Structural and kinetic basis for heightened immunogenicity of T cell vaccines

Structural studies were performed using soluble, purified TCR (6) from the NY–ESO-1_157–165–specific 1G4 CTL clone (hereafter 1G4 TCR). The 1G4 TCR–A2–ESO 9C complex crystal diffracted to 1.9 Å resolution and contained one complex per asymmetric unit (Table I). A 1.4-Å resolution crystal structure was also determined for the unliganded 1G4 TCR. In the complex structure the orthogonal binding orientation of the 1G4 TCR with respect to the long axis of the ESO 9C peptide (69°) lies at one extreme of the range observed, to date, for TCR–pMHC class I interactions and is very similar to the orientation observed for the JM22 TCR–A2–influenza matrix peptide complex (20). These two structures, plus that recently reported for the TCR AHIII–A2–p1049 complex (21), establish that there is significant overlap between the binding orientations for class II– and class I–restricted TCR recognition.

The 1G4–A2–ESO 9C complex displays a prominent methionine–tryptophan (MW) “peg” in the central P4/P5 portion of the peptide, which forms a large protrusion from the surface of the pMHC (Fig. 1, A and B). The 1G4 TCR envelops this protrusion between its complementarity determining region (CDR) loops, making contacts via the CDR3α,CDR3β and CDR1α loop residues. In previous pMHC–TCR complex structures, a single upward pointing residue from the peptide has often formed a peg about which the TCR CDR loop residues cluster making contacts conferring specificity and affinity (5, 22, 23). In this case, the peg is composed of a double side chain motif. This bulky and hydrophobic structure provides the majority of interactions between the epitope and the 1G4 TCR. More generally, given its central position within the TCR recognition surface, it seems probable that the MW peg constitutes the dominant antigenic feature of this peptide (Fig. 1 A).

The total solvent-accessible surface area buried at the 1G4 TCR–A2–ESO 9C interface is ∼1,440 Ų with the TCR contribution being split relatively equally between α and β chains, 46% and 54%, respectively. Of this total 730 Ų (∼50%) results from TCR–peptide interactions with 49% of the buried surface area between the ESO 9C peptide and the 1G4 TCR accounted for solely by interactions involving the MW side chains (360 Ų). Analyses of the shape complementarity at the interface between the 1G4 TCR and (a) the pMHC (S_c = 0.71), (b) the entire ESO 9C peptide (S_c = 0.83), and (c) specifically the MW side chains (S_c = 0.87) demonstrate that there is a very high degree of shape complementarity to the epitope, in particular to the MW peg. Rudolph and colleagues calculated S_c values ranging from 0.41 to 0.66 (with most clustered between 0.61–0.66) for the nine TCR–pMHC complex structures available in 2002 (5). Where the resolution of the x-ray diffraction data is sufficient for conserved water molecules to be included in the structure, these are frequently found to be integral to the pMHC–TCR interactions (12, 20, 24). For example, in the JM22–flu matrix_58–66–HLA–A2 complex S_c equals 0.73, with interfacial waters included in the calculation, and 0.64 without (20). The 1G4 TCR–A2–ESO 9C interface is notable in that it achieves high S_c values without any contribution from interfacial water, because the interface between the MW side chains and the TCR CDR3αCDR3β and CDR1α loops contains no trapped water molecules. The core of the 1G4 TCR–A2–ESO 9C interface is stabilized by hydrophobic and ring-stacking interactions, consistent with the hydrophobic and aromatic nature of the peptide M and W side chains (Fig. 1, C and D). Instead, water molecules cluster at the periphery of this pMHC–TCR interface, forming a belt associated with the peptide, TCR and A2, which may contribute to the thermodynamic properties of 1G4 TCR recognition (unpublished data).

