TL;DR
This paper reveals conserved structural interactions between T cell receptor (TCR) loops that shape antigen binding. It also demonstrates that TCR alpha and beta chains are not independently selected within antigen-specific repertoires, showing a significant sequence constraint that can be leveraged by co-evolutionary models to predict TCR pairing.
What Problem Does This Paper Address?
T cell receptors (TCRs) are crucial for adaptive immunity, recognizing peptide-MHC complexes to initiate immune responses. While the complementarity-determining regions (CDRs) of TCRs are known to interact with the peptide-MHC (pMHC) and determine specificity, the intricate structural contacts between the TCR's own alpha and beta chains, and how these contacts influence sequence selection, remain less understood. This lack of understanding limits our ability to predict TCR function and design TCR-based therapies. The paper aims to systematically map these intra- and inter-chain CDR contacts and investigate whether sequence constraints exist between TCR alpha and beta chains within antigen-specific repertoires.
Methods
The researchers employed a multi-pronged approach:
- Structural Analysis: They analyzed a large dataset of available TCR-pMHC crystal structures to map conserved intra-chain (within a single TCR chain) and inter-chain (between alpha and beta chains) contacts between CDR residues. Contacts were defined by a minimum distance threshold of 5Å between non-hydrogen atoms.
- Sequence Analysis and Mutual Information (MI): To quantify sequence constraints, they used a mutual information framework. This statistical measure assesses the dependence between two variables, in this case, different regions of the TCR. They calculated MI between various TCR components (V genes, J genes, CDR3 sequences) to understand how information about one region restricts possibilities in another.
- Co-evolutionary Pairing Algorithms: Leveraging the idea that interacting protein families often co-evolve, they adapted and applied three co-evolution-based methods (MI-based Iterative Pairing Algorithm - MI-IPA, Graph Alignment - GA, and a combined GA+MI-IPA) to predict TCR alpha-beta pairing within antigen-specific repertoires. These methods were used in a "de novo" setting, meaning they were not trained on known TCR pairs for each epitope.
- Benchmarking: The performance of their co-evolutionary models was compared against existing methods for epitope prediction (TULIP) and TCR sequence embedding (SCEPTR).
Datasets
The study utilized several publicly available datasets:
- The Structural T-Cell Receptor Database (STCRDab): Used for analyzing existing TCR-pMHC crystal structures.
- VDJdb database: A comprehensive database of TCR sequences and their associated epitopes, used for analyzing paired-chain TCR data and training/evaluating pairing algorithms.
- Observed T Cell Receptor Space (OTS) database: Used for analyzing polyclonal TCR subsets.
- Data from Tanno et al., 2020: Used for a control analysis of sorted naïve T cells.
No new datasets were generated for this study.
Key Insights and Ideas
- CDR Loops Form a Unified Binding Surface: The analysis of TCR structures revealed that the six CDR loops, not just those directly contacting the pMHC, form a continuous and conserved surface. This suggests a coordinated structural arrangement that likely influences antigen binding.
- Inter-chain CDR3 Interactions are Key: While germline loops (CDR1, CDR2) showed conserved inter-chain contacts, the most striking observation was the extensive and conserved interaction between the CDR3 regions of the alpha and beta chains. This "X-shaped" pattern suggests that the highly variable CDR3s are not just interacting with the pMHC but also with each other, imposing constraints.
- Antigen-Specific TCRs Aren't Randomly Paired: Contrary to previous findings that suggested weak or no constraints on TCR alpha-beta pairing at a repertoire level, this study found significant sequence constraints within antigen-specific repertoires. This implies that the selection of TCR alpha chains is not independent of the selection of TCR beta chains when targeting a specific antigen.
- Co-evolutionary Models Can Predict Pairing: The adapted co-evolutionary algorithms (MI-IPA, GA, GA+MI-IPA) demonstrated a significant ability to predict correct TCR alpha-beta pairings, outperforming existing methods and performing better than random chance. This suggests that the inter-chain constraints identified carry a learnable signal.
Key Results
- Conserved CDR Contacts: Analysis of 340 TCR structures revealed conserved intra- and inter-chain contacts between CDR loops, particularly between CDR1 and CDR3, and between the CDR3 regions of the alpha and beta chains.
- Mutual Information (MI) Signals Constraints: MI calculations showed significant statistical dependence between TCR alpha and beta chains in antigen-specific repertoires, indicating sequence constraints. This dependence was generally higher than in background or naïve repertoires.
- Co-evolutionary Pairing Performance: The MI-IPA, GA, and GA+MI-IPA models achieved significantly better precision in predicting TCR alpha-beta pairings than random chance across most epitopes. For example, the GA model showed better than chance performance in 19/22 epitopes.
- Model Performance Correlates with Repertoire Characteristics: The performance of the pairing models was found to correlate with factors like repertoire size, effective set size, and sequence similarity within the repertoire. Smaller effective set sizes and higher sequence clustering tended to correlate with higher MI and better model performance.
Limitations and Future Work
- Limited Epitope Diversity: The study analyzed 22 epitopes. A much larger selection of pMHCs would be needed to fully understand the factors determining the relationship between TCR interactions and pMHC binding.
- Data-Driven Limitations: The authors acknowledge that the pairing algorithms' performance is likely limited by the available data and the inherent signal strength. They posit that the theoretical maximum pairing accuracy is well below 1 for many repertoires, suggesting that the data only captures some of the pairing rules.
- Model Extrapolation: The MI-IPA, even with correct pairs as training, underperformed the theoretical limit, suggesting limitations in the algorithm's ability to extrapolate all learned rules.
- Functional Validation: While the study identifies structural and sequence constraints, further experimental work is needed to validate these predictions and understand their impact on TCR binding kinetics, thermodynamics, and overall function.
- Beyond Sequence: The study primarily relies on sequence data. Incorporating other information, such as clonotype abundance, might improve pairing predictions.
- Future Directions: The authors suggest exploring better ways to initialize the MI-IPA using GA results, investigating the impact of non-sequence information, and expanding the analysis to a larger set of pMHCs to fully understand epitope-specific constraints. They also highlight the potential for TCR sequence embedding methods and binding prediction models to capture pairing signals, suggesting future work in this area.
Summary updated