Imidazole-[1,5‑a]‑Pyridines Can Occupy the EGFR Allostery with a Strong Polar Interaction

Abstract

The imidazole­[1,5-a]­pyridine derivatives were recently synthesized and showed remarkable bioactivity against three cancer cell lines, but an understanding of their activities is still missing. To prompt a detailed investigation into their molecular binding mechanisms, we carried out a comprehensive computational workflow for structure-based drug design. The protocol encompasses (i) ligand polarization using quantum chemical calculations, (ii) docking algorithms to generate initial protein–ligand conformations, (iii) molecular dynamics simulations to evaluate ligand diffusion within the protein pocket, and (iv) binding free energy calculations through the umbrella sampling molecular dynamics method. The inherent flexibility of the epidermal growth factor receptor (EGFR) kinase protein in an aqueous environment challenges the stability of ligand–protein associations. Of the 15 imidazole-pyridine compounds considered, after completing a total simulation time of 1.1 μs, only three compounds have been found to possess strong interactions with critical residues in the allosteric pocket of the EGFR inactive conformation in which the electrostatic potential energies play an important role. Notably, the compound named 3h389 carries out an exceptionally large binding free energy, outperforming the well-known allosteric inhibitor EAI045 considered a reference. More interestingly, the EGFR conformational changes under both 3h389 and EAI045 binding are similar. These results promote the imidazole­[1,5-a] compound denoted as 3h389 to be a highly strong candidate for a preclinical evaluation in cancer therapy. If such an evaluation can be performed, it underlines the utility of our computational pipeline for drug discovery efforts.

1. Introduction

The overexpression of the epidermal growth factor receptor (EGFR) has been known to be strongly implicated in various cancer types, making it a critical target for therapeutic interventions. Of the most effective strategies, an inhibition of the EGFR tyrosine kinase domain using small-molecule inhibitors was often implemented, , and significantly improved survival rates have been observed, particularly for patients having a non-small-cell lung cancer (NSCLC). , This approach has already driven the development of four generations of tyrosine kinase inhibitors (TKIs) to address the resistance caused by emerging mutations. The first-generation TKIs, such as gefitinib or erlotinib, act as ATP-competitive inhibitors by reversibly occupying the ATP-binding site, thereby preventing an EGFR dimerization and downstream signaling. These inhibitors are highly effective in the initial stages of treatment, with most patients responding positively until resistance arises due to mutations. , In recent years, the mutations noted as L858R, T790M and C797S emerged as the most prevalent EGFR variants. Of these mutations, the T790M and C797S mutations are particularly problematic because they become resistant to the first-generation TKIs (gefitinib and erlotinib). Specifically, the T790M mutation involves substitution of tyrosine (TYR) with methionine (MET) at residue 790, which increases the ATP-binding affinity and diminishes the inhibitory efficacy of these drugs at the ATP-binding site. To address this resistance, second-generation TKIs were developed, including afatinib. The afatinib drug brings in a strong binding to the ATP-binding site of T790M mutant EGFR. However, it was discovered that afatinib also covalently binds to residue C797, resulting in an irreversible interaction. Such a covalent binding contributes to high toxicity and significantly limits the clinical application of afatinib, because it may also inhibit the proliferation of normal cells. To combat the emergence of T790M-mediated resistance, the osimertinib drug was introduced as a reversible TKI third-generation. Osimertinib effectively targets the ATP-binding site of both wild-type (WT) EGFR and mutant variants, providing a broader therapeutic spectrum. , Although the use of currently approved TKIs has been highly encouraging, a resistance inevitably develops over time due to the interplay between both the ATP competition and ongoing EGFR mutations. ,

The past decade has witnessed the emergence of the next generation of TKIs, often referred to as allosteric EGFR inhibitors. These inhibitors, which target an allosteric pocket adjacent to the ATP-binding site, offer a promising approach to address an inherent resistance associated with ATP-competitive mechanisms and mutations. −

