Structural basis for oligoclonal T cell recognition of a shared p53 cancer neoantigen

Adoptive cell therapy (ACT) with tumor-specific T cells can mediate cancer regression. The main target of tumor-specific T cells are neoantigens arising from mutations in self-proteins. Although the majority of cancer neoantigens are unique to each patient, and therefore not broadly useful for ACT, some are shared. We studied oligoclonal T-cell receptors (TCRs) that recognize a shared neoepitope arising from a driver mutation in the p53 oncogene (p53R175H) presented by HLA-A2. Here we report structures of wild-type and mutant p53–HLA-A2 ligands, as well as structures of three tumor-specific TCRs bound to p53R175H–HLA-A2. These structures reveal how a driver mutation in p53 rendered a self-peptide visible to T cells. The TCRs employ structurally distinct strategies that are highly focused on the mutation to discriminate between mutant and wild-type p53. The TCR–p53R175H–HLA-A2 complexes provide a framework for designing TCRs to improve potency for ACT without sacrificing specificity. Developing broadly applicable neoantigen-directed adoptive cell therapies (ACTs) is challenging because each cancer patient has an unique neoantigen repertoire. Here, the authors present the crystal structures of tumor-specific T cell receptors (TCRs) that recognize a shared neoepitope arising from the R175H driver mutation in the p53 oncogene (p53R175H) alone and bound to p53R175H–HLA-A2, which are of interest for the structure-guided design of TCRs to improve T cell potency for ACT.

A doptive cell therapy (ACT) with ex vivo-expanded tumorinfiltrating lymphocytes (TILs) can mediate durable cancer regression in patients with metastatic melanoma, cervix, bile duct, colon, and breast cancers [1][2][3][4][5] . This therapeutic effect is mediated mainly by CD8 + cytotoxic T cells 6 but CD4 + T cells are also likely to contribute 5 . The prime target of tumorspecific T cells are neoantigens that arise as a consequence of DNA alterations during malignant transformation 7 . Recent technological advances in mass spectrometry and highthroughput T cell-based assays have greatly accelerated the identification of neoantigens resulting from somatic mutations, as well as the T cells that recognize them, in individual patients.
A major challenge in developing broadly applicable neoantigen-directed ACT is the unique neoantigen repertoire of each cancer patient 7 . There are few shared mutated targets among patients, even among patients with similar cancers. For example, in a study of patients with gastrointestinal cancers, 99% of neoantigenic determinants recognized by neoantigen-reactive TILs were unique (private) and not shared (public) between any two patients 8 . Nevertheless, it has been possible to identify a limited number of shared cancer neoantigens. Of particular interest are neoantigens derived from oncogenes bearing driver mutations because these mutations are tumor-specific, biologically important for tumor progression, and likely to be expressed by all tumor cells 9 . In a seminal study of ACT, a patient with metastatic colorectal cancer was treated effectively with four distinct CD8 + T cell clones that specifically targeted a neoepitope arising from the KRAS G12D driver mutation in an HLA-C*08:02-restricted manner 2 .
Other shared mutated neoantigens expressed by cancers of unrelated patients have now been identified. TP53, which encodes the tumor suppressor p53, is the most frequently mutated gene across all cancer types 10 . Indeed, TP53 mutations are found in 40-50% of cancer patients, and effect most of the hallmarks of cancer cells, including genomic instability, proliferation, and metastasis 11,12 . Mutant p53 predisposes to cancer development and is associated with ineffective therapeutic responses and unfavorable prognoses 10 . Despite these effects, no drug to abrogate the oncogenic functions of mutant p53 has yet been approved for any cancer treatment. A substantial portion of TP53 mutations occur at hotspot positions R175, G245, R248, R249, R273, and R282 10 . Because mutations at these sites confer a growth advantage to tumor cells and are associated with malignant progression, they are attractive candidates for targeted immunotherapy.
