Peptide agonist design from GPCR sequences

Abstract

Simplifying protein agonists into peptides, or identifying alternative peptide agonists, could substantially enhance the pharmaceutical utility of many GPCR and is therefore highly desirable for biomedical applications. However, de novo peptide design for these receptors remains challenging, as agonism requires both binding to and activation of the targe; such ontarget activity is difficult to incorporate by current approaches that primarily aims at binding. Here we report that based on cross-attention mechanism, our new model GRADE accurately predicts agonist-GPCR pairing using only sequences, and designed peptide agonists across multiple targets with success rates of 20%-40%. Further with the CCR/CXCR family as a focal group, GRADE resulted in agonist with expanded agonistic spectrum than natural chemokines, as well as target-specific agonists. Structural determination revealed GRADE-designed agonists bind to CCR5 orthosteric pocket with novel, reversed polarity of natural chemokines. In a murine model of tumor immune clearance, selected agonists activated CCR5-expressing CD8+ T cells and anti-tumor immunity, in particular the CCR5-specific agonist removed crossreactivity that is inherent to CCL5. With drastically reduced model complexity, GRADE offers an accurate and flexible method for GPCR peptide agonist design using only target sequences, and demonstrates the potential of cross-attention mechanism in future protein ligand design for expanded range of targets.