The allosteric pocket, which is located adjacent to the ATP-binding site, appears only in an inactive conformation of the EGFR protein when the C-helix loop moves outward. The tyrosine kinase domain can spontaneously activate its dimerization by forming an asymmetric structure with another monomer. However, this process requires an intrinsic conformational shift from the inactive to the active state. , A legitimate question on how the allosteric conformation operates has been a subject of significant research effort. , Small ligands have thus appeared as effective agents to prevent EGFR from transformation, and inhibit EGFR signaling. The EAI001, EAI045, and DDC4002 drugs were among the earliest allosteric inhibitors identified. Their effectiveness in targeting wild-type EGFR has however been limited due to the inherent behavior of EGFR. Despite this unfortunate situation, allosteric inhibitors have shown a remarkable efficacy in treating the L858R/T790M and L858R/T790M/C797S mutations in non-small-cell lung cancer (NSCLC). More recently, the JBJ-125 and JBJ-036 mutations have been highlighted as potent analogs of EAI045 in part due to their strong binding affinities. Nevertheless, identification of effective inhibitors remains a challenging task, and the discovery of novel inhibitor compounds continues to be a highly demanding and attractive area of study.

Computational approaches, particularly molecular dynamics (MD) simulations, have provided researchers with powerful tools to gain physical insights into macromolecular systems. Typically, an MD project aims to achieve two major objectives, namely, sampling of conformations and evaluation of binding affinities. Previous investigations of the thermodynamic features of wild-type EGFR in its active (2GS6) and inactive (2GS7) states included four replicates of 200 ns all-atom MD simulations. A 2011 study by Wan et al. highlighted the critical role of the C-helix orientation and formation of salt bridge K745-E762. At the molecular level, substantial studies carried out by the Shan and Nusinov groups focused on thermodynamic and mechanical properties of EGFR in both active and inactive states. , Additionally, studies of various crystallographic structures (e.g., 7JXM, 7K1H, and 7K1I) conducted by Beyett et al. in 2022 revealed the contributions of K745, E762, and the DFG-loop in mediating interactions between EGFR mutants and inhibitors such as EAI045. In 2019, the binding free energies of EAI045 to different mutations were determined using the molecular mechanics MM-PBSA method, yielding values of from 8.7 and −11.0 to −11.7 kcal/mol. In early 2024, Bhanja et al. elucidated the resistance mechanisms of various EGFR mutations against EAI045 in the allosteric region using 1000 ns MD simulations. Notably, a dual inhibition of the EGFR kinase domain with JBJ and osimertinib generated great hope for efficient treatment. Parallel efforts by Al Rubaye et al. examined allosteric inhibitors in screening a class of organic compounds, the carbothioamides. Utilizing the 6DUK crystal structure, the latter team developed a strategy combining docking, MM-PBSA calculations, and 100 ns MD simulations to compare in silico results with in vitro tests. These studies introduced some candidates for antitumor activities. Motivated by these findings, we set out to study employing atomistic level simulations to elucidate the interaction mechanisms between a series of novel compounds, the imidazole-pyridine derivatives with some key residues in the allosteric pocket.

Classical MD simulations allow the ligands moving from an initially unbound position to a bound state to be observed if such binding occurs. However, the chance as to whether such a binding is successful depends heavily on simulation duration and system’s inherent dynamics. While classical MD is a powerful tool for producing reliable results, processes that occur over extended time scales still present significant challenges. This limitation makes classical MD simulations not quite suitable for screening large databases of molecules for identifying potential candidates. To address this issue, we propose an optimized computational protocol tailored for rational drug design. In a prior investigation, a series of 15 newly synthesized compounds demonstrated some promising cytotoxic activity against non-small-cell lung cancer cell line (SK-LU-1). The EGFR, which is widely acknowledged as a central regulator of tumorigenesis and progression in NSCLC, represents a critical molecular target for understanding the disease and, thereby, its treatment. However, despite the observed cytotoxic potential of these compounds, their interactions with key cellular factors remain to be comprehensively elucidated. In this context, a primary objective of this study is to explore the binding capabilities of the new imidazole-pyridine compounds within the allosteric region of the T790M/C797S/V948R EGFR variant.

