The molecular bases of δ/αβ T cell–mediated antigen recognition

We previously identified CD1d-restricted, α-galactosylceramide (α-GalCer)–reactive Vδ1⁺ T cells (Uldrich et al., 2013). Further analysis of human Vδ1⁺ CD1d–α-GalCer tetramer⁺ T cells using a different anti-Vδ1 clone (A13) compared with our previous study (clone TS8.2; Uldrich et al., 2013) revealed that some Vδ1⁺ cells coexpressed a TCR-β chain rather than a TCR-γ chain (Fig. 1 a). Single cell TCR sequencing of individual Vδ1⁺ TCR-β⁺ CD1d–α-GalCer tetramer⁺ T cells confirmed the expression of the Vδ1 gene recombined to Jα gene and Cα constant region (Fig. 1 b). TCR sequence analysis of these cells from five different donors revealed nine distinct hybrid Vδ1-Jα-Cα-TCR-β chains from within the Vδ1⁺ CD1d–α-GalCer tetramer⁺ population (Fig. 1 b). Thus, the Vδ1-specific antibody (clone A13) was capable of recognizing TCRs where the Vδ1 gene is recombined to Jα and Cα genes, whereas the other Vδ1-specific antibody (clone TS8.2) did not bind to these hybrid TCRs (not depicted). Accordingly, we termed CD1d–α-GalCer tetramer⁺ cells that express δ/αβTCRs as CD1d-restricted δ/αβ T cells.

In contrast to many of the CD1d-restricted Vδ1⁺ γδ T cells identified in our previous study that were only partially dependent on CD1d-bound lipid Ag (Uldrich et al., 2013), CD1d-restricted δ/αβ T cells showed an absolute requirement for CD1d-Ag, as “unloaded” (“CD1d/Endo”) human CD1d tetramers failed to bind these cells (Fig. 1 a). Analysis of in vitro expanded type I NKT cells, CD1d-restricted γδ T cells, and CD1d-restricted δ/αβ T cells revealed that the CD1d-restricted δ/αβ T cells typically expressed low levels of CD161 (not depicted). Despite the ability of Vδ1 to pair with several different TCR-β chains (Fig. 1 b), the TCR-β chain (Vβ11) common to type I NKT cells, was not detected on CD1d-restricted δ/αβ T cells (Fig. 1 a). CD4 and CD8 expression by δ/αβ T cells varied between donors and they were mostly CD4⁺CD8⁻ or CD4⁻CD8⁺ (Fig. 1 a), and this typically differed from CD4 and CD8 expression on type I NKT cells from within the same donors. Similar to CD1d-restricted γδ T cells (Uldrich et al., 2013), human CD1d-restricted δ/αβ T cells did not stain with mouse CD1d–α-GalCer tetramer, unlike human type I NKT cells which showed strong cross-reactivity to mouse CD1d (not depicted; Brossay et al., 1998). We detected a clear population of CD1d-restricted δ/αβ T cells within the expanded CD1d–α-GalCer–reactive T cell population in 13 out of 30 donors (Fig. 1 c). In most cases, CD1d-restricted δ/αβ T cells represented <1% of total CD1d–α-GalCer–reactive cells, although they were higher in some individuals, including one individual in which they were over 50% of CD1d–α-GalCer–reactive cells (Fig. 1 c). Accordingly, CD1d-restricted δ/αβ T cells represent a novel subset of human CD1d–α-GalCer–restricted T cells, which are distinct from both type I NKT cells and CD1d-restricted γδ T cells.

To evaluate the Ag specificity of CD1d-restricted δ/αβ T cells, CD1d–α-GalCer tetramer⁺ cells were isolated and expanded from six different donor PBMC samples and stained with CD1d tetramers loaded with α-GalCer, α-glucosylceramide (α-GlcCer), sulfatide, 3′-deoxy-α-GalCer, 4′-deoxy-α-GalCer, and OCH, known NKT cell ligands (Wun et al., 2011; Rossjohn et al., 2012). Tetramer staining of CD1d-restricted δ/αβ T cells was compared with tetramer staining of type I NKT cells from the same donors (Fig. 2, a and b). CD1d-restricted δ/αβ T cells bound α-GalCer–loaded CD1d tetramer in an Ag-dependent manner as these cells failed to stain with unloaded CD1d tetramer containing endogenous (“endo”) Ag (Fig. 2, a and b). Most CD1d-restricted δ/αβ T cells, with the exception of these cells from donor 2, did not bind to CD1d–α-GlcCer tetramer, which is in stark contrast to type I NKT cells from the same donors that stained brightly with this tetramer (Fig. 2, a and b). The CD1d-restricted δ/αβ T cells were also unreactive to sulfatide-loaded CD1d tetramer, whereas reactivity against 3′-deoxy- and 4′-deoxy-α-GalCer was highly variable between individual donors (Fig. 2, a and b). We have previously established that human type I NKT cells are very sensitive to the loss of the 3′-OH group on α-GalCer (Wun et al., 2012). In contrast, we show here that in at least some individuals, human CD1d-restricted δ/αβ T cells (from donors 1, 3, and 4) were clearly stained by the CD1d-3′-deoxy-α-GalCer tetramer, whereas most of the type I NKT cells from those same donors were not (Fig. 2, a and b). Conversely, CD1d-restricted δ/αβ T cells from donors 4 and 6 stained poorly with CD1d-4′-deoxy-α-GalCer tetramers, whereas the type I NKT cells from the same donors were brightly labeled (Fig. 2, a and b). Furthermore, CD1d-restricted δ/αβ T cells from donor 5 showed considerably brighter staining with 4′-deoxy-α-GalCer compared with CD1d–α-GalCer tetramer (Fig. 2, a and b). Reactivity of CD1d-restricted δ/αβ T cells against the OCH Ag showed that δ/αβ T cells from four donors (1, 3, 4, and 5) were capable of binding to this Ag (Fig. 2, a and b), whereas, as expected, most type I NKT cells failed to bind OCH (Matulis et al., 2010; Wun et al., 2011). Collectively, these data highlight that CD1d-restricted δ/αβ T cells are capable of recognizing glycolipids presented by CD1d, and the δ/αβTCR composition imbues these cells with a different pattern of glycolipid Ag specificity that distinguishes them from type I NKT cells from the same donors.

