Structural Basis for Self-Discrimination by Neoantigen-Specific TCRs

Physical interactions between T cell receptors (TCRs) and mutation-derived tumour neoantigens (neoAg) presented by major histocompatibility class-I (MHC-I) enable sensitive and specific cytolysis of tumour cells. Adoptive transfer of neoAg-reactive T cells in patients is correlated with response to immunotherapy; however, the structural and cellular mechanisms of neoAg recognition remain poorly understood. We have identified multiple cognate neoAg:TCRs from B16F10, a common murine implantable tumour model of melanoma. We identified a high affinity TCR targeting H2-Db-restricted Hsf2K72N that conferred specific recognition of B16F10 in vitro and in vivo. Structural characterization of the peptide-MHC (pMHC) binary and pMHC:TCR ternary complexes yielded high-resolution crystal structures, revealing the formation of a solvent-exposed hydrophobic arch in H2-Db that enables multiple intermolecular contacts between pMHC and the TCR. These features of structural stability strikingly mimic that of a previously published influenza peptide-H2-Db complex and its corresponding TCR, suggesting that there are shared structural motifs between neoantigens and viral peptides that explain their shared immunogenicity.


Main
The TCR is a variable heterodimeric protein complex that non-covalently binds to the surface-bound peptide-major histocompatibility complex (pMHC), which presents peptide antigens derived from degraded intracellular proteins 1 . Anti-tumour T cell immunity is mediated by the physical interaction between T cell receptors (TCR) and tumour antigens presented by pMHC on tumour cells 2 . Tumour cells accumulate somatic non-synonymous mutations encoding variant proteins that degrade to form mutation-derived tumour neo-antigens (neoAg) 3 .
Analogous to pathogens, tumours evolve in hosts under selective pressure from endogenous and treatment-induced immunity 4 . However, immunogenic neoAg can persist despite selective immunoediting and are increasingly recognised as the primary target of tumour-reactive TCRs [5][6][7] .
Studies demonstrate a causal relationship between neoAg-reactive T cells and radiographic regression of established tumours and/or prolonged disease-free and overall survival 16 . High functional avidity/structural affinity has emerged as a recurrent feature of neoAg-reactive TCRs and may be necessary to recognise tumour cells naturally selected for low target antigen surface density. In early examples, this has been shown to potentially derive from TCR recognition of structural differences between mutation-derived neoAg and the corresponding wild-type product, but the broader generalizability of these findings remains unknown 24 . Many other core questions remain unanswered, such as why non-synonymous mutations are rarely recognised by TCRs; and how some neoAg-reactive TCRs selectively recognise mutated peptides and do not cross-react with the corresponding wild-type peptides, whereas others exhibit significant cross-reactivity. Structure-guided mechanistic answers to these questions might enable the prediction of neoAg-reactive TCR activity as well as potential toxicities resulting from cross-reactivity, thus enabling the rapid translation of safe and effective neoAg-reactive TCRs into the clinic.
To systematically address these and other questions pertaining to neoAg-reactive TCRs, we employed the B16F10 murine melanoma cell line. B16F10 is an orthotopic implantable tumour model syngeneic to C57BL/6 mice that exhibits limited spontaneous immunogenicity and is refractory to multiple types of immunotherapy. We reasoned that neoAgs identified in this model would appropriately represent neoAgs in advanced human cancers, thereby improving the translational relevance of our findings. To identify expressed non-synonymous mutations, we first performed exome and transcriptomic sequencing of the B16F10 and selected a subset of these putative neoAg based on published ranking criteria for in vivo validation 25 . We then characterised the vaccine-induced CD8 + T cell response to seven novel neoAgs. Finally, using biochemical and cellular assays in combination with high-resolution crystal structures of the prototype murine neoAg Hsf2 p.K72N neoAg-MHC complex, with and without a corresponding reactive TCR, we determined the structural requirements for TCR antigen recognition and selectivity, as well as conditions for tumour cell recognition in vitro and tumour growth control in vivo.

Identification of Model Neoantigens in the B16F10 Model
To identify B16F10 neoAg, we performed paired exome sequencing of cultured B16F10 murine melanoma cells and C57BL/6 splenocytes, as well as bulk RNASeq of resected B16F10 tumours (Fig. 1a). We then integrated these data and identified putative neoAg using published methods 25,26 . We then performed murine immunization studies using a peptide-based vaccine 27,28 .