Our workflow begins by docking each of the 15 imidazole-pyridine compounds, shown in Figure , into the allosteric pocket, followed by multiple rounds of MD simulations for each protein–ligand complex. To assess the binding affinities of the most promising imidazole-pyridine compounds, the binding free energies are calculated using the Umbrella Sampling (US) method. This stepwise screening strategy, including the successive docking, MD simulations, and binding free energy calculations, has been demonstrated to be an effective approach for identifying top candidates from a moderate-sized compound library. Though the workflow of docking refinement with MD computing binding free energy with US or MM-PBSA method has been standard in the computational drug discovery field for years, the key innovation in this protocol is the use of multistage MD sampling which allows ligands to repeatedly search for a stable binding conformation. This is critical, as the self-breathing motion of the protein can otherwise eject the ligand from the binding site. An enhanced MD sampling ensures that the most promising compounds progress into the computationally intensive US stage, as carrying out US calculations for all compounds from the outset would be prohibitively resource-intensive. After conducting multiple MD simulations with varied initial velocities, only three imidazole-pyridine compounds, named as 3h389, 3h409, and 3h416 (cf. Figure ), remain stably bound within the allosteric pocket. Binding free energies of these compounds are determined using US methods. Notably, compound 3h389 exhibits an exceptionally low binding free energy of −25.6 kcal/mol, significantly outperforming the known EAI045, which was characterized by a binding free energy of −12.5 kcal/mol. This result provides a compelling evidence for promoting the imidazole­[1.5-a]­pyridine 3h389 (cf. Figure ) as a potential fourth-generation EGFR inhibitor for cancer treatment. Additionally, our present study highlights the structural changes in EGFR conformations in the absence and presence of inhibitors, offering deeper insights into the dynamics of protein–ligand interactions.

1.

Structure of the “inactive” EGFR kinase domain, with the helix loop conformation “out,” visualized using PyMOL (A) and the HOMO (B) and LUMO (C) of 3h389 chemical compound. The allosteric pocket-bound ligand EAI045 is shown as green spheres. Residues K745 and E762 are depicted as yellow sticks, while the DFG motif is highlighted in magenta. The ATP molecule (gray spheres, aligned from the crystal structure PDB ID: 7JXM) located in the ATP-binding region is excluded from this study. Chemical structures of 15 novel imidazole­[1,5-a]­pyridine compounds considered and EAI045 are plotted and displayed in this Figure.

2. Materials and Methods

2.1. System Preparation

The T790M/C797S/V948R EGFR kinase domain mutant and its complex with EAI045 are obtained from the Protein Data Bank (PDB ID: 5ZWJ). Missing residues are modeled using the Modeler program. , Each protein in the ErbB family contains an extracellular domain (ECD) with an N-terminal glycosylated sequence of about 620 amino acids that includes the ligand-binding site. A hydrophobic transmembrane domain (TMD) links the extracellular region to the intracellular region through a juxtamembrane domain (JMD). The intracellular region of ErbB proteins comprises around 540 amino acids and includes a tyrosine kinase domain (TKD) and a carboxyterminal tail (CTT) of approximately 230 amino acids. , The mutant protein, consisting of 347 amino acids, namely, residues 675–1022, is structurally defined by C-loop and N-loop regions connected via a hinge. Unlike the active conformation, the inactive conformation exposes an allosteric pocket adjacent to the ATP-binding region. Notably, the K745–E762 salt bridge and the DFG motif (D855, F856, G857) play pivotal roles in stabilizing the active state by obscuring the allosteric region.

Hydrogen atoms are added to the receptor using the gmx pdb 2gmx package with the Amber ff99SB-ILDN force field. In the narrow case of the EGFR exams, previous simulations, including those by Shaw in 2013, Martínez-Ortega in 2025, Zhang in 2025, Xu in 2025, still employed the AMBER99SB-ILDN force field. The choice of a common force field facilitates a consistent comparison of our current results with those of previous ones. The TIP3P water model is applied. For the ligands, geometry optimizations and atomic partial charge calculations are performed using the Gaussian 16 program at the HF/6–31G level. Atomic net charges are derived using the RESP fitting technique implemented in AmberTools. , Force field parameters for the reference EAI045 and 15 novel imidazole-pyridine compounds are generated using the Antechamber program with the GAFF force field model. Protein–ligand complexes are centered within a cubic box with dimensions 10 nm × 10 nm × 10 nm, ensuring sufficient distance between the protein and the periodic box boundaries to prevent self-interactions. The system is solvated with water molecules and neutralized by adding counterions. The prepared system, containing approximately 100,000 atoms, undergoes the following steps including energy minimization, 1 ns NVT equilibration, and 1 ns NPT equilibration. Multistage MD simulations, detailed in the screening strategy given below, are performed. To address edge effects, periodic boundary conditions are employed. Long-range electrostatic interactions are computed using the particle mesh Ewald (PME) method, and hydrogen atom covalent bonds are constrained using the SHAKE algorithm. , All simulations were generated by the support of PAWSEY HPC and the Gromacs package version 2022. Snapshots are saved every 1000 steps with a 2 fs time step for each.