The immunogenicity of p53 mutations in patients with cancer was recently demonstrated by the detection of T responses against several shared p53 neoantigens, notably R175H and R248W 13,14 . Both of these driver mutations are located in the DNA-binding domain of p53 and alter its DNA-binding capacity 10 . Several TCRs have been isolated from TILs of epithelial cancer patients that recognize a neoepitope corresponding to residues 168-176 of p53R175H. This neoepitope includes the arginine-to-histidine mutation at position 175 (HMTEVVRHC) 13,14 . The TCRs are restricted by the common MHC class I allele HLA-A*02:01. These oligoclonal TCRs, transduced at high frequency into a patient's peripheral blood lymphocytes for ACT, may prove effective in eliminating tumors expressing HLA-A*02:01 and the p53R175H mutation 13,14 .
With the goal of understanding TCR recognition of cancer neoantigens at the atomic level, we determined crystal structures of three p53R175H-specific TCRs (12-6, 38-10, and 1a2) 13,14 in complex with HLA-A*02:01 and the shared neoantigen p53R175H. In previous studies, we determined structures of human melanoma-specific TCRs bound to a neoepitope from mutant triose phosphate isomerase (mutTPI) and HLA-DR1 15,16 . However, that neoepitope, unlike KRAS G12D 2 or p53R175H 10 , was only expressed in a single melanoma patient. The unique, rather than shared, nature of mutTPI, a feature that also characterizes the vast majority of cancer neoantigens discovered to date 7,8 , precludes broad use of mutTPI-specific or similar TCRs in ACT. Structures have also been reported of TCRs in complex with epitopes from the tumor-associated antigens NY-ESO-1 and MART-1 bound to HLA-A2 17,18 . However, NY-ESO-1 and MART-1, unlike mutTPI 15 or p53R175H 13,14 , are not neoantigens, but instead non-mutated self-antigens that are selectively expressed in certain cancer types. By contrast, the TCR-p53R175H-HLA-A2 structures described here involve a shared cancer neoantigen. The structures reveal how oligoclonal TCRs 12-6, 38-10, and 1a2 discriminate between wild-type and mutated p53, and demonstrate that there are multiple distinct solutions to recognizing the p53R175H neoepitope with sufficient affinity to mediate tumor cell killing.
Differences between p53 and p53R175H are confined to mutation site. To understand how the conservative arginine-tohistidine mutation in p53R175H, which replaces one positively charged amino acid by another, renders this peptide immunogenic, we determined the structures of the wild-type p53-HLA-A2 and mutant p53R175H-HLA-A2 complexes to 2.37 and 2.38 Å resolution, respectively (Supplementary Table 1) (Fig. 2a). Clear and continuous electron density extending along the entire length of both MHC-bound peptides allowed confident identification of all peptide atoms ( Supplementary Fig. 1). In both p53-HLA-A2 and p53R175H-HLA-A2, the peptide is bound in conventional orientation with the side chains of P2 Met and P9 Cys accommodated in pockets B and F, respectively, in the peptide-binding groove (Fig. 2b). Methionine and cysteine are among the most common residues at primary anchor positions P2 (Leu > Thr > Met ∼ Val > Ile) and P9 (Val > Ile > Thr > Ala > Cys > Leu), and are known to confer high affinity for HLA-A*02:01 20 . The solvent-exposed side chains of P1 His, P4 Glu, P7 Arg, and P8 Arg/His project away from the peptide-binding groove and compose a highly featured surface for potential interactions with TCR. By contrast, the side chains of P5 Val and P6 Val present a relatively featureless, non-protruding surface that may be a difficult target for TCR recognition. The p53-HLA-A2 and p53R175H-HLA-A2 complexes exhibit very little structural deviation from each other (Fig. 2a). In particular, the wild-type and mutant p53 peptides are highly superimposable, except at the P8 mutation site (Fig. 2b). The root-mean-square difference (r.m.s.d.) for α-carbon atoms in the peptide chains is 0.18 Å, while for all atoms, excluding the P8 Arg and His side chains, it is 0.45 Å. Therefore, structural differences between the p53-HLA-A2 and p53R175H-HLA-A2 complexes that disclosed the naturally altered p53 self-peptide to the T cells of cancer patients 13,14 are restricted to the mutation site at P8.
