TCRs Used in Cancer Gene Therapy Cross-React with MART-1/Melan-A Tumor Antigens via Distinct Mechanisms

The identification of tumor-associated Ags preferentially presented by human cancers has led to the development of immunotherapeutic strategies for cancer such as peptide vaccines and adoptive T cell transfer. In adoptive T cell transfer, tumor Ag-specific T cells are activated ex vivo and transplanted back into a lymphodepleted patient. Although clinical trials with adoptive transfer have been promising (reviewed in Ref. 1), a liability is that variation in T cell repertoires impacts the likelihood of any individual producing a highly avid TCR specific for a given tumor Ag. That most tumor Ags are nonmutated self-Ags against which T cells will likely be negatively selected compounds this liability. A recent development that can address these concerns is the transfer of T cells genetically engineered to express tumor Ag-specific TCRs with defined recognition properties.

The first two trials examining the use of genetically engineered T cells in humans were published recently (2, 3). Both trials targeted the MART-1 protein (also referred to as Melan-A), upregulated in the majority of melanomas. The two trials used different class I MHC-restricted TCRs: DMF4 and DMF5. The two receptors are unrelated, using different Vα and Vβ segments and possessing different CDR3 loops (Table I). In functional assays, DMF5 T cells are more avid than DMF4, and DMF5 T cells are more efficiently stained with MART-1/HLA-A*0201 (HLA-A2) tetramers (4). Although the clinical trials were small, use of DMF5 resulted in an improved rate of cancer regression (13% with DMF4 versus 30% with DMF5). Use of DMF5 was also associated with incidences of eye, ear, and skin autoimmune toxicity not reported with DMF4. Expanded trials with DMF5-engineered T cells are underway, and DMF5 continues to be exploited as a model receptor for the development of T cell-based gene therapy of cancer (5–7).

Despite this progress, Ag recognition by MART-1–specific TCRs in general is complex and poorly understood. Most MART-1–specific TCRs examined cross-react between the decameric epitope spanning residues 26–35 (EAAGIGILTV), as well as the nonameric epitope spanning residues 27–35 (AAGIGILTV), both presented by the class I MHC (HLA-A2). Compared with the nonamer, the additional amino acid in the decamer forces the peptide to bulge and zigzag in the HLA-A2 peptide-binding groove, resulting in the presentation of different surfaces to the T cell repertoire (8). In addition to highlighting the capacity for TCRs to cross-react with structurally diverse ligands (9), nonamer/decamer cross-reactivity is likely to be important in melanoma immunotherapy. The nonamer is believed to be the clinically relevant Ag in HLA-A2⁺ individuals (10–13). Yet due to poor binding of the nonamer to HLA-A2 and the inability to generate a superior heteroclitic nonamer that maintains the nonameric conformation in the HLA-A2 peptide-binding groove (8, 14), the majority of efforts targeting MART-1 have made use of the stronger binding decamer or a decameric variant modified at position 2 (ELAGIGILTV) to select, assay, and activate MART-1–specific T cells. The decamer (or its anchor-modified variant) was among the first peptides to be used in clinical trials of peptide-based cancer vaccines and remains a component of many candidate cancer vaccine formulations (e.g., Ref. 15).

We studied MART-1 nonamer/decamer recognition by the DMF4 and DMF5 TCRs, determining the structural basis for cross-reactivity between the nonamer and decamer peptide/HLA-A2 complexes. We found that the two receptors cross-react via fundamentally different mechanisms. DMF4 cross-reacts with a complex mechanism, altering its orientation over the peptide/MHC complex to accommodate the differences in the peptides. In contrast, DMF5 binds the two ligands identically, accommodating the differences through the use of a permissive architecture that is preformed in the free receptor. The simpler mode of cross-reactivity for DMF5 is associated with higher affinity toward both ligands, helping to explain DMF5’s stronger functional avidity. In addition to providing a foundation for further developments in cancer immunotherapy, the results contribute to our understanding of the mechanisms underlying TCR cross-reactivity, demonstrating that different TCRs can use different mechanisms to cross-react with the same two ligands and that TCR binding orientation can be determined by peptide alone.

