells help to protect the host from infections.

Indirect i NKT-cell activation

Cytokines and endogenous antigens can mediate iNKT-cell activation. How do iNKT cells, with such limited TCR diversity, respond to so many different infectious agents? Recent studies have shed light on the mechanisms by which iNKT cells are activated during infections (Fig. 4), including cases in which a microbial antigen for the invariant TCR is not present. It was reported that human Vα24i NKT cells produced IFN-γ in response to Salmonella typhimurium when cultured with DCs66. S. typhimurium lipopolysaccharide (LPS) or recombinant IL-12 could induce IFN-γ production by iNKT-cell clones in DC co-cultures, and IFN-γ production was inhibited by anti-IL-12 antibodies. Mouse DCs that are deficient for the adaptor myeloid differentiation primary-response gene 88 (MyD88), which interferes with much of the signalling by toll-like receptors (TLRs), failed to induce IFN-γ. These data and the results from in vivo experiments indicate that TLR engagement, mediated in this case by Salmonella spp. infection or by LPS, induces IL-12 synthesis that is crucial for the activation of iNKT cells. This IL-12 was necessary but not sufficient, as the response could also be blocked with anti-CD1d antibodies66.

Figure 4: Invariant natural killer T (iNKT) cells have different pathways leading to their activation.
figure 4

a–c | Indirect activation. The three indirect pathways do not depend on recognition by the iNKT-cell T-cell receptor (TCR) of a microbial antigen, but depend on cytokine release by activated dendritic cells (DCs) and/or the recognition of endogenous glycolipid ligands. a | Cytokine- and endogenous-antigen-mediated activation. During Salmonella typhimurium infection, lipopolysaccharide (LPS) stimulates Toll-like receptors (TLRs) on DCs and induces interleukin (IL)-12 release. iNKT cells are activated by the combination of IL-12 produced by LPS-stimulated DCs and recognition of endogenous antigen presented by CD1d. It has not been determined if LPS induces upregulation of endogenous antigen in DCs. b | Endogenous-antigen-mediated activation. Schistosoma mansoni egg-sensitized DCs induce IFN-γ and IL-4 production by iNKT cells. In this response, TLR-mediated activation of DCs is not involved. However, recognition of endogenous antigen is required. It has not been determined if endogenous antigen is upregulated in S. mansoni egg-sensitized DCs. c | Escherichia coli-LPS-stimulated DCs for IL-12 and IL-18 release. These cytokines are sufficient for IFN-γ production by iNKT cells, and recognition of endogenous antigen presented by CD1d is not necessary for iNKT-cell activation. d | Microbial-antigen-mediated direct activation. Glycosphingolipids from Sphingomonas spp. and galactosyl-diacylglycerols from Borrelia burgdorferi induce iNKT-cell activation by engaging their invariant TCRs. TLR-mediated DC activation, inflammatory cytokines such as IL-12 or recognition of endogenous antigen are not involved in this response. Modified with permission from Ref. 102 © (2007) Macmillan Publishers Ltd.

More recently, it was shown that DCs from Cd1d−/− mice could not induce IFN-γ production by Vα14i NKT cells in response to S. typhimurium67. Neither could mice deficient for β-hexosaminidase (HEXB), an enzyme that is required for the synthesis of isoglobotrihexosylceramide (iGb3)68, the first candidate endogenous antigen. Because S. typhimurium does not contain lipid antigens for the iNKT-cell TCR, these data indicate that iNKT cells were activated in this system by the combination of IL-12, induced by LPS engagement of TLRs expressed by DCs, and the recognition of relatively weak self ligands, such as iGb3, presented by CD1d (Fig. 4a). In this response, TCR recognition of a foreign microbial antigen is not involved, and therefore we refer to this pathway as an 'indirect mechanism' for iNKT-cell activation (Fig. 4a–c).

Bacterial infections can alter CD1D expression. Although the recognition of endogenous antigens is required for iNKT-cell activation in response to S. typhimurium, does bacterial infection increase CD1D expression and/or the presentation of stimulatory endogenous glycolipid ligands? Increased synthesis of the autologous GSL antigens including sulphatide, presented by CD1a or CD1b, and the ganglioside GM1, presented by CD1b, has been shown following the stimulation of monocytes with bacteria or bacterial components69. Regarding the CD1d isoform, there are several reports of increased CD1D expression following exposure to bacteria or bacterial products. CD1D expression by DCs was increased when cultured in vitro with S. typhimurium or LPS from Escherichia coli 70, although the increased expression of CD1d on DCs was not observed in vivo during oral infection with S. typhimurium.