Having determined that some δ/αβ T cells were present within the CD1d–α-GalCer–reactive T cell population, we next examined δ/αβTCRs within the general population of Vδ1⁺ T cells in humans by analyzing freshly isolated PBMCs. Using the anti-Vδ1 antibody clone A13, we determined that many Vδ1⁺ T cells coexpress a TCR-β chain, rather than a TCR-γ chain, and are therefore δ/αβ T cells rather than γδ T cells (Fig. 3 a). The ratio of δ/αβ to γδ T cells within the Vδ1⁺ population varied widely, from <5% to >80% δ/αβ T cells, with a mean of ∼45% (Fig. 3 b). Further investigation of the cell surface phenotype of δ/αβ T cells, in comparison with αβ T cells, revealed that δ/αβ T cells can coexpress CD4 or CD8, although the ratio of CD4 to CD8 was generally different from that observed for αβ T cells, with more CD8⁺ and less CD4⁺ δ/αβ T cells than αβ T cells (P < 0.05; Wilcoxon paired ranked test). γδ T cells from the same donors were predominantly CD4⁻CD8⁻ or CD8⁺ (Fig. 3 c). TCR-Vβ profiling of these cells indicated that they express a broad range of TCR-β chains, and moreover, the representation of Vβ chains used by δ/αβ T cells did not necessarily parallel that of the αβ T cells from the same donors (Fig. 3 d). This is exemplified by the Vβ8 population, which ranged from 3 to 15% of αβ T cells in four donors, whereas δ/αβ T cells from the same four donors were <3% Vβ8⁺. Thus, these data highlight that Vδ1⁺ δ/αβ T cells are present in similar frequency to Vδ1⁺ γδ T cells in human peripheral blood. Moreover, aside from their unusual TCR, δ/αβ T cells are also distinct from both αβ and γδ T cells with regard to their CD4 or CD8 coexpression profiles and TCR-β repertoire.

Using flow cytometric sorting, we purified δ/αβ, αβ, and γδ T cells from several human donors and stimulated them in vitro for 3 d in the presence of anti-CD3– and CD28-coated beads to measure their relative ability to produce cytokines in response to TCR cross-linking. These results (Fig. 3 e) demonstrated that δ/αβ T cells are responsive to TCR cross-linking, resulting in abundant cytokine production, including IL-2, IL-4, IL-13, IL-17, IFN-γ, GM-CSF, and TNF, at levels comparable with that of αβ T cells and clearly distinct from that observed for both Vδ1⁺ and Vδ1⁻ γδ T cells from matched donors. The response from the γδ T cells was generally lower than that of the δ/αβ T cells for IL-2, IFN-γ, and TNF, whereas the Vδ1⁺ γδ T cells produced comparable GM-CSF and IL-13 to the δ/αβ T cells after anti-CD3/CD28 stimulation (Fig. 3 e). To examine the cytokine-producing potential of these different cell types further, we compared all four subsets after a brief (4 h) stimulation with PMA and ionomycin and measured IFN-γ and TNF by intracellular cytokine staining (Fig. 3, f and g). These data indicated that both Vδ1⁺ and Vδ1⁻ γδ T cells can produce abundant IFN-γ and TNF and thus were functionally similar using this method of stimulation.

We next wanted to determine whether direct ligation of the δ/αβTCR with defined Ag could result in cellular activation. To achieve this, we stably transduced a CD1d-restricted δ/αβTCR (clone 9B4) and an irrelevant pHLA-reactive TCR into the αβTCR-deficient Jurkat-76 cell line and cultured these cells in the presence of α-GalCer or α-GlcCer. As these cell lines express CD1d (not depicted), no Ag-presenting cells were added. CD69 up-regulation was used as an indicator of Ag-mediated cellular activation (Fig. 4, a and b). Type I NKT TCR (NKT15)–transduced SKW3 cells, which also express CD1d, were included as a positive control in this assay. The control HLA-restricted TCR-transduced cells did not respond to either glycolipid Ag. These experiments demonstrated that the δ/αβTCR was capable of recognizing α-GalCer Ag presented by CD1d and, furthermore, that this recognition event could transmit cellular activation signals (Fig. 4 a). Also, although NKT15 TCR-transduced cells were capable of recognizing α-GalCer and α-GlcCer equally, the δ/αβTCR-transduced cells were only capable of responding to α-GalCer (Fig. 4 b), consistent with our CD1d tetramer staining of CD1d-restricted δ/αβ T cells from in vitro expanded PBMCs (Fig. 2). Furthermore, the δ/αβTCR-transduced cells responded to three different analogues of α-GalCer with different acyl chains, including C24:1 (Fig. 4 a), C20:2 (Fig. 4 b), and C26 analogues (not depicted). Collectively, these data indicate that δ/αβTCRs are capable of transmitting activating signals in response to specific glycolipid Ags, resulting in cellular activation and diverse cytokine production.