Recombinant soluble TCRs and peptide/HLA-A2 molecules were refolded from bacterially expressed inclusion bodies using established procedures (16). Peptides were purchased from Genscript or synthesized locally using an ABI 433A instrument and verified by mass spectrometry. All structure and binding experiments with the MART-1 decamer used the anchor-modified ELAGIGILTV variant. Recombinant DMF4 and DMF5 used an engineered disulfide bond in the constant domains to enhance stability (17).

Crystals of the DMF4-peptide/HLA-A2 complexes were grown from 15% PEG4000, 0.2 M MgCl₂ buffered with 0.1 M Tris (pH 8.5) at 25°C. Crystals of the DMF5-peptide/HLA-A2 complexes were grown from 20% PEG4000 buffered with 0.1 M HEPES (pH 7.5), with the addition of 10% propanol at 25°C. Crystals of free DMF5 were grown from 15% PEG 3350, 0.2 M MgCl₂ buffered with 0.1 M Tris (pH 8.5) at 25°C. Crystallization was performed using sitting drop/vapor diffusion. Streak seeding was used to obtain higher-quality crystals. For cryoprotection, crystals were transferred into 20% glycerol/80% mother liquor for 30 s and immediately frozen in liquid nitrogen. Diffraction data were collected at the 19BM, 19ID, 21ID, and 31ID beamlines at the Advanced Photon Source, Argonne National Laboratories. Data reduction was performed with HKL2000 (18). The ternary complexes were solved by molecular replacement using MOLREP or Phaser using Protein Data Bank (PDB) entry 2GJ6 (19) as a search model, with the coordinates of peptides, solvent, and CDR loops removed. The structure of free DMF5 was solved using the coordinates of the TCR from PDB entry 1AO7 (20) as a search model with solvent and CDR loops removed. Rigid body refinement, followed by translation/libration/screw (TLS) refinement and multiple steps of restrained refinement were performed with Refmac5 (21). TLS groups were chosen as previously described (19). Once defined, TLS parameters were included in all subsequent steps of the refinement. Anisotropic and bulk solvent corrections were taken into account throughout refinement. After TLS refinement, it was possible to unambiguously trace the position of peptides and TCR CDR loops in all structures against σ_A-weighted 2F_o-F_c maps. Waters were added using ARP/wARP (22). Evaluation of models and fitting to maps were performed using Coot (23) and XtalView (24). Procheck (25), the template structure check in WHATIF (26), and MolProbity (27) were used to evaluate the structures during and after refinement. Hydrogen bonds were determined with the PISA Web server and validated with distance and geometry criteria (28). Intermolecular contacts were tabulated using a cutoff of 4 Å. Measurements of TCR-docking angle followed the recommended procedure (29). Surface complementarities are the Sc statistic of Lawrence and Colman (30). Note that the peptides in the decamer complexes are numbered from 1 to 10, in contrast with our previous structure of the decamer/HLA-A2 complex, in which the peptide was numbered from 0 to 9 (8). PDB entries for the structures are listed in Table II.

Surface plasmon resonance experiments were performed using a Biacore 3000 instrument, as previously described (16). The TCR was coupled to the sensor surface using amine coupling. Data were corrected for bulk solvent effects using a blank flow cell. For experiments with the nonamer, improved accuracy was obtained by fixing the activity of the surface at values predetermined with the decamer (31). Flow rates were 5 μl/min. All injections were repeated twice, and affinity measurements reflect simultaneous fits to both datasets. Solution conditions were 10 mM HEPES, 150 mM NaCl, 3 mM EDTA, 0.005% surfactant P-20 (pH 7.4), 25°C. Data were processed with Biaevaluation 4.1 (GE Healthcare) and fit with OriginPro 7.5 (OriginLabs).

