Adaptability of the semi‐invariant natural killer T‐cell receptor towards structurally diverse CD1d‐restricted ligands

William C Florence, Chengfeng Xia, Laura E Gordy, Wenlan Chen, Yalong Zhang, James Scott‐Browne, Yuki Kinjo, Karl O A Yu, Santosh Keshipeddy, Daniel G Pellicci, Onisha Patel, Lars Kjer‐Nielsen, James McCluskey, Dale I Godfrey, Jamie Rossjohn, Stewart K Richardson, Steven A Porcelli, Amy R Howell, Kyoko Hayakawa, Laurent Gapin, Dirk M Zajonc, Peng George Wang, Sebastian Joyce

Author Affiliations

  1. William C Florence1,,
  2. Chengfeng Xia2,,
  3. Laura E Gordy1,
  4. Wenlan Chen2,
  5. Yalong Zhang2,
  6. James Scott‐Browne3,
  7. Yuki Kinjo4,
  8. Karl O A Yu5,
  9. Santosh Keshipeddy6,
  10. Daniel G Pellicci7,
  11. Onisha Patel8,
  12. Lars Kjer‐Nielsen7,
  13. James McCluskey8,
  14. Dale I Godfrey7,
  15. Jamie Rossjohn8,
  16. Stewart K Richardson6,
  17. Steven A Porcelli5,
  18. Amy R Howell6,
  19. Kyoko Hayakawa9,
  20. Laurent Gapin3,
  21. Dirk M Zajonc*,10,
  22. Peng George Wang*,2 and
  23. Sebastian Joyce*,1
  1. 1 Department of Microbiology and Immunology, Vanderbilt University School of Medicine, Nashville, TN, USA
  2. 2 Department of Chemistry and Biochemistry, Ohio State University, Columbus, OH, USA
  3. 3 National Jewish Centre for Allergy and Immunology Research, Denver, CO, USA
  4. 4 Division of Developmental Immunology, La Jolla Institute for Allergy and Immunology, La Jolla, CA, USA
  5. 5 Department of Microbiology and Immunology, Albert Einstein College of Medicine, Bronx, NY, USA
  6. 6 Department of Chemistry, University of Connecticut, Storrs, CT, USA
  7. 7 Department of Microbiology and Immunology, University of Melbourne, Melbourne, Victoria, Australia
  8. 8 Department of Biochemistry and Molecular Biology, Monash University, Melbourne, Victoria, Australia
  9. 9 Fox Chase Cancer Centre, Philadelphia, PA, USA
  10. 10 Division of Cell Biology, La Jolla Institute for Allergy and Immunology, La Jolla, CA, USA
  1. *Corresponding authors: Department of Microbiology and Immunology, Vanderbilt University School of Medicine, A4223 Medical Centre North, 1161 21st Avenue South, Nashville, TN 37232, USA. Tel.: +1 615 322 1472; Fax: +1 615 343 7392; E-mail: sebastian.joyce{at}vanderbilt.eduDivision of Cell Biology, La Jolla Institute for Allergy and Immunology, La Jolla, CA 92037, USA. Tel.: +1 858 752 6605; Fax: +1 858 752 6994; E-mail: dzajonc{at}liai.orgDepartment of Chemistry and Biochemistry, Ohio State University, Columbus, OH 43210, USA. Tel.: +1 614 292 9884; Fax: +1 614 688 3106; E-mail: wang.892{at}
  1. These authors contributed equally to this work

View Full Text

This article has a correction. Please see:


The semi‐invariant natural killer (NK) T‐cell receptor (NKTcr) recognises structurally diverse glycolipid antigens presented by the monomorphic CD1d molecule. While the α‐chain of the NKTcr is invariant, the β‐chain is more diverse, but how this diversity enables the NKTcr to recognise diverse antigens, such as an α‐linked monosaccharide (α‐galactosylceramide and α‐galactosyldiacylglycerol) and the β‐linked trisaccharide (isoglobotriaosylceramide), is unclear. We demonstrate here that NKTcrs, which varied in their β‐chain usage, recognised diverse glycolipid antigens with a similar binding mode on CD1d. Nevertheless, the NKTcrs recognised distinct epitopic sites within these antigens, including α‐galactosylceramide, the structurally similar α‐galactosyldiacylglycerol and the very distinct isoglobotriaosylceramide. We also show that the relative roles of the CDR loops within the NKTcr β‐chain varied as a function of the antigen. Thus, while NKTcrs characteristically use a conserved docking mode, the NKTcr β‐chain allows these cells to recognise unique aspects of structurally diverse CD1d‐restricted ligands.


The antigen receptor of αβ T cells faces a daunting recognition problem because each T‐cell receptor (TCR) interfaces multiple ligands during its lifetime and yet maintains specificity. They recognise self‐ligand (peptide or lipid) in the context of a self‐antigen‐presenting molecule (major histocompatibility complex (MHC) class I, class II or CD1) in the thymus; foreign ligand in the context of a self‐MHC/CD1 molecule in the periphery; and in some cases, non‐self‐ligands in the context of non‐self‐MHC molecules. This recognition conundrum is much more daunting for the semi‐invariant natural killer (NK) T‐cell receptor (NKTcr) because its recognition landscape is predominantly germline‐encoded and, hence, highly conserved (Borg et al, 2007; Scott‐Browne et al, 2007; Wun et al, 2008), while the antigenic landscapes of CD1d–lipid complexes are highly diverse (Koch et al, 2005; Zajonc et al, 2005, 2008; Kinjo et al, 2006). Understanding how NKT cells see extremely diverse lipid antigens is important because this innate T lymphocyte regulates early immune responses to pathogens and tumours, and is a target for immunotherapy against some autoimmune diseases (Van Kaer, 2005).