However, whereas the unusual docking geometries of these latter TCRs are incompatible with signaling, that of 38-10 allows a robust response to antigen 13 .
All three TCRs are shifted towards the C-terminus of the p53R175H peptide, which is the site of the driver mutation at P8. To quantitate the shifts, we projected the positions of the TCR centers onto the pMHC plane, where the x-axis is aligned with the peptide and a more positive x value indicates a shift toward the peptide C-terminus (Supplementary Table 4). Remarkably, TCR 38-10 exhibits the third-highest C-terminal shift among 136 TCR-pMHC class I structures reported to date, with 12-6 (22ndhighest) and 1a2 (26th-highest also quite shifted. Only two reversed-polarity TCRs 23 are more skewed along the peptide than 38-10. The C-terminal shift of TCRs 38-10, 12-6, and 1a2 is key to their ability to discriminate between wild-type and mutant p53 peptides (see below).
As depicted by the footprints of TCRs 12-6 and 1a2 on the pMHC surface (Fig. 4d, f), both TCRs, which were derived from different cancer patients 13,14 , establish contacts with the Cterminal half of the p53R175H peptide mainly via the CDR3β loop. By contrast, TCR 38-10, which was isolated from the same patient as 12-6 13 , engages the C-terminal half of the p53R175H peptide mostly through CDR3α (Fig. 4e). Overall, the footprints of 12-6 and 1a2 on pMHC resemble each other more closely than either footprint resembles that of 38-10.
TCR 1a2 relies heavily on the somatically-generated CDR3α loop for MHC recognition (Fig. 5e). Indeed, the percentage contribution of this loop to interactions with MHC (49% of contacts) exceeds that of any other CDR in any of the three complexes (Table 3). Five consecutive residues at the tip of the 1a2 CDR3α loop (Leu94α-Ser98α) pack tightly against the HLA-A2 α1 helix, with two hydrogen bonds providing additional stabilization (Supplementary Table 5).
TCRs target p53 driver mutation. Upon binding p53R175H-HLA-A2, TCRs 12-6, 38-10, and 1a2 bury 71% (303 Å 2 ), 76% (336 Å 2 ), and 76% (304 Å 2 ), respectively, of the peptide solvent-accessible surface, which is typical for TCR-pMHC complexes 25 . However, the large majority of interactions between these TCRs and the p53R175H peptide involves C-terminal residues P7 Arg and P8 His: 44 of 52 van der Waals contacts and 7 of 9 hydrogen bonds for 12-6, 57 of 74 van der Waals contacts and 5 of 8 hydrogen bonds for 38-10, and 37 of 53 van der Waals contacts and 6 of 8 hydrogen bonds for 1a2 (Supplementary Table 6) (Fig. 6). These interactions are about evenly distributed between P7 Arg and P8 His, which suggests the functional importance of both residues for TCR binding. This conclusion is supported by binding energy calculations using Rosetta 26 to predict changes in TCR affinity upon alanine substitution of all peptide residues in the three complexes. In each case, the largest ΔΔG values, ranging from 1.7 to 3.6 kcal/mol, were observed for residues P7 and P8 (Supplementary Table 7). Therefore, all three TCRs focus on the C-terminal portion of the antigenic peptide for binding, in sharp contrast to most other TCRs, which preferentially target the central portion of peptides, corresponding to residues P4-P6 25 . The TCRs discriminate between mutant and wild-type p53 by minimizing interactions with the central and Nterminal portions of p53R175H, which are structurally identical in the wild-type peptide (Fig. 2b). Interactions between 12-6 and the p53R175H peptide are mediated almost exclusively by CDR3β (Supplementary Table 6), whereas 38-10 and 1a2 employ both CDR3α and CDRβ for peptide recognition (Fig. 6a, b). Consistent with differences in α/β chain pairing and docking geometry (Fig. 4), TCRs 12-6, 38-10, and 1a2 use distinct strategies to achieve highly specific recognition of the mutant p53 peptide relative to wild-type, as demonstrated by SPR (Fig. 1). However, the TCRs share a pronounced skewing toward the  Table 4) as a result of positioning one or both of their CDR3 loops directly over P8 His (Fig. 6a-c). In the unbound p53R175H-HLA-A2 structure (Fig. 2a), the P8 His imidazole ring has one face against the side chain of Val76H of the HLA-A2 α1 helix, leaving its other face and most of its edge available for TCR binding. Each TCR provides close contacts that tightly sandwich the imidazole between Val76H and a specific CDR3 side chain: CDR3β Gln99 in 12-6, CDR3α Tyr103 in 38-10, and CDR3β Gln97 in 1a2 (Fig. 6c). In addition, the phenyl ring of CDR3α Tyr103 forms ππ stacking interactions with the imidazole ring of P8 His. Further selectivity for the mutant p53 peptide arises from hydrogen bonds with the P8 His side chain: 12-6 Glu95β Oε2-Nδ1 P8 His, 12-6 Trp98β Νε1-Nε2 P8 His, 38-10 Tyr31β OH-Nδ1 P8 His, and 1a2 Ser98α Oγ-Nε2 P8 His (Supplementary Table 6).