The structures of the DMF4 and DMF5 TCRs bound to the MART-1 27–35 nonamer (AAGIGILTV) and anchor-modified 26–35 decamer (ELAGIGILTV) were determined at resolutions between 2.3 and 2.8 Å (Table II). All four ternary complexes displayed the diagonal docking mode traditionally seen in TCR recognition of foreign Ags. This and other structural descriptors, such as buried surface area and shape complementarity, were within the range seen for other TCR–pMHC interactions (29) and are summarized in Table III. Electron-density images for key regions of each structure are shown in Supplemental Fig. 1. The structures are described and compared in detail below, beginning with the more complex DMF4 structures.

The structures of the DMF4 TCR bound to the MART-1 nonamer and decamer peptide/HLA-A2 complexes showed that the TCR binds the two ligands differently (Fig. 1A). When the HLA-A2 peptide-binding domains in the two structures are superimposed, the root mean square deviation (RMSD) between the TCR variable domains is 5.1 Å. Viewed from the top through the TCR, each loop, with the exception of CDR3β, is arranged differently over the pMHCs (Fig. 1B). This is reflected in a 15° difference in docking angle, with the TCR positioned more diagonally over the nonamer (44°) than the decamer (29°). Other than CDR3α, each CDR loop remains in the same conformation. Thus, DMF4 engages the nonameric and decameric MART-1/HLA-A2 complexes with geometries that differ predominantly by a rigid-body rotation over the pMHC, with the pivot point of rotation centered on CDR3β (Fig. 1C).

The different geometries by which DMF4 binds the nonamer and decamer complexes result in different contacts made by various CDR loop amino acids to positions on HLA-A2 (Fig. 2, see Supplemental Fig. 2 for a more detailed list). One illustration of these differences is in the TCR–HLA-A2 hydrogen-bonding patterns: only two hydrogen bonds are formed to HLA-A2 in the DMF4–decamer complex. In contrast, six TCR–HLA-A2 hydrogen bonds are formed in the nonamer complex.

Examining the DMF4–HLA-A2 interfaces in more detail, the differences in environments due to the change in TCR orientation can be broken down into three general classes: placement of TCR and HLA-A2 atoms into different environments with the formation of wholly new interatomic interactions; a mimicking of the general chemical environment around HLA-A2 residues but using atoms from different TCR amino acids; and retention of environment with only small changes in interatomic interactions. Instances of each class are shown in Fig. 3. The most dramatic change in environment occurs with Thr163 in the HLA-A2 α2 helix. In the nonamer structure, Thr163 hydrogen bonds with Asn29 of CDR1α and Arg68 of HV4α. However, in the decamer structure, Thr163 forms only a single long-range van der Waals contact with Asn29α, with Asn29α and Arg68α instead interacting with the peptide (Fig. 3A).

An example in which the HLA-A2 chemical environment is mimicked using different TCR amino acids is seen with Arg65 in the HLA-A2 α1 helix, which hydrogen bonds with Thr92α of CDR3α in the decamer complex but with Gly93 of CDR3α in the nonamer complex (Fig. 3B). In addition to rotation of the DMF4 TCR, the change in environment around Arg65 is driven by a shift in CDR3α conformation (Fig. 1C). This conformational change seems to occur solely for the TCR to hydrogen bond with Arg65, as CDR3α forms no contacts with the peptide in either the nonamer or decamer structure and there are no steric clashes that would force a conformational change in the loop if DMF4 were to bind the nonamer with a decamer-like orientation. The importance of position 65 in TCR recognition of class I MHC, and HLA-A2 in particular, was noted previously (32–34) and is likely reflected in this case in how the need to hydrogen bond with the TCR forces a conformational change CDR3α.