The TCR α and β‐chains each comprise variable (V) and constant (C) domains, in which the V domains contain the complementarity determining region‐1 (CDR1), CDR2 and CDR3, that interact with the peptide‐laden MHC (pMHC) molecules. While variations within V‐gene segments confer diversity to CDR1 and CDR2, imprecise joining and non‐templated nucleotide additions during V(D)J recombination generate the much more diverse CDR3 (Rudolph et al, 2006). TCR specificity is not limited to peptidic antigens. NKT cells express a semi‐invariant TCR that recognises lipid‐based antigens presented by the monomorphic CD1d molecule. The NKTcr is made up of an invariant TCR α‐chain, which is generated by TRAV11*02 (mouse and rat Vα14) to TRAJ18 (Jα18) or orthologous TRAV10 (human Vα24) to TRAJ18 rearrangement. Despite the stochastic nature of V(D)J recombination, stringent intrathymic selection steps permit only one CDR3α for NKT cells. Thus, NKT cells express a completely invariant TCR α‐chain that is highly conserved across species (Koseki et al, 1990; Porcelli et al, 1993; Dellabona et al, 1994; Lantz and Bendelac, 1994). In contrast, CDR3β of TRBV13‐2*01 (mouse Vβ8.2) and the orthologous TRBV25‐1 (human Vβ11), which pair with the invariant Vα14 and Vα24 α‐chains, respectively, are highly diverse (Porcelli et al, 1993; Lantz and Bendelac, 1994). So too are the CDR3β of TRBV29*02 (Vβ7) and TRBV1 (Vβ2) β‐chains, which also pair with Vα14 and Vα24 α‐chains in mice (Lantz and Bendelac, 1994; Capone et al, 2003). Thus, NKT cells have an invariant TCR α‐chain, but maintain some diversity in their TCR β‐chain, which is why they are referred to as semi‐invariant.

The NKTcr binds structurally diverse CD1d‐restricted ligands that range from α‐linked monohexosylceramides such as the prototypic NKT‐cell antigen α‐galactosylceramide (αGalCer; Kawano et al, 1997; Burdin et al, 1998) and the closely related α‐galacturonosylceramide (αGalACer; Sphingomonas spp.; Kinjo et al, 2005; Mattner et al, 2005; Sriram et al, 2005); α‐galactosyldiacylglycerol (αGalDAG; Borrelia burgdorferi) (Kinjo et al, 2006), to the trihexosylceramide isoglobotriaosylceramide (iGb3) Zhou et al, 2004). Despite this apparent diversity, most mammalian glycolipids are not recognised by the NKTcr, including lactosylceramide (LacCer), globotriaosylceramide (Gb3; α1,4Gal‐LacCer), which is closely related to iGb3 (α1,3Gal‐LacCer), monosialylganglioside GM1 (α2,3sialyl‐LacCer) or isoglobotetraosylceramide (iGb4) (Zhou et al, 2004). A small subset of NKTcrs also recognise phosphatidylethanolamine (Rauch et al, 2003), phosphatidylinositol (Gumperz et al, 2000), β‐galactosylceramide (Ortaldo et al, 2004; Parekh et al, 2004), the disialylganglioside GD3 (Wu et al, 2003) and microbial phosphatidylinositol‐mannoside (PIM4; Mycobacterium tuberculosis (Fischer et al, 2004). Crystal structures of several CD1d–antigen complexes showed that these lipids bind to CD1d through their long‐chain base (LCB) and/or fatty acyl chain(s) so that the polar head groups protrude out of the CD1d antigen‐binding site (ABS). Furthermore, the disposition of the head group depends on the unique pattern of hydrogen‐bond network formed by residues of CD1d α1 and α2‐helices with the polar aspects of the bound ligand (Giabbai et al, 2005; Koch et al, 2005; Zajonc et al, 2005, 2006). Thus, the α‐linked galactose of αGalCer lies just above the entrance, flat and parallel to the plain of CD1d's ABS (Koch et al, 2005; Zajonc et al, 2005). In striking contrast, the β‐anomeric linkage of the first sugar orients the glycan α1,3Gal–β1,4Gal–β1,1′Glc of iGb3 perpendicular to the ABS (Zajonc et al, 2008). How the semi‐invariant NKTcr recognises multiple diverse ligands in the context of CD1d remains an enigma.

A major advance in understanding how the NKTcr recognises its cognate antigen came from the crystal structure of the human NKTcr/CD1d–αGalCer co‐complex (Borg et al, 2007). It showed that the Vα24 NKTcr interfaces CD1d–αGalCer in an unusual parallel docking mode by engaging germline‐encoded CDR1α, CDR3α and CDR2β residues above the F’ pocket (Borg et al, 2007; Wun et al, 2008). Further, alanine‐scanning mutagenesis of mouse and human NKTcrs revealed that they interact with CD1d–αGalCer similarly by using germline‐encoded hotspots within CDR1α, CDR3α and CDR2β (Scott‐Browne et al, 2007; Wun et al, 2008). One of these studies also revealed that the Vα14 NKTcr bound to CD1d–αGalCer and the very different CD1d–iGb3 landscapes in a conserved manner, using similar germline‐encoded hotspots (Scott‐Browne et al, 2007). The NKTcr showed very little conformational change upon ligation (Gadola et al, 2006; Kjer‐Nielsen et al, 2006; Borg et al, 2007; Zajonc et al, 2008), consistent with a ‘lock and key’ recognition model where neither temperature nor ionic strength affected the CD1d–αGalCer interaction (Cantu et al, 2003). These findings suggest that the NKTcr acts like a pattern‐recognition receptor and also explain why the NKTcr ligand recognition is CD1d‐restricted. Yet, these studies provide little insight into ligand specificity and the molecular features of the diverse epitopes recognised, and do not explain the basis for distinct TCR β‐chain usage for antigen recognition by the NKTcr.

As an approach to understand the role of β‐chain in CD1d‐restricted antigen recognition, we probed a panel of NKT‐cell‐derived hybridomas that express different TCR β‐chains, as well as a panel of NKTcr point mutants that have altered CDR, with a variety of CD1d‐restricted glycolipid ligands and variants with modifications of their sugar moieties. From the recognition patterns that emerged, we concluded that the NKTcr was highly adaptable, depending on the β‐chain used, to recognise unique aspects of structurally diverse ligands within a conserved binding footprint on CD1d molecules.


