JMIR Publications

ABSTRACT

Human cytomegalovirus (HCMV) is a ubiquitous beta herpesvirus that is usually controlled, but never cleared, by the immune system. It remains a major cause of morbidity in transplant recipients and the leading infectious cause of congenital disability. Cellular immunity, particularly CD8⁺ and CD4⁺ T cells, is the central arm of CMV control, and for three decades the tegument phosphoprotein pp65 (UL83) and the immediate early protein IE1 (UL123) have dominated T-cell epitope discovery, monitoring assays, adoptive therapy and vaccine design. Proteome wide mapping, immunopeptidomics and viral deletion studies now show that CMV specific T cells target a far broader set of viral proteins than this paradigm implies, including IE2, pp150, UL48, UL36, gB, UL86, UL138 and products of noncanonical open reading frames. In parallel, HLA specific machine learning predictors trained on mass spectrometry eluted ligands, protein language models and structure prediction networks have transformed the ability to nominate candidate epitopes in silico. Prediction, however, is not immunogenicity, and HLA binding is not natural presentation. This review (i) summarizes the evidence for CMV T- cell epitopes beyond pp65 and IE1, grading antigens by the strength of experimental support; (ii) compares experimental and computational discovery pipelines; (iii) discusses HLA diversity and population coverage, including the need to evaluate candidates in genetically diverse populations such as those of South Asia; (iv) evaluates multiantigen and multiepitope vaccine strategies; and (v) proposes a ten step framework (CMVAITCell) that couples multiparameter in silico prioritization to mandatory experimental validation. We conclude that expanding antigenic breadth is biologically justified, that AI/ML can enlarge and rank the candidate epitope space, and that demonstrated natural presentation and T-cell recognition remain the nonnegotiable currency of credible CMV vaccine design.