Previous studies that have compared TCR structures in unbound and complexed states (25–29) report remarkably large (up to 15 Å) conformational changes in the CDR3 α and β loops. Although the 1G4 TCR uses a substantial cavity to accommodate the MW peg between its CDR1α, CDR3α, and CDR3β loops (Fig. 1 C), this cavity is preformed (Fig. 1 E). In contrast with the mechanisms of induced TCR fit reported for other pMHC–TCR complexes, a well optimized binding surface is achieved by relatively small structural changes. Comparison of the unliganded and liganded 1G4 TCR structures reveals only relatively minor main chain shifts in all the CDRβ loops (for example, <1.6 Å for V96β at the tip of the CDR3β; Fig. 1, C and E). The largest main chain shifts (up to 4.2 Å) in the Vα domain also occur at the tip of the CDR3 loop resulting from a change in the main chain conformation of residues 95–99. Only subtle adjustments in preexisting (unliganded) side chain conformations appear to be required to optimize the interaction of Y100α with the peptide at M4 and ring-stacking interactions between Y31α and the W5 side chain. However, these TCR residues show a considerable reduction in flexibility on engagement with the MW peg (average crystallographic temperature factors for the side chain atoms of Y31α are 36 Ų unliganded and 23 Ų liganded, and for Y100α the values are 50 Ų unliganded and 22 Ų liganded).

Because the 1G4 TCR recognition appears to be so focused on the central region of the peptide, we wished to assess whether substitutions of the carboxyl-terminal residue of the NY–ESO-1 peptide could alter the peptide structure and affinity of interaction with the 1G4 TCR. Because we have previously shown that substitution of cysteine to valine at position 165 of the NY–ESO-1_157–165 epitope enhanced its ability to sensitize target cells for killing by the 1G4 CTL clone (19), we decided to focus further structural studies on the 9V NY–ESO-1 peptide analogue (ESO 9V).

The crystal structure of the 1G4 TCR–A2–ESO 9V complex was determined at 1.7 Å resolution from crystals of the same space group and essentially identical unit cell dimensions to those used for the 1G4–A2–ESO 9C structure. To a first approximation the two complex structures are very similar. Structural comparison reveals that substitution of the cysteine to valine causes little difference in the general orientation of the side chain within the F pocket of the A2 peptide binding groove. Changes in the positions of the A2 side chains constituting the F pocket are also minimal. However, the relatively high resolution of both the 1G4–A2–ESO 9C and 1G4–A2–ESO 9V crystal structures permits a detailed analysis (Fig. 1 F). It is evident that the ESO 9V side chain is buried deeper into the F-pocket and occupies a slightly larger volume than the cysteine side chain. Both the shape complementarity and the buried surface within the F-pocket is greater for ESO 9V compared with ESO 9C (side chain surface areas: ESO 9V = 191 Ų, ESO 9C = 162 Ų and shape complementarities: ESO 9V = 0.81, ESO 9C= 0.72). These changes are sufficient to enhance significantly the van der Waal's contacts between the altered peptide ligand and A2, in line with this MHC class I molecule's reported preference for valine as COOH-terminal anchor residue (30).

The changes in the position of the COOH-terminal anchor residue, resulting from the C9V substitution, propagate through the peptide main chain resulting in the P6–P8 region being positioned slightly deeper (0.3 Å) within the A2 peptide binding groove. Although the P6–P8 region of the peptide has fewer contacts with the 1G4 TCR than the MW peg at P4 and P5, the structural difference in the peptide is clearly transmitted to the TCR (Fig. 1 F). The angle of engagement made by the 1G4 TCR differs between the peptide complexes by some 0.9°. This is essentially a change in the TCR's tilt, relative to an axis drawn perpendicular to the pMHC binding surface, in response to the lower position of the main chain COOH-terminal half of the ESO 9V peptide. Numerous small changes are discernible at the complex interface. These changes are focused on residue Q155 of the A2 α2 helix. For the A2–ESO 9V complex the Q155 side chain is well ordered and makes hydrogen bonds to the Oγ atoms of S52 and S53 side chains. In contrast, for the A2–ESO 9C complex the Q155 side chain is relatively disordered and appears to be sampling an additional conformation, forming a hydrogen bond to T95α on the CDR3α loop of the 1G4 TCR. The change in TCR tilt of the A2–ESO 9V complex precludes the latter option and the CDR3α loop adopts a less well ordered structure. The average B-factor for residues in the tip of CDR3α loop (all atoms in Ser96 α and Gly97 α) are 33.93 and 42.46 for the A2–ESO 9C and A2–ESO 9V complexes, respectively. Analysis of the shape complementarity between the 1G4 TCR and the pMHCs suggests a slightly better overall fit is achieved with A2 /ESO 9V (S_c = 0.73) compared with A2–ESO 9C (S_c = 0.71). The significance of this difference in S_c is discussed in Materials and methods. The difference includes tighter interactions between TCR and variant peptide (ESO 9V peptide S_c = 0.85; ESO 9C peptide S_c = 0.83).

