Poor correspondence between predicted and experimental binding of peptides to class I MHC molecules

Naturally processed peptides presented by class I major histocompatibility complex (MHC) molecules display a characteristic allele specific motif of two or more essential amino acid side chains, the so-called peptide anchor residues, in the context of an 8-10 amino acid long peptide. Knowledge of the peptide binding motif of individual class I MHC molecules permits the selection of potential peptide antigens from proteins of infectious organisms that could induce protective T-cell-mediated immunity. Several methods have been developed for the prediction of potential class I MHC binding peptides. One is based on a simple scanning for the presence of primary peptide anchor residues in the sequence of interest. A more sophisticated technology is the utilization of predictive computer algorithms. Here, we have analyzed the experimental binding of 84 peptides selected on the basis of the presence of peptide binding motifs for individual class I MHC molecules. The actual binding was compared with the results obtained when analyzing the same peptides by two well-known, publicly available computer algorithms. We conclude that there is no strong correlation between actual and predicted binding when using predictive computer algorithms. Furthermore, we found a high number of false-negatives when using a predictive algorithm compared to simple scanning for the presence of primary anchor residues. We conclude that the peptide binding assay remains an important step in the identification of cytotoxic T lymphocyte (CTL) epitopes which can not be substituted by predictive algorithms.