The upregulation of CD1d was also observed on macrophages transferred into Mycobacterium tuberculosis infected mice71. When mouse macrophages were cultured with M. tuberculosis bacteria, bacterial lipids or the synthetic TLR2 agonist Pam3Cys (tripalmitoyl-S-glyceryl cysteine), Cd1d expression was increased, but only if recombinant IFN-γ was also added71. Recently, it was reported that L. monocytogenes induces increased Cd1d expression on macrophages and DCs in vitro and in vivo, which could be inhibited by anti-IFN-β antibodies72. Macrophages with increased Cd1d expression following exposure to bacteria or bacterial products were more effective at stimulating CD1d-dependent cytokine release by iNKT cells, even in the absence of exogenous antigen. These data indicate that antigen-presenting cells (APCs) stimulated with bacteria or TLR agonists can induce the activation of iNKT cells through the increased presentation of endogenous antigens by CD1d combined with IL-12 from activated DCs.

Endogenous antigen-mediated iNKT-cell activation. Another mechanism of indirect iNKT-cell activation was shown in a study of the response to Schistosoma mansoni . iNKT cells were activated during S. mansoni infection and had an important role in the augmentation of the TH2 response73. When liver mononuclear cells (LMNCs) were cultured with S. mansoni-egg-sensitized DCs, the production of IFN-γ and IL-4 was significantly higher in cells from wild-type mice compared with cells from Jα18−/− mice or Cd1d−/− mice. S. mansoni eggs do not contain glycolipid antigens for the iNKT-cell TCR, and DCs from Hexb−/− mice, unable to synthesize the endogenous antigen iGb3, failed to induce IFN-γ and IL-4 production by LMNCs74. These data indicate that iNKT-cell activation by S. mansoni-egg-sensitized DCs was mediated by the recognition of iGb3. However, it remains controversial whether iGb3 is in fact the sole or crucial endogenous ligand for iNKT cells. Indeed, that DCs from Hexb−/− mice are inhibited in their stimulation of iNKT cells could be due to disruption of the endolysosomal vesicles where CD1d is loaded with antigen, as opposed to a deficiency in iGb3 synthesis75.

Surprisingly, DCs from Il12−/− mice or MyD88−/− mice could induce cytokine production by LMNCs in a similar manner to wild-type DCs. These data suggest that, in response to S. mansoni egg antigens, iNKT cells can be activated by self glycolipids presented by CD1d, even in the absence of TLR signalling and IL-12 (Fig. 4b). It has not yet been determined, however, if S. mansoni egg extract increases the synthesis of endogenous glycolipid antigens and/or Cd1d expression. Similarly, it is not known if S. mansoni-egg-sensitized DCs are activated in a TLR-independent manner to produce innate immune cytokines, other than IL-12, that might contribute to the activation of iNKT cells.

Cytokine-mediated iNKT-cell activation. It has recently been shown that Vα14i NKT cells also can be activated by IL-12 and IL-18 produced by DCs that have been activated by E. coli LPS, even in the absence of TCR stimulation by endogenous antigens presented by CD1d76. This purely cytokine-driven response constitutes a third type of indirect iNKT-cell activation. When mice were injected with E. coli LPS, Vα14i NKT cells produced IFN-γ but not IL-4, consistent with an inflammatory response that might enable iNKT cells to contribute to the induction of TH1-type responses. Purified Vα14i NKT cells produced IFN-γ in response to E. coli LPS when cultured with wild-type DCs, but not with Il12−/− or Il18−/− DCs. Surprisingly, Cd1d−/− DCs were also able to induce IFN-γ production by Vα14i NKT cells. Furthermore, physiological concentrations of IL-12 and IL-18 together could induce IFN-γ production by purified Vα14i NKT cells, even in the absence of DCs. These data indicate that inflammatory cytokines such as IL-12 and IL-18, which are produced by TLR-stimulated DCs and other APCs, are necessary and sufficient for the induction of Vα14i NKT-cell activation (Fig. 4c). Finally, a recent paper indicates a fourth indirect activation pathway, in which the cytokine and the TCR signals needed for iNKT-cell activation are delivered by two different DC subsets77.

We conclude that iNKT cells can be activated by infections indirectly through the activation of APCs even in the absence of microbial glycolipid antigens. By using any one of the indirect mechanisms described above, iNKT cells can sense the presence of many types of microorganism, including viruses that do not encode unique glycolipid antigens. This allows iNKT cells to make a rapid response to a wide range of infectious agents, similar to other cells that participate in the innate immune response. It is likely that a relatively weak TCR signal, delivered by self-glycolipid ligands presented by CD1d, can contribute to the signal delivered by innate-immune-cell cytokines. Furthermore, under some circumstances, such as those observed with exposure to Schistosome spp. egg extracts, the TCR signal in reaction to self glycolipids presented by CD1d might be sufficiently potent to mediate iNKT-cell activation. Although the indirect pathways are probably important, they do not account for the specificity and conservation of the invariant TCR.