Thr163 and Arg65 of HLA-A2 lie toward the N-terminal end of the peptide in the HLA-A2–binding groove, where the differences in environment are magnified because they are most distant from the CDR3β pivot point (blue highlights in Fig. 1C). Thus, positions on HLA-A2 closer to CDR3β retain more of their chemical environments in the two complexes. Indeed, the DMF4–HLA-A2 contacts near CDR3β are the only ones shared in the two DMF4 structures. Of particular interest are shared contacts between germline CDR loops and HLA-A2. Gln155 maintains the greatest number of these (Fig. 3C), forming eight with Tyr49 in CDR2α. Both Gln155 and tyrosines in CDR2α have been suggested to play a key role in TCR recognition of class I MHC (32, 35), and the close alignment of Gln155 with Tyr49α, despite the different docking angle, could indicate such a role. However, Gln155 also forms a hydrogen bond with Gln100 of CDR3α in both structures (Fig. 3C), complicating such an interpretation.

We next compared the structures of the DMF4-bound pMHC complexes with those of the previously solved free pMHCs (8, 36). No changes occur in either peptide or MHC upon TCR recognition of the decamer (Fig. 4A). However, upon recognition of the nonamer, a large shift occurs in the center of the peptide, bringing the conformation of the center closer to that of the decamer (Fig. 4B). The shift extends from the carbonyl oxygen of Ile4 to the amide nitrogen of Ile6 and is maximal at the amide nitrogen of Gly5, which moves 2.7 Å toward the HLA-A2 α2 helix. The shift in the nonamer is similar to a recent description of “induced molecular mimicry” upon TCR binding (37). However, due to the presence of the additional amino acid in the decamer there are still conformational differences between the nonamer and decamer, with the peptides out of alignment and register at Ile4 (nonamer) and Ile5 (decamer) (Fig. 4C).

Closer examination of the DMF4-peptide/HLA-A2 interfaces shows how the repositioning of DMF4 over the two pMHC molecules allows the TCR to accommodate the remaining structural differences in the peptides. Beginning with the peptide N terminus, multiple electrostatic interactions link DMF4 to the decamer (Fig. 5A): Arg68 of the HV4α loop forms a salt bridge with the N-terminal glutamate, and an interfacial water links Asn29 of CDR1α and Thr92 of CDR3α to the carbonyl oxygen of Gly4. None of these interactions is present in the structure with the nonamer (Fig. 5B): without a hydrogen-bonding partner the water molecule is absent, and most importantly, the side chain of Ile4 in the nonamer complex occupies the position of the Gly4 backbone in the decamer complex, forcing a repositioning of the CDR1α loop. Without movement of CDR1α, steric clashes would occur between the side chains of Ile4 and Asn29α (Fig. 5C). These clashes are avoided by the more diagonal placement of DMF4 over the nonamer, which moves CDR1α out of the way of the Ile4 side chain. The clashes between Ile4 and Asn29α are the only clashes that occur when the pMHC from the nonameric complex is superimposed onto that of the decameric complex. Because the conformation of CDR1α is unchanged despite the different position of the TCR, the surprising conclusion is that the energetic cost for the TCR to bind in a different orientation is less than that for moving CDR1α out of the way via a conformational change.

After Ile4/5, the nonamer and decamer peptides begin to move into alignment and are superimposable at Ile6/7. At this point, both peptides interact with CDR3β, which, as the pivot point for the TCR, maintains its position in the two structures. CDR3β is aligned parallel to the C-terminal halves of the peptides, forming a motif similar to that of an antiparallel β-sheet (Fig. 5A, 5B). A hydrogen bond is formed between the amide nitrogen of Val98β and the carbonyl oxygen of Ile6/7 in both DMF4 complexes, and the Val98 side chain forms several van der Waals interactions with the peptides. The position of Val98β appears to drive the conformational change that occurs in the nonamer peptide, as steric clashes would occur between Val98β and the backbone of Gly5 of the nonamer if the peptide did not move. Two residues down the CDR3β loop, Val96 hydrogen bonds with Thr8/9.

