Universität zu Köln

Rabipour, Mina (2026). Mapping the Allosteric Landscape of Full-Length Lyn Kinase: Integrative Molecular Dynamics and Network Perspectives on Cancer-Associated Mutations. PhD thesis, Universität zu Köln.

PDF Dissertation_Mina_Rabipour_2026.pdf - Published Version Download (180MB)

Abstract

Protein kinases regulate cellular signal transduction and are key therapeutic targets. Among Src family kinases (SFKs), Lyn plays a critical role in immune signaling and cancer but remains less well characterized. To probe its regulatory mechanisms, a full-length model of Lyn was constructed, comprising the SH3, SH2, and SH1 (kinase) domains. The conformational dynamics of this model were analyzed using molecular dynamics (MD) simulations, network-based analysis, and machine learning. Apo wild-type, ATP-bound, dasatinib-bound, and six cancer-associated mutants (K275A, E290D/K, I364L/N) were simulated for a total of 78 μs. Principal component and cross-correlation analyses revealed that ATP stabilizes domain interfaces and strengthens SH2–SH1 coupling. In contrast, dasatinib and oncogenic mutations weaken long-range correlations and partially decouple SH3 and SH2 modules from the catalytic domain. Local flexibility changes in the αC helix and hinge region were found to align with Molecular Mechanics Poisson–Boltzmann Surface Area (MM/PBSA) estimates and ligand interactions. Through network analysis, 44 dynamically central residues were identified, forming a collective allosteric connector spanning the SH3, SH2, SH1, and linker regions. These connectors mediate interdomain communication, preserve activation loop contacts, and anchor critical salt bridges, such as K275–E290 and K275–D385. The fragmentation of these networks under inhibitor binding or mutation reflects a disruption of regulatory communication. A random forest classifier was trained on MD-derived features, including interdomain distances, regulatory spine integrity, and salt-bridge occupancies, to distinguish active-like from inactive-like ensembles, achieving high predictive accuracy (AUC = 0.98). Interdomain coupling and activation loop contacts were identified as primary contributors by feature importance analysis, consistent with network-derived insights. Comparative sequence analysis across SFKs showed that many allosteric connectors are conserved, particularly in the activation loop, αC helix, and SH2–SH1 linker. However, several Lyn-specific residues, such as P73 in the SH3 domain, linker residue I227, and K343, D346, and I475 within the αE and αF helices, were found to diverge across the family and serve as high-scoring connectors. Structural mapping revealed overlap between Lyn connectors and a known distant allosteric inhibitor binding pocket in ABL1 kinase (the Asciminib site). The αE and αF helices at the SH1C–SH2 interface emerged as Lyn-specific, flexible hotspots responsive to mutations, suggesting potential druggable regulatory sites beyond the traditional ATP cleft. Together, these findings support the concept of selective allosteric modulation of Lyn kinase. Through the integrated use of long-timescale MD, network analysis, and machine learning, a general framework is provided for uncovering hidden regulatory nodes in kinases and guiding structure-based drug discovery. This approach is broadly applicable to dissecting allosteric control in multidomain signaling proteins across the kinome.