TCR β‐chain usage dictates diverse glycolipid antigen recognition by the NKTcr

Unlike the CDR3α of the NKTcr, the CDR3β is highly diverse, both in length and primary structure (e.g., Supplementary Table S1). Additionally, CDR1β and CDR2β of distinct β‐chains differ from each other. Because mouse NKTcr use multiple different β‐chains (Vβ8.2, Vβ7 and Vβ2, as well as Vβ14 and Vβ6; Porcelli et al, 1993; Lantz and Bendelac, 1994), we hypothesised that differences within CDR1β, CDR2β and CDR3β account for the ability of NKTcr to recognise diverse lipid antigens. Using a panel of NKT hybridomas expressing identical Vα14 α‐chain paired with four commonly used β‐chains (Vβ8.2, Vβ14, Vβ7 and Vβ6; Lantz and Bendelac, 1994; Gui et al, 2001), we determined whether TCR β‐chain usage influences recognition of diverse glycolipid ligands (Supplementary Figure S1A). Note that Vβ8.2, Vβ14, Vβ7 and Vβ6 have diverse CDR1β, but partly conserved, CDR2β sequences (e.g., Y46 and/or Y48 residues that were critical for interaction with the α1‐helix of CD1d; Borg et al, 2007; Scott‐Browne et al, 2007; Wun et al, 2008; Mallevaey et al, 2009; Pellicci et al, 2009). Additionally, the NKT hybridomas expressed diverse CDR3β regions in association with the same Vβ‐region and, hence, the same CDR1β and CDR2β (Supplementary Table S1).

Bone marrow‐derived dendritic cells (BMDCs) pulsed with αGalCer, iGb3 or αGalDAG were co‐cultivated with the above NKT hybridoma panel and IL‐2 secretion was assessed as a measure of activation. The data showed that the Vβ8.2+, Vβ14+ and Vβ7+ hybridomas recognised αGalCer whereas the Vβ6+ hybridoma did not (Figure 1A), despite the fact that these hybridomas showed comparable response to antigen‐independent stimulation through CD3ε (Figure 1D). These data are consistent with previous reports, which demonstrated that several NKTcr β‐chains are permissive to αGalCer recognition (Mallevaey et al, 2009). The Vβ8.2Jβ2.6+ hybridoma was autoreactive (i.e., in the absence of exogenously added lipid) to BMDCs and was included as a positive control for the assay (Figure 1C).

Figure 1.

TCRβ usage impacts Vα14 TCR antigen recognition. (A) BMDCs pulsed with 0.1–100 ng/ml αGalCer overnight were co‐cultivated with a panel of NKT hybridomas expressing Vα14 α‐chain paired with unique TCR Vβ‐chains (Supplementary Table S1). After 24 h, ELISA measured IL‐2 secreted into the culture medium in response to NKT hybridoma activation. The data are represented as the half‐maximal response (top panel) or half‐maximal stimulatory concentration (EC50; bottom panel) of αGalCer. (B, C) BMDCs pulsed with 10 μg/ml iGb3 (B) or 20 μg/ml αGalDAG (C) overnight were co‐cultivated with the hybridoma panel and activation was measured by ELISA after 24 h stimulation. (D) NKT hybridomas expressing different β‐chains were stimulated with 0.1–27 μg (in three‐fold dilution) plate‐bound anti‐CD3ε mAb. After ∼24 h, IL‐2 ELISA was performed on culture supernatants to determine the sensitivity of each hybridoma to direct TCR stimulation. The data are represented as the EC50 of anti‐CD3ε mAb. Data are representative of duplicate (B–D) or triplicate (A) experiments, each performed in triplicate.

Compared with the αGalCer recognition pattern, the recognition of iGb3 and αGalDAG by this hybridoma panel was clearly distinct (Figure 1). With the exception of the Vβ8.2Jβ2.2+ hybridoma, most Vβ8.2+ NKT hybridomas recognised iGb3 (Figure 1B). While the Vβ7+ hybridoma also recognised iGb3, the Vβ14+ and Vβ6+ hybridomas did not (Figure 1B). Note that a titration experiment showed that the hybridoma panel required a minimum of 10 μg of iGb3 and 20 μg of αGalDAG for recognition (data not shown). Therefore, because micelle formation and bioavailability influence lipid antigen presentation at higher concentrations, the functional avidity of the different Vβ‐expressing NKTcrs for iGb3 and αGalDAG could not be determined.

As the aforementioned experiment used phytosphingosine‐containing iGb3 (Supplementary Figure S1A), we determined whether the chemical make‐up of the LCB influenced the recognition pattern. iGb3 made up of sphingosine‐, sphinganine (4,5‐dihydro)‐ and 4,5‐dibromosphinganine‐containing LCB were recognised just as well as phytosphingosine‐based iGb3 (data not shown), suggesting that the chemical composition of the sphingoid backbone had very little influence on iGb3 recognition. Finally, all Vβ8.2+, the Vβ7+ and only the Vβ14Jβ1.2i+ hybridomas recognised αGalDAG (Figure 1C), whereas the remaining three Vβ14+ and the Vβ6+ hybridomas were unresponsive to this antigen (Figure 1C).

The varied antigen‐recognition pattern could have resulted from variation in TCR expression and differences in the activation threshold of the different NKT hybridomas. Therefore, we determined the IL‐2 response of each NKT hybridoma to antigen‐independent stimulation using anti‐CD3ε monoclonal antibody (mAb). All hybridomas responded similarly to this stimulation (Figure 1D). This result suggests that the recognition pattern of NKT hybridomas was intrinsic to the NKTcr and not due to variation in TCR expression levels or activation threshold. Taken together, the above data suggest that CDR3β together with CDR1β and CDR2β dictate NKTcr's ability to recognise structurally diverse ligands. Alternatively, although less likely, other cell‐surface molecules differentially expressed by the hybridomas could result in the observed differences in responses.

