Pan-prediction of MHC-II restricted epitopes across species via an AlphaFold-based quantification scheme

Controlling infectious diseases—especially zoonoses—is critical for both the economy and public health. Developing safe, effective vaccines remains the most reliable strategy for disease control, but a key challenge in vaccine design lies in identifying effective T cell epitopes within pathogen proteomes. T cell epitopes are peptide ligands that bind to major histocompatibility complex (MHC) molecules: CD8⁺ T cell epitopes are presented by MHC class I (MHC-I), while CD4⁺ T cell epitopes are presented by MHC class II (MHC-II). Predicting MHC-II restricted epitopes is notably more difficult than MHC-I due to inherent differences in their structure and peptide-binding properties.

While the accumulation of MHC-II immunopeptidome mass spectrometry data and advances in sequence-based tools (e.g., NetMHCpan) have improved MHC-II epitope prediction, these resources primarily focus on human MHC-II (human leukocyte antigen class II [HLA-II]). Even pan-prediction tools like NetMHCpan, which rely on amino acid similarity in MHC binding grooves, perform poorly when predicting non-human MHC-II restricted epitopes.

To address this gap, this study developed AF-pred (AlphaFold-prediction), a structure-based tool for quantitative MHC-II epitope prediction. The tool first selects the optimal model for the target peptide-MHC-II (pMHC-II) complex by evaluating structural confidence metrics, then automatically aligns it with resolved reference pMHC-II structures to enable precise predictions.

To validate AF-pred’s ability to overcome species barriers and MHC-II heterodimer diversity, MixMHC2pred—trained on limited cattle and chicken MHC-II data for basic cross-species use—served as a reference. Direct comparisons were made using predictions for porcine MHC-II (SLA-DRA0101_SLA-DRB0101) with African swine fever virus (ASFV) P72 protein, and bovine MHC-II (BoLA-DRA0101_BoLA-DRB31601) with foot-and-mouth disease virus (FMDV) polyprotein. Results showed significant differences between AF2M-pred, AF3-pred, and MixMHC2pred, highlighting AF-pred’s superior cross-species performance.

Further validation used bat MHC-II (EF-DRA and EF-DRB1) paired with an overlapping peptide library from the SARS-CoV-2 spike protein’s RBD region. AF3-pred produced more rational predictions of anchoring residues and pockets, while AF2M-pred excelled when peptide-binding motifs were weakly restricted. AF3-pred, focusing on complex formation constraints, better suited tasks with clear restrictive interactions.

The crystal structure of the complex between SARS-CoV-2 spike peptide P1008 and Bat-MHC-II_EF was resolved. Comparing predicted and experimental pMHC-II structures revealed AF-pred tends to generate peptide conformations with more plausible interactions within a reasonable range.

Analysis of AF2M-pred and AF3-pred predictions for an avian MHC-II (BL2*019:01)-derived peptide (PGDSDIIRSMPEQTSEK) and its mutants binding Bat-MHC-II_EF confirmed AF-pred's ability to explore structural interactions and optimize peptides via stable conformations. AF2M-pred explored broader conformations, while AF3-pred better identified binding restrictions. Simplifying predictions to focus on the 9-amino-acid binding core (ignoring unstable flanking regions) reduced complexity and improved accuracy.

AF-pred predicts MHC-II epitopes without relying on massive immunopeptidome data, leveraging AlphaFold's structural prediction power and MHC-II’s conserved binding patterns—simplifying prediction and enabling reliable evaluation. This study confirms structure-based tools'feasibility for MHC-II prediction, with advantages in cross-species generalization and interpretability. Iterative optimization of AF-pred will enhance accuracy, supporting T cell epitope vaccine development and strengthening public health protection.

About Author:

Nianzhi Zhang he serves as an Associate Professor at the College of Veterinary Medicine, China Agricultural University, and is the director of the Preventive Veterinary Medicine discipline at the College of Veterinary Medicine, China Agricultural University, as well as the Key Open Laboratory of Preventive Veterinary Medicine of the Ministry of Agriculture.

Nianzhi Zhang has repeatedly determined the crystal structures of MHC molecules from different species and clarified the rules of antigen presentation by the corresponding MHC molecules. He has presided over one Young Scientists Fund project of the National Natural Science Foundation of China, one general project of the National Natural Science Foundation of China, and one Basic Scientific Research Project of Central Universities. He has also participated in multiple research projects, including the National Major Project on Genetically Modified Organisms, 863 Program projects, 973 Program projects, and projects funded by the National Natural Science Foundation of China.

To date, he has published 7 SCI and EI indexed papers in academic journals such as Immunity, The Journal of Immunology, and Journal of Virology.