To assess the effect of replacing P8 His by Arg, which corresponds to reversion to the wild-type p53 peptide, we performed in silico mutagenesis using Rosetta 27 . A similar modeling protocol was previously used to predict binding effects of TCR-pMHC interface mutations 28 . Peptide substitutions were modeled in each X-ray complex structure, followed by side-chain packing and energetics-based scoring to calculate ΔΔG. Predicted ΔΔG values were 1.6, 1.2, and 2.0 Rosetta energy units (REU; analogous to kcal/mol) for 12-6, 38-10, and 1a2, respectively, suggesting substantial losses in TCR binding affinity for wild-type p53 peptide (Supplementary Table 7), as observed experimentally by SPR (Fig. 1). To assess possible structural defects leading to TCR affinity loss for the p53 revertant peptide, we calculated TCR-pMHC shape complementarity statistics (S c ) for the X-ray and modeled p53 revertant interfaces. S c values for p53175H interfaces are 0.72, 0.64, and 0.69 for 12-6, 38-10, and 1a2, respectively, commensurate with other MHC class I TCR-pMHC structures in the TCR3d database 29 , while they are 0.69, 0.57, and 0.71 for p53 revertants and the same respective TCRs, indicating loss of shape complementarity for the 12-6 and 38-10 TCR interfaces, and less predicted effect on the shape complementarity of the 1a2 interface. To further investigate the mechanistic basis of peptide specificity for these TCRs, the individual Rosetta scoring function terms comprising the predicted ΔΔG values noted above and in Supplementary Table 7 were obtained (Supplementary Table 8). This revealed that loss of favorable van der Waals interactions dominated the change in predicted binding affinity for the 12-6 TCR, whereas disruptions of side chain-side chain hydrogen bond interactions involving P8 His were primarily responsible for predicted 38-10 and 1a2 TCR affinity losses.
To assess ligand-induced conformational changes in the TCRs, we determined the structures of 12-6 and 1a2 in unbound form to 2.36 and 1.83 Å resolution, respectively (TCR 38-10 did not crystallize) (Supplementary Table 3). Superpositions of the free and bound TCR 12-6 and 1a2 structures are shown in Supplementary Fig. 3. Superposition of the VαVβ domains of free 12-6 onto those of 12-6 in complex with p53R175H-HLA-A2   NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-020-16755-y ARTICLE revealed structural differences in the CDR3 loops (Fig. 7d). Whereas conformational adjustments in CDR3β were restricted mainly to shifts in side-chain orientation, CDR3α underwent a large movement (r.m.s.d. in α-carbon positions of 3.7 Å for residues 93-97), which allowed CDR3α Gly94 to hydrogen bond with P4 Glu and Arg65H, CDR3α Tyr95 to hydrogen bond with CDR3β Leu97 and contact the HLA-A2 α1 helix, and CDR3α Gln96 to hydrogen bond with Arg65H. CDR3α Tyr95 showed the largest individual displacement (6.5 Å in its α-carbon position). Ligand-induced conformational changes were also observed in the CDR3 loops of 1a2 (Fig. 7e). CDR3β underwent a rearrangement (r.m.s.d. in α-carbon positions of 1.8 Å for residues 96-102) that resulted in formation of four hydrogen bonds and 33 van der Waals contacts with P7 Arg and P8 His at the critical C-terminus of p53R175H. CDR3α experienced a considerably larger movement (r.m.s.d. of 5.5 Å in α-carbon positions for residues 93-102), with CDR3α Glu96 undergoing an α-carbon displacement of 11.5 Å. This rearrangement allowed CDR3α to form a β-hairpin whose tip engages the HLA-A2 α1 helix and p53R175H, thereby optimizing TCR interactions with both MHC and peptide (Fig. 7f).