Our comparison of the ESO 9C- and ESO 9V-based complex structures revealed the substitution of cysteine at position 9 by valine provides an improved fit to the A2 F pocket, a result that might be expected on the basis of established anchor preferences. However, this substitution, at a position buried within the binding groove, was also revealed to trigger subtle, but significant, changes at the TCR–pMHC interface. To correlate these structural features of the TCR–pMHC complexes with kinetics of binding, we assessed the affinity of binding of the ESO 9V and ESO 9C peptides to A2 molecules, and the affinity of 1G4 TCR binding to A2 molecules loaded with either the ESO 9V peptide or the wild-type ESO 9C peptide. Binding of ESO 9V peptide to A2 molecules was compared with that of ESO 9C by measuring peptide binding to metabolically labeled A2 molecules in T2 cells (31). Measurement of the optical density of metabolically labeled A2 SDS gel bands showed that ESO 9V peptide can stabilize a higher proportion of A2 molecules than the ESO 9C peptide (Fig. 2 A).

To confirm these results, experiments were performed using the 3M4E5 Fab antibody specific to NY–ESO-1_157–165–A2 complexes (32). Initial Biacore measurements demonstrated that the 3M4E5 Fab antibody recognizes A2 molecules loaded with either ESO 9V or ESO 9C peptide with a similar affinity of ∼60 nM (unpublished data). FACS staining with the 3M4E5 antibody of T2 cells pulsed with different concentrations of ESO 9V and ESO 9C peptides confirmed that ESO 9V peptide binds to A2 molecules with greater efficiency than the ESO 9C peptide (Fig. 2 B).

Surface plasmon resonance affinity measurements were performed using the soluble form of 1G4 TCR in order to compare the affinity of binding of a NY–ESO-1_157–165–specific TCR to A2 molecules loaded with either the wild-type ESO 9C peptide or the ESO 9V analogue. The experiments demonstrated a higher affinity of binding for the 1G4 TCR to the A2–ESO 9V complex than to the A2–ESO 9C complex (Fig. 2 C). Analysis of the binding curves of conformation-specific anti-A2–peptide and anti-β2M antibodies confirmed that equivalent amounts of correctly refolded A2–peptide complexes were immobilized to the Biacore chips (Fig. S1, A and B). Good agreement was observed between the affinities determined kinetically (the ratio of K_off to K_on) and those determined by equilibrium measurements. The equilibrium binding constants (K_d) at 25°C for the soluble 1G4 TCR binding to A2 loaded with the ESO 9V or ESO 9C peptide were determined to be 5 and 13.3 μM, respectively (Fig. 2 C). Kinetic measurements demonstrated that the higher affinity of binding was predominantly due to a slower off rate for the 1G4 TCR when bound to the A2–ESO 9V complex compared with the A2–ESO 9C complex. The slower off-rate is in line with the structural data, which indicate that more pMHC–TCR contacts form at the ESO 9V interface than at the ESO 9C interface.

To assess the functional consequences of the differences in pMHC and TCR–pMHC binding affinities, we analyzed the rate of polarization of cytotoxic T cell granules on stimulation of a NY–ESO-1_157–165–specific CTL clone with target cells pulsed with either the ESO 9C or the ESO 9V peptides. These experiments were performed with the same 1G4 CTL clone for which the structural and kinetic studies had been derived.

When T2 cells are pulsed with 1 μM ESO 9C peptide (Video S1 and Fig. 3, A–E), lytic granule polarization is slow. Fig. 3, A–E, shows two CTLs from contact to polarization. In one case (white arrow) granule polarization from the periphery (Fig. 3 B) to the synapse (Fig. 3 E) occurs within 17 min after the initial contact. In the other example (red arrow), the CTL first crawls under one target (Fig. 3, A and B) before contacting a new target (Fig. 3 C) without polarization of the lytic granules toward either target in >22 min. In contrast, granule polarization is rapid when T2 targets are pulsed with 1 μM ESO 9V peptide (Video 2 and Fig. 3, F–L). For one CTL, from the first encounter (Fig. 3 F, green arrowhead) to granule polarization (Fig. 3 L) the elapsed time is 4 min. Other CTLs in the field being filmed (Fig. 3 L, black arrowheads) had already polarized their granules toward targets by the time images were captured.

