Pharmacophore profiling highlights diverging binding modes of nemtabrutinib and ibrutinib across tyrosine kinases

Background. Commanding kinase selectivity remains a major bottleneck in the discovery of targeted anticancer drugs, with homologous ATP-binding pockets leading to off-target inhibition, toxicity, and reduced clinical effectiveness. Addressing these difficulties, we present a streamlined computational workflow allowing rationalizing selective compound activity by capturing subtle, transient structural differences across many homologous kinases. We demonstrate our approach by characterizing the diverging mechanisms of action of nemtabrutinib and ibrutinib in cell lines derived from two specific B-cell lymphoma subtypes (activated B-cell like [ABC] and germinal-center B-cell like [GCB] DLBCL), correlating in vitro anti-tumor activity with in silico selectivity scores computed across 20 tyrosine kinases. Methods. To enable high-throughput, cost-efficient parallel processing of ATP-binding pocket variants, as a preliminary step we compiled a subset of metastable pocket conformations adopted by tyrosine kinases (DFG motif and associated dihedral) with and without generic compound binding (templating with AlphaFold2 and publicly available data). The structural alignment of this conformational dictionary and its processing via structure-based pharmacophore methods (CDP:Kit) produced rich maps of kinase pocket interactomes. Fitting compounds into these maps then enabled scoring matching pharmacophore features and computing a relative selectivity score between kinases for each compound (CDP:Kit and lightweight binding energy estimations) using BTK as baseline for confirmed targeting. Results. We deciphered the synergistic interplay of 10 residue locations within the extended ATP-binding pocket, responsible for the selective binding of nemtabrutinib and ibrutinib across tyrosine kinases. As aligned to human BTK sequence and specifying potential for modified interactions, we highlight reference locations: Q412 (loop-mediated covalent binding, none), V416 (hydrophobic, H-bond), F442 and M449 (hydrophobic, sulfur-aromatic, cation-pi), L460 and I472 (hydrophobic, sulfur-aromatic, pi-stacking), T474 (hydrophobic, halogen-bond, H-bond, sulfur-aromatic, pi-stacking), C481 (covalent binding, H-bond, none), N484 (H-bond, none) and L542 (hydrophobic, sulfur-aromatic, none). Conclusions. We demonstrate a general strategy for rational, structure-guided characterization of selective compound activity across conserved kinase families. Our approach enabled characterizing the diverging binding modes of nemtabrutinib and ibrutinib across 20 kinases, in agreement with anti-tumor activity data and expression levels in cell lines derived from two B-cell lymphoma subtypes. We identified FYN, FRK and MAST1 as likely targets of nemtabrutinib and not of Ibrutinib, responsible for GCB DLBCL non-proliferation.