Direct activation and bacterial antigens

Phospholipid antigens. Several groups have searched for microbial glycolipid antigens that directly activate the iNKT cells by engaging their invariant TCR. The first antigen reported was the glycophosphatidylinositol (GPI) anchor of proteins from Plasmodium spp.and Trypanosoma spp. It was asserted that the IgG responses to GPI-anchored surface antigens of Plasmodium spp.and Trypanosoma spp. were regulated through CD1d-restricted recognition of the GPI moiety by IL-4-producing NKT cells78. However, those findings remain controversial, as two subsequent studies did not reproduce the results79,80.

Amprey et al. later showed that a subset of liver Vα14i NKT cells could be activated by a lipophosphoglycan (LPG) extract from L. donovani42. L. donovani is part of the genus Leishmania, the members of which are responsible for a wide array of diseases, ranging from self-healing cutaneous lesions to destructive skin and mucosal disease, and ultimately to fatal visceral infection. The authors observed an increase in parasite burden and a defect in the granulomatous response in L. donovani-infected mice deficient in Vα14i NKT cells. They also showed that a small percentage (3–6%) of the iNKT cells in the liver produced IFN-γ early after visceral L. donovani infection. They were able to show that this IFN-γ production by iNKT cells was IL-12 independent but CD1d dependent, indicating direct microbial antigen recognition by the iNKT-cell TCR instead of indirect recognition. L. donovani possesses a dense surface glycocalyx formed mainly by related glycoinositol phospholipids (GIPLs) and LPG. In a competitive assay, both purified GIPLs and LPG could inhibit α-GalCer-induced IL-2 production by Vα14i NKT-cell hybridomas, indicating that GIPLs and LPG can bind to CD1d and be potential microbial antigens. However, LPG–CD1d tetramer staining of liver mononuclear cells could not be achieved, and injection of purified LPG could stimulate only 1.4% of the liver NKT cells in vivo. Therefore, these data indicate that LPG could activate only a subset of Vα14i NKT cells.

A third early report on a potential microbial antigen for iNKT cells showed that several mammalian lipids, namely phosphatidylglycerol (PG) and phosphatidylinositol (PI), could specifically bind to CD1d81. Moreover, PI, but not PG, could induce IFN-γ production by splenic T cells from Vα14-transgenic mice. Those results prompted the authors to see if the same were true for a mycobacterial variant of PI. They found that a subset of mouse Vα14i NKT cells and human Vα24i NKT cells reacted to a variant of PI, a purified glycolipid extracted from a mycobacterium cell-wall fraction and enriched for PI tetramannosides (PIM4) (Fig. 3b). Analysis with PIM4–CD1d tetramers showed that the reactive cells comprised only a minority of the α-GalCer–CD1d tetramer-positive cells, especially in the mouse liver, indicating that PIM4 reactivity, similar to the Leishmania spp. LPG antigen, might be dependent on special features of the more variable β-chain of the semi-invariant TCR, as the α-chains are identical. Another limitation of this study was that only purified glycolipid extracted from mycobacteria was used, leaving open the possibility that a minor constituent of the purified product was responsible for the Vα14i NKT-cell activation. Indeed, a later article reported that a synthetic version of PIM4 did not stimulate iNKT cells, either in vitro or in vivo82.

Natural glycosphingolipid antigens from Sphingomonas spp. The studies on Leishmania spp.- and mycobacteria-derived phospholipids helped establish the principle that the invariant TCR of iNKT cells recognizes glycolipids from microbial agents. However, because only a minority of the cells reacted, these specificities could not account for the conservation and selection of the invariant TCR. In 2005, the publication of three articles on GSLs from Sphingomonas spp. described antigens that are capable of activating essentially all iNKT cells67,83,84. Sphingomonas spp. are Gram-negative bacteria that lack LPS and are highly abundant in the environment, including soil, sea water and plants85. Prior to these studies, analyses of Sphingomonas spp. bacteria had revealed the presence of glycosylceramides in the cell wall, with structures similar to α-GalCer, including the rather unusual α-linkage of the sugar to the sphingosine-containing lipid86,87. The structures of two of these are shown in Fig. 3c. The abundance of these bacteria in the oceans indicates that they could have been in the original marine-sponge sample that was used to identify α-GalCer, and their structural similarity to this potent antigen made them good candidates for antigens able to activate the population of iNKT cells.