CTL and target cells were mixed with different concentrations of the two peptides, incubated at 37°C for 30 min and fixed in order to provide quantitative data with a large number of conjugates. Lytic granules were labeled with antibodies against cathepsin D (Fig. 4 A, a–c, green) and the p58 Golgi protein (Fig. 4 A, a–c, red). Both the Golgi apparatus and granules polarize upon CTL recognition of target (33), and the degree of polarization in CTL conjugated to targets were counted according to whether granules and Golgi proteins were located (a) at the rear of the CTL (Fig. 4 A, a, unpolarized); (b) at the side of the CTL (Fig. 4 A, b, polarizing); or (c) polarized at the immunological synapse (Fig. 4 A c). When T2 cells were pulsed with the ESO 9C peptide polarization was poor (Fig. 4 B) and 1 μM of peptide was required to see 30% of conjugates with tightly polarized granules at the synapse. By comparison, for ESO 9V granule polarization is much more efficient and only 0.01 μM ESO 9V peptide is required to see 30% of polarized CTL (Fig. 4 C). The higher percentage of target cell killing (19) and secretion of lymphokines (Fig. 5) may therefore reflect the shorter times required for CTL to recognize and destroy individual targets.

We sought to assess whether the CD8 requirement of the 1G4 NY–ESO-1_157–165–specific CTL clone differed for the recognition of wild-type ESO 9C or ESO 9V analogue peptides (Fig. 5). We show that recognition of the ESO 9V peptide is less CD8 dependent than recognition of the ESO 9C peptide, because CTL recognition of ESO 9C was significantly more efficient in the presence of A2 molecules bearing a CD8 binding site. These results also confirm that the 1G4 CTL clone is up to 100 times more sensitive to the ESO 9V peptide compared with the ESO 9C peptide.

We assessed whether the ESO 9V peptide is capable of expanding in A2 transgenic mice greater numbers of NY–ESO-1_157–165–specific CTL that are cross reactive to the native epitope. The results show that recombinant vaccines encoding the ESO 9V peptide are capable of inducing stronger NY–ESO-1_157–165 responses than vaccines encoding the wild-type ESO 9C peptide, both in terms of percentages of responding mice and frequency of the T cell response induced (Fig. 6 A). Furthermore, ESO 9V–primed mice were capable of recognizing the wild-type ESO 9C peptide, as defined by ex vivo tetramer staining with A2K^b tetramers loaded with the ESO 9C peptide (Fig. 6 B). The proportion of NY–ESO-1_157–165–specific CTL, stained by A2K^b tetramers loaded with either the ESO 9C or the ESO 9V peptide was very similar. Consistent with these findings, in vitro killing assays showed that NY–ESO-1_157–165–specific CTL, generated from ESO 9V–immunized mice, were capable of lysing target cells loaded with wild-type ESO 9C peptide, as well as target cells loaded with the ESO 9V analogue (Fig. 6 C). IFN-γ ELISPOT assay was used to assess further cross reactivity and the function of these CTL. CTL generated from animals immunized with the ESO 9V sequence were capable of producing IFN-γ in response to ESO 9C peptide. As expected, responses were enhanced in the presence of the reducing agent TCEP, which prevents cysteinylation and dimerization of ESO 9C peptide (unpublished data (19, 34). Finally, a larger proportion of ESO 9V–primed T cells was capable of secreting IFN-γ, when stimulated by target cells pulsed with ESO 9C peptide, than ESO 9C primed T cells (Fig. 6 D). Enhanced IFN-γ response by T cells from ESO 9V–primed mice was not due to the higher numbers of ESO 9V–primed T cells, because the numbers of splenocytes from ESO 9V and ESO 9C primed mice added to the ELISPOT plates were normalized for the percentage of ESO 9C tetramer positive T cells.

Previous studies (35) have raised the issue that changes in peptide anchor residues can directly affect TCR recognition, as well as pMHC stability. Such effects have important implications for T cell cancer vaccines. A2–ESO 9C represents a clinically relevant system in which we can assess the factors that will need to be considered when engineering vaccine epitopes. Because it is known that cysteinylation may reduce antigenicity of synthetic peptides binding to MHC class I molecules (34), we sought to study whether modification of cysteine 165 in the NY–ESO-1_157-165 epitope would increase its immunogenicity. Our results demonstrate that modification of cysteine to valine at position 165 of NY–ESO-1_157–165 both augments epitope binding to MHC and heightens TCR affinity. These two properties act in concert to generate stronger T cell cytolysis and activation, without compromising cross-reactivity with the wild-type peptide. Valine, while optimizing the anchoring of the peptide in the A2 F pocket, does not alter the orientation and height of the peptide's prominent methionine–tryptophan TCR binding motif. However, subtle repositioning of the peptide main chain introduces some small but significant changes at the pMHC–TCR interface which appear sufficient to generate a lower TCR K_off rate.