2.2. MD Outcome Analysis

Key computational parameters, including hydrogen bonds, contact numbers, and root-mean-square deviation (RMSD) over time, are analyzed. A hydrogen bond is defined as having a donor–acceptor distance smaller than 0.35 nm and an angle greater than 135°. A contact is defined when the distance between any protein and the ligand atom is smaller than 0.6 nm. Protein–ligand interaction energies are evaluated by decomposition of the electrostatic and van der Waals potential energies. The gmx-energy package of Gromacs helps us to generate the Coulomb and Vdw potentials from the snapshots that are extracted from each of 1000 MD simulated steps. Special focus is placed on interactions between the ligand and key residues inside the EGFR allosteric pocket, to emphasize how these residues interact with different chemical fragments.

The free energy surface (FES) is constructed based on two parameters including the number of contacts and the total interaction energy. Representative protein–ligand conformations, corresponding to global minima on the FES, are selected for binding free energy calculations using the Umbrella Sampling (US) method.

2.2.1. Potential of Mean Force via Umbrella Sampling

In this study, we apply the Umbrella Sampling method, which not only computes the absolute binding free energy between small ligand and receptor but also explores how the ligand could translate from a bound state into unbound one during an escape tunnel. However, the US method cannot by itself generate a releasing process; thus, it needs the support of the steered molecular dynamics (SMD) simulation method. Starting with four representative structures of the 3h389, 3h409, and 3h416 compounds and EAI045 obtained from previous parts, an external force is employed to pull the ligand slowly out of the allosteric active site. In the SMD performance, the pulling velocity and the spring constant are chosen at 1­(m/s), 600 (kJ/mol/nm²). During each pulling process, more than 120 snapshots are collected along the move of the ligand, for each moving step of 0.25 Å when ligand–protein interaction is still strong and 0.5 Å when ligand–protein interaction becomes weak. After running three typical steps of minimization, equilibrium, including EM, NVT, and NPT, the ligand and protein in each snapshot are subsequently simulated under a 10 ns classical molecular dynamics simulation. When the conformation space is sampled enough, the free energy profile is then constructed by the support of the gmx wham package of Gromacs program, and the binding free energies can be thereby computed.

3. Results and Discussion

3.1. Chemical Structures and Initial Docking Positions

Fifteen novel compounds synthesized (cf. Figure ) are potential anticancer agents by integrating two highly bioactive scaffolds, a β-carboline and a bisindole. The β-carboline moiety is well documented for its diverse biological activities, including DNA intercalation, inhibition of topoisomerases, and induction of apoptosis, making it a prominent framework in anticancer research. − Similarly, the bisindole scaffold has widely been recognized for its role in modulating key molecular targets associated with tumor progression and cell survival. HUMO and LUMO (in Table S4, SI file) and the spatial structure of compounds are presented in Figure .

The docking energies of 15 compounds with the receptor are calculated by using the AutoDock Vina program and summarized in Table S1 (SI file). Each ligand is placed in the allosteric region, and the resulting key metrics such as the number of contacts, hydrogen bonds, Coulomb potential energy, and van der Waals potential energy are initially computed based on docking conformations (cf. Table S1, SI file). Notable results include some remarkable values including 2.1 hydrogen bonds in the 3h382 complex, 2650 contacts in the 3h388 and 3h390 complexes, or/and interaction energies of −137.6 and −146.1 kcal/mol for the 3h382 and 3h389 complexes, respectively. However, docking data alone are inherently limited in reliability due to the fact that protein side-chain flexibility, hydrogen atoms, and solvent effects are all neglected. Consequently, thermodynamic processes must be considered to assess the stability of the protein–ligand interactions. Proteins can dynamically alter ligand engagement through mechanisms such as self-breathing, as explored in subsequent sections.

