De novo design of small cyclic peptide inhibitors from non-canonical amino acids by cofolding-guided search

Cyclic peptides hold the potential to interact with targets beyond the scope of small molecules, yet their discovery has traditionally depended on libraries reused for various targets, motif-based diversification, or computational methods limited to a restricted set of non-canonical amino acids. In a groundbreaking development, researchers have unveiled nCycle-Forge, a simulated annealing framework that ingeniously designs cyclic peptide binders.

This innovative method iteratively substitutes building blocks and evaluates each proposal by cofolding it with the target.

Through the application of nCycle-Forge, a library of 1,472 cyclic peptides was designed, synthesized, and screened against the active-site pocket of Thrombin and the shallow p53-binding surface of MDM2. Remarkably, these designs yielded chemically diverse hits against both targets in a single round, sans experimental optimization. The hit rates were comparable to or up to 24 times higher than those observed in larger experimental libraries.

The purified designs exhibited sub- to low-micromolar potency, with the most potent Thrombin inhibitor boasting a Ki of 0.14 M and the most effective MDM2 binder registering an EC50 of 2.4 M. Surface plasmon resonance experiments confirmed direct binding, competition with orthonosteric inhibitors at both sites, and weaker binding under conditions that would typically linearize the cyclic designs.

It is noteworthy that nCycle-Forge requires no building block-specific parameterization, conformer generation, or retraining, thus presenting a target-focused approach to generating functional cyclic peptides that transcend the limitations of the conventional canonical amino acid alphabet.