CD4+ T-cell epitope prediction by combined analysis of antigen conformational flexibility and peptide-MHCII binding affinity

Tysheena P. Charles, Daniel L. Moss, Pawan Bhat, Peyton W. Moore, Nicholas Kummer, Avik Bhattacharya, Ramgopal R. Mettu, Samuel J. LandrybioRxiv 2020 · DOI

TL;DR

This paper introduces a more accurate method for predicting CD4+ T-cell epitopes by combining antigen processing likelihood (APL) with peptide-MHCII binding affinity. This combined approach significantly improves epitope prediction accuracy, which is crucial for rational vaccine design and understanding adaptive immunity.

What Problem Does This Paper Address?

Predicting CD4+ T-cell epitopes is a critical but challenging task in immunology and vaccine development. CD4+ T-cells recognize peptide fragments presented by MHC class II molecules, and their activation is influenced by a complex interplay of antigen processing, protein structure, and peptide-MHCII binding affinity. Current prediction methods often focus on just one or two of these factors, leading to suboptimal accuracy, especially in predicting dominant epitopes (those that elicit the strongest T-cell response). This paper aims to improve epitope prediction by integrating structural information about antigen flexibility and protease sensitivity with peptide-MHCII binding predictions.

Methods

The authors developed and refined an Antigen Processing Likelihood (APL) algorithm. This algorithm leverages structural data from X-ray crystallography (like b-factors and solvent-accessible surface area) and biochemical data (like linear antibody epitope frequency and protease sensitivity) to predict regions of an antigen that are likely to be processed into peptides. They then combined this APL score with predictions of peptide-MHCII binding affinity (using the NetMHCII 2.3 Server). The prediction accuracy was evaluated using positive predictive value (PPV) and receiver-operator characteristic (ROC) curves. They also explored using biochemical data (linear antibody epitope frequency and protease sensitivity) as a proxy for structural flexibility when crystal structures were unavailable.

Datasets

The study utilized data from several sources:

  • F1-V protein: This plague vaccine candidate was used as a primary model antigen for detailed epitope mapping and method validation.
  • Existing epitope mapping data: The authors compiled data from published studies on 13 diverse antigens in C57BL/6 mice and a set of antigens from human subjects. These datasets were used for training and optimizing the APL algorithm and for evaluating the combined prediction methods.

Key Insights and Ideas

  • Structure matters for processing: The conformational flexibility and protease sensitivity of an antigen, as revealed by its structure, significantly influence which epitopes are generated and presented.
  • APL as a crucial component: The APL algorithm, which incorporates structural and biochemical flexibility, is a powerful predictor of CD4+ T-cell epitopes, particularly in eliminating false positives.
  • Synergy of APL and MHCII binding: Combining APL with peptide-MHCII binding affinity provides a more accurate prediction of dominant epitopes than either method alone.
  • Biochemical data as a structural proxy: Linear antibody epitope frequency and protease sensitivity can effectively substitute for crystallographic data in APL calculations when crystal structures are not available.
  • Dominant epitopes at the edges: Dominant CD4+ T-cell epitopes often occur at the boundaries between stable and unstable (flexible) regions of an antigen, a pattern that the APL algorithm is designed to capture.

Key Results

  • For a set of 13 diverse antigens in C57BL/6 mice, the combined APL and MHCII binding prediction achieved an accuracy of ~40%, significantly outperforming APL alone (~18%) and MHCII binding alone (~27%).
  • In human subjects, APL alone achieved a significant PPV of 27% when a broader threshold was applied, and the ROC-AUC was 0.71.
  • For the F1-V antigen, the combined prediction method correctly identified 4 hits for 7 peptides predicted at the top 10th percentile in C57BL/6 mice, compared to 2 hits for APL alone and 2 for MHCII binding alone.
  • The study demonstrated that using only biochemical data (linear antibody epitope frequency and protease sensitivity) for APL calculations could achieve significant accuracy in predicting F1-V CD4+ epitopes in CBA mice.

Limitations and Future Work

  • Structural requirement for APL: A key limitation of the APL algorithm is its reliance on high-resolution 3D structures, typically X-ray crystal structures. While homology modeling can help, it's not always feasible. The authors suggest that natively unfolded proteins are likely poorly immunogenic anyway, mitigating this limitation somewhat.
  • Parameter optimization: The APL algorithm requires careful parameter optimization, which can be computationally intensive.
  • Future directions: The authors suggest further investigation into strategies to improve APL scoring, particularly for regions with modest local stability. They also highlight the potential for APL to be a major asset in predicting CD4+ epitopes in human immune responses.

Summary updated

Abstract

Abstract Antigen processing in the class II MHC pathway depends on conventional proteolytic enzymes, potentially acting on antigens in native-like conformational states. CD4+ epitope dominance arises from a competition between antigen folding, proteolysis, and MHCII binding. Protease-sensitive sites, linear antibody epitopes, and CD4+ T-cell epitopes were mapped in the plague vaccine candidate F1-V to evaluate the various contributions to CD4+ epitope dominance. Using X-ray crystal structures, antigen processing likelihood (APL) predicts CD4+ epitopes with significant accuracy without considering peptide-MHCII binding affinity. The profiles of conformational flexibility derived from the X-ray crystal structures of the F1-V proteins, Caf1 and LcrV, were similar to the biochemical profiles of linear antibody epitope reactivity and protease-sensitivity, suggesting that the role of structure in proteolysis was captured by the analysis of the crystal structures. The patterns of CD4+ T-cell epitope dominance in C57BL/6, CBA, and BALB/c mice were compared to epitope predictions based on APL, peptide binding to MHCII proteins, or both. For a sample of 13 diverse antigens larger than 200 residues, accuracy of epitope prediction by the combination of APL and I-A b -MHCII-peptide affinity approached 40%. When MHCII allele specificity is also diverse, such as in human immunity, prediction of dominant epitopes by APL alone approached 40%. Since dominant CD4+ epitopes tend to occur in conformationally stable antigen domains, crystal structures typically are available for analysis by APL; and thus, the requirement for a crystal structure is not a severe limitation.
BibTeX
@misc{charles2020cd,
  author = {Tysheena P. Charles and Daniel L. Moss and Pawan Bhat and Peyton W. Moore and Nicholas Kummer and Avik Bhattacharya and Ramgopal R. Mettu and Samuel J. Landry},
  title = {{CD4+ T-cell epitope prediction by combined analysis of antigen conformational flexibility and peptide-MHCII binding affinity}},
  year = {2020},
  doi = {10.1101/2020.05.21.109967},
  howpublished = {bioRxiv},
}
Figure