Biophysical analysis of H2-D b /Hsf2 p.K72N68-76
We and others have previously shown that the position of the mutated amino acid with respect to peptide length can be used to organ neoAg into two principal classes [32][33] . Namely, neoAg in which the mutated amino acid side-chain is solvent facing and may form intermolecular bonds with incoming TCR directly (class I); versus neoAg in which the mutated amino acid side-chain is buried within the MHC-I binding pocket and therefore variably interacts with incoming TCR (class II).
Hsf2 p.K72N68-76 is a H2-D b restricted peptide (68YGFRNVVHI76) derived from Heat shock factor 2 (Hsf2, Uniprot: P38533). The underlying point mutation results in substitution of a basic Lys/K residue at position 5 of wild type Hsf268-76 (pK5; 'p' indicating peptide residue, with the number designating the position of the residue in the peptide starting from the N-terminus) for a polar non-charged residue Asn/N (pN5). In silico binding analysis demonstrated that both Hsf2 p.K72N68-76 [0.007 %Rank] and Hsf268-76 [0.29 %Rank] were predicted to bind H2-D b , but a significant affinity differential exists between the mutant (MT) and wild type (WT) peptide ( Fig. 1f). Cell-based RMA-S binding confirmed this with observed half-maximal stabilization of  (Fig. 3a). Based on the observed binding affinity differential we hypothesised that the immunogenicity of Hsf2 p.K72N is primarily derived from physical interactions between the mutant residue pN5 with H2-D b .
Thus, Hsf2 p.K72N is a protype murine class II neoAg, and, understanding the nature of these interactions may elucidate the mechanism leading to the immunogenicity of Hsf2 p.K72N68-76, and class II neoAg generally.
Analysis of published H2-D b crystal structures demonstrates a conserved peptide binding mode mediated by primary hydrophobic interactions between conserved residues lining the H2-D b A-B-D and F-pockets and peptide N-/C-terminal anchor residues 34 . Additionally, H2-D b bound peptides form characteristic secondary polar interactions within the MHC-I C-pocket, mediated by bi-directional hydrogen bonds between asparagine residues H2-D b N97 and peptide position 5 (pN5). The hydrogen bond mediated by H2-D b N97 leads to biased presentation of peptides with pN5. Lastly, H2-D b is defined by a conserved hydrophobic bridge formed by the side chains of W73 (a1-helix), W147, and Y156 (a2-helix) that runs perpendicular to the binding cleft and imparts an arched solvent-accessible conformation to residues in p6-p8 of H2-D b bound peptides which is absent from H2-K b bound peptides 34 . We hypothesised that Hsf2 p.K72N68-76, but not WT Hsf268-76, satisfied the pN5 requirement imposed by H2-D b , and that the p6-p8 residues would form a solvent-exposed ridge accessible to incoming TCR. To validate this, we produced the soluble H2-D b /Hsf2 p.K72N68-76 (YGFRNVVHI) molecules as previously described, then solved the crystal structure of the binary peptide-MHC complex to 1.74Å ( Fig.   3b) 35,36 . We observed a typical H2-D b peptide binding mode mediated by hydrophobic interactions between the H2-D b A, B, D, and F-pockets and buried peptide residues pY1-pF3 and pI9, respectively-all of which are conserved between both MT/WT peptides and therefore unlikely to produce differential immunogenicity (Fig. 3c,d).  (Fig. 3f). Overall, this arrangement suggests H2-D b /Hsf2 p.K72N68-76 is stabilised primarily at the C-terminus, via the conserved pIle9 anchor as well as the pAsn5 anchor created by the p.K72N mutation with limited contribution from the N-terminal segment. Additionally, the C-terminal region's thermal rigidity and solvent exposure suggest this segment may be preferentially targeted by H2-D b /Hsf2 p.K72N68-76 reactive TCR 34 . The combined structural and biochemical binding affinity data suggest that WT Hsf268-76 with an N-terminal segment that is poorly compatible with H2-D b , and lacking pN5 anchor would fail to stabilise H2-D b , thereby explaining the differential binding affinity of the MT and WT peptides. Furthermore, the absence of the pN5 anchor would fail to stabilise the p6-p8 solvent exposed peptide ridge for interaction with incoming TCR. Finally, the fact that this feature results from the neoAg-specific pN5 anchor residue provides a plausible structural correlate for the published association between neoAg peptide-MHC stability in solution and CD8 + T cell reactivity 37 .