By using purified or synthetic versions of these GSLs in several different in vitro assays, it was shown that a galacturonic-acid-containing GSL (GalAGSL), and one containing glucuronic acid (GlcAGSL)83 (Fig. 3c) and similar compounds67,84 could bind to CD1d. Moreover, although not as potent as α-GalCer, the Sphingomonas spp. GSLs could specifically activate mouse Vα14i NKT cells as well as human Vα24i NKT cells67,83. Hexb−/− DCs also could stimulate Vα14i NKT cells in response to Sphingomonas spp. These data indicate that the recognition of the putative self ligand iGb3 is not required for iNKT-cell activation in response to Sphingomonas spp.67 Furthermore, TLR signalling and IL-12 were not required for GSL-mediated iNKT-cell activation, indicating that the observed activation was due to direct recognition of the microbial antigen by the TCR as opposed to the indirect pathway67,83. Direct recognition by the invariant TCR was confirmed by staining with CD1d tetramers loaded with the Sphingomonas spp. GSL. Indeed, in contrast to the LPG and PIM4 studies, CD1d tetramers loaded with synthetic GalAGSL were able to detect at least 50% of the Vα14i NKT cells, defined as cells reacting with α-GalCer–CD1d tetramers. Moreover, the reactive cells were absent in Cd1d−/− and Jα18−/− mice, showing overlap between the α-GalCer-reactive and Sphingomonas-spp.-GSL-reactive populations83. Ehrlichia spp., which also belong to the class of α-proteobacteria, can activate iNKT cells independently of TLR signalling and iGb3 synthesis as well; this indicates the direct recognition of microbial antigen, although this antigen has not yet been defined67.

Mice deficient for Vα14i NKT cells had a reduced clearance of Sphingomonas spp. bacteria at days 1–3 following infection67,83. The effect of Vα14i NKT cells on clearance was most evident in the liver83, where these cells are most prevalent, but delayed clearance was also evident in the lung67; eventually, however, the bacteria could be cleared even in mice lacking Vα14i NKT cells. Therefore, direct recognition of GSLs by iNKT cells can contribute to the early stages of host defence against LPS-negative microorganisms. However, when a high dose of Sphingomonas spp. bacteria were administered, mice with iNKT cells died from shock whereas iNKT-cell-deficient mice were protected. This result provides another example in which an overactive protective iNKT-cell response is detrimental. The α-linked GSLs are believed to be unique to Sphingomonas spp. and related bacteria, but these bacteria are not highly virulent or pathogenic in humans, although infections in immunocompromised patients have been reported88,89. It remained uncertain whether iNKT cells can recognize other classes of glycolipids that might be derived from pathogenic microorganisms.

Galactosyl diacylglycerol antigens from Borrelia burgdorferi. B. burgdorferi has recently been shown to have a different category of glycolipid antigen that activates the TCR of iNKT cells79. These spirochetes lack LPS but, contrary to Sphingomonas spp., they are pathogenic. B. burgdorferi is the causative agent of Lyme disease, which, with 15,000 cases each year, is the most common vector-borne disease in the USA90. Cd1d−/− mice infected with B. burgdorferi have been reported to have an increased bacterial burden, and they develop increased thickening of the tibiotarsal joint, indicative of arthritis40. During B. burgdorferi infection, either after injection of live spirochetes or by using infected ticks, Vα14i NKT cells were activated, as seen by an increase in both CD25 and CD69 expression on α-GalCer–CD1d tetramer-positive cells79. Similar data were obtained when DCs pulsed with bacterial lysates were injected into the mice. B. burgdorferi expresses two abundant glycolipids (B. burgdorferi glycolipid (BbGL)-I and BbGL-II)91. BbGL-I is a cholesteryl 6-O-acyl-β-galactoside (structure not shown) and BbGL-II is a 1,2-diacyl-3-O-α-galactosyl-sn-glycerol (Fig. 3d). BbGL-II, but not BbGL-I, was able to induce IL-2 secretion by Vα14i NKT-cell hybridomas, although it was not as potent as α-GalCer or the Sphingomonas spp.-derived GalAGSL. The BbGL-II isolated from the bacteria contained a mixture of C14:0, C16:0, C18:0, C18:1 and C18:2 fatty acids, with C16:0 and C18:1 being the most abundant, and it was uncertain which fatty acid(s) was linked to the sn1 position of the glycerol and which was linked to the sn2 position. Therefore, to further define iNKT-cell specificity, eight chemically synthesized variants of BbGL-II were tested in several assays. BbGL-IIc, which consists of an oleic acid in the sn1 position and a palmitic acid in the sn2 position (Fig. 3d), was by far the most potent antigen82.