Unlike DMF4, the DMF5 TCR binds the MART-1 nonamer and decamer peptide/HLA-A2 complexes identically (Fig. 1D). In the two DMF5 complexes, the backbones of the TCR Vα/Vβ domains, common residues of the peptides, and the HLA-A2 peptide-binding domains superimpose with an RMSD of only 0.5 Å, and the conformations of the CDR loops are the same (Fig. 1E). The key interresidue contacts within the DMF5-peptide/HLA-A2 interfaces are listed in Fig. 2; a more detailed list of contacts is given in Supplemental Fig. 2. As expected from the near-identical structures, the participation of HLA-A2 amino acids in the two DMF5 interfaces is essentially the same.

We next compared the structures of the DMF5-bound pMHC complexes with those of the free. As with DMF4, no changes occur in either peptide or MHC upon DMF5 recognition of the decamer (Fig. 4D). However, upon recognition of the nonamer, a shift occurs in the center of the peptide, bringing the conformation of the center closer to that of the decamer (Fig. 4E). The shift is nearly identical to that seen with DMF4: it extends from the carbonyl oxygen of Ile4 to the amide nitrogen of Ile6 and is maximal at the amide nitrogen of Gly5, which moves 2.7 Å toward the HLA-A2 α2 helix. Again, although the backbones are closer, the peptides remain out of alignment and register at Ile4/5 (Fig. 4F).

A close inspection of the two DMF5-peptide/HLA-A2 interfaces reveals how DMF5 is able to recognize the decamer and the shifted nonamer without requiring the changes in TCR-binding orientation or CDR-loop conformation required for DMF4. Beginning with the N termini of the peptides, the side chain of Gln30 of CDR1α hydrogen bonds to the carbonyl oxygen of Leu2 in the decamer and Ala2 in the nonamer (Fig. 5D, 5E). Although the conformations of the peptides begin to diverge after this hydrogen bond, they are close enough to permit the side chain of Gln30α to form a second hydrogen bond to the amide nitrogen of Gly5 (decamer) and Ile4 (nonamer). The DMF5 TCR does not form a hydrogen bond or salt bridge with the N-terminal glutamate in the decamer, making only long-range van der Waals contacts to the glutamate side chain.

The structural differences between the nonamer and decamer become more significant following the second hydrogen bond made by Gln30 of CDR1α. After this hydrogen bond, the backbone of the decamer bulges up toward the TCR. This bulge does not occur in the nonamer, but the β carbon of Ile4 of the nonamer occupies the same position as the carbonyl carbon of Gly4 of the decamer. Both the bulge in the decamer and the side chain of Ile4 in the nonamer are accommodated by a wide slot in the TCR that is walled by the side chains of Gln30 of CDR1α and Phe100 of CDR3β and roofed by the triple-glycine motif in the center of CDR3α (Fig. 5E, 5F). The slot is large enough to accommodate both peptides without any compensatory adjustments. Indeed, the CDR3α “roof” is high enough such that CDR3α forms no contacts to the decamer and only three, long-range van der Waals contacts to the nonamer (Supplemental Fig. 3). Thus, an accommodating architecture is one component of how DMF5 recognizes both the MART-1 nonamer and decamer.

After exiting the slot, the side chain of Ile5 in the decamer extends toward the HLA-A2 α2 helix, occupying space at the periphery of the interface that is empty in the nonamer structure. At this point, the peptide backbones are closer in alignment and are linked to the TCR via a water molecule that serves as the hub of a network of hydrogen bonds between CDR3β and the centers of the peptides. In both structures, the water links the backbone of Ile6 (nonamer) and Ile7 (decamer) with the backbone of Phe100β and the side chain of Ser99β (Fig. 5C, 5D). An additional hydrogen bond is made to Gly6 in the decamer but not to Gly5 in the nonamer due to lingering structural differences in the peptides. This network of hydrogen bonds explains the need for the structural shift that occurs in the center of the nonamer upon binding; if the nonamer did not adopt a conformation closer to the decamer at this point, there would be no room for the bridging water molecule, preventing the formation of the hydrogen bonds between CDR3β and the peptide.