3.2. MD Simulations Eliminate 12/15 Compounds When Going into Further Steps

Molecular dynamics (MD) simulations have proven to be effective tools for studying the dynamics of small ligands within allosteric pockets. Based on the results from the initial 100 ns MD simulations, we analyze the time-dependent minimum distances between the ligands and key residues within the allosteric pocket. Notably, residues K745 and E762 are known to form a salt bridge that closes the allosteric pocket. Averaged values from the last 20 ns of these simulations are calculated and summarized in Tables and S2 (SI file).

1. Results Obtained from the Last 20 ns of the First Round, 100 ns MD Simulations .

				mindist to K745	mindist to S797	coulomb potential	VDW potential	interaction energy
N	system	number of contacts	H-bond	(nm)	(nm)	(kcal/mol)	(kcal/mol)	(kcal/mol)
1	3h382	2710	1.53	0.16	1.07	–120.1	–39.0	–159.1
2	3h384	642	0.32	1.51	2.44	–38.0	–11.1	–49.1
3	3h385	1950	0.94	0.21	1.22	–48.3	–40.4	–88.7
4	3h387	2380	2.07	0.18	0.85	–80.4	–38.5	–118.9
5	3h388	1350	1.3	0.38	1.28	–42.1	–22.9	–65.0
6	3h390	1050	0.21	0.43	0.88	–54.3	–15.8	–70.1
7	3h393	946	0.08	1.52	2.16	–46.7	–15.6	–62.3
8	3h412	1070	0	0.26	2.42	–24.9	–16.1	–41.0
9	3h414	1840	2.68	0.84	1.76	–62.9	–30.1	–93.0
10	3h417	2470	0.045	0.18	1.1	–61.5	–39.0	–100.5
11	3h418	1820	0.71	0.19	0.82	–60.0	–27.8	–87.8
12	3h419	1590	1.85	0.49	1.63	–67.5	–24.4	–91.9
13	EAI045	1210	2.58	0.19	0.93	–87.1	–16.9	–104,0

^aThe averaged minimum distances between the ligand and two key residues (K745 and S797) are shown as references to evaluate whether the ligand has exited the EGFR allosteric region. Additional distance values for other residues surrounding the allosteric region are provided in Table S2 (SI file). Interaction energy values (kcal/mol) represent the sum of Coulombic and van der Waals potential energies. Information on only 12 out of 16 systems considered that have fallen in the first and second testing steps is given. Results concerning EAI045 are also given.

In Figure , the minimum distances between ligand (LIG) and residues K745 and S797 for the 3h393 system are plotted. These results reveal that compound 3h393 completely exits the allosteric pocket at approximately 40 ns, with initial movement beginning at 20 ns. Similarly, compound 3h384 moves earlier, starting at 8 ns, and completely exits at 10 ns (cf. Figures S1 and S2 in SI file). Conventional metrics such as protein–ligand number of contacts or root-mean-square deviations (RMSD) are often insufficient to determine as whether a ligand leaves the binding pocket. Despite high docking energies of −9.1 kcal/mol (3h384) and −9.8 kcal/mol (3h393), thermodynamic mechanisms ultimately ruled out the stable binding of these compounds to EGFR, as illustrated in Figure concerning the equilibrium release profile of compound 3h393.

2.

Minimum distances between the ligand and two selected residues, K745 (red) and S797 (black), over time. As K745 resides deep within the allosteric pocket, a significant increase in its distance from the ligand suggests that the ligand has exited the protein’s allosteric site. Data were plotted from the 100 ns MD simulation of the 3h393 system. Similar results for other complexes are provided in Figure S1 in the Supporting Information.

3.

Time-evolution snapshots of the protein–ligand configuration. The protein surface is displayed in gray, while flexible regions are highlighted: residues 720–724 (green), residues 746–749 (yellow), residues 860–863 (blue), the DFG motif, residues 755–757 (magenta), and the helix loop, residues 752–769 (red). Compound 3h393 is represented as a light-orange stick model.