Vβ usage impacts the recognition of 3′ and 4′ αGalCer analogues

To gain insight into how the NKTcr recognises αGalCer, we used αGalCer analogues, which contain modifications at the 3′‐ and 4′‐hydroxyls of galactose (Supplementary Figure S1B). Note that the αGalCer analogues contained C8 acyl chain, which does not alter the conformation of CD1d upon binding (Zajonc et al, 2005; Pellicci et al, 2009). The 3′‐ and 4′‐hydroxyls of αGalCer interact with NKTcr's CDR1α and/or CDR3α residues (Supplementary Figure S2) (Pellicci et al, 2009) yet their deoxy forms were reported not to affect recognition (Wun et al, 2008). Therefore, we determined whether chemical modifications at the 3′ (‐amino, ‐azido and ‐N‐acetyl) and 4′ (‐O‐methyl, ‐O‐ethanol and ‐N‐acetyl) of galactose are recognised by NKTcr, and whether Vβ usage influences this recognition using the approach described above. Replacement of the 3′‐hydroxyl with an ‐amino, ‐azido or ‐N‐acetyl group prevented recognition by the NKT hybridoma panel regardless of the NKTcr's β‐chain composition (Figure 2A). Therefore, although the 3′‐hydroxyl itself is dispensable (Wun et al, 2008), chemical substitutions at this position are detrimental for NKTcr recognition.

Figure 2.

NKTcrs recognise 4′‐hydroxy variants in a distinct manner. (A) BMDCs pulsed with 500 ng/ml of 3′‐hydroxy variants (Supplementary Figure S1B) overnight were co‐cultivated with the same panel of NKT hybridomas described in Figure 1. After 24 h, ELISA measured IL‐2 secreted into the culture medium in response to NKT hybridoma activation. (B) BMDCs were pulsed with the indicated concentrations of αGalCer or its 4′‐hydroxy variants and used to stimulate NKT hybridomas. Top row: Vβ8.2Jβ1.1 (circles); Vβ8.2Jβ2.1 (diamonds); Vβ8.2Jβ2.5 (triangles) and Vβ8.2Jβ2.6 (squares); middle row: Vβ14Jβ1.2i (circles); Vβ14Jβ1.2ii (diamonds); Vβ14Jβ2.5 (triangles) and Vβ14Jβ2.6 (squares); bottom row: Vβ7 (diamonds) and Vβ6 (squares). ELISA measured IL‐2 secreted into the culture medium in response to NKT hybridoma activation. (C) Schematic rendition of NKTcr recognition pattern. Amino‐acid sequence of CDR1β, CDR2β and CDR3β (upper case) of each Vα14 TCR is indicated on the left; lowercase indicates residues flanking each CDR. Vβ sequences were obtained from IMGT ( Data in panels A and B are representative of two independent experiments performed in triplicate.

On the other hand, the recognition of the 4′ α‐GalCer analogues was quite varied. All Vβ8.2+ hybridomas recognised these αGalCer variants, but to different degrees, indicating that CDR3β sequences might influence this recognition (Figure 2B, top row panel 1). The Vβ8.2+ hybridomas recognised 4′‐O‐methyl and 4′‐O‐ethanol αGalCer analogues to a similar extent (Figure 2B, top row panels 2 and 3), whereas the 4′‐N‐acetyl variant was less antigenic when compared with the recognition of αGalCer (Figure 2B, top row panel 4). All Vβ14+ NKT hybridomas recognised αGalCer but not to the same extent as the Vβ8.2+ hybridomas (Figure 2B, middle row panel 1). Recognition of the 4′ analogues by Vβ14+ NKT hybridomas was Jβ‐dependent, such that only the Vβ14Jβ2.6+ hybridoma recognised both 4′‐O‐methyl and 4′‐O‐ethanol analogues, whereas the Vβ14Jβ1.2i+ hybridoma recognised only the 4′‐O‐ethanol variant (Figure 2B, middle row panels 2 and 3). The 4′‐N‐acetyl variant was not recognised by any of the Vβ14+ hybridomas (Figure 2B, middle row panel 4). Furthermore, the Vβ7+ NKT hybridoma recognised αGalCer and each of the 4′ variants to similar extent, whereas the Vβ6+ hybridoma failed to respond to any of these lipid antigens (Figure 2B, bottom row). These data suggest that the 4′‐N‐acetyl variant is a less potent antigen than the 4′‐O‐methyl and ‐O‐ethanol analogues, thereby revealing an agonistic hierarchy in NKTcr ligands: αGalCer>4′‐O‐methyl∼4′‐O‐ethanol≫4′‐N‐acetyl. As the 4′‐N‐acetyl analogue is not recognised by Vβ14+ hybridomas (Figure 3B), we conclude that the recognition of weak NKTcr ligands is influenced not only by CDR3β but also by CDR1β and CDR2β loops.

Figure 3.

NKTcr interfaces CD1d–iGb3 with a distinct recognition logic. (A) BMDCs pulsed with 10 μg/ml iGb3, 2″′‐deoxy‐iGb3, 3″′‐deoxy‐iGb3, 4″′‐deoxy‐iGb3 or 6″′‐deoxy‐iGb3 were used to stimulate a panel of NKT hybridomas described in Figure 1. After 24 h, ELISA measured IL‐2 secreted into the culture medium in response to NKT hybridoma activation. Data are representative of two independent experiments performed in triplicate. (B) Schematic rendition of NKTcr recognition pattern. Amino‐acid sequence of CDR1β, CDR2β and CDR3β (upper case) of each Vα14 TCR is indicated on the left; lowercase indicates residues flanking each CDR (for sequences see

Vβ usage also impacts iGb3 recognition

Low affinity of the CD1d–iGb3/NKTcr interaction has complicated structural, biochemical and functional studies of this ligand–receptor complex (Zhou et al, 2004; Zajonc et al, 2008). We approached the structural analyses of this complex using a combination of iGb3 variants and the aforementioned hybridoma panel expressing Vα14 α‐chain paired with Vβ8.2, Vβ14, Vβ7 or Vβ6 β‐chain. For this, 2″′, 3″′, 4″′ or 6″′ deoxy variants of the terminal galactose were incorporated into the phytosphingosine‐based ceramide backbone during syntheses (Supplementary Figure S1C).