To study the molecular basis for the binding of a δ/αβTCR to CD1d-Ag, we first determined the specificity of an isolated δ/αβTCR (9B4) for CD1d tetramer–glycolipid, using the same panel of glycolipid Ags as shown in Fig. 2. The pattern was clearly distinct from that of cells transduced with type I NKT TCR (clone NKT15), and cells transduced with an irrelevant pHLA-specific TCR showed no staining with any of the CD1d-Ag tetramers tested. Namely, 9B4 δ/αβTCR-transduced Jurkat cells recapitulated the pattern of reactivity observed for the δ/αβ T cells from donor 1 (Fig. 2), demonstrating strong staining with CD1d tetramers loaded with α-GalCer and 4′-deoxy-α-GalCer, moderate staining with 3′-deoxy-α-GalCer and OCH, and very little reactivity with α-GlcCer and sulfatide-loaded CD1d tetramer (Fig. 5 a).

We next generated a soluble version of this TCR using previously described methods (Kjer-Nielsen et al., 2006), and the soluble 9B4 δ/αβTCR exhibited a molecular mass of 48 kD and reacted with an anti-Vδ1 mAb and an anti-TCR Cα mAb, thereby indicating that the δ/αβTCR had refolded properly (not depicted). Surface plasmon resonance (SPR) was used to investigate the affinity of the interaction between the 9B4 δ/αβTCR and CD1d loaded with C26:0 α-GalCer (Fig. 5 b). The human NKT15 type I Vα24⁺ αβTCR (Kjer-Nielsen et al., 2006) was included as a positive control (Fig. 5 b). No notable autoreactivity to CD1d-endogenous was observed for the 9B4 δ/αβTCR, in stark contrast to the autoreactivity previously observed for the CD1d-restricted 9C2 γδTCR (Uldrich et al., 2013). The 9B4 δ/αβTCR bound to CD1d–α-GalCer with very high affinity (K_d = 66 nM), even compared with the type I NKT15 αβTCR (K_d = 190 nM; Fig. 5 b), and markedly higher than the 9C2 Vδ1⁺ γδTCR (K_d of 15 µM; Uldrich et al., 2013). This higher affinity of the 9B4 δ/αβTCR is partly caused by a slower dissociation rate (half-life of 24 s compared with 3.5 s for the NKT15 αβTCR). Thus, the 9B4 δ/αβTCR binds strongly to CD1d–α-GalCer and is highly sensitive to the type of CD1d-restricted lipid Ags presented.

To understand how the 9B4 δ/αβTCR recognized CD1d–α-GalCer, we determined the crystal structure of the ternary complex (Fig. 6 a and Table S1) as well as the nonliganded 9B4 TCR (Table S1). The 9B4 TCR docked orthogonally (73°) above the A′ pocket of CD1d, relative to the CD1d cleft, burying ∼1,050 Ų, and interacted with residues ranging from 58 to 79 of the α1 helix and from 153 to 171 of the α2 helix of CD1d (Fig. 6, a and b). This mode of recognition differed markedly from type I NKT TCR–CD1d–α-GalCer recognition, whereupon the Vα24-Vβ11 TCR sat parallel over the F′ pocket of CD1d (Fig. 6, a and b; Borg et al., 2007). Accordingly, the interatomic contacts and roles of the individual CDR loops at the type I NKT TCR–CD1d–α-GalCer interface were markedly different to those at the 9B4 TCR–CD1d–α-GalCer interface (Fig. 6 c). Indeed, the δ/αβTCR–CD1d–α-GalCer complex was more similar to the 9C2 γδTCR–CD1d–α-GalCer complex (Fig. 6, a–c), which sat orthogonally (83°) over the A′ pocket (buried surface area [BSA] of ∼950 Ų), although the 9B4 TCR variable regions were shifted more toward to the center of the CD1d cleft (shift of 7 Å for the TCR-β chain and 3 Å for the TCR-Vδ/α chain, relative to the 9C2 TCR-γ and -δ chains, respectively), thereby bringing the δ/αβTCR closer to the α-GalCer headgroup.

The δ/αβTCR-mediated recognition of CD1d was dominated by the Vδ1 chain (66% of the total BSA), relative to the Vβ2 chain (33% of the total BSA; Table S2). For the TCR-β chain, both the CDR1β and CDR3β made contributions to the interface (13% and 11% of the total BSA, respectively; Figs. 6 c and 7 a). Namely, the aliphatic moiety of Arg79 of CD1d was pincered by Gln29β from the CDR1β loop and the framework residue Leu77β (Fig. 7 a). In addition, Arg79 contacted Thr31β, which, together with the adjacent Thr32β, packed against Val72 from CD1d, with Thr32β and Asn57β also interacting with His68 of CD1d (Fig. 7 a). The CDR3δ/α loop (13% of the total BSA) extended toward the α2 helix of CD1d, with its main chain packed against α-GalCer (discussed below) and contacted residues from the α1 and α2 helices (Fig. 7 b). Namely, Trp160 of CD1d was positioned between Val108α and Thr109α, the latter of which also contacted Ile69 and Trp153 of CD1d (Fig. 7 b). Trp153 is a position that differs in mouse CD1d (Gly155; Godfrey et al., 2005), thereby providing a basis of why the δ/αβTCR does not cross-react onto mouse CD1d–α-GalCer (not depicted).