V100β E95β
E103β G94α  Fig. 6 Interactions of TCRs with the p53R175H peptide. a Interactions between 12-6 (left), 38-10 (center), and 1a2 (right) and the p53R175H peptide. The side chains of contacting residues are shown in stick representation with carbon atoms in green (TCR α chain), violet (TCR β chain), orange (p53R175H), or cyan (mutated P8 His), nitrogen atoms in dark blue, oxygen atoms in red, and sulfur atoms in yellow. Peptide residues are identified by one-letter amino acid designation followed by position (p) number. Hydrogen bonds are indicated by red dashed lines. b Comparison of interactions between 12-6 (left), 38-10 (center), and 1a2 (right) and the p53R175H peptide. Hydrogen bonds are red dotted lines and van der Waals contacts are black dotted lines. For clarity, not all van der Waals contacts are shown. c Close-up of interactions between 12-6 (left), 38-10 (center), and 1a2 (right) and P8 His.
T cells are provided to the patient to achieve this objective 7 . However, just how structurally different a neoantigen must be from its wild-type counterpart in order to overcome self-tolerance and elicit a T-cell response is not well understood. Bioinformatic profiling of cancer neoepitopes indicates that mutations at anchor positions which improve peptide binding to MHC molecules are associated with immunogenicity and tumor rejection 30,31 . However, most mutations in neoepitopes do not involve anchor residues and do not appreciably affect peptide binding to MHC. The p53R175H neoepitope studied here represents such a case.
Detailed comparison of the mutant p53R175H-HLA-A2 and wild-type p53-HLA-A2 structures revealed that their conformations differ only at the P8 mutation site, where one positively charged residue (histidine) replaces another (arginine). This substitution was sufficient to render the p53R175H peptide immunogenic in cancer patients 13,14 . Although replacement of histidine by arginine is conservative with respect to charge, these two amino acids differ markedly with respect to size and shape, which enables TCRs to distinguish between them. Assuming that p53 is expressed in the thymus and that the R175H mutation occurred after thymic development (i.e., during malignant transformation), the escape of p53R175H-specific T cells from negative selection is most likely explained by low affinity of the TCRs for wild-type p53 peptide. In support of this hypothesis, we were unable to detect any interaction of 12-6, 38-10, and 1a2 with p53-HLA-A2 by SPR, whereas these TCRs bound p53R175H-HLA-A2 with micromolar K D s, which manifested as an overwhelming preference for histidine over arginine at P8 in T-cell activation assays 13,14 . These results are consistent with previous structural studies of a unique melanoma neoepitope arising from a threonine-to-isoleucine mutation in a peptide derived from triose phosphate isomerase that produced only subtle changes in the binding surface for TCR 15,16 . Therefore, cancer neoantigens need differ only slightly from their wild-type counterparts for them to be immunogenic in patients. TCRs 12-6, 38-10, and 1a2 achieve high specificity for p53R175H by concentrating on the driver mutation at the C-terminal portion of the neoepitope, while avoiding extensive interactions with the Nterminal and central portions, which are shared with wildtype p53.