Following the water-bridged hydrogen bonds to the centers of the peptides, the conformations of the nonamer and decamer peptides are identical. In both structures, a final TCR–peptide hydrogen bond is made by the side chain of Asn33 of CDR1β to the side chain of Thr8/9 (Fig. 5D, 5E).

We next determined the structure of the free DMF5 TCR to 2.1 Å resolution in a crystal form with two molecules per asymmetric unit (Table II; see Supplemental Fig. 1 for electron-density images). The two copies of the molecule superimpose closely (RMSD for superimposition of the backbones of the variable domains is 0.8 Å). Each CDR loop adopts the same overall conformation in the two copies of the molecule (Fig. 6A). However, the positions at the tip of CDR3α differ by 2.1 and 1.4 Å displacements at the backbone carbonyls of Gly93 and Gly94, respectively (Fig. 6B), indicating that the triple-glycine motif of Gly93, Gly94, and Gly95 imparts a degree of flexibility to CDR3α. The conformation of CDR3α in the first molecule in the asymmetric unit is closest to the conformation seen in the bound state of the receptor.

The position of CDR1α also differs slightly in the two molecules in the asymmetric unit of the free DMF5 structure, with a rotation around Gly28 that impacts the path and position of the C-terminal end of the loop (Fig. 6C). The C-terminal end of CDR1α in the bound state of the receptor is displaced slightly away from the two conformations seen in the unbound state. The N-terminal half of CDR1α is largely the same in the bound and unbound conformations, although modest changes are needed in the Gln30 side chain torsion angles for it to engage peptide.

Although the structure indicates some flexibility for CDR1α and CDR3α, the conformational adaptations needed to bind ligand are small in context, less than the average seen for CDR1α and CDR3α in a recent comparison of bound and free TCRs (38). Thus, the major elements of DMF5 used to bind ligand appear largely preconfigured in the free receptor. This is illustrated in Fig. 6D, which shows how free DMF5 sits over the decamer peptide when superimposed onto the TCR in the DMF5-decamer/HLA-A2 complex, emphasizing the slot needed to accommodate the peptide.

We next examined the interactions of the DMF4 and DMF5 TCRs with the two MART-1 ligands using surface plasmon resonance (Fig. 7). The affinity of DMF5 toward the decamer and nonamer ligand was 6 and 40 μM, respectively. The affinity of DMF4 toward the decamer and nonamer was 29 and 170 μM, respectively. Thus, both TCRs bind the decamer more strongly than the nonamer, with DMF5 possessing stronger affinity for both. The stronger affinities toward decamer are consistent with the need for the nonamer to undergo a structural shift upon binding of both TCRs.

Interestingly, although DMF5 binds both the nonamer and decamer more tightly than does DMF4, the difference in binding free energy between recognition of the nonamer and decamer is identical within error for the two TCRs (ΔΔG° = 1.2 ± 0.3 kcal/mol for decamer, 1.1 ± 0.1 kcal/mol for nonamer). Thus, from a free energy perspective, although DMF5 binds both nonamer and decamer with higher affinity, its mechanism of cross-reactivity is not superior to DMF4’s. Lastly, although we attempted kinetic measurements, dissociation rates for all cases were fast (>0.5 s^-1), precluding accurate measurements of binding kinetics.

Recent clinical trials demonstrated that the adoptive transfer of genetically redirected T cells can lead to cancer regression in humans (2, 3). The first two TCRs used in this approach, DMF4 and DMF5, both recognize the overlapping, but structurally diverging, 26–35 (decamer) and 27–35 (nonamer) epitopes from the MART-1/Melan-A protein. Although the trials were small in size, clinical outcomes differed with the two receptors. Use of DMF4 led to a 13% rate of cancer regression, whereas use of DMF5 led to a 30% rate of regression and associated eye, ear, and skin toxicity. The DMF5 TCR is currently in use in larger clinical trials and continues to be used as a model TCR for improvements in T cell-based gene therapy of cancer (5–7).