To elicit effective antitumor responses, it is important to develop strategies capable of reversing T cell tolerance to tumor antigens and jump starting the immune response to expand rapidly greater numbers of tumor-specific CTL. Although previous papers have correlated higher affinity of peptide binding to MHC class I molecules and TCR with increased immunogenicity (36–40), none of the previous studies have compared functional data to structural and kinetic analyses.

The use of a Fab antibody specific to the A2/NY–ESO-1_157–165 peptide complex (32) permits comparison of the efficiency of ESO 9C and ESO 9V binding to surface expressed A2 molecules (Fig. 2 B). The results of these experiments are consistent with ∼100-fold difference in binding affinity to A2 molecules between the ESO 9C and ESO 9V peptides. They also show that, within the range of peptide concentrations used for the analysis of granule polarization (Figs. 3 and 4) and cytokine secretion (Fig. 5), cells pulsed with ESO 9V peptide present to the 1G4 CTL clone approximately three- to fivefold more peptide than cells pulsed with the ESO 9C peptide. These results suggest that the enhanced effector functions of the 1G4 CTL clone stimulated by ESO 9V peptide can be accounted for by a combined effect of higher peptide affinity to A2 molecules and slower off rate of the 1G4 TCR from the ESO 9V–A2 complex. These results are consistent with previously published data demonstrating enhanced antigen-specific antitumor immunity with altered peptide ligands (40), although for the system studied by Slansky and colleagues the effect is entirely due to a threefold slower pMHC–TCR off rate. Although lytic granule polarization formed in response to maximal pMHC densities has been extensively characterized, fewer reports have examined variations in the efficiency of lytic granule polarization with peptide density (41). The results of our confocal microscopy experiments (Fig. 4) highlight variations in the efficiency of lytic granule polarization over a range of different peptide concentrations and different stability of TCR–pMHC complexes.

Increased immunogenicity in A2 transgenic mice and enhanced expansion of NY–ESO-1–specific CTL from melanoma patients' PBL demonstrate that the enhanced stimulation of the clone 1G4 can be extended to a polyclonal population. Although we cannot rule out the possibility that in vivo results can mainly be accounted for by the higher affinity of binding of the ESO 9V peptide to A2 molecules, it is of interest that NY–ESO-1_157–165–specific T cells from ESO 9V–primed mice have an enhanced ability to recognize target cells pulsed with the ESO 9C peptides, as compared with NY–ESO-1_157–165–specific T cells from ESO 9C primed mice (Fig. 6 D). It is possible that this enhanced T cell activity could be due to the increased stability of the ESO 9V–A2 complex resulting in differences at the level of the immunological synapse at the T cell–DC interface and in the expansion of better quality T cells. Consistent with this possibility, differences in the molecular composition of the immunological synapse between CTL and target cells have recently been described at different peptide concentrations (41), resulting in different signaling pathways and activation thresholds in individual CTL.

It appears, therefore, that the “consensus optimization” of the NY–ESO-1_157–165 F-pocket anchor side chain to valine can achieve increased T cell stimulation by augmenting the peptide affinity to A2, while maintaining, and in at least one case enhancing TCR recognition by decreasing the dissociation rate constant for TCR binding. Although the enhanced stability of the A2–ESO 9V variant was expected, our combined structural and functional analysis underscores the importance of also assessing the indirect effect of anchor residue changes on the TCR recognition surface.