As described above, we found that only Vβ7+ and Vβ8.2+, with the exception of Vβ8.2Jβ2.2+, NKT hybridomas recognised iGb3 (Figure 1A), a pattern similar to 4′‐N‐acetyl‐αGalCer recognition (Figure 2B). This recognition required the 2″′‐hydroxyl group within the terminal galactose by most (Vβ8.2Jβ2.5, Vβ8.2Jβ2.2, Vβ8.2Jβ2.4 and Vβ8.2Jβ2.2), and the 3″′‐hydroxyl by some (Vβ8.2Jβ2.5 and Vβ8.2Jβ2.2 only), Vβ8.2 NKT hybridomas (Figure 3). None of the Vβ14+ hybridomas recognised iGb3 or its deoxy variants (Figure 3). Therefore, despite their obvious structural differences, the three CDRβ loops impact iGb3 recognition in a manner similar to 4′‐N‐acetyl αGalCer recognition.

CDR1β and CDR3β residues, in addition to previously identified ‘hotspots’ influence the recognition of αGalCer variants

The crystal structure of human CD1d–αGalCer/Vα24–Vβ11 NKTcr (Borg et al, 2007), as well as mouse CD1d–αGalCer/Vα14–Vβ8.2 and CD1d–αGalCer/Vα14–Vβ7 NKTcrs (Pellicci et al, 2009), combined with mutagenesis studies (Scott‐Browne et al, 2007; Wun et al, 2008; Mallevaey et al, 2009), have shown that CDR1α, CDR3α and CDR2β residues are critical for CD1d–αGalCer recognition. As the 4′‐αGalCer analogues were differentially recognised by different hybridomas (Figure 2), these variants were used in combination with a panel of Vα14Jα18/Vβ8.2Jβ1.1‐derived mutants with individual amino‐acid alteration within their CDRs to determine the CDRβ residues that influence recognition of the variant part of the antigen. The rationale behind this approach was that some of the Vα14 NKTcr mutants may gain recognition and reveal the CDR residues that influence recognition either directly or indirectly. From the Vα14‐Vβ8.2 NKTcr/CD1d–αGalCer crystal structure (Pellicci et al, 2009), we were able to precisely gauge whether the effect of the alanine‐scan mutants were likely a result of adverse affect on the conformation of the NKTcr (indirect effects, namely CDR1α: V26A, P28A, H31A, L32A, R33A; CDR1β: H27A, N29A and CDR2β: S47A, G49A) or on CD1d‐antigen recognition (direct effects). Also note that none of the mutants become autoreactive to BMDCs, the APC used for antigen presentation (data not shown).

As reported previously (Borg et al, 2007; Scott‐Browne et al, 2007; Wun et al, 2008), our results revealed that CDR1α, CDR3α and CDR2β residues were critical for αGalCer recognition (Figure 4). For the most part, residues critical for αGalCer recognition were also critical for 4′‐O‐methyl‐ and 4′‐O‐ethanol‐αGalCer recognition (Figure 4), suggesting that the footprint of the NKTcr binding to CD1d molecules remains essentially the same for these two variants. Exceptions were CDR1α mutants R33A, which did not recognise 4′‐O‐methyl or 4′‐O‐ethanol‐αGalCer, and H31A, which did not recognise 4′‐O‐methyl‐αGalCer, but partially recognised 4′‐O‐ethanol‐αGalCer (Figure 4). The effects of H31A and R33A are perhaps indirect because they are thought to alter the fold of the NKTcr (Borg et al, 2007; Scott‐Browne et al, 2007; Wun et al, 2008; Pellicci et al, 2009).

Figure 4.

The 4′‐N‐acetyl αGalCer variant interacts with Vα14 TCR with a unique register. (A) BMDCs pulsed with 500 ng/ml αGalCer, 4′‐O‐methyl, 4′‐O‐ethanol or 4′‐N‐acetyl variants were used to stimulate a panel of NKTcr mutants consisting of a single point mutation within CDR1, CDR2 and CDR3 of both TCR α‐ and β‐chains. The Vα3.2+ NKT hybridoma was used as the negative control. After 24 h, ELISA measured IL‐2 secreted into the culture medium in response to NKT hybridoma activation. As the wt hybridoma and the derived mutants were equally sensitive to antigen‐independent stimulation (Scott‐Browne et al, 2007), the data were normalised to wt IL‐2 response to αGalCer‐pulsed BMDCs and represented as mean±s.e.m. of three independent experiments performed in triplicate. (B) Schematic rendition of NKTcr recognition pattern.

The 4′‐N‐acetyl variant was poorly (∼20%) recognised by Vα14Jα18/Vβ8.2Jβ1.1 TCR compared with recognition of αGalCer (Figures 2B and 4), but alteration of residues within CDR1α, CDR2α, CDR1β and CDR3β restored recognition of this variant (Figure 4). Specifically, CDR1α mutants D29A and N30A, as well as CDR2α mutants V50A, D51A and D54A, overcame poor recognition of the 4′‐N‐acetyl variant (Figure 4). Similarly, CDR1β mutants N26A and N29A and CDR3β mutant G94A also rescued defective recognition of the 4′‐N‐acetyl analogue (Figure 4). Note that the differential recognition pattern of the mutant NKT hybridomas was not due to differences in TCR expression levels or activation thresholds because they were similar for all hybridomas as previously reported (Scott‐Browne et al, 2007). Together, these data suggest that the family of NKTcrs is afforded plasticity through different TCR β‐chains, which, therefore, permit binding of structurally variant glycolipid ligands. Furthermore, in addition to the CDR1α, CDR3α and CDR2β residues that form a common NKTcr footprint (Scott‐Browne et al, 2007; Wun et al, 2008; Mallevaey et al, 2009; Pellicci et al, 2009), recognition of weak agonists such as 4′‐N‐acetyl‐αGalCer may be influenced by CDR residues outside these ‘hotspots’. Consistent with this, the β‐chain of the Vβ7+ and Vβ6+ NKTcrs was shown to have a greater role in interacting with CD1d–αGalCer in comparison with the Vβ8.2+ NKTcr (Mallevaey et al, 2009; Pellicci et al, 2009).