Early work assumed the MART-1 nonamer and decamer were structurally equivalent because of the high frequency of cross-reactive T cells in HLA-A2⁺ individuals (39). However, comparative structures of the two peptide/HLA-A2 complexes demonstrated that this is not the case, with the decamer adopting a bulged conformation as a result of the additional amino acid (8). Although the mechanisms underlying nonamer/decamer cross-reactivity are of interest given the fundamental role that T cell cross-reactivity plays in cellular immunity (9), MART-1 nonamer/decamer cross-reactivity may also be important in immunotherapy. The nonamer is believed to be the physiologically relevant epitope in HLA-A2⁺ individuals (10–13); however, because of the poor binding of the nonamer to HLA-A2, the decamer or its anchor-modified variant is regularly used to identify and activate MART-1–specific T cells. The decamer or its variant also continues to be used as a chief component of many cancer-vaccine formulations.

In cross-reacting between the MART-1 nonamer and decamer, both DMF4 and DMF5 require the nonamer to shift its backbone into a more decamer-like conformation, explaining the higher affinity toward decamer for both TCRs. Binding-induced conformational changes in peptide backbones have been observed previously in TCR recognition (e.g., Refs. 20, 40, 41), but it is interesting that in this study, the changes are observed in the nonamer rather than in the longer and more extensively bulged decamer. An earlier analysis of MART-1 bound to HLA-A2 suggested that the nonamer possesses greater intrinsic flexibility than does the decamer (8). Both DMF4 and DMF5 apparently use this flexibility in engaging nonamer, albeit for different reasons: DMF4 to avoid steric clashes and DMF5 to form hydrogen bonds.

After shifting the nonamer, the methods of DMF4 and DMF5 cross-reactivity diverge. The DMF4 TCR uses a complex mechanism, alternating between a binding orientation that, judging by hydrogen bonds, optimizes interactions with the peptide at the expense of the MHC (decamer) or optimizes interactions with the MHC at the expense of the peptide (nonamer). In contrast, DMF5 engages both ligands almost identically, using an open architecture apparently preformed in the free TCR. Although the structures do not readily indicate why, DMF5’s simpler mode of cross-reactivity is associated with improved affinity toward both ligands.

The differences in affinity between the DMF4 and DMF5 TCRs are relatively modest: 5-fold for the decamer and 4-fold for the nonamer. These results are consistent with the differences in functional avidity with the original DMF4 and DMF5 T cell clones (4) and could help to explain the reported differences in clinical outcomes with the two TCRs. Enhancing TCR affinity has been suggested as a means for improving immunotherapy (42), and TCRs with ≥1000-fold gains in affinity have been generated through molecular-evolution techniques (e.g., Refs. 43, 44). The small differences in affinity between DMF4 and DMF5 raise the possibility that such large enhancements may not be needed to impact clinical results. This may be significant given that losses in peptide specificity have been observed with some very high-affinity TCRs (45, 46), an outcome that would clearly be detrimental for Ag-specific immunotherapy (again though, we emphasize that the clinical trials with DMF4 and DMF5 were small, and other factors, such as differences in α/β-chain pairing in the transduced TCRs, could have contributed to the outcomes).

Structure-guided computational design has also been used to enhance TCR affinity (47), in principle providing a degree of control absent in molecular evolution and reducing the risk of losses in Ag specificity. The structures with the DMF5 TCR provide a starting point for pursuing such an approach. The fact that DMF5 engages both the MART-1 nonamer and decamer identically using a preformed architecture raises the likelihood of successfully enhancing affinity toward both nonamer and decamer without losses in specificity, reducing concerns that a MART-1–specific TCR with enhanced affinity toward the nonamer would have reduced affinity toward the decamer or vice versa, either of which would negatively impact immunotherapy using decamer-based peptides to identify or activate nonamer-specific T cells. A natural concern is that higher-affinity TCRs targeting MART-1 may induce even stronger autoimmune toxicity if used clinically, requiring more significant interventions than previously used (2).