The Vδ1-mediated contacts with CD1d are dominated by the CDR1δ loop (38% of the total BSA), with further contributions from the framework region (15% BSA; Table S2). Although the CDR2δ loop did not interact with CD1d, the flanking framework regions did (Fig. 7 c). Here, Arg55δ extended down toward the α2 helix to form a salt bridge with Glu156, whereas Asp66δ and Lys71δ mediated contacts with the glycosylation site attached to Asn163 of CD1d (Fig. 7 c). These framework-mediated interactions with CD1d were analogous to those observed in the 9C2 γδTCR–CD1d-Ag complex (Uldrich et al., 2013).

The CDR1δ loop sat over the distal end of the A′ pocket of CD1d and exclusively contacted CD1d (Fig. 7 d). The interactions were dominated by two Trp residues (Trp29 and Trp30) that were flanked by Ser residues (Ser28δ and Ser31δ) whose side chains formed vdw interactions with CD1d, while their main chains hydrogen bonded with CD1d residues (Gln62, Gln168, and Trp160; Fig. 7 d). Trp29δ was wedged between the α1 and α2 helices of CD1d, packed against the aliphatic side chains of Gln62, Thr65, and Leu66 and the aromatic ring of Trp160 (Fig. 7 d). Trp30δ was positioned at the periphery of CD1d, stacked principally against Gln168 (Fig. 7 d). These CDR1δ loop–mediated contacts by the δ/αβTCR were very similar to those observed in the 9C2 γδTCR–CD1d-Ag structure (Fig. 7 e). Thus, the germline-encoded regions of the Vδ1 domain of the δ/αβTCR adopted a very similar mode of CD1d-Ag recognition as the γδTCR. This showed that Vδ1 is not only capable of binding CD1d when rearranged with a Dδ-Jδ–encoded CDR3δ motif (in the context of a γδTCR), but also when rearranged with a permissive Jα segment in the context of a δ/αβTCR.

The structure of 9B4 TCR in the unligated state enabled us to assess the extent of plasticity of the CDR loops upon CD1d–α-GalCer engagement. Upon ligation, the 9B4 TCR variable regions did not significantly change conformation (root-mean-square deviation [rmsd] 0.5 Å for both Vδ1 and Vβ2); however, surprisingly, the positioning of the constant regions was significantly altered by approximately a 17° shift in the elbow angle between the V and C domains in a direction toward the F′ pocket of CD1d (not depicted). In contrast, there was minimal change in the conformation of the CDR loops upon ligation (rmsd ranging from 0.3 to 1 Å), with the biggest alteration observed for Trp30 of the CDR1δ loop, which flipped upon ligation to avoid clashing with the α2 helix of CD1d (Fig. 7 f). Thus, analogous to αβTCR and γδTCR recognition of CD1d–α-GalCer, a rigid “lock-and-key” mechanism underpinned δ/αβTCR ligation (Borg et al., 2007; Pellicci et al., 2009; Uldrich et al., 2013).

The electron density at the interface and for the α-GalCer Ag in the 9B4 ternary complex was unambiguous (Fig. 8 a), thereby permitting a comparison of how three TCRs arising from distinct lineages, the αβTCR, the γδTCR, and the δ/αβTCR, interacted with the same Ag-presenting molecule bound to the same ligand. Although the 9B4 TCR-Vδ1 chain dominated the interaction with CD1d, α-GalCer recognition was mediated entirely by the TCR-β chain, specifically being sequestered by the CDR1β and CDR3β loops via a polar interaction network. This recognition is notably dominated by water-mediated contacts and interactions that involved the main chain of the δ/αβTCR (Fig. 8 b). Namely, all of the hydroxyl moieties of the α-GalCer headgroup, with the exception of the 2′-OH moiety, were involved in hydrogen bonding with the TCR-β chain of the δ/αβTCR. Specifically, Pro107β hydrogen bonded to the 4′-OH, whereas three water-mediated hydrogen bonds were formed between Gln29β, Thr31β, and Gly109β and the 4′-OH, 6′-OH, and 3′-OH moieties of α-GalCer, respectively (Fig. 8 b).

The structural basis for α-GalCer recognition by the δ/αβTCR was markedly distinct to that mediated via the 9C2 γδTCR and the NKT15 αβTCR (Fig. 8, b–d; Borg et al., 2007; Pellicci et al., 2009; Uldrich et al., 2013). Within the γδTCR–CD1d–α-GalCer complex, the CDR3γ loop exclusively contacted the α-GalCer headgroup, with Arg103γ and Tyr111γ hydrogen bonded to the 3′-OH and 4′-OH moieties (Fig. 8 c; Uldrich et al., 2013). In the NKT15 complex, the interactions with α-GalCer were exclusively mediated via the invariant TCR-α chain, where the α-GalCer headgroup sat underneath the CDR1α loop and abutted the CDR3α loop (Fig. 8 d; Borg et al., 2007; Pellicci et al., 2009). Here, the galactose 2′-OH, 3′-OH, and 4′-OH moieties were involved in polar-mediated contacts with the αβTCR. Accordingly, the αβTCR, γδTCR, and δ/αβTCRs contact the same lipid Ag via markedly distinct binding mechanisms.