Biophysical analysis of TCR-47BE7:H2-D b /Hsf2 p.K72N
Having identified a plausible physical mechanism for enhanced binding of the neoAg Hsf2 p.K72N68-76 to H2-D b , we sought to understand the physical basis for TCR recognition. We hypothesised that TCR recognition was mediated by the structural interactions between the TCR and mutant peptide specific structural feature dependent on the pN5 anchor, the solvent-exposed p6-p8 ridge. TCR 47BE7 (Vα7-1:Jα21, Vβ2:Dβ2:Jβ2-1) is derived from a vaccine-induced cytotoxic CD8 + T cell clone that recognises the H2-D b /Hsf2 p.K72N68-76 with sub-nanomolar functional avidity (EC50 5.61pM, 95%CI 5.15-6.11pM). To understand the structural basis for 47BE7 binding, we synthesised recombinant soluble TCR as previously described 38 . Correct folding and preserved substrate recognition in solution were determined by measuring interaction kinetics between 47BE7 and H2-D b /Hsf2 p.K72N68-76 by biolayer interferometry (BLI). 47BE7 bound immobilised H2-D b /Hsf2p.K72N68-76 with affinity typical for non-self-reactive TCR (KD 2.7±0.3 µM) (Fig. 4a) 39 . Complex on and off-rates were too fast to be determined with precision.
To identify the physical basis for TCR binding, we crystallised and solved the 47BE7/H2-D b /Hsf2 p.K72N68-76 ternary structure to 2.6Å (Fig. 4b). The crystal structure exhibited defined electron density at the TCR:pMHC interface, allowing unequivocal placement of all critical amino acid side chains at the complex interface. TCR 47BE7 exhibited conventional oblique docking geometry (13.13º incident-angle, 57.49º crossing-angles). The TCR centroid was biased towards the peptide C-terminus, overlying the solvent-exposed and structurally rigid p6-p8 segment identified in the pMHC binary structure and hypothesised to contribute to TCR binding ( Fig. 4c). The total interface BSA was 842.3Å 2 (364.9Å 2 TCRα, 477.4Å 2 TCRβ); of which 63.54% and 36.46% was made with H2-D b and peptide, respectively. Interfacial contacts between 47BE7 and H2-D b /Hsf2 p.K72N68-76 were mediated by complementarity determining region (CDR) loops CDR1α, CDR2α, CDR3α, CDR2β, and CDR3β, with little contribution from CDR1β. (Fig. 4d,e). The bound and unbound structure of H2-D b /Hsf2 p.K72N68-76 was relatively preserved (0.6Å Cα RMSD), arguing against gross structural re-organization of the peptide-MHC on TCR complexation (Extended Data Fig. 3a). The buried residue H2-D b Y159 projected into the E pocket in the unbound structure but was observed to rotate 83.54º towards the D pocket in the bound structure, and the space previously occupied by Y159 is instead filled with glycerol (Extended Data Fig. 3b). The presence of the glycerol contaminant in this position did not significantly alter other amino acid side chains, preserved the expected p6-p8 arched peptide confirmation, and had no appreciable impact on the TCR interface.
Binding of the TCR to H2-D b was mediated by the solvent-exposed residues H2-D b Q72, R75, R80 (H2-D b α1 helix), as well as G18-E19 (H2-D b loop A) which undergo re-organization on complexation. (Extended Data Fig. 3c). The H2-D b loop A translates 4.8Å, and concomitantly the side-chain of E18 rotates 161.14º vis-à-vis the backbone Cα towards while E19 rotates 104º away from the H2-D b α1 helix. This places E18 within the hydrogen bond distance of R75 and R79, which positions R75 to form hydrogen bonds with the hydroxyl and carboxyl groups of CDR2β S51 and the Y52 carboxyl (Extended Data Fig. 3d). Additional interfacial TCR-MHC polar interactions include hydrogen bonds between CDR2β M56-Q72 and CDR3β Q97 and N80, as well as additional salt bridges between CDR3β E97 and K146 (H2-D b α2 helix). Polar interactions between TCRα and H2-D b were limited to a single hydrogen bond between CDR1α Y32-S150 (H2-D b α2 helix) (Extended Data Fig. 3d).
