Linking ATP and allosteric sites to achieve superadditive binding with bivalent EGFR kinase inhibitors

The study addresses how linker structure governs the efficacy of bivalent small molecules that simultaneously engage two adjacent sites in a target protein. Although fragment linking and bivalent strategies can, in principle, deliver superadditive binding, successful cases are rare and linker optimization is largely empirical. EGFR kinase provides a model with adjacent ATP (orthosteric) and allosteric pockets previously shown to support cooperative co-binding of separate inhibitors. The research question is whether changing the point of connection and chemistry of the linker bridging an ATP-site trisubstituted imidazole and an allosteric dibenzodiazepinone can produce superadditive potency and what structural factors underlie such gains. The purpose is to design, synthesize, and characterize alternatively linked ATP–allosteric EGFR inhibitors, define their binding modes structurally and computationally, and establish principles for efficient linker optimization. This has importance for kinase drug discovery and more broadly for fragment linking and bivalent design, where rational, structure-guided linker strategies are needed.

Prior work has established diverse heterobifunctional strategies (e.g., molecular glues, degraders, dimerizers) and fragment-based drug discovery where linking can, in theory, yield superadditive affinity. However, decades of efforts show superadditivity is scarce, with outcomes sensitive to linker length, geometry, and flexibility. In kinases, both ATP-competitive and allosteric inhibitors exist, and for EGFR the allosteric pocket lies adjacent to the ATP site, enabling cooperative co-binding and synergy between orthosteric and allosteric ligands. Preclinical studies show dual ATP/allosteric inhibition can enhance tumor regression and delay resistance. Recent reports demonstrated ATP–allosteric bivalent molecules for EGFR, though challenges include high molecular weight. The literature emphasizes the need for structural strategies to guide linker design beyond trial-and-error and suggests that matching parent fragment binding modes and enabling appropriate flexibility can be critical for successful linking.

• Chemical synthesis: Designed and synthesized bivalent EGFR inhibitors linking an ATP-site trisubstituted imidazole to a dibenzodiazepinone allosteric motif with two connection strategies: N-linked methylene (compound 1) and C-linked amide (compounds 2–4; 4 includes a Michael acceptor for covalent C797 targeting). N-linked 1 was assembled via cross-coupling, Miyaura borylation, and Suzuki couplings followed by deprotection. C-linked series used Buchwald–Hartwig amination to form the dibenzodiazepinone core, saponification, and amide coupling to appropriate anilines from the orthosteric motifs, then deprotection. Matching C-linked benzo analogues (10–12) were also synthesized for comparison.
• Biochemical assays: Employed HTRF-based kinase activity assays to determine IC50 values against EGFR WT and mutants L858R (LR), L858R/T790M (LRTM), and L858R/T790M/C797S (LRTMCS) using purified kinase domains at defined enzyme concentrations (WT 10 nM, LR 0.1 nM, LRTM and LRTMCS 0.02 nM; ATP 100 µM). Dose–response curves (11-point, triplicate) were fit by nonlinear regression. Time-dependent covalent inhibition kinetics for compound 4 were measured using a Sox-based fluorescence assay to extract kinact, KI, and kinact/KI for WT, LR, and LRTM.
• Structural biology: Determined X-ray cocrystal structures of EGFR(T790M/V948R) with compounds 1 (2.1 Å, PDB 8FV3) and 2 (2.2 Å, PDB 8FV4). Crystals were obtained by soaking kinase crystals (inactive αC-helix out state) with ligands and AMP-PNP/Mg2+. Data collection at NSLS-II 17-ID-2 and APS 24-ID-C; structures solved by molecular replacement and refined with PHENIX/COOT.
• Computational studies: Conducted 20 μs of MD simulations (Desmond, OPLS4; 10×1 μs replicas per compound for 1 and 2) to analyze protein–ligand interaction patterns, water-mediated H-bonds, torsional distributions, and mobility. Performed docking for covalent 4 (GLIDE) to propose a binding pose consistent with covalent C797 orientation. Calculated MM-GBSA energies (Prime) and WaterMap analyses to quantify hydration site thermodynamics and water displacement upon binding. MacroModel conformational searches and energy profiles were used to assess ligand conformational favorability relative to crystal poses.
• Cellular studies: Western blot assays measured EGFR pY1068 and downstream pERK1/2 and pAKT in NSCLC cell lines (H1975 L858R/T790M; H3255 L858R; HCC827 exon19del E746-A750) after 6 h compound treatments. Antiproliferative effects were assessed in H1975 and HCC827 (MTT) and Ba/F3 cells expressing WT or mutant EGFR (resazurin). Kinome selectivity profiling produced an S(35) score. Metabolic stability was evaluated in human liver microsomes with LC-MS analysis.
• Data availability and validation: Electron density supported ligand placement; Ramachandran statistics reported. MD, MM-GBSA, and WaterMap datasets are deposited in Zenodo. Assays were run in triplicate or repeated as specified.

