Contact-Network Phenotyping of the CDK Family Reveals Selective Distal C-Lobe Contact Redistribution by Modern CDK5 Inhibitors and a Quantitative Selectivity Landscape against CDK2 and CDK1

Selective inhibition of CDK5 over CDK1, CDK2, CDK4, and CDK6 remains a central medicinal-chemistry challenge because pocket-centric methods capture local similarity but not full-domain structural phenotypes. Here we introduce kinase-aware contact-network phenotyping, a calculation-light approach that combines systematic Cα contact-map comparison with automated kinase topology annotation across 30 curated CDK crystal structures spanning CDK1, CDK2, CDK3, CDK4, CD5, and CD6 with 21 pairwise comparisons. Three findings emerge. First, the apo CDK5 to selective inhibitor transition (1H4L to 7VDP) produces 33.3% ligand-adjacent gained contacts (13 of 39), compared with 2.8% (1 of 36) for the apo to nonselective transition (1H4L to 1UNL); the apo-anchored ratio of 11.9-fold quantifies the selective-specific reorganization while controlling for generic pocket-occupancy effects, with the largest distance shifts (up to 14.6 Å) concentrated in the distal C-lobe core (residues 221 to 292). Second, the DFG aspartate D144 has the most atypical contact environment in CDK5 by the Miyazawa-Jernigan knowledge-based statistical-potential Z-score (Z = +1.38 in apo and all four selective complexes), and a noncircular cross-family structural observable, the per-kinase D144 contact-shell changed-contact count, rises systematically with phylogenetic distance from CDK5 (zero across three within-CDK5 pairwise comparisons; 1.5 ± 1.0 for CDK2, 3.0 ± 0.0 for CDK1, 6.8 ± 1.5 for CDK6). Third, contact-network divergence follows a strict hierarchy: internal CDK5 evolution (∼6%) is much smaller than CDK5-to-cross-family divergence (∼22 to 37%), while region-burden profiling identifies a 5-fold hinge differential and a 165-contact C-lobe advantage of CDK5 over CDK2. These results define a quantitative selectivity landscape derived from static crystal-structure analysis and identify structural correlates inaccessible to binding-site fingerprints or RMSD-based methods.