Using Global Analysis to Extend the Accuracy and Precision of Binding Measurements with T cell Receptors and Their Peptide/MHC Ligands

Abstract

In cellular immunity, clonally distributed T cell receptors (TCRs) engage complexes of peptides bound to major histocompatibility complex proteins (pMHCs). In the interactions of TCRs with pMHCs, regions of restricted and variable diversity align in a structurally complex fashion. Many studies have used mutagenesis to attempt to understand the "roles" played by various interface components in determining TCR recognition properties such as specificity and cross-reactivity. However, these measurements are often complicated or even compromised by the weak affinities TCRs maintain toward pMHC. Here, we demonstrate how global analysis of multiple datasets can be used to significantly extend the accuracy and precision of such TCR binding experiments. Application of this approach should positively impact efforts to understand TCR recognition and facilitate the creation of mutational databases to help engineer TCRs with tuned molecular recognition properties. We also show how global analysis can be used to analyze double mutant cycles in TCR-pMHC interfaces, which can lead to new insights into immune recognition.

Keywords: T cell receptor; binding; double mutant cycle; global analysis; mutagenesis; peptide/MHC.

Figure 1

TCR-pMHC structural overview and simulated binding data. (A) Structure of a TCR-pMHC complex, showing the TCR bound to the composite pMHC ligand. (B) Simulated TCR-pMHC binding data. The wild-type data is simulated with a K_D of 2 μM and reaches 99% saturation, and mutant data with a K_D of 100 μM reaches ~50% saturation.

Figure 2

Global analysis of simulated wild-type and mutant datasets perturbed with added noise outperforms single analyses. While the average value of both is close to the real value, the average error of the global fits is 0.11 kcal/mol, compared to 0.51 kcal/mol for the individual fits. The columns correspond to increasing amounts of added noise as indicated; “Ref” refers to the actual ΔΔG° of 2.32 kcal/mol.

Figure 3

Fitting wild-type and mutant experimental data individually or globally can lead to significant differences in the ΔΔG°. The absolute value of the difference between individual and global fits, referred to as in the text as |Δ(ΔΔG°)| is shown for 25 pairs of experimental datasets. The dashed line represents a Z-score of 1.64 (90% cutoff of one standard deviation), revealing the eight datasets where the difference between individual and global fitting was most significant (noted by asterisks).

Figure 4

Perturbations to experimental datasets can significantly impact accuracy and precision whereas global analysis is substantially more robust. (A) The impact of added noise and deletion of data points for dataset 17 in Figure 3. Unlike individual fitting, global fitting handles the perturbations robustly. (B) A similar result is seen for dataset 2 in Figure 3. Of note are the instances in which perturbations to the last three wild type data points were removed (W3) or the middle three mutants removed (M3), in which the ΔΔG°-values were dramatically impacted for single fitting. The same perturbations however had little impact with global fitting.

Figure 5

Individual analysis of mutant datasets tends to overestimate degree of saturation, inflating K_D and reducing ΔΔG°. (A) Individual fits of the DMF5 TCR binding to wild-type (black) and mutant (red) pMHC. Although the exact same surface was used for both titrations, the fitted RU_max differed between the wild-type and mutant experiments. The lower value for the mutant interaction led to an inflated K_D and smaller ΔΔG°. (B) When both datasets are fit globally with RU_max as a shared parameter, the reported value is the same as the high affinity value in the wild-type single analysis. The mutant K_D was thus weaker and ΔΔG° smaller.

Figure 6

Example of a double mutant cycle analysis in a TCR-pMHC interface. When the double mutant cycle exploring the interaction between Trp101 of the A6 TCR and Ala69 of HLA-A2 was fit globally, the reported interaction free energy was 0.0 ± 0.3 kcal/mol. If the four datasets making up the interaction were individually fit, the resulting interaction free energy was −1.0 ± 0.2 kcal/mol.