DynaDom: structure-based prediction of T cell receptor inter-domain and T cell receptor-peptide-MHC (class I) association angles

Abstract

Background: T cell receptor (TCR) molecules are involved in the adaptive immune response as they distinguish between self- and foreign-peptides, presented in major histocompatibility complex molecules (pMHC). Former studies showed that the association angles of the TCR variable domains (Vα/Vβ) can differ significantly and change upon binding to the pMHC complex. These changes can be described as a rotation of the domains around a general Center of Rotation, characterized by the interaction of two highly conserved glutamine residues.

Methods: We developed a computational method, DynaDom, for the prediction of TCR Vα/Vβ inter-domain and TCR/pMHC orientations in TCRpMHC complexes, which allows predicting the orientation of multiple protein-domains. In addition, we implemented a new approach to predict the correct orientation of the carboxamide endgroups in glutamine and asparagine residues, which can also be used as an external, independent tool.

Results: The approach was evaluated for the remodeling of 75 and 53 experimental structures of TCR and TCRpMHC (class I) complexes, respectively. We show that the DynaDom method predicts the correct orientation of the TCR Vα/Vβ angles in 96 and 89% of the cases, for the poses with the best RMSD and best interaction energy, respectively. For the concurrent prediction of the TCR Vα/Vβ and pMHC orientations, the respective rates reached 74 and 72%. Through an exhaustive analysis, we could show that the pMHC placement can be further improved by a straightforward, yet very time intensive extension of the current approach.

Conclusions: The results obtained in the present remodeling study prove the suitability of our approach for interdomain-angle optimization. In addition, the high prediction rate obtained specifically for the energetically highest ranked poses further demonstrates that our method is a powerful candidate for blind prediction. Therefore it should be well suited as part of any accurate atomistic modeling pipeline for TCRpMHC complexes and potentially other large molecular assemblies.

Keywords: Adoptive T-cell therapy; Epitope prediction; Glutamine side chain prediction; Immunoinformatics; Protein domain association angles; T-cell recognition; TCR structural modeling; Vaccine design.

Fig. 1

Representation of the TCRpMHC complex (PDB-ID 2bnq). The MHC class I molecule is depicted in green (i.e., α1, α2, and α3 chains). The β-microglobulin is colored in cyan and the peptide bound to MHC in magenta. The two chains of TCR, α and β, are represented in blue and red colors, respectively. In the present application, the domains shown as transparent are removed from the structure, and only the two variable domains of TCR (i.e., Vα and Vβ), the α1 and α2 chains of MHC as well as the peptide are modeled. In addition, the two centers of rotations CoR_β and CoR_μ, are respectively represented by an orange and a black colored ball

Fig. 2

TCR and TCRpMHC complexes modeling pipeline. Center column: standard pipeline (see Methods) for the remodeling of the TCR Vα/Vβ association angles and for the pMHC positioning with respect to the TCR. Blue highlighted steps are performed in both modeling pipelines: only Vβ and combined Vβ/pMHC placement. Green highlighted steps are performed only if the pMHC is included in the remodeling process. The left and the right columns illustrate the individual steps of the pipeline. Steps with numbers circled in black: TCR Vα/Vβ association angle modeling pipeline, steps with numbers circled in green: combined Vβ/pMHC modeling pipeline. Steps 3 to 7 are performed for each of the 11 starting conformations. The protein domains represented in blue, red, and green color correspond to the Vα, Vβ, and pMHC units, respectively. In step 1 (both for TCR and TCRpMHC modeling), the different protein domains are described by unified cuboids and assembled. The illustration of steps 2 and 5 show the Q-flip correction/optimization. At step 2, each glutamine residue is optimized with respect to its direct environment only (only the corresponding variable domain is accounted for). Whereas in step 5, the two glutamine residues are optimized simultaneously, thus accounting for the whole TCR environment. In step 4 (only for TCRpMHC modeling), the pMHC unit is pre-placed, translated away from the TCR and optimized with respect to the fixed TCR variable domains. At step 6 (both for TCR and TCRpMHC modeling), the position of all cuboids as well as the orientation of the glutamine residues are optimized concurrently. The latter illustrations show the structure before and after optimization, with the target crystal structure depicted in gray

Fig. 3

Remodeling of the 2f53 structure. a Superposition of the 11 Vβ starting orientations with respect to the Vα domain (represented in blue color). The average conformation of Vβ is shown in red color. b Hydrogen bonds of the conserved Q-Q interaction at the CoR_β position. Left: misassigned conformation in the experimental crystal structure. Right: proper orientation of the Q residues after application of the Q-flip correction. The picture shows that the interaction between the two residues has been improved as well as the interaction of the residues with their respective environment. c Modeling of the ternary TCRpMHC complex. The Vα, Vβ, and pMHC units are represented in blue, red, and green colors, respectively. The reference crystal structure is depicted in gray color. Left: initial assembly of the complex. Right: final model with an RMSD of 0.61 Å with respect to the crystal structure. Magnifications lenses: conformation of the conserved Q-Q interaction between the Vα and the Vβ domain

Fig. 4

Percentage of structures among the 75*11 models with an RMSD value lower than 3, 2 and 1 Å. The total set of 75*11 structures is separated into structures for which the Q residues were originally paired or mispaired within their corresponding crystal structure. Each histogram box corresponds to a different setting of the modeling procedure, i.e. with only distance restraint (M_T01), only Q-flip correction (M_T10), both (M_T11), or none of them (M_T00). The percentage of structures with an RMSD value lower than 3, 2 and 1 Å are presented on the left, middle, and right plots, respectively. The right plot shows that for the structures presenting an originally wrong orientation of the Q residues, the Q-flip correction significantly improves the quality of the resulting model (i.e., M_T10 and M_T11)