From the 100 ns simulations, both the electrostatic (Coulomb) and van der Waals interaction energies between the ligands and the receptor are calculated. For both 3h384 and 3h393 compounds, these data are negligible due to interactions occurring outside the binding site. In contrast, seven compounds, including the 3h385, 3h388, 3h390, 3h412, 3h414, 3h418, and 3h419, are characterized by average interaction energies of −88.7, −65, −70.1, −41, −93, −87.8, and −91.9 kcal/mol, respectively, during the final 20 ns of simulations. However, these values are substantially weaker than those observed for EAI045 (−104 kcal/mol), suggesting a limited binding potential. Given computational resource constraints, these compounds are now excluded from further analysis.

To test again the ligand–protein binding stability, a second round of MD simulations is performed with 500 ns trajectories for the series of compounds 3h382, 3h387, 3h389, 3h409, 3h416, and 3h417 and the reference compound EAI045. Of these compounds, three systems, including 3h382, 3h387, and 3h417, fail to maintain a stabilized binding within the allosteric region. Notably, the 3h387 system exhibits an unexpected behavior, as shown in Figure . While compound 3h387 initially interacted strongly with EGFR giving rise to an interaction energy of approximately −125 kcal/mol during the first 200 ns, the protein–ligand association is abruptly disrupted. This is evidenced by the rapid increase and significant fluctuation in the minimum distances between the ligand and residues K745 and S797 (cf. Figure ). This observation suggests that the ligand is naturally released, likely due to a conformational change in the protein–ligand complex. Future studies with improved configuration sampling are necessary to explore the structural evolution of EGFR in greater detail. Numerical results related to the minimum distances between key residues and compounds 3h382, 3h384, and 3h387 are provided in Table S3 (SI file).

4.

Top: Time-dependent ligand–protein interaction energy. Bottom: Minimum distance between compound 3h387 and residues K745 (red) and S797 (black).

The inherent flexibility of EGFR in an aqueous environment tends to contribute to various kinetic motions, including translation, rotation, flexing, and breathing, which can induce structural changes. These motions challenge the stability of ligand–protein engagement, potentially transforming a bound state into an unbound state at varying rates or, alternatively, reinforcing the association. Ligands can be displaced due to suboptimal initial configurations generated during docking, underscoring an inherent limitation of computational drug design. Although this approach may exclude some promising candidates, compounds that endure multiple rounds of extended MD simulations are likely to possess stronger potentials. Based on the present results, three compounds, including 3h389, 3h409, and 3h416, emerge as the most realistic candidates for further investigation.

3.2.1. Electrostatic Energies Dominantly Contribute to Interactions of the 3h389, 3h409, 3h416, and EAI045 Systems

Four compounds, including 3h389, 3h409, 3h416, and EAI045, successfully complete every three independent trajectories, maintaining a bound state with the EGFR protein. A total simulation time of 1.1 μs, including 100 ns of 1-round, 500 ns 2-round, and 500 ns 3-round, is completed per system. The goal of this comparison is to identify compounds that remain bound to EGFR across all MD trajectories and exhibit stronger binding affinities than EAI045. For each complex, we collect ∼ 4000 snapshots from the last 200 ns of two independent 500 ns MD trajectories. The latter information is used to calculate the average number of hydrogen bonds, the number of contacts, and the polar/nonpolar potential energies whose numerical results are shown in Table . This part aims not only to explore the ligand interactions with key residues in the EGFR allosteric pocket but also to quantify the contributions of three fragments to these energies.

2. Averaged Values of the Number of Contacts, Hydrogen Bonds, and Interaction Energies (Including Electrostatic Potential Energy [Elec] and van der Waals Potential Energy [VdW]) between the Ligand and the Protein are Collected from the Last 200 ns of Two Independent 500 ns MD Trajectories .

N		num count	H-bond	interaction energy (kcal/mol)	electrostatic (kcal/mol)	Vdw (kcal/mol)	EGFR residues forming hydrogen bonds
1	3h389	1650	0.27	–198	–184.4	–13.6	F723 – K745 – R748 – I759 – R841 – F856 – L858 – A859 – K860
2	3h409	2700	1.24	–122.1	–81.1	–41	K745 – R831 – D855 – F856 – G857 – L858 – A859
3	3h416	1780	1.24	–162.5	–143.7	–18.8	A722 – K745 – R748 – E749 – Q758 – Q791 – D761 – E762 – N842 – T854 – D855 – F856 – G857 – L858 – K860 – K867
4	EAI045	1585	3.97	–143.8	–125.2	–18.6	K745 – R748 – T854 – D855 – F856 – L858 – K860

^aRGFR residues forming hydrogen bonds with the ligand are also listed.