CDR2α, CDR1β and CDR3β residues impact on iGb3 recognition

The NKTcr mutant hybridomas were probed with iGb3 and its four deoxy analogues (Supplementary Figure S1C), focusing on the 2″′‐ and 3″′‐deoxy variants that were poorly recognised by the Vα14Jα18/Vβ8.2Jβ1.1 NKTcrs (Figure 5). In agreement with a previous study (Scott‐Browne et al, 2007), we also observed that iGb3 recognition involved the same germline‐encoded ‘hotspots’, which include residues within CDR1α, CDR3α and CDR2β (Figure 5). Furthermore, mutations within all three β‐chain CDRs also affected recognition of iGb3, none of which, except CDR3β G94A, were rescued by the 2″′‐ and 3″′‐deoxy variants (Figure 5). Most interestingly, alanine mutations either partially (CDR2α V50A) or fully (CDR2α K53A and CDR3β G94A) rescued the recognition of 2″′‐deoxy‐iGb3 and 3″′‐deoxy‐iGb3, which were poorly recognised by the wt NKTcr (Figure 5). From these data we conclude that iGb3 recognition by the NKTcr is distinct from αGalCer but shares some characteristics with that of 4′‐N‐acetyl‐αGalCer. Furthermore, the data suggest a potential role of CDR1β and CDR3β, in addition to the CDR2α, loop in the recognition of iGb3.

Figure 5.

NKTcr recognises CD1d–iGb3 with a distinct recognition logic. (A) BMDCs were pulsed with iGb3 or its variants (as described in Figure 3A) and used to stimulate a panel of NKTcr mutants, and activation was monitored (as described in Figure 3A). Data were normalised as in Figure 4A and represented as mean±s.e.m. of at least three independent experiments performed in triplicate. (B) Schematic rendition of NKTcr recognition pattern.

CDR2α and CDR1β residues impact on αGalDAG recognition

Having determined that the αGalDAG recognition pattern by naturally occurring Vα14+ hybridomas was distinct from that of αGalCer and iGb3 (Figure 1), we tested how the Vα14 NKTcr mutants recognise this antigen. The responses of the mutant hybridomas again revealed that αGalDAG recognition involved the germline‐encoded ‘hotspots’ within CDR1α, CDR3α and CDR2β (Figure 6). While none of the CDR3α mutants recognised αGalCer or iGb3, we found that CDR3α D93A uniquely recognised αGalDAG (Figures 4, 5 and 6). Further, although Y48A mutant does not recognise the three ligands, the remainder of CDR2β residues were differentially required for recognition: none were required for αGalCer, but S47, G51 and S52 were required for iGb3 while S47 and S52 were required for αGalDAG recognition (Figures 4, 5 and 6). A similar discordance was also observed with CDR3β mutants: none were critical for αGalCer and αGalDAG recognition (Figures 4 and 6), but many (S93, T95, T96 and T98) were critical for iGb3 recognition (Figure 5). Moreover, CDR2α residues V50 and K53, which were required for 2″′‐deoxy iGb3 recognition, along with D54, were critical for αGalDAG recognition, but were irrelevant for αGalCer recognition (Figures 4, 5 and 6). These data indicate that the galactose head group of αGalDAG when bound to CD1d is disposed differently from the galactose of αGalCer. Thus, while maintaining a common NKTcr footprint mediated by the ‘hotspots’, effective recognition of αGalDAG involves different contact residues within these hotspots, and additionally requires CDR2α and CDR1β residues.

Figure 6.

The αGalDAG recognition logic is distinct from that of αGalCer or iGb3 recognition by NKTcr. BMDCs pulsed with 20 μg/ml αGalDAG were used to stimulate a panel of NKTcr mutants and activation was measured as described in Figure 4A. Data were normalised as described in Figure 4A and represent the mean±s.e.m. of at least three independent experiments performed in triplicate.


Our current knowledge of how NKTcr recognises its cognate ligands is based on the crystal structure of Vα24–Vβ11 NKTcr/CD1d–αGalCer (Borg et al, 2007) as well as the recently solved crystal structures of Vα14–Vβ8.2 and Vα14–Vβ7 NKTcr/CD1d–αGalCer (Pellicci et al, 2009) complexes. Moreover these structures have been complemented by mutational studies on the Vα14 and Vα24 NKTcrs (Borg et al, 2007; Scott‐Browne et al, 2007; Wun et al, 2008; Mallevaey et al, 2009). Collectively, these structures have revealed a similar docking mode between the Vα14–Vβ8.2, Vα14–Vβ7 and Vα24–Vβ11 NKTcrs. These structures also revealed that differences in Vβ‐chain usage nevertheless impacted on CD1d–αGalCer recognition. How differing Vβ usage would impact on recognising CD1d‐restricted antigens other than αGalCer was less clear. In the absence of the crystal structures of NKTcr/CD1d–iGb3 and NKTcr/CD1d–αGalDAG, we have used an approach based on functional recognition to delineate how semi‐invariant NKTcrs recognise diverse ligands. The approach involved probing a panel of naturally occurring Vα14 TCRs and derived point mutants that have altered CDRs with a panel of CD1d‐restricted glycolipids and their sugar variants. Importantly, the recently solved crystal structure of the Vα14–Vβ8.2 NKTcr/CD1d–αGalCer structure enabled us to ascertain the mutations that are likely to exert indirect effects (e.g., CDR1α: V26A, P28A, H31A, L32A, R33A; CDR1β: H27A, N29A; CDR2β: S47A; G49A (Pellicci et al, 2009) versus mutations that would impact directly on recognition.

Our results revealed several hitherto unknown key features of diverse ligand recognition by NKTcr: first, we found that distinct epitopic sites in αGalCer and iGb3 were critical for recognition. Second, recognition of weak CD1d‐restricted NKTcr agonists critically depended on the β‐chain that paired with the Vα14 α‐chain. Third, we discovered that while the NKTcrs dock on CD1d with a conserved footprint, they are malleable enough to recognise αGalDAG, which is structurally related to αGalCer, and the structurally most divergent iGb3 (Figure 7B). Fourth, as we will discuss below, our data best fit the ‘squash’ model for iGb3 recognition by the NKTcr (Figure 7A).

Figure 7.