The central role of the MW peg in our complex structures is consistent with results obtained for an alanine scan of the ESO 9C peptide using NY–ESO-1_157–165–specific T cell lines (42). A recent paper by Webb and colleagues confirmed the presence of this feature in the isolated A2–ESO 9C structure (43). We have extended the analysis of Webb et al. by determining a high resolution structure (1.5 Å) of unliganded A2 molecules loaded with the NY–ESO-1_157–165 peptide analogue bearing a leucine at position 165 (A2–ESO 9L), which demonstrates that the overall structure of the ESO 9L epitope in A2 is similar to the structure of the A2–ESO 9V and A2–ESO 9C complexes (unpublished data). It is of interest, however, that for isolated structures the MW side chains are more disordered than in the TCR complexes, consistent with the possibility that binding of the 1G4 TCR stabilizes the MW peg. Our study has primarily focused on TCR binding in assessing factors to be considered when engineering peptide analogs. However, CD8 binding to class I molecules is important for stabilizing the interaction between low affinity TCR and class I molecules (44) and for controlling the dynamics of CTL activation and immunological synapse formation (45). Consistent with these findings, we have shown that higher affinity recognition of the ESO 9V peptide by the 1G4 TCR is less CD8 dependent than recognition of the ESO 9C peptide. This alteration in CD8 dependency does not require any change in TCR docking orientation and therefore does not conform to the model suggested by Buslepp and colleagues (46).

The amount of signal that T cells receive by interacting with APCs is determined by several parameters including the concentration of pMHC complex (class I–peptide binding affinity) and the duration of the interaction between T cells and APCs (TCR binding affinity). From this concept it follows that the success of vaccine strategies will be dependent on our ability to fine tune several parameters to induce strong T cell expansions. The use of superagonist analogs in clinical applications provides an opportunity to address some of the shortcomings of current vaccination strategies, as it has been shown that peptide analogs can induce stronger responses than wild-type peptides in vivo (36–40). Peptide analogs are particularly relevant in the context of recombinant viral vector vaccination strategies, because immunodominance of viral vector–specific responses often impairs their ability to elicit immune responses specific for recombinant epitopes. It is possible that the low frequency of recombinant epitope–specific CTL responses may be due to competition from immunodominant viral epitope–specific responses, underscoring the need for engineering high-affinity peptide analogs for recombinant virus vaccine approaches.

In conclusion, our results highlight the indirect effects of anchor residue substitutions on TCR binding, effects that are of substantial significance for the rational optimization of vaccination strategies. The sensitivity of TCR recognition can extend to detect differences in buried anchor side chains, resulting in altered TCR binding kinetics and different levels of activation signals. Our results suggest the importance of incorporating structural data into the process of identifying candidate superagonist peptide analogs. Identification of highly antigenic NY–ESO-1 peptide analogs is of importance for the development of vaccines capable of expanding NY–ESO-1–specific CTL in cancer patients. Phase 1 clinical trials using NY–ESO-1 synthetic protein, peptides and recombinant viruses are already in progress aimed at eliciting tumor-specific T cell responses. The use of the NY–ESO-1 9V analogue should be considered in future clinical trials, because it could result in a more efficient induction of NY–ESO-1_157–165–specific CTL in cancer patients.

Residues 1–278 of the A2 heavy chain with the COOH-terminal BirA tag were expressed in Escherichia coli as inclusion bodies and refolded and purified with the NY–ESO-1 peptides SLLMWITQC or SLLMWITQV and β₂ m as described previously (for example see reference 47).

Expression and purification of the 1G4 NY–ESO-1 TCR. The 1G4 TCR was refolded and purified from E. coli derived inclusion bodies as described in reference 6.

Initial crystallization screens for the 1G4 TCR (10 mg/ml), 1G4–A2-ESO 9C or 1G4–A2-ESO 9V complexes (1:1 TCR–pMHC ratio at 10 mg/ml) used 100 nl + 100 nl drops dispensed by a Cartesian Technology Microsys MIC4000 (using Greiner 96-well protein crystallization plates), were stored in a automated vault and imaged with Veeco imaging systems. All crystallizations were done by the sitting drop vapor diffusion technique. Crystals of the isolated 1G4 TCR were grown at room temperature (21°C) from 2 μl + 2 μl protein + mother liquor (0.2 M sodium nitrate, 20% PEG 3350, pH 6.8) to dimensions of 180 μm × 130 μm × 70 μm. Crystals of the 1G4-A2-ESO 9C and 1G4-A2-ESO 9V complexes were grown at room temperature (21°C) from 2 μl + 2 μl protein + mother liquor (0.2 M potassium sodium tartarate tetrahydrate, 20% PEG 3350, pH 7.2) to dimensions of ∼130 μm × 80 μm × 70 μm.