The finding that the binding orientation of DMF4 differed with the nonamer and decamer was surprising. The switch in binding orientation seems attributable to two structural features: the need to avoid steric clashes between CDR1α and the center of the nonamer and the ability of HV4α to form a salt bridge with the N-terminal glutamate of the decamer. Different orientations of a single TCR over two different ligands were previously observed with the murine 2C TCR in complex with the dEV8 and the QL9 peptides (48). However, in that case, the two peptides were presented by different MHC proteins (H-2L^d and H-2K^b, respectively). The results with DMF4 extend this finding by demonstrating that TCR-binding orientations can be dictated by peptide alone. This peptide-determined binding mode necessitates a structurally and energetically permissive relationship between the germline elements of DMF4 and HLA-A2. Such permissiveness may be a fundamental mechanism of TCR cross-reactivity and is likely further illustrated in the observation of unusual TCR-binding modes seen with TCRs bound to self-Ags associated with autoimmunity (reviewed in Ref. 49).

Although the changes in binding orientation seen with DMF4 alter the interactions between the TCR germline elements and HLA-A2, some interactions are conserved, most notably those between Tyr49 of CDR2α and Gln155 of HLA-A2. Both tyrosines in TCR CDR2 loops and Gln155 in class I MHC molecules have been proposed to play key roles in TCR binding, with tyrosines in particular implicated in encoding a genetic bias of TCRs toward MHC proteins (32, 35, 50). The two DMF4 structures, with different binding orientations despite the same variable domains and MHC, provide a new opportunity to test this hypothesis with structure-guided mutations. It is notable that, in addition to interacting with Tyr49 of CDR2α, Gln155 hydrogen bonds with Gln100 of CDR3α, highlighting possible cooperativity in the interactions of the germline and nongermline elements with HLA-A2.

Recognition of MART-1 Ags in HLA-A2⁺ individuals is characterized by a strong bias toward TCRs using the Vα 12-2 variable domain (51, 52). Based on the structure of the Vα 12-2 Mel5 TCR with the MART-1 nonamer presented by HLA-A2, Cole et al. (53) proposed that this bias was attributable to interactions between the germline CDR1α loop and the peptide, describing this as “innate-like” recognition of Ag. The DMF5 TCR forms the same CDR1α-peptide interactions as does Mel5, using Gln30 to form two hydrogen bonds to the peptide backbone (Supplemental Fig. 5A). Interestingly, CDR1α of the well-characterized TCR A6, which also uses Vα 12-2, forms similar interactions with the Tax, Tel1p, and HuD peptides (20, 54, 55). Although these peptides are unrelated to those of MART-1, their N-terminal conformations are very similar when bound to HLA-A2. Thus, CDR1α of Vα 12-2 appears optimally positioned to interact with this peptide conformation. However, the extent to which CDR1α-peptide interactions underlie the Vα 12-2 bias in MART-1–specific TCRs remains uncertain, as there are also conserved patterns of van der Waals interactions between the Vα 12-2 germline loops and HLA-A2 in the various structures (Supplemental Fig. 5B). Determining the energetic balance between these two sets of interactions will again require more probing investigations. Lastly, MART-1–specific Vα 12-2 TCRs also show a weak conservation in the length and sequence of CDR3α and CDR3β (56). The two CDR3 loops of Mel5 form a similar slot as in DMF5 to accommodate the bulge in the MART-1 decamer (Supplemental Fig. 6); the weak bias in CDR3α/CDR3β composition may thus reflect that only a subset of possible CDR3 loops is compatible with this architecture.

We thank Cynthia Piepenbrink for outstanding technical assistance.

The authors have no financial conflicts of interest.