The recognition logic. (A) iGb3 modelled onto the Vα14Vβ8.2/CD1d–αGalCer crystal structure shows clashes with the CDR1α and CDR3α loops. CDR1α (orange); CDR2α (green); CDR3α (blue); CD1d (grey) and iGb3 (yellow). (B) The Vα14Vβ8.2 NKTcr surface showing CDR loops (top panel). Mutations that decreased ligand binding are shown in red (middle panel). The NKTcr surface with αGalCer (left), iGb3 (middle) and αGalDAG (right). Mutations that increased ligand binding are shown in dark blue (bottom panel). The NKTcr surface with 4′‐N‐acetyl analogue of αGalCer (left) and 2″′‐deoxy analogue of iGb3 (right). TCR surface (pale cyan); CDR1α (orange); CDR2α (green); CDR3α (blue); CDR1β (pale green); CDR2β (pink) and CD1d helices (dark grey); CDR3β is not shown.

We found that the Vα14 NKTcr is intolerant to chemical modifications at the 3′ position of αGalCer even though the 3′‐hydroxyl is dispensable for NKTcr recognition (Wun et al, 2008). On the other hand, the Vα14 NKTcr was tolerant to multiple 4′ modifications, consistent with the previous finding that the 4′‐hydroxyl is dispensable for recognition (Wun et al, 2008). Nonetheless, a unique modification of αGalCer with 4′‐N‐acetyl was very poorly recognised by Vα14Jα18–Vβ8.2Jβ1.1 TCR. Its recognition was rescued by several mutations within CDR1α, CDR2α, CDR1β and CDR3β, many of which are surprisingly outside the common binding footprint. Thus, the CDRβ loops appear to complement the recognition landscape formed by the α‐chain CDRs.

The 2″′‐ and 3″′‐hydroxyls of the terminal α1,3Gal of iGb3 were essential for Vα14Jα18–Vβ8.2Jβ1.1 recognition. As with the 4′‐N‐acetyl‐αGalCer variant, mutations within CDR1α, CDR2α and CDR3β rescued recognition of both the deoxy‐iGb3 variants. Because this and previous mutagenesis studies revealed germline‐encoded CDR1α, CDR3α and CDR2β ‘hotspots’ in αGalCer and iGb3 recognition (Scott‐Browne et al, 2007), we assume that the Vα14 NKTcr docks on both CD1d–αGalCer and CD1d–iGb3 complexes in a similar mode. The perpendicular disposition of the trihexoses of iGb3 in an unliganded state (Zajonc et al, 2008) as compared with the parallel orientation of the monohexose in CD1d–αGalCer (Zajonc et al, 2005) structure poses a serious recognition problem, and would result in steric clashes with the NKTcr (Figure 7A) if one assumed a docking mode similar with respect to CD1d–αGalCer recognition. Thus, we suggest that the glycan moiety of iGb3 is flattened upon NKTcr engagement, which is analogous to the TCR‐induced flattening of extremely bulged pMHC class‐I ligands (Tynan et al, 2007). However, with our current data, it is not possible to speculate on precisely how iGb3 might be flattened to fit into an NKTcr/CD1d–antigen complex. Notwithstanding, because LacCer and Gb3, which are related to iGb3, are not recognised by the NKTcr (Zhou et al, 2004), it appears as though the terminal α1‐3Gal in iGb3, in its squashed position, provides critical new contact(s) that are required for binding to CD1d, NKTcr or both.

CD1d–αGalDAG recognition by the Vα14 NKTcr also involves the previously reported germline‐encoded ‘hotspots’ made up of residues within CDR1α, CDR3α and CDR2β. In addition, it involves additional unique residues contributed by other CDR many of which are dispensable for αGalCer and iGb3 recognition. Most importantly, CDR2α V50, K53 and D54 were critical for αGalDAG recognition. These results suggest that the positioning of the galactosyl group in αGalDAG differs from that of αGalCer. In the recently reported model, the diacylglycerol backbone of αGalDAG is predicted not to form the critical hydrogen bond network with CD1d R79 and D80. Hence, it is expected that the galactose in αGalDAG is raised above the plane of the galactose in αGalCer and is moved away from the α2‐helix (Kinjo et al, 2006). A similar disposition of the galactose is observed when human CD1d displays αGalCer that is caused by a species variation (human W153, which is equivalent to mouse G155; Koch et al, 2005; Zajonc et al, 2005). The putative differences in the disposition between the galactoses of αGalDAG and αGalCer when displayed by CD1d might explain the distinct logic used by the NKTcr for αGalDAG and αGalCer recognition.

We found that the β‐chain played a significant role in 4′‐N‐acetyl‐αGalCer, iGb3, iGb3 analogues and αGalDAG recognition because none of the Vβ14+ NKTcr recognised these ligands. This could partly be because Vβ14 lacks the CDR2β Y48 residue, which is critically important for the binding of Vβ8.2‐containing NKTcr to CD1d (Borg et al, 2007; Scott‐Browne et al, 2007; Wun et al, 2008; Mallevaey et al, 2009). Equally intriguing was the finding that despite the lack of Y48, Vβ14+ NKTcr recognised αGalCer. This is perhaps because Y46 present in Vβ14 compensates for the absence of Y48. Notwithstanding, Vβ14+ NKTcr did not recognise iGb3 or αGalDAG. As well as because CDR2β Y46‐containing Vβ6+ NKTcr does not recognise any of the ligands tested, we predict that CDR3β contributes significantly to the energetic landscape for low‐affinity ligand recognition (4′N‐acetyl‐αGalCer, αGalDAG and iGb3), perhaps by influencing the structure of the closely disposed CDRα loops. A recent report identified an I/LxxPI/L (I, L and P are isoleucine, leucine and proline; x is any amino acid) motif within CDR3β as a potential contributor for Vβ6+ NKTcr antigen recognition (Mallevaey et al, 2009). Such a recognition logic would then explain the exclusion of the majority of the β‐chains from the NKTcr repertoire during positive selection (Mallevaey et al, 2009), which is presumably mediated by low‐affinity self‐ligands. Thus, we conclude that the β‐chain CDRs fine‐tune the affinity threshold for NKTcr ligand recognition.

