Structural Homology of HHV-6B Epitopes as Candidates for Molecular Mimicry Triggers of Type One Diabetes Mellitus Onset

Molecular mimicry is a mechanism by which an infectious agent may trigger an autoimmune response through structural or sequence similarity to self-antigens. Identifying potential molecular mimics is critical to understanding autoimmune disorders, including type 1 diabetes mellitus (T1DM). Human herpesvirus 6B (HHV-6B), a common childhood virus implicated in multiple autoimmune conditions, has been epidemiologically linked to a significant increased risk of T1DM for individuals with a genetic predisposition. However, the connection between HHV-6B and T1DM remains unresolved, as known HHV-6B epitopes show minimal sequence homology with T1DM autoantigens. This study explores whether HHV-6B can act as a molecular mimic by assessing structural and biding similarities between HHV-6B-derived and T1DM autoantigen-derived epitopes. Epitope peptide structures and their interactions with T1DM autoantigens were modeled using Boltz-2. Several HHV-6B epitopes studied here demonstrated high structural alignment with T1DM epitopes, but even when structural homology is lacking, the HHV-6B and T1DM epitopes fit into the same binding region of HLA molecules, suggesting a plausible mechanism for T cell cross-reactivity. These findings provide computational evidence that HHV-6B may act as a molecular mimic contributing to autoimmune responses in T1DM genetically susceptible individuals. This study demonstrates that structural modeling is a useful tool for identifying potential mimicry candidates that sequence-based methods may not find, underscoring the importance of integrating structure-based modeling including docking into molecular mimicry prediction pipelines. Structural modeling of both the 50 epitopes from the 25 candidate pairs from the Suleman studies and those three from Bach’s paper was performed with Boltz-2 (v.2.2.0), an open-source deep learning method that predicts biomolecular structures and binding sites with great accuracy. Boltz-2 requires protein inputs of at least ten amino acids, and many of the epitopes considered here (Table 1 and 2) did not meet this requirement, so each epitope sequence was expanded symmetrically using the full protein sequences from UniProt (https://www.uniprot.org/). The extended sequences are given in Table S1 of the Supplementary maerial. In all cases selected for analysis, model 0 from Boltz-2 consistently exhibited the highest confidence scores, and these models were used in subsequent analyses. The confidence scores for all the calculations average 0.74 for the structures of the isolated epitopes and 0.87 for the structures of the epitopes bound to the HLA molecules. These values are within the range of good Boltz-2 prediction.