As expected, binding of the TCR to the Hsf2 p.K72N68-76 peptide was mediated by the Cterminal epitope comprised of pV6-pH8, which formed a rigid solvent exposed arch. In agreement, a comparison of solvent-exposed surface area (SASA) by peptide residue of the bound and unbound pMHC complex show significant SASA reduction for pR4 and pV6-pH8, which comprise the peptide contribution to the core epitope buried by TCR 47BE7 (Fig. 4f). For the remaining peptide residues-including the mutated residue pN5-SASA does not change significantly on complexation, suggesting these side-chains do not contribute to ternary complex formation. Additionally, the observed polar interactions were limited to the CDR3α Y94 hydroxy and the pR4 backbone carboxyl; the CDR3β Y101 hydroxyl and the pH8 side chain imidazole and backbone amide; as well as a salt bridge between CDR3β E97 and pH8. Other electrostatic interactions were observed in the form of water bridging between CDR2β D57 as well as CDR3α N95 and pR4. Finally, non-polar van der Waals (VdW) contacts form between CDR1α Q31 and Y32 and pV6 and, as well as CDR3α Y32, CDR3β Y101 and pVal7. Collectively, these data suggest that the TCR 47BE7 epitope is biased towards the peptide C-terminus, in a region of pre-existing structural rigidity within H2-D b /Hsf2 p.K72N68-76 binary complex. The epitope surface chemistry is defined by sparse peripheral polar contacts with solvent-exposed amino acids pR4 and pH8.
Centrally, there is a densely packed hydrophobic patch stabilised by multiple VdW interactions with the peptide arch formed by sequential pV6-pV7 residues. (Extended Data Fig. 3e).
Surprisingly, while the mutant pAsn5 residue lies within the core epitope, it makes no significant side-chain contacts with TCR-47BE7, suggesting that it has a minimal direct impact on TCR binding. Instead, the pN5 anchor indirectly stabilises the critical components of the core epitope required for TCR binding.

Structural Basis for Neoantigen-specific TCR Selectivity
Selectivity for the mutant peptide is an important characteristic of neoantigen-reactive TCRs that is hypothesised to provide an increased therapeutic window relative to TCR responsive to non-mutated tumour antigens. In cell-based cytokine production assays, we found TCR 47BE7 to be approximately 1.55×10 6 fold more sensitive to the mutant peptide (EC50, 5.6pM, 95%CI, 5.2-6.1pM), relative to the WT peptide (EC50, 8.7µM 95%CI, NR) (Fig. 2d). This difference was several orders of magnitude larger than the 175-fold difference in binding affinity observed in RMA-S MHC-I stabilization assays (Fig. 3a). Moreover, in response to saturating peptide concentrations, we observed significantly greater cytokine production on a per-cell basis when exposed to MT versus the WT peptide, suggesting that the WT peptide functions as a partial agonist only. Notably, TCR 47BE7 binds via an epitope comprised of the solvent-exposed , which are conserved between Hsf2 p.K72N68-76 (YGFRNVVHI) and Hsf268-76 (YGFRKVVHI). The non-conserved MT residue pN5 side-chain provides stabilization to the pMHC but is not essential for peptide-presentation and does not directly contribute to TCR bond formation (Fig. 4d). This suggests that secondary structural factors indirectly related to the p.K72N amino acid substitution, such as stabilization of the pV6-pH8 arch, contribute to antigen discrimination by TCR-47BE7.