We determined that there are many other δ/αβTCRs in healthy humans (Fig. 3), although the specificity of these cells is unclear. Nevertheless, previously, we had described a panel of T cell clones that were restricted to HLA-B*35:01 presenting a 9-mer epitope (IPSINVHHY [IPS]) originating from the pp65 Ag of CMV (Amir et al., 2011). We noted that one of these T cell clones, clone 12, expressed the same Vδ1-Jα52-Cα rearrangement as the 9B4 TCR, albeit with a different CDR3δ/α sequence and TCR-β chain gene usage (Vβ5.1-Jβ2.1). Thus, we formally investigated how this Vδ1⁺ δ/αβTCR mediated an MHC-restricted response.

Using SPR, we established that the clone 12 δ/αβTCR exhibited a 15 µM affinity for the HLA-B*35:01–IPS complex (not depicted), which fell within the normal affinity range observed for αβTCR–MHC-I interactions (Gras et al., 2012). Next, we solved the structure of the clone 12 δ/αβTCR in complex with HLA-B*35:01–IPS Ag at 3.0 Å resolution (Fig. 9 and Table S1). The clone 12 TCR docked 56° across the HLA-B*35:01–IPS cleft (Fig. 9 a), within the range of standard αβTCR–pHLA-I complexes (Gras et al., 2012). In addition, the clone 12 TCR ternary complex possessed features typically associated with αβTCR–pHLA-I complexes (Gras et al., 2012). For example, the BSA of the TCR interface was 1,014 Ų (total BSA of the TCR/pHLA interface = 2,040 Ų, within the range of αβTCR–pMHC-I interactions). The clone 12 δ/αβTCR predominantly used its TCR-δ chain (65% BSA) to engage the pHLA complex, where the interaction was dominated by the CDR1δ (27%), CDR3δ/α (20%), and Vδ1 framework (18% BSA) residues, whereas the CDR3β loop was the principal contributor (19% BSA) from the β chain (Fig. 9 b). Although the CDR3β loop mainly contacted the peptide Ag (discussed below), Tyr112β H bonded to Gln155, an MHC residue which is frequently contacted by αβTCRs (Burrows et al., 2010). The other contacts mediated via the β chain included a salt bridge (Arg30β to Glu76) and a hydrogen bond with Asn80 and two residues from the CDR2β loop mediating contacts. Thus the germline-encoded TCR-β chain–MHC contacts were very limited (Table S3 and not depicted).

The long CDR3δ/α loop (17 residues) formed an extended hairpin structure that sat above the HLA α1 helix (residues 65–72), while pointing away from the peptide (Fig. 10 a). Here, two residues from the N-region and two residues from the Jα gene segment were involved in hydrophobic interactions with HLA-B*35:01 (Table S3). Additionally, Thr115δ hydrogen bonded to Thr69 of the α1 helix (Fig. 10 a and Table S3). Interestingly, the clone 12 TCR framework residue Arg55δ interacted with the HLA-B*35:01 molecule in a similar fashion to that observed for the 9B4 interaction with the CD1d molecule (Fig. 8 c), whereby the Arg55δ points toward the α2 helix and hydrogen bonded to Ala150 (Fig. 10 b). In addition, Glu67δ formed a salt bridge with Arg151, and Asp66δ made hydrophobic interactions with Glu154, Gln155, and Ala158 (Fig. 10 b).

The CDR1δ loop, with its two large Trp residues (Trp29δ and Trp30δ), contacted the N-terminal side of the Ag-binding cleft of the HLA-I molecule on both helices (Fig. 10 c). Trp29δ packed against Arg62, whereas Trp30δ extended toward the α2 helix, stacking between the short aliphatic chains of Ala158 and Leu163. Furthermore, another aromatic residue from the CDR1δ loop, Tyr33δ, packed against Gln155. Accordingly, the CDR1δ loop played an extensive role in enabling this HLA-I–restricted response. Nevertheless, the placement of the CDR1δ loop atop the HLA-I and CD1d differed markedly (Fig. 10 d), indicating the versatility of the same germline-encoded region to interact with diverse Ag presentation platforms.

Although the TCR-δ chain of clone 12 TCR dominated the interaction with the HLA-I molecule, it made very limited contacts with the IPS peptide. Specifically, the contacts from the δ chain were focused solely onto P4-Ile, which formed hydrophobic contacts with CDR1δ and CDR3δ/α (Fig. 10 e). Notably, however, these Vδ1-driven contacts appeared to play a key role in interacting with P4-Ile, as mutation of the P4-Ile to an Ala dramatically decreased the binding affinity of the δ/αβTCR (K_d > 200 µM; not depicted). The majority of contacts to the IPS epitope were made by the Vβ chain, principally a framework residue (Arg66β) and the CDR3β loop (Fig. 10 f). Arg66β contacted the P8-His, and interestingly, Arg66β is found in the Vβ5, Vβ1, and Vβ6.8 genes only. The structure of Vβ1⁺ TCR also showed how this framework TCR-β chain residue directly contacted the Epstein-Barr virus–derived epitope (Gras et al., 2010). The CDR3β loop formed most of the interactions with the IPS peptide, contacting the P4-Ile, P6-Val, P7-His, and P8-His residues, and many of these peptide residues were important for the interaction (Fig. 10 f and Table S3). Here, the Tyr112β inserted its aromatic ring between the main chain of the P4-Ile and P6-Val, whereas Tyr113β sat atop the P7-His and contacted the P6-Val (Fig. 10 f). Furthermore, Glu109β contacted P7-His and hydrogen bonded to the main chain of P8-His (Fig. 10 f). Thus, the TCR-β chain plays a major role in mediating contacts with the peptide, with a key contribution from the CDR1δ loop. In summary, we have demonstrated that T cells expressing δ/αβTCRs can recognize both CD1d–glycolipid Ags as well as MHC–peptide Ags and have provided the molecular bases for how these interactions occur.