Structural studies of TCR-pMHC complexes involving common V segments and MHC alleles have revealed conservation of specific TCR-MHC interactions [32][33][34] . These conserved interactions, which occur between germline-encoded CDR1 and CDR2 loops and MHC, support the hypothesis that the canonical diagonal docking orientation of TCR on MHC, which is maintained in the TCR-p53R175H-HLA-A2 complexes, is the result of coevolution of TCR and MHC molecules. Surprisingly, however, while matches to at least one germline gene were found for TCRs 12-6, 38-10, and 1a2 among MHC class I-restricted TCRs in the structural database, inspection of the corresponding TCR-pMHC complexes indicated no matches to these TCRs in germline loop engagement of MHC. This unexpected lack of conserved TCR-MHC interactions applied even to TCRs restricted to HLA-A2, as well as to closely related MHC alleles such as HLA-B7 and HLA-B35. Therefore, considerably greater  flexibility exists for germline-encoded contacts between TCR and MHC than generally appreciated [32][33][34] , supporting the notion that peptide "editing" can lead to variability in germline contacts with MHC 16,35 . This flexibility is exposed in the TCR-p53R175H-HLA-A2 complexes through the unusually high tilt of two of these TCRs over MHC and the shift of all three TCRs toward the peptide C-terminus relative to other MHC class I-restricted TCRs, a set which already exhibits a peptide Cterminal shift compared to MHC class II-restricted TCRs 36 . Also of note, no MHC class I-restricted TCRs were identified in the structural database that used the TRAV38-1 germline gene of TCR 38-10, and none that used the same TRAV/TRBV gene combinations as 12-6, 38-10, or 1a2. These newly described structures highlight that, in spite of the sizable number of TCR-pMHC complex structures determined to date, many functional germline interactions and docking geometries likely have yet to be revealed.
Considerable efforts have been made to engineer TCRs with improved affinity for cancer-associated antigens for use in ACT [37][38][39] . However, large gains in TCR affinity may lead to increased cross-reactivity 40 , resulting in adverse clinical events 41 . In a striking case, an affinity-matured TCR targeting the MAGE-A3 melanoma antigen unexpectedly cross-reacted with an epitope from the muscle protein titin, causing cardiovascular toxicity and deaths 42 . Such off-target TCR recognition has prompted new structure-guided efforts to engineer therapeutic TCRs for enhanced specificity while maintaining optimal on-target affinity 41 . In one promising approach, the MART-1-specific TCR DMF5 was modified to promote stronger binding to the peptide portion of its pMHC ligand, which resulted in reduced crossreactivity with MART-1 homologs 43 . An attractive feature of p53R175H-specific TCRs such as 12-6, 38-10, and 1a2 is that the parental T cells survived negative selection, thereby minimizing the possibility of cross-reactivity with self-antigens. However, we do not know whether these TCRs possess optimal on-target affinities. Accordingly, future efforts will be directed at rational design of p53R175H-specific TCRs to optimize on-target affinity without compromising neoepitope specificity. Alternatively, it was recently demonstrated that antigen-specific TCR function could be enhanced by structure-based mutations in the Vα and Vβ domains outside the CDR loops that increase the level of cell surface expression 44 , an approach that could be attempted with the p53R175H-specific TCRs described here.
To prepare biotinylated HLA-A2, a 17-amino acid tag (GGGLNDIFEAQKIEW HE) was added to the C-terminus of the HLA-A*0201 heavy chain. The tagged p53-HLA-A2 and p53R175H-HLA-A2 proteins were produced as described above. Biotinylation was carried out using BirA biotin ligase (Avidity). Biotinylated protein was separated from excess biotin with a Superdex 200 column (50 mM sodium phosphate (pH 7.0), 100 mM NaCl).
Structure determination and refinement. Before structure determination and refinement, all data reductions were performed using the CCP4 software suite 46 . Structures were determined by molecular replacement with the program Phaser 47 and refined with Phenix 48 and Refmac 49 . The models were further refined by manual model building with Coot 50 based on 2F o -F c and F o -F c maps. The α chain of anti-EBV TCR RL42 (PDB accession code 3SJV) 51 , the β chain of anti-EBV TCR SB27 (2AK4) 52 , and NLV-HLA-A2 (5D2L) 53 were used as search models with the CDRs and peptide removed to determine the orientation and position of the 12-6-p53R175H-HLA-A2 complex. The orientation and position parameters of unbound TCR 12-6, p53-HLA-A2, and p53R175H-HLA-A2 were obtained using the corresponding components of the 12-6-p53R175H-HLA-A2 complex.