To further characterise the biochemical basis for antigen discrimination we created a positional scanning peptide library. Each position in Hsf2 p.K72N68-76 was replaced with the remaining 19 protein-coding amino acids then cytokine production by 47BE7-expressing tgTCR CD8 + T cells was assessed (Fig. 4g). Side-chain substitution of the non-core anchor residues pY1-pG2 and pI9 were generally tolerated. A limited set of conservative, primarily aromatic substitutions was tolerated at the non-core residue pF3. Therefore, these residues are nonessential for TCR binding. Conversely, side-chain substitutions within the core epitope pArg4- This observation could be explained by both the lack of significant interfacial contacts between 47BE7 and the side chain of pN5, as well as the possibility that additional P5 residues may allow stabilization of the core epitope. Consistent with this, analysis of available PDB H2-D b structures demonstrates that alternative p5 residues, including glycine, alanine, aspartate, histidine and methionine, adopt conformations similar to that of asparagine (Extended Data Fig. 4a,b).
Hypothesizing that a stable peptide-MHC interaction is necessary for TCR binding, we performed in silico binding affinity analysis of all position 5 substituted peptides. We observed a weak direct correlation between predicted binding affinity to MHC-I and TCR-47BE7 reactivity.
Notably, the WT Hsf268-66 (YGFRKVVHI) was a significant outlier to this trend, suggesting differential peptide-MHC binding affinity alone was not sufficient to explain the selectivity of TCR-47BE7 (Fig. 4h).
To identify alternative structural explanations, we performed in silico docking studies of Hsf268-76 using the H2-D b /Hsf2 p.K72N binary structure as a template as previously described 6 .
Initial models suggested that the H2-D b C pocket is unable to accommodate the large pK5 without steric clashes with H2-D b residues lining the binding pocket. Thus, binding of WT Additionally, the extended pK5 is sterically incompatible with H2-D b W73 (a1 helix). As a result, W73 which is buried in the reference structure (SASA 5.695Å 2 ), rotates 180º (9.57Å) to become solvent-exposed (53.137Å 2 , SASA) (Fig. 4i). This rotation disrupts the conserved H2-D b W73-W147-Y156 bridge and changes the surface topology and hydrophobicity in this region of the core TCR epitope (Fig 4j). Specifically, in the TCR-47BE7/H2-D b /Hsf2 p.K72N68-76 ternary crystal structure the affected area is a cavity bounded by the pR4 and pH8, H2-D b E72 and R75 which collectively coordinate four water molecules. We previously observed these water molecules to form water bridges with TCR-47BE7 CDR3α N95. Rotation of the H2-D b Trp73 hydrophobic side-chain into this position likely expels these waters and abrogates electrostatic interactions between CDR3α N95 and H2-D b E72. Collectively, our modeling suggests that TCR-47BE7 specificity towards H2-D b /Hsf2 p.K72N is not solely attributable to robust peptide-MHC stability, but also to changes in surface topology and hydrophobicity at the TCR-peptide/MHC binding interface.

Discussion
There is significant interest in the study of neoantigens and corresponding reactive TCR due to published association with clinical outcomes in patients treated with tumour immunotherapy. However, clinical translation of these findings is limited by the lack of relevant pre-clinical models for testing fundamental assumptions of neoantigen biology. The experiments we present were designed to provide a preclinical model for studying neoAg-reactive TCR structure-activity relationships.
We first identified and performed a basic characterization of several TCR-antigen combinations from the widely-utilised B16F10 melanoma model and present these to the community for further study. We then completed in-depth biochemical and structural studies of the prototype anchor-residue modified neoantigen H2-D b /Hsf2 p.K72N and the corresponding monoclonal TCR 47BE7. We selected H2-D b /Hsf2 p.K72N68-76 and 47BE7 for characterization due to demonstrable in vitro and in vivo activity in a challenging tumour model. Furthermore, 47BE7 exhibited high functional avidity and limited cross-reactivity in our in vitro studies, which we hypothesised could provide insight into mechanisms underlying binding and crossreactivity of neoAg-reactive TCRs.