Since the discovery of the TCR genes in the 1980s (Hedrick et al., 1984), considerable insight into αβ T cell–mediated immunity, and to a lesser extent γδ T cell immunity, has been gleaned. αβ T cells are frequently associated with the adaptive immune response, whereas γδ T cells are considered to exhibit more innate-like features (Vantourout and Hayday, 2013). However, such functional division of these T cell subsets has become blurred. For example, although it was traditionally considered that αβ T cells recognized peptides presented by MHC molecules, innate-like T cells expressing semi-invariant αβTCRs can also recognize distinct nonpeptide-based Ags. Namely, CD1d-restricted type I NKT cells and MR1-restricted MAIT cells, upon activation by lipids and vitamin B metabolites, respectively, rapidly produce an array of cytokines (Rossjohn et al., 2012; Birkinshaw et al., 2014; Gapin, 2014). Furthermore, it is emerging that TCRs can adopt distinct, as well as similar, docking strategies in binding to peptide, lipid, or small molecule metabolites that are bound to their respective Ag-presenting molecules, thereby highlighting the adaptability of the αβTCR scaffold (Bhati et al., 2014).

Whereas most T cells either fall into the αβTCR⁺ or the γδTCR⁺ fraction, the existence of TCR gene rearrangements involving unusual combinations of α, β, γ, and δ chain genes have sporadically been reported in the literature (Hochstenbach and Brenner, 1989; Miossec et al., 1990, 1991; Peyrat et al., 1995; Bowen et al., 2014), although the significance of these unusual gene recombination events, the Ags to which they responded to, and the molecular bases for these interactions are unclear. Here, we have identified populations of Vδ1⁺ δ/αβ T cells, demonstrating that they are present with a similar frequency to Vδ1⁺ γδ T cells. Importantly, δ/αβTCRs appear to be functional, capable of transmitting activation signals to the T cells that express them. To understand the molecular bases for how these TCRs recognize Ag, we have focused on two distinct δ/αβTCRs, a CD1d-restricted α-GalCer–reactive δ/αβTCR and an HLA-restricted viral peptide–reactive δ/αβTCR.

We show that the δ/αβTCR architecture resembles that of αβTCRs and γδTCRs and provide a basis for how the Vδ1 chain can pair with the TCR-β chain. The binding of the δ/αβTCR to CD1d was dependent on the nature of the Ag bound, and the mode of recognition was markedly distinct from that of type I NKT TCR–CD1d–α-GalCer docking, more closely resembling that of the γδTCR–CD1d–α-GalCer complex (Uldrich et al., 2013). Indeed, the CDR1δ loop–mediated contacts between the δ/αβTCR and γδTCR were very similar (Uldrich et al., 2013), suggesting a critical role for this region of the TCR in the CD1d restriction of these cells. In contrast, the interactions of the δ/αβTCR with α-GalCer were markedly different from the γδTCR–CD1d–α-GalCer complex. Thus, although the CDR1δ loop appears to represent a focal point for the CD1d-restricted response, the nature of the docking mode is nevertheless fine-tuned by the pairing of the TCR-β or TCR-γ chain and the hypervariable CDR3 loops, in a manner which is strikingly unique to this hybrid TCR. Furthermore, this is the first example of a CD1d–α-GalCer–reactive TCR that does not bind with a parallel docking mode over the F′ pocket of CD1d, a characteristic of all αβTCR⁺ type I NKT cells (Rossjohn et al., 2012).

We also demonstrate how a δ/αβTCR can engage an HLA-I–peptide complex. Notably, the δ/αβTCR docking on CD1d-lipid and HLA-I–peptide differed, despite these two δ/αβTCRs sharing the same Vδ1-Jα52⁺ chain. Nevertheless, in both cases, the Vδ1 framework region, coupled with the Trp-rich CDR1δ loop, played a prominent role in contacting the respective Ag-presenting molecules. These interactions differed in their positioning on the Ag-binding platforms, thereby highlighting how the same germline-encoded region can play key, yet disparate, roles in binding to polymorphic and monomorphic Ag-presenting molecules.

In summary, our study highlights the existence, and molecular bases for Ag recognition, of T cells that express hybrid TCRs comprised of the Vδ1 variable region rearranged to diverse Jα genes and the Cα constant region and paired to a diverse range of TCR-β chains. These hybrid δ/αβTCR⁺ T cells are surprisingly abundant in humans and have unique characteristics both at the cellular and molecular level, thus adding a level of diversity that extends beyond the αβ T cell and γδ T cell lineages. Collectively, these findings represent a large conceptual advance in our understanding of T cell–mediated immunity.

The 9B4 TCR monomer, 9B4 TCR–CD1d–α-GalCer ternary complex, and clone 12 TCR–HLA-B*3501–IPS ternary complex were deposited in the Protein Data Bank (PDB) under the accession codes 4WNQ, 4WO4, and 4QRR, respectively.

Healthy human PBMCs were obtained from the Australian Red Cross, ethics approval 13-04VIC-07 (Australian Red Cross) and 1035100.1 (University of Melbourne). PBMCs were isolated using a histopaque-1077 (Sigma-Aldrich) density gradient. CD3⁺ CD1d–α-GalCer tetramer⁺ cells were MACS (Miltenyi Biotec) and FACS sort-enriched using a FACSAria III (BD). Enriched CD3⁺ CD1d–α-GalCer tetramer⁺ cells were expanded in vitro, essentially as previously described (Uldrich et al., 2013).