Crystals were cryoprotected in reservoir solutions containing 10 and 20% glycerol; they were then flash-cooled, and maintained at 100 K, using a cryostream (Oxford Cryosystems). High-resolution data sets (1.4 Å for the 1G4 TCR, 1.9 Å for the 1G4–A2-ESO 9C complex and 1.7 Å for the 1G4–A2-ESO 9V complex) were collected at station ID14 EH2 of the European Synchrotron Radiation Facility (ESRF, Grenoble, France) with an Area Detector Systems Corporation Quantum 4 CCD detector. The 1G4–A2-ESO 9C and the 1G4–A2–ESO 9V complex crystals belonged to space group P2₁. Data were auto-indexed and integrated with the program DENZO (48), and scaled with the program SCALEPACK (48; Table I).

The structures of the 1G4 TCR, 1G4–A2-ESO 9C, and the 1G4–A2-ESO 9V complexes were determined by molecular replacement with the program EPMR (49). For the two pMHC–TCR complexes, initial solutions were obtained for the HLA-A2 heavy chain, and β₂M followed by a second round of molecular replacement with the A6 TCR used as a search probe while the A2 solution was kept fixed. The 1G4 molecular replacement was solved with the A6 TCR in a single round of molecular replacement. The program SWISS-MODEL (www.expasy.org/swissmod/SWISS-MODEL.html) was used to replace the A6 TCR sequence with that of 1G4 before refinement.

Using CNS (50), the models were refined as rigid body domains (α1 and α2, α3, β₂M, peptide, Vα, Vβ, Cα, and Cβ for the complexes and the Vα, Vβ, Cα, and Cβ TCR domains for the 1G4 data). Subsequent minimization procedures included positional refinement, simulated annealing and restrained individual B factor refinement with bulk solvent correction and overall anisotropic B factor scaling. Manual refitting of the models was done with the interactive graphics program O (51), and water molecules were added at the last stages of refinement on the basis of peaks that were at least 3.0σ in height in F_obs – F_calc electron density maps. The CDR loops of the isolated TCR showed significantly higher B factors than when complexed with pMHC, but the electron density was of sufficient quality to allow positioning of a single dominant main chain conformation for all of these loops.

We assessed the stereochemistry of the refined structures with the program PROCHECK (52). Surface complementarity calculations were done with the CCP4 program Sc (53; probe size 1.7 Å). To allow comparison of similar S_c values between structures it was necessary to estimate the standard uncertainty in atomic positions in the structures and how this translates to uncertainty in S_c value. For the 1G4–A2-ESO 9C complex (the lowest resolution structure) the standard uncertainty in atomic positions derived by Refmac (54) using the R_free was 0.148 Å. By running Sc 100 times on an ensemble of structures for this complex with the x, y, and z coordinates of each atom randomly perturbed using a Gaussian distribution of standard deviation 0.148 Å, we estimate the standard uncertainty in S_c to be 0.017. Thus, although the difference in S_c values for the two complexes is small, we believe it to be significant. Solvent accessibility calculations were done with Naccess (http://wolf.bms.umist.ac.uk/naccess/; probe size 1.4 Å). Graphical representations of the structure were prepared using programs Bobscript (55) and CSC Chemdraw (Cambridge Scientific Computing, Inc.).

Surface plasmon resonance experiments were performed by using a Biacore3000 (Biacore). Biotinylated soluble HLA (ligand) was immobilized on Streptavidin-coated CM5 chip (Biacore) at the level of 1,000 RU (response units) per flow cell. Equilibrium binding was measured at the flow rate of 5 μl/min, starting from the lowest analyte concentration. The data points were plotted using Origin software. K_d values were calculated using the standard hyperbolic model fitting kinetics experiments. Kinetics of TCR–pMHC interactions was measured at 30 μl/min. The curves were then fitted using a simultaneous k_on/k_off fitting model (Bioevaluation software; Biacore).

This assay was performed as previously described (31).

3 × 10⁴ T2 cells were pulsed with different concentrations of peptides for 2 h at 37°C. Cells were incubated with 40 μg/ml biotinylated Fab antibodies for 30 min at 4°C and visualized by Streptavidin-conjugated R-PE (Sigma-Aldrich) for 20 min at room temperature.