We found that the lysine to asparagine substitution at the position five anchor residue results in a 175-fold improvement in the surface presentation of H2-D b /Hsf2 p.K72N68-76 relative to the wild-type Hsf268-76. The crystal structure of H2-D b /Hsf2 p.K72N demonstrates that the mutated pAsn5 residue is directly responsible for this effect, due to stabilizing polar interactions between pN5 and H2-D b N97. As an additional direct consequence of the anchor residue mutation, the carboxy-terminal segment of the neoAg peptide distal to pAsn5 forms a rigid solvent-exposed hydrophobic arch which is essential for binding to TCR 47BE7. The structural stability of neoAg pMHC has been repeatedly associated with immunogenicity 29,30 .Our data suggest the plausibility of this association being driven by increased pMHC surface abundance secondary to slow peptide disassociation kinetics. We expand on these findings by showing that pMHC stability measures may indirectly capture fine structural features associated with immunogenicity, including rigid structural elements necessary for TCR recognition. We propose that when these Recently, several groups have published structural studies of human neo-reactive TCR 22,24,40 . The structural data we present supports and expands on these earlier findings in several noteworthy ways. First, we observed high-level commonalities between TCR 47BE7/Hsf2 p.K72N and TCR9a/TCR10, as well as TCR4, which bind to the class II (anchor-residue modified) neoAg HLA-C*08:02/Kras p.G12D 22,41 . These TCRs employ a similar binding mode characterised by multiple intermolecular contacts distributed across the TCR:pMHC interface.
Experimental modification of contacted residues within the core TCR epitope eliminates TCR reactivity, suggesting that the totality of the interface is necessary in both instances for TCR binding, similar to Hsf2 p.K72N. This binding mode contrasts notably with that employed by neoAg-reactive TCR that bind to class I (solvent-exposed residue modified) neoAg such as TCR12-6/TCR38-10, which bind to the neoAg HLA-A*02:01/TP53 p.KR175H 39 . In this latter circumstance the observed TCR contacts are biased towards the solvent-exposed mutant residue and avoid contacts with the remaining peptide surface. While we observe some bias towards the peptide C-terminus in the footprint of TCR 47BE7, the eccentricity is not as extreme as has been described for class I neoAg. These observations, while speculative, suggest that there may be stereotyped binding modes exhibited by neoAg-reactive TCR targeting class I and class II neoantigens, and by extension, predictable, albeit noisy, rules governing neoAg immunogenicity.
Further elucidation of the rules governing these interactions may enable rapid clinical translation of safe and effective neoAg-reactive TCRs.

Protein crystallization and Data Collection
TCR 47BE7 and H2-D b /68YGFRNVVHI were refolded and purified separately as described above. Protein crystallization was performed by sitting drop vapor diffusion. 96-well Intelliplates (Art Robbins Instruments) were seeded with a Mosquito crystallization Robot (SPT Labtech) utilizing a 1:1 v/v protein to precipitant ratio then incubated at 18°C until crystal formation. H2-D b /YGFRNVVHI formed prism-shaped crystals (Tris-HCl 0.1 M, pH 8.5, sodium acetate 0.2 M, PEG3350 20-25%). Crystals were cryoprotected with the same mother liquor, supplemented with ethylene glycol 25% v/v then flash-frozen in liquid nitrogen and stored until use.

Structure Solution and Refinement
X-ray diffraction data were collected at Argonne APS beamline 19BM. Data were indexed, integrated, and scaled using HKL3000 and the AIMLESS/CCP4 program suite. The crystal structure of H2-D b /YGFRNVVHI was solved using PHASER with the reference search model (PDB: 5OPI) and refined using REFMAC and COOT. The H2-D b /68YGFRNVVHI binary structure coordinates and structure factors are accessible via Protein Data Bank accession code (PDB: 7N9J). The structure of TCR 47BE7/H2-D b /68YGFRNVVHI ternary complex was solved using the binary complex as a search model, the 2FoFc map of which was used to sequentially build and refine the TCRα and TCRβ chains for TCR_47BE7 heterodimer by alternating refinement in REFMAC with model building and refinement in COOT. The 47BE7/H-2-Db/68YGFRNVVHI ternary complex coordinates and structure factors are accessible via Protein Data Bank accession code (PDB: 7NA5)

Structural Modeling
Structural modeling of wild-type H2-D b /Hsf268-76 was performed using Rosetta. The high resolution crystal structure of H2-D b /Hsf2 p.K72N68-76 served as the template, into which the position 5 arginine to lysine mutation was introduced using the PyMol mutagenesis function. Energy minimization was then performed using FlexPepDock (59). The top scoring models were visually inspected in PyMol, Ca and side-chain positioning was determined to be similar. The top scoring model was selected for display. Images were generated using PyMol.