Similarly, the α chain of an anti-HCV TCR (5YXN), the β chain MART-1specific TCR DMF4 (3QEQ) 54 and p53R175H-HLA-A2 with the CDRs and peptide removed were used as search models to determine the orientation and position of the 38-10-p53R175H-HLA-A2 complex. The α chain of preproinsulinspecific TCR 1E6 (3UTP) 55 and the β chain of Nef-specific TCR T36-5 (3VXT) 56 with the CDRs removed were used as search models for molecular replacement to determine the structure of TCR 1a2. The structure of the 1a2-p53R175H-HLA-A2 complex was solved using TCR 1a2 and p53R175H-HLA-A2 as search models. Refinement statistics are summarized in Supplementary Tables 1-3. Contact residues were identified with the CONTACT program 46 and were defined as residues containing an atom 4.0 Å or less from a residue of the binding partner. The PyMOL program (https://pymol.org/) was used to prepare figures.
Surface plasmon resonance analysis. The interaction of TCRs 12-6, 38-10, and 1a2 with p53-HLA-A2 and p53R175H-HLA-A2 was assessed by surface plasmon resonance (SPR) using a BIAcore T100 biosensor at 25°C. Biotinylated p53-HLA-A2 or p53R175H-HLA-A2 was immobilized on a streptavidin-coated BIAcore SA chip (GE Healthcare) at 3000 resonance units (RU). The remaining streptavidin sites were blocked with 20 μM biotin solution. An additional flow cell was injected with free biotin alone to serve as a blank control. For analysis of TCR binding, solutions containing different concentrations of 12-6, 38-10, or 1a2 were flowed sequentially over chips immobilized with p53-HLA-A2, p53R175H-HLA-A2, or the blank. Both equilibrium and kinetic data were fitted with a 1:1 binding model using BIA evaluation 3.1 software.
Computational structural analysis. Previously determined structures of TCR complexes and their binding parameters were obtained from TCR3d (https://tcr3d. ibbr.umd.edu) 29 . The set of 151 MHC class I complex structures from TCR3d was filtered to retain only complexes with αβTCRs, and to remove redundant complexes with identical TCR CDR loop and epitope sequences; this resulted in a set of 133 complex structures that was used for comparisons of docking orientations, positions, and contacts. Calculation of docking and incident angles was performed as previously described 22 . Calculation of ΔΔG for peptide point mutations was performed using Rosetta (release 2019.45), following a previously reported computational mutagenesis protocol 27 . This protocol was executed as a Rosetta Script, for which the code is available on Github on the Kortemme Lab ddg repository (https://github.com/Kortemme-Lab/ddg/), as part of the "alanine-scanning" protocol capture. The updated "REF15" scoring function in Rosetta 57 was used for packing and minimization during computational mutagenesis, and interaction ΔΔG were calculated with Rosetta's "interface" weights. Calculations of solventaccessible surface areas were performed using the naccess program (http://wolf. bms.umist.ac.uk/naccess/). S c shape complementarity values were computed by the "sc" program in the CCP4 suite 46 .
Calculation of TCR centers. TCR-pMHC complexes were oriented into a common reference frame centered at average Cα atom position of MHC helices, with helix residues as defined previously 21 , and rotated such that the x-y plane is parallel with the helices, and the x-axis is parallel to peptide groove, with greater x value corresponding to peptide C-terminus. Complex structures in this reference frame are downloadable from the TCR3d database (https://tcr3d.ibbr.umd.edu/ downloads). TCR variable domain centers were calculated by taking centers of individual variable domains by average positions of Sγ atoms of conserved Cys residues (or Cα atoms at corresponding positions where Cys residues are not present in the TCR), and then calculating the mean position of TCR Vα and Vβ centers. X position (x pos) and y position (y pos) values represent projections into the x-y plane, and thus the MHC plane, of these centers.
Reporting summary. Further information on research design is available in the Nature Research Reporting Summary linked to this article.