T2 target cells were pulsed with 1 μM, 100 nM, 10 nM, 5 nM, and 1 nM of either ESO 9C or ESO 9V peptide for 1 h at 37°C and washed twice in PBS. 1G4 CTL and T2 cells were washed in PBS and each cell pellet was resuspended to a final concentration of 5 × 10⁵ cells/ml in RPMI. 1G4 CTL and target were mixed 1:1, left for 5 min in suspension, and the plated on glass multiwell slides and incubated for 30 min at 37°C. Cells were fixed in 100% methanol precooled to −20°C, washed in PBS, and blocked in PBS, 2% BSA (Sigma-Aldrich). The slides were mounted in PBS containing 90% glycerol and 2.5% DABCO. Samples were examined using a Bio-Rad Radiance 2000 MP laser scanning microscope and the conjugation rate and granule polarization were quantified using an Axioplan 2 epifluorescent microscope (Zeiss).

2 × 10⁴ 1G4 CTL were incubated with 60 nM lysotracker green DND-26 (Molecular Probes) for 1 h at 37°C in RPMI, 5% human serum, and washed once in PBS and resuspended in 100 ml RPMI without phenol red, 5% human serum, 1 mM Hepes. T2 target cells were pulsed with 1 μM of either ESO 9C peptide or ESO 9V peptide for 1 h at 37°C and washed twice in PBS. 2 × 10⁴ target cells were allowed to adhere on a glass coverslip mounted in a temperature-controlled chamber for 10 min at 37°C in RPMI without serum, without phenol red plus Hepes. Lysotracker green DND-26–labeled 1G4 NY–ESO-1_157–165 CTL were added to the T2 cells in the chamber. Sequential confocal images were acquired every 20 s.

1G4 CTL and pulsed B cell targets were incubated at 37°C for 4 h at an effector–target ratio of 1:1. After 4 h the supernatant was harvested and assayed for MIP-1β and IFN-γ by ELISA (R&D Systems). Standard deviation from the mean of two duplicate assays is shown.

A2K^b transgenic mice were vaccinated using a DNA-prime/recombinant vaccinia virus–boost strategy (56). Mice were injected intramuscularly with 50 μg plasmid DNA encoding either ESO 9V or ESO 9C peptide followed by intravenous injection with 5 × 10⁶ plaque forming units recombinant vaccinia virus containing the same antigens on day 10. 10 d after injection, fresh PBL were isolated from blood taken from the tail vein using red cell lysis buffer (Invitrogen). A2K^b tetramer for ex vivo staining were synthesized as previously described (57). Approval of care and use of animals was obtained from the Clinical Medicine Ethical Review Committee (University of Oxford, Oxford, UK) and by the UK Home Office.

ELISPOT analysis for mouse IFN-γ secretion was performed with cells incubated overnight according to the protocol provided by the manufacturer (Mabtech). Mice splenocytes were generated using the method described above and rested in RPMI supplemented with 10% FCS at 4°C overnight.

In the experiment shown in Fig. 6 D, ESO 9C peptide was treated with 20 mM TCEP at room temperature for 1 h before being added to the wells at the final concentration of 200 μM.

Atomic coordinates and structure factor amplitudes for the 1G4 TCR–ESO 9C–A2 complex, 1G4 TCR–ESO 9V–A2 complex, and the 1G4 TCR have been deposited in the Protein Data Bank under accession numbers 2BNR, 2BNQ, and 2BNU, respectively.

For surface plasmon resonance, the antibody binding assay shown in Fig. S1, 10 serial dilutions in duplicates of BB7.2 and BBM.1 were made. For live cell video microscopy (Videos 1 and 2) a Nikon TE300 microscope attached to a Bio-Rad Radiance 200 MP laser scanning microscope was used with a 488-nm laser for epifluorescence and Nomarski differential interference contrast for transmitted light. The images were processed using MetaMorph version 4.5 software.

The authors wish to thank Dawn Shepherd, Ian Hermans, Uzi Gileadi, Jonathan Silk, Michael Palmowski, Fareed Mirza, and David Stuart for helpful discussions and assistance. We are also grateful to Karl Harlos and the staff of the ESRF and EMBL in Grenoble for assistance with X-ray data collection. We acknowledge use of crystallization facilities provided by the MRC-funded Oxford Protein Production Facility and The European Commission Integrated Programme (SPINE, Grant code QLRT-2001-00988).

E.Y. Jones is a Cancer Research UK Principal Research Fellow. This work was funded by Cancer Research UK (C399-A2291), the US Cancer Research Institute, the UK Medical Research Council, and National Translational Cancer Research Network. G. Held was supported by the German Research Foundation. G. Griffiths is funded by the Wellcome Trust.

The authors have no conflicting financial interests.