• Potency depends strongly on linker connection: The N-linked methylene bivalent 1 is weak (IC50: LR 1300 ± 100 nM; WT, LRTM, LRTMCS >10,000 nM). In contrast, C-linked amide bivalents 2 and 3 are highly potent: LR 1.5 ± 0.1 nM (2) and 1.2 ± 0.09 nM (3); LRTM 0.059 ± 0.005 nM (2) and 0.051 ± 0.005 nM (3); LRTMCS 0.064 ± 0.004 nM (2) and 0.063 ± 0.005 nM (3).
• Covalent analogue 4 (C-linked with Michael acceptor) shows reversible inhibition of LRTMCS with IC50 1.8 ± 0.3 nM and exhibits time-dependent covalent kinetics: kinact/KI (M−1 s−1) WT 2500 ± 30, LR 14100 ± 300, LRTM 1070 ± 40; kinact (min−1) WT 0.18 ± 0.004, LR 0.52 ± 0.02, LRTM 0.10 ± 0.004; KI (μM) WT 1.20 ± 0.04, LR 0.61 ± 0.03, LRTM 1.6 ± 0.1.
• Superadditivity relative to parent fragments: Parent ATP-site imidazoles 5–6 are weak against resistant mutants (e.g., LRTM 5800 ± 300 nM for 5; >10,000 nM for 6), and allosteric benzo inhibitors 8–9 show IC50 ~39–59 nM (LRTM/LRTMCS). Thus, C-linked bivalents 2–3 are ~103–106-fold more potent than parents. Matching C-linked benzo analogues 10–12 are inactive or weak (IC50 ≥10 μM), underscoring that the amide linker and point of connection drive the potency gains.
• Linking coefficients estimated from IC50 values indicate superadditivity for C-linked amides (upper limits ~0.5–1.0 M−1; lower-limit <1 consistent with superadditivity), whereas N-linked 1 shows distinctively higher estimated coefficients (>2.0 × 10^7 M−1), consistent with its poor activity profile and stark contrast to C-linked outcomes.
• Structural basis: Cocrystals reveal distinct allosteric-site conformations: 1 binds with an “outward” benzo conformation; 2 adopts an “inward” conformation enabling additional H-bonds (T854, D855) and a K745 side-chain swing that accommodates a coordinated water at the ATP site. The benzo conformation influences A-loop positioning. Despite similar linker length, the point of connection dictates conformation.
• Computational corroboration: MD simulations recapitulate interaction patterns and reveal water-mediated contacts. MM-GBSA indicates a ΔΔG ≈ −9.5 kcal/mol favoring 2 over 1 and lower ligand strain energy for 2, consistent with tighter binding. WaterMap shows 2 fully displaces energetically unfavorable waters that 1 does not. A single unique rotatable C–C bond in the C-linked linker increases mobility of the back-pocket phenyl in the allosteric site, underpinning enhanced complementarity and potency.
• Cellular activity and selectivity: The covalent C-linked 4 suppresses EGFR pY1068 in H1975, H3255, and notably HCC827 (exon19del) cells, and reduces pERK and pAKT in H1975. Antiproliferative activity of 4 is ~100–500 nM (about 60-fold less potent than AZD9291) and Ba/F3 LR EC50 ~220 nM, improved versus earlier bivalents. 4 exhibits broad kinome selectivity (S(35) = 0.084) and metabolic stability in human liver microsomes.

The findings demonstrate that altering the linker’s point of connection between ATP and allosteric fragments can shift binding mode within the allosteric pocket, enabling new H-bonds, favorable side-chain rearrangements, and displacement of unfavorable waters. This conformational control yields superadditive potency gains (>10^6-fold) for C-linked amide bivalents versus the N-linked analogue, despite similar linker lengths. The structural and computational analyses converge on a unifying mechanism: the C-linked amide creates an “inward” benzo conformation with increased mobility of a key phenyl ring and improved pocket complementarity, aligning with the binding mode of the parent allosteric fragment. These results directly address the challenge of fragment linking by showing that exploring alternative points of connection can produce substantial potency improvements beyond traditional adjustments of length or functionality. Biologically, the bivalents—particularly the covalent C-linked 4—engage cellular EGFR across relevant oncogenic mutants with selectivity, though further medicinal chemistry is needed to improve cellular potency and pharmacologic properties.

This work introduces ATP–allosteric bivalent EGFR inhibitors in which a re-engineered, C-linked amide linker drives superadditive binding and picomolar to low-nanomolar potency against drug-resistant EGFR mutants (L858R/T790M and L858R/T790M/C797S). X-ray structures, MD, MM-GBSA, and hydration analyses elucidate how the linker’s point of connection controls allosteric-site conformation, interactions, and water displacement, accounting for the potency gains over N-linked designs and parent fragments. The study proposes a generalizable design principle: early exploration of alternative points of connection between fragments can be a highly effective strategy for linker optimization in fragment linking and bivalent inhibitor development. Future work should optimize physicochemical properties for improved cellular efficacy and pharmacokinetics, expand evaluation across the kinome, and further clarify structural determinants of activity in exon19 deletion contexts.

• Cellular potency is limited relative to standard-of-care EGFR inhibitors (e.g., AZD9291), likely due to high molecular weight and permeability constraints of bivalents.
• Structural characterization utilized EGFR(T790M/V948R) to capture the inactive conformation; while informative, this engineered variant may not fully reflect all physiological conformations.
• Linking coefficient estimates rely on upper/lower limits constrained by weak activity of matched allosteric analogues (e.g., 10–12), introducing uncertainty.
• The structural mechanism for activity against exon19 deletion mutants (HCC827) remains underdeveloped and requires further studies.
• Generalizability of the “point of connection” strategy across diverse targets has not yet been validated experimentally beyond EGFR.
• Synthesis of alternative connection variants can be resource-intensive, potentially limiting rapid exploration.