While CDR3β did not appear to play a key role in recognition of αGalCer, variations in its length and composition did impact on the recognition of analogues of αGalCer and iGb3, as well as αGalDAG. Existing structural data (Borg et al, 2007; Pellicci et al, 2009) have not supported a key role for CDR3β in direct recognition of CD1d–αGalCer; however, some CDR3β conformations appear to be more permissive than others for CD1d–glycolipid‐antigen recognition (Kjer‐Nielsen et al, 2006; Scott‐Browne et al, 2007; Mallevaey et al, 2009). Taken together, the impact of CDR3β conformation seems to vary as a function of the glycolipid antigen. It is difficult to speculate on precisely how CDR3β is contributing to recognition in these instances, although some CDR3β conformations might inhibit binding through steric hindrance, as suggested for the human NKT‐15 TCR (Kjer‐Nielsen et al, 2006), and some may indirectly affect recognition by altering the juxtaposition of other CDR regions that are involved in direct binding as has been observed in the structure of Vα14–Vβ7 NKTcr/CD1d–αGalCer complex in comparison to that of Vα14–Vβ8.2 NKTcr/CD1d–αGalCer complex (Pellicci et al, 2009).

In sum, the second site‐suppression strategy allowed us to gain new insight into how the NKTcr recognises multiple ligands, especially of CD1d–iGb3 and CD1d–αGalDAG complexes for which crystal structures of the receptor/antigen co‐complexes are yet to be solved. Our data reinforce the previously reported common docking footprint, but extend the recognition logic significantly further for the recognition of weak/low‐affinity agonists. Such weak agonists require other areas of the recognition landscape (e.g., CDR2α, CDR1β and CDR3β) to achieve the energy necessary for NKT‐cell activation. Therefore, we conclude that there is a consistent logic with which the NKTcr positions its ‘hotspot’ to find the epitope. The malleable nature of NKTcrs described herein then determines whether the orientation with which they ‘perch’ on the antigens, especially weak agonists, can be stabilised sufficiently to generate an activation signal. These new insights can be used to design altered glycolipid ligands that will have the potential to steer downstream immune responses in a desired manner.

Materials and methods


C57BL/6 mice were purchased from the Jackson Laboratory and used in compliance with Vanderbilt Institutional Animal Care and Use Committee‐approved regulations.


The syntheses and structure validation of all αGalCer and iGb3 and their variants are described elsewhere (Chen et al, 2007; Xia et al, 2009). αGalCer and iGb3 and their sugar variants were dissolved in DMSO at 1 mg/ml and used at the indicated concentrations. αGalDAG preparation and use has been reported (Kinjo et al, 2006).

Preparation of BMDC

BM cells were harvested from the femurs of 8 to 10‐week‐old C57BL/6 mice and differentiated into DC by cultivating precursors in petridishes for 7 days in RPMI+ (RPMI‐1640 supplemented with 10% fetal calf serum (FCS), 1% l‐glutamine, 1 mM penicillin/streptomycin; CellGrow) containing 20 ng/ml recombinant murine colony‐stimulating factor‐2 (rCsf‐2; PeproTech). On days 3 and 5 of culture, 70% of the medium containing non‐adherent cells was removed and replaced with fresh medium containing 20 ng/ml of rCsf‐2. On day 7, the loosely adherent and non‐adherent cells largely containing BMDCs and some granulocytes were harvested and used immediately for antigen loading.

NKT hybridoma stimulation and ELISA

All Vα14+ mouse NKT hybridomas N38‐3C3, DN32.D3, N37‐1H5a, N38‐2C12, N57‐2C12, N57‐2B6, H41‐2C9, H41‐3C5, H41‐2D9, N38‐2H4, Vα14‐DOβ and its alanine mutants have been described (Lantz and Bendelac, 1994; Gui et al, 2001; Scott‐Browne et al, 2007). TCR α‐ and β‐chain usage and CDR sequences are shown in Supplementary Table S1. They were maintained in RPMI+ containing 55 μM 2‐mercaptoethanol. BMDCs were dispensed into round‐bottomed microtitre plates at a density of 3 × 104 cells/well and incubated with increasing concentrations of αGalCer and its 3′ and 4′ variants, (0.1–500 ng/ml), iGb3 and its deoxy variants (10 μg/ml) or αGalDAG (20 μg/ml) for 18–24 h at 37°C. Lipid‐loaded BMDCs were washed twice with warm PBS and co‐cultivated with 6 × 104 NKT hybridoma cells per well for 24 h at 37°C. Cell‐free supernatant was harvested and IL‐2 secreted during co‐culture was determined by ELISA.

TCR modelling

The NKTcr/CD1d–iGb3 model was prepared based on the previously reported crystal structure of CD1d–iGb3 complex (Zajonc et al, 2008) and the Vα14Vβ8.2 crystal structure (Pellicci et al, 2009). All NKTcr–CD1d models were prepared in an identical docking orientation as observed in the Vα14Vβ8.2/CD1d–αGalCer crystal structure (Pellicci et al, 2009).

Supplementary data

Supplementary data are available at The EMBO Journal Online (

SA Porcelli serves as a paid consultant for Vaccinex Inc., Rochester, NY. The remaining authors have no conflicting financial interests.

Supplementary Information

Supplementary Materials and Methods [emboj2009286-sup-0001.pdf]

Supplementary Figures [emboj2009286-sup-0002.pdf]

Review Process File [emboj2009286-sup-0003.pdf]


We thank L Van Kaer for critical evaluation of the data and helpful suggestions on the writing of this paper. This work was supported by grants from the NIH HL069765 (WCF, LEG) AI007611 (LEG), AI45889 (SAP), AI057519 (ARH), AI074952 (DMZ), AI048224 and AI061721 (SJ), as well as by the Medical Scientist Training Program of the Albert Einstein College of Medicine (KAOY), the Cancer Research Institute's Investigator Award (DMZ) and an endowment from the Ohio State University (PGW).

Author contributions: WCF, LEG and SJ designed experiments. WCF and LEG performed experiments. CX, WC, and YZ synthesised glycolipids; OP, LK‐N and RJ provided modelled structures based on mutagenesis data; JS‐B, YK, KOAY, SK, SKR, SAP, ARH, KH, and LG provided critical reagents; WCF, CX, JR, DIG, DMZ, PGW and SJ wrote and WCF, DGP, DIG, JR, LG, SAP, ARH, DMZ, PGW and SJ edited the paper.


View Abstract