Researchers at Stanford used AI protein design tools to engineer a tiny, rigid four-helix protein that mimics how T cell receptors recognize cancer-specific peptides on tumor cells. Their mini-TCR mimic bound the NY-ESO-1 tumor antigen presented by HLA-A*02 with 9.5 nM affinity and showed no detectable binding to a control peptide on the same MHC molecule. A crystal structure at 2.05 Å resolution confirmed the binder docks at a TCR-like 40-degree angle and makes focused contacts with the peptide's protruding side chains. A computational screen of over 14,000 HLA-A*02 peptides correctly predicted the only two real off-target peptides. As a T cell engager format, it killed cancer cells selectively in cytotoxicity assays.