By using MyD88-deficient mice or TRIF-deficient TrifLps2/Lps2 mice, the authors showed that BbGL-IIc induced in vitro proliferation, as well as in vivo activation of Vα14i NKT cells independently of the presence of MyD88 or TRIF, indicating that Vα14i NKT-cell activation induced by BbGL-IIc is not dependent on TLR signals and the indirect pathway, but rather is due to direct recognition of the antigen by the TCR of the Vα14i NKT cells. This broad reactivity by the TCRs expressed by iNKT cells was confirmed by tetramer staining. CD1d tetramers loaded with BbGL-IIc detected approximately 23% of the liver Vα14i NKT cells compared with α-GalCer-loaded tetramers. This might be an underestimate of the extent of reactivity to this relatively weak antigen in the Vα14i NKT-cell population, as BbGL-IIc was able to activate all of the Vα14i NKT-cell hybridomas tested. Therefore, a substantial fraction of the Vα14i NKT cells can probably directly recognize α-galactosyl diacylglycerols derived from B. burgdorferi by direct engagement of their invariant TCR.

The iNKT-cell response to B. burgdorferi glycolipids is conserved in humans. Vα24i human NKT cell lines produced IFN-γ and IL-4 after culture with cells transfected with CD1d in the presence of the synthetic galactosyl diacylglycerol compounds82. The response pattern was different from that observed with mouse Vα14 iNKT cells; minimal cytokine release was induced by BbGL-IIc and maximal responses were observed after culture with compounds having a higher degree of unsaturation in the acyl chains, in particular BbGL-IIf (Fig. 3d).

Because BbGL-IIc is a diacylglycerol-based molecule and not a GSL, these data have several important implications for understanding the biology of iNKT-cell responses to infectious agents. First, because bacteria other than B. burgdorferi have glycoglycerol lipids, the invariant TCR could then have a broad reactivity to various microorganisms, which might in part explain the evolutionary selection for this TCR specificity. Second, by altering the degree of unsaturation of the acyl chains, we speculate that bacteria could evade recognition by iNKT cells by producing glycolipids that can bind to CD1d but do not activate the invariant TCR.

Conclusions and perspectives

The recent findings on iNKT cells show that they have important roles in host defence against various pathogens. Regarding the direct-recognition pathway, bacterial glycolipid antigens are recognized by the invariant TCR, including glycosphingolipids found in Sphingomonas spp., and glycerol-based antigens that have a wider distribution. It is certain, however, that some bacteria, such as Salmonella spp. and E. coli, do not have glycolipids for the invariant TCR. It therefore remains to be determined which types of microorganism have such antigens, and if this is confined predominantly to certain types — for example, extracellular bacteria, those that are not Gram-positive or those lacking LPS. Moreover, parasite antigens that unequivocally activate the invariant TCR have not been reported, although it is likely that such antigens exist in some species. It is also unknown if microbial modulation of the composition of the fatty-acid chains is an effective mechanism for the evasion of the iNKT-cell response, or if some bacteria inhibit CD1d expression, similar to viruses. Finally, whereas the conservation of the iNKT-cell specificity is striking, there is still relatively little evidence that human iNKT cells are important for microbial clearance, and additional studies are clearly needed. Moreover, some species such as cattle lack CD1d and iNKT cells. This might reflect exposure to different pathogens or the redundant functions of group I CD1 molecules.

Ultimately, however, if iNKT cells are important, it might be possible to incorporate glycolipids that activate iNKT cells into vaccines. Because CD1d molecules are not polymorphic, and iNKT cells respond similarly in different individuals, it should be possible to design 'one-size-fits-all' glycolipid agents.

The indirect activation pathway allows iNKT cells to respond to many infectious agents. Future studies will probably delineate the different mechanisms whereby iNKT cells are activated by viral as opposed to bacterial infections, the relevant APC types, the innate sensors, including TLRs and other innate sensors used to detect the different infections, and the cytokine and antigen-presentation pathways that communicate with the iNKT cell, leading to activation. A better definition of the relevant self antigens is also required.

In addition to host defence, iNKT cells have been implicated in several chronic inflammatory conditions, including asthma92 and even atherosclerosis93,94. It will be important to determine if iNKT-cell activation in these contexts depends on direct activation mediated by specific microbial glycolipid antigens from commensal or pathogenic microorganisms, or alternatively, whether iNKT-cell activation depends predominantly on cytokines from innate immune cells.