Enriched CD3⁺ CD1d–α-GalCer tetramer⁺ cells were stained with Vδ1 (clone A13 supernatant, which can bind to Vδ1 when incorporated in hybrid Vδ1-Jα-Cα TCR chains, was produced in L. Moretta’s laboratory), anti–mouse IgG (clone Poly 4053; BioLegend), 5% normal mouse serum, and then with antibodies specific for αβTCR (clone T10B9.1A-31; BD), CD3ε (clone UCHT1; BD), CD8α (SK1; BD), CD4 (RPA-T4; BD), Vβ11 (C21; Beckman Coulter), CD69 (FN50; BD), γδTCR (11F2; BD), and CD161 (191B8; Miltenyi Biotec). Cells were costained with 7-aminoactinomycin D viability dye (Sigma-Aldrich) and with human CD1d tetramers (produced in-house), as previously described (Uldrich et al., 2013). TCR-Vβ repertoire analysis was performed using a TCR-Vβ repertoire kit (Beckman Coulter). Cells were analyzed using an LSR Fortessa (BD), and data were analyzed using FlowJo software (Tree Star Inc.).

Lipids were dissolved in 0.5% Tyloxapol (Sigma-Aldrich) or Tween-sucrose-histidine buffer containing 0.5% vol/vol Tween-20, 57 mg/ml sucrose, and 7.5 mg/ml histidine. Lipids included α-GalCer C24:1 (PBS-44; provided by P. Savage, Brigham Young University, Provo, UT); α-GalCer C26 (Alexis); α-GalCer (C20:2), α-GlcCer C20:2, OCH, 3′,4″-dideoxy-α-GalCer (3′-deoxy-α-GalCer), 4′,4″-dideoxy-α-GalCer (4′-deoxy-α-GalCer; the deoxy analogues were provided by S. Keshipeddy and S. Richardson, University of Connecticut, Storrs, CT; Wun et al., 2011); and sulfatide C24:1 (Avanti Polar Lipids). For CD1d tetramer experiments, lipids were loaded into human CD1d at 3:1 or 6:1 lipid/protein molar ratio by overnight incubation.

CD1d–α-GalCer tetramer Vδ1⁺ CD3⁺ γδTCR⁻ cells were single cell sorted and cDNA isolated using SuperScript VILO (Invitrogen) as per the manufacturer’s instructions. Transcripts encoding for Vδ1 were amplified by two rounds of nested PCR, using Vδ1 external primer 5′-CAAGCCCAGTCATCAGTATCC-3′, Vδ1 internal primer 5′-CAACTTCCCAGCAAAGAGATG-3′, Cα external primer 5′-GACCAGCTTGACATCACAG-3′, and Cα internal primer 5′-TGTTGCTCTTGAAGTCCATAG-3′. Transcripts encoding for the different Vβ genes were identified using multiplex nested PCR as previously described (Wang et al., 2012). PCR fragments were separated using a 1.5% agarose gel and DNA sequenced by Molecular Diagnostics, University of Melbourne.

CD1d-restricted δ/αβTCR⁺ Jurkat cells (9B4), CD1d-restricted type I NKT TCR⁺ Jurkat or SKW3 cells (NKT15), and control HLA peptide–specific TCR⁺ Jurkat cells were generated essentially as previously described (Uldrich et al., 2013). Parental Jurkat-76 cells and SKW3 cells were provided by L. Kjer-Nielsen (University of Melbourne, Parkville, Victoria, Australia).

SPR experiments were conducted at 25°C on either a BIAcore 3000 (Biacore) or ProteOn XPR36 (Bio-Rad Laboratories) instrument with HBS buffer (10 mM Hepes, pH 7.4, 150 mM NaCl, and 0.005% surfactant P-20). In some instances the HBS buffer was supplemented with 1% bovine serum albumin to prevent nonspecific binding. The human αβTCR–specific mAb (clone 12H8; Borg et al., 2005) or streptavidin was coupled to CM5 or GLC chips with standard amine coupling, respectively. Experiments were conducted as previously described (Gras et al., 2009; Pellicci et al., 2009). BIAevaluation version 3.1 (Biacore) or ProteOn Manager version 2.1 (Bio-Rad Laboratories) software was used for data analysis with 1:1 Langmuir binding model. The equilibrium data were analyzed using Prism (GraphPad Software).

Soluble TCRs, including CD1d-restricted δ/αβTCR (9B4), type I NKT TCR (NKT15), and HLA-B*35–IPS (clone 12), were generated as previously described (Kjer-Nielsen et al., 2006). In brief, gene fragments encoding the TCR ectodomains were cloned into pET30 expression vector (EMD Millipore) and expressed in BL-21 Escherichia coli cells. Inclusion body protein was prepared and refolded similar to that previously described in Clements et al. (2002) except in the presence of 5 M urea. Human CD1d with and without a Bir-A tag were expressed and purified as previously described (Borg et al., 2007; Uldrich et al., 2013). Soluble pHLA-B*35:01 was prepared as described previously (Gras et al., 2010). The 9B4 TCR–CD1d–α-GalCer and clone 12 TCR–HLA-B*35:01–IPS ternary complexes were purified by gel filtration from mixtures of the respective monomers.

