Computational Design of Lysine Targeting Covalent Binders Using Rosetta

Abstract

Chemical probes that form a covalent bond with their target protein have been established as a powerful tool for investigating proteins and modulating their activity, but until recently were mostly targeting cysteine residues. Covalent binders that target lysine residues are increasingly reported. Covalent binding to lysine involves challenges such as the increased pK_a of the side chain and its considerable flexibility. Here, we describe two computational methods to derivatize lysine-binding covalent small-molecules based on known noncovalent binders, approaching the design problem from two opposite directions. In a "ligand-side" approach, we scan different ligand positions to install an electrophile and dock these derivatized ligands into the target protein. In a "protein-side" approach, we install an electrophile on the target lysine and model its conformational space to find suitable installation vectors on the ligand. We applied both of these protocols retrospectively to a data set of electrophilic ligands and to a data set of vitamin B6 covalently bound to a receptor lysine residue. Our ligand-side protocol successfully identified the known covalent binder in 80% and 86% of cases, while the protein-side protocol achieved identification rates of 56% and 82%, respectively. We prospectively validated these protocols by designing and testing a new lysine-targeting MKK7 inhibitor. Mass-spectrometry and crystallography validated the covalent binding to the target lysine. Applying these protocols to a data set of known kinase inhibitors identified high-confidence covalent candidates for more than 200 human kinases, demonstrating the potential impact of our protocols.

1

Outline of the ligand-side protocol. (A) We start from a known noncovalent ligand that binds near a receptor lysine residue. (B) We generate a library of electrophilic analogs by installing different warheads on different ligand positions. (C) We dock each analog into the protein while applying covalent constraints to enforce covalent bond formation to the target lysine, and we select high-confidence analogs (green) that yielded top-10 scoring models with constraint score <2 and RMSD < 1.5 Å to the crystallographic conformation (white). Other analogs (pink) are discarded.

2

Success rate of modeling lysine covalent adducts. (A) Percent of starting structures for which our ligand-side protocol sampled the correct binding conformation (light blue), ranked it among the ten top-scoring models (blue) or ranked it as the top-scoring model (dark blue), shown for each data set. Similarly, (B) shows the percent of starting structures for which our protein-side protocol sampled the correct ring conformation, ranked it among the 10% top-scoring conformers or ranked it among the 1% top-scoring conformers. (C) Native conformation (beige) and a top-10 scoring model (blue) produced by our ligand-side protocol, with ligand RMSD = 1.5 Å, of PDB ID 3EWU. Similarly, (D) shows a top-10% scoring model produced by our protein-side protocol, with ring RMSD = 1.0 Å, of the same structure.

3

Outline of our protein-side, covalent small-molecules design protocol. (A) We start from a crystal structure of a noncovalent ligand that binds near a receptor lysine. (B) The ligand is removed from the structure and the target lysine is modified with an electrophile installed on a ring moiety, such as an aryl-sulfonyl fluoride. (C) We model the possible rotamers of the modified lysine in the context of the binding pocket. (D) We compare the top-scoring rotamers to the crystallographic conformation to identify models with ring RMSD < 1 Å. (E) High-confidence covalent analogs are suggested accordingly.

4

Our protocols successfully identify a new lysine-targeting MKK7 binder. (A) Chemical structures of reported MKK7 inhibitors MKK-COV-3 and K00007 and their derivatized lysine targeting covalent candidates. (B) The modeling prediction of our ligand-side protocol (green) and our protein-side protocol (blue) overlaid on the template binder (beige, PDB ID: 5Z1D). (C) LC/MS spectra of compound 1 after overnight incubation at 4 °C with the protein, and after 1 h reduction with sodium borohydride at room temperature. (D) Inhibition of MKK7 by compounds 1–4 in a kinase activity assay. (E) Co-crystal structure of MKK7 with compound 1 (beige), overlaid on the closest modeling prediction within the top-scoring models of our ligand-side protocol (green) and our protein-side protocol (blue).