Results summarized in Table reveal several types of interactions between the compounds considered and the EGFR allosteric region. In terms of hydrogen bonds, EAI045 forms the largest number, with an average of 3.97 hydrogen bonds per frame, whereas 3h389 has the smallest number, with only 0.27 hydrogen bonds per frame.

Figure and Table show the characteristic conformation of the EAI045-EGFR complex, where EAI045 preferentially forms hydrogen bonds with the EGFR residues K745, R748, and L858. Despite a relatively low interaction energy of −143.8 kcal/mol for EAI045, this value is less negative than those of 3h416 (−162.5 kcal/mol) and 3h389 (−198 kcal/mol). However, the larger number of hydrogen bonds remains a distinguished feature of EAI045.

6.

Interaction maps of the four effective systems considered as inhibitors plotted with support from Discovery Studio.

The compound 3h389 interacts with the EGFR mutant predominantly through electrostatic potential energy, with the lowest Coulomb potential energy recorded to be −184.4 kcal/mol and the lowest van der Waals (Vdw) potential energy of −13.6 kcal/mol. Similar values are also observed for 3h416 and the reference EAI045 where the electrostatic energies are also notably strong (being −143.7 and −125.2 kcal/mol, respectively), while the Vdw energies turn out to be minimal (−18.8 and −18.6 kcal/mol, respectively). Consequently, the dominance of polar interactions results in a small difference in the number of contacts between these compounds, namely, with 1650, 1780, and 1585 contacts per frame for 3h389, 3h416, and EAI045, respectively. In contrast, compound 3h409, with a Vdw energy of −41 kcal/mol, forms an average of 2700 contacts per frame.

The large electrostatic energies observed for interactions between the imidazole-pyridine compounds and the EGFR variant can be attributed to an involvement of polar residues within the allosteric region. Specifically, the three compounds 3h389, 3h409, and 3h416 interact with key residues such as K745, E762, and D855, which have previously been identified as important for intermolecular binding. As shown in Table , K745 contributes −46.5, −32.1, −20.6, and −30.3 kcal/mol of electrostatic energy to the 3h389, 3h409, 3h416, and EAI045 systems, respectively. Similarly, E762 contributes −21.3, −17.4, −25.4, and −25.7 kcal/mol to the same systems. These findings suggest that, similarly to EAI045, the three compounds 3h389, 3h409, and 3h416 successfully prevent the formation of a salt bridge between K745 and E762. Notably, the strongest interaction of −82.7 kcal/mol is observed between both 3h389 and D855. In the case of 3h409, both D855 and K860 (located in the A-loop of the EGFR protein) are in nonbinding contact with the compound, despite relatively suitable Coulomb interaction energies, as detailed in Table .

3. Electrostatic Potential Energy Values between Specific Residues in the Allosteric Pocket and Four Different Compounds .

N0	residue	3h389	3h409	3h416	EAI045
1	F723	–0.2	–0.1	0	0.1
2	K745	–46.5	–32.1	–20.6	–30.3
3	L747	0	0	–0.2	–0.2
4	R748	0	0	–0.3	–14.7
5	E749	0	–11	–7.2	0
6	E758	0.1	–0.1	–6.6	–0.4
7	I759	0	0.1	0.1	0
8	E762	–21.3	–17.4	–25.4	–25.7
9	T854	–0.4	–3.2	0	–1.2
10	A763	0	0	0	0
11	D855	–82.7	–2.8	–19.6	–14.5
12	F856	–4.2	–3	–0.1	–0.5
13	M766	0.1	–0.2	0	–0.5
14	G857	–0.2	–6.1	–1.5	0
15	C775	0	–0.6	0	–1.6
16	L858	–3.6	–3.2	–1.5	–3.4
17	A859	–0.4	–0.3	–0.2	0.2
18	K860	–22.9	0.1	–29.3	–31.2
19	L777	0	–0.1	0	0
20	L788	0	–0.3	–0.1	–0.1
21	M790	0	–0.9	–0.2	–0.2
22	K867	0	0	–6.6	0
23	E868	0	0	–19.7	0
24	R836	0	0	–3.2	0
25	D837	–0.4	0	–2.4	–1

^aSignificant values are highlighted in bold. The van der Waals energy component is presented separately in Tables S4–S6 (SI file) due to its relatively minor contribution to the total interaction energy.