Solutions of the 9B4 TCR monomer (at 5 mg/ml in 10 mM Tris-HCl, pH 8.0, and 150 mM NaCl) and 9B4 TCR-CD1d–α-GalCer ternary complex (at 10 mg/ml in 10 mM Tris-HCl, pH 8.7, and 150 mM NaCl) were crystallized at 20°C in 27% PEG 1500 plus 0.1 M sodium malonate/imidazole/borate buffer (at a 2:3:3 molar ratio), pH 8.1, in hanging drops and 3% PEG 3350 in sitting drops, respectively. Plate-like crystals of the clone 12 TCR–HLA-B*35:01–IPS complex (at 4 mg/ml) were grown at 20°C with 0.3 M potassium thiocyanate, 24% PEG 3350, 2% ethylene glycol, 10 mM l-Proline, and 10 mM spermidine. Crystals of 9B4 monomer and CD1d ternary complex were flash frozen in well solution plus 2-methyl-2,4-pentanediol, whereas the clone 12–IPS ternary complex crystals were soaked in mother liquor with 25% PEG 3350 and 5% (wt/vol) ethylene glycol as cryoprotectant and then flash frozen in liquid nitrogen. Data were collected at 100 K on the MX1 and MX2 beam lines at the Australian Synchrotron (Melbourne). Data were integrated using iMosflm or XDS (Kabsch, 2010) software and scaled using the SCALA package of the CCP4 suite. The 9B4 TCR–CD1d–α-GalCer ternary complex was solved by molecular replacement using PHASER with three separate search ensembles: CD1d without α-GalCer, TCR minus the Vα domain and CDRβ loops, and Vδ1 minus the CDRδ loops (derived from PDB codes 1ZT4 [Koch et al., 2005], 4DZB [Reantragoon et al., 2012], and 4LFH [Uldrich et al., 2013], respectively). The structure of unligated 9B4 TCR was solved by molecular replacement using the refined 9B4 TCR structure derived from the ternary complex. There were two TCR monomers in each asymmetric unit with an rmsd of 0.1 Å over 432 Cα atoms; thus, structural analysis was restricted to one TCR molecule. The clone 12 TCR–HLA-B*35:01–IPS complex was solved as above, using the 9C2 TCR (Uldrich et al., 2013) and HLA-B*3501 without the peptide (PDB code 3LKN [Gras et al., 2010]) as search models. All structures were refined into the experimental maps by iterative rounds of model building using COOT (Emsley et al., 2010) and refinement with BUSTER (Bricogne et al., 2011). The electron density at the Ag interfaces was clear, with unbiased electron density for IPS peptide, the α-GalCer headgroup, and all CDR loops of the 9B4 and clone 12 TCR. Structural models were validated at the Research Collaboratory for Structural Bioinformatics Protein Data Bank Data Validation and Deposition Services website, and molecular illustrations were generated using the PyMOL package (DeLano, 2002). Domain superpositions and rmsd calculations were based on Cα atoms.

Table S1 shows data collection and refinement statistics. Table S2 shows 9B4 δ/αβTCR contacts with CD1d–α-GalCer. Table S3 shows clone 12 δ/αβTCR contacts with HLA-B*35:01–IPS.

We thank John Waddington and Marcin Ciula at the University of Melbourne (Parkville, Victoria, Australia) for assistance with protein production, the staff at the Australian Synchrotron for assistance with data collection, and the Macromolecular Crystallization Facility at Monash University for technical assistance. We thank Dr. Richard Birkinshaw for assistance with data collection and integration, Professor Paul Savage for kindly providing the PBS-44 α-GalCer analogue, and Santosh Keshipeddy and Stewart Richardson for the deoxy-α-GalCer analogues.

This work was supported by the Australian Research Council (ARC; DP110104124, CE140100011, and LE110100106), the National Health and Medical Research Council of Australia (NHMRC; #1013667), the Cancer Council of Victoria (#1042866), and National Institutes of Health RO1 grant (GM 087136). D.G. Pellicci is supported by an NHMRC Biomedical fellowship, A.P. Uldrich by an ARC Future Fellowship (FT140100278), S. Gras by an ARC Future Fellowship (FT120100416), D.I. Godfrey by an NHMRC Senior Principal Research Fellowship (#1020770), and J. Rossjohn by an NHMRC Australia Fellowship (AF50).

The authors declare no competing financial interests.

Author contributions: D.G. Pellicci and A.P. Uldrich identified and performed cellular and molecular characterization of δ/αβ T cells. S.B.G. Eckle, R. de Boer, D.G. Pellicci, A.P. Uldrich, S. Gras, and E. Chabrol produced TCR protein complexes for crystallographic experiments. A.P. Uldrich, J. Le Nours, E. Chabrol, and S. Gras solved the crystal structures. E. Chabrol, M.H.M. Heemskerk, R. de Boer, S.B.G. Eckle, and J. McCluskey undertook or supervised experiments on the MHC-restricted δ/αβTCR. A.P. Uldrich, D.G. Pellicci, and S. Gras undertook SPR investigations, wrote the paper, and generated figures. D.G. Pellicci, A.P. Uldrich, F. Ross, K. McPherson, and R.T. Lim performed cell-based experiments, including glycolipid specificity and functional studies. L. Moretta, A.R. Howell, and G. Besra provided reagents that were crucial to this study. J. Rossjohn and D.I. Godfrey were joint senior authors: they co-led the investigation, devised the project, and wrote the paper.