3.3. Similar EGFR Conformation Changes Are Induced in the Presence of 3h389 and EAI045 Interactions

An important issue concerns the configuration change in the EGFR protein backbone induced by the presence of ligands. Every snapshot extracted from the whole of 4000 ns MD simulations of four systems, including 3h389, 3h409, 3h416 compounds, and the EAI045 reference, are collected and clustered based on two components, the root-mean-square deviation (RMSD) and the radius of gyration (R _g). This means that ∼40,000 snapshots are implemented to construct the free energy surface as plotted in Figure . To identify the local minima in the free energy surface, the gmx-cluster package of Gromacs is employed and the RMSD and R _g cutoff are chosen at 0.01 nm. In this work, two deepest local minima are found and denoted as M1 and M2 in Figure , which was plotted under the support of Gnuplot open-source sofware.

5.

Free energy landscapes in 2D (A) and 3D (B) views are constructed from 40,000 snapshots of EGFR backbone, extracted from a total of 4 ns MD simulations (3h389, 3h409, 3h416, and EAI045 systems). Two local minima, M1 (3h389 and EAI045 located) and M2 (3h409 and 3h416 located), are also identified. Separate free energy profiles for each ligand were plotted in Figure S6. (C) Plot of the differences of conformation changes between M1 (in magenta) and M2 (in green) local minima; ligands are hidden to simplify the picture.

There are exactly 482 snapshots located in the M1 region, equal to 1.2% of the total ones. From the view side of the M2 region, we counted 550 snapshots in the M2, which is approximately 13.8% of the total snapshots. The most interesting feature in this section worth mentioning is that we recognize two representative conformations of 3h389 and EAI045 systems that are chosen independently from its performance, which are placed in the M1 location, whereas two similar ones of 3h409 and 3h416 belong to the M2 region. This observation has indicated that the EGFR backbone conformation changes induced by 3h389 and EAI045 are similar to each other. It is the same in the case of 3h409 and 3h416 when the related structures belong to the M2 location. Moreover, the lower panel in Figure visualizes the difference in EGFR conformation change. According to the pictures displayed below and numerical data for interaction energies between ligands and key amino acids (Tables S4 and S6, SI file), we can also demonstrate that the interactions between 3h389 and EGFR allosteric pocket, including Coulomb and VdW potential energies, are mostly comparable to those of EAI045. This indicates that the 3h389 compound can be considered as a strong candidate that emerged from this design of EGFR inhibitors.

When 482 and 550 snapshots located in the M1 and M2 regions, respectively, were taken into account, the gmx do_dssp module helped us to have a look at the protein secondary structures. Calculations revealed that the number of residues in β formation is unchanged across the five β-strands (β1−β5). On the other hand, when we explored 18 residues that make up the αC-helix loop, from S752 to V769, there was a significant difference. Only an average of 11.5 residues (per 18 ones, or 64%) were in helix formation when 3h409 and 3h416 bound. In the case of the 3h389 and EAI045 systems, the corresponding value is 16.1, or 89%. This suggests that these compounds can have the ability to significantly stretch the αC-helix loop structure, which is known to be a crucial component of EGFR dimerization. The impact of this phenomenon warrants more research.

3.3.1. Binding Free Energy between Candidates and EGFR Variant

In Figure , the 2D ligand-residue interaction maps are produced from four representative structures that are extracted separately in each simulation. Related free energy surfaces are shown in Figure S3 (SI file). These structures are now used in computations with the Umbrella Sampling MD method to determine the absolute binding free energy between a ligand and the protein. The steered molecular dynamics (SMD) simulations with low pulling velocity that are produced over 120 windows in each system, enhance the sampling conformation of transition states. The panels in Figure S4 (SI file) display some pertinent diagrams.

The free energy profiles from four systems selected are shown in Figure , where the binding free energy values indicate the free energy differences between the bound and unbound states. In comparison to the EAI045 reference system, which has an absolute binding free energy of −12.1 kcal/mol, the 3h389 compound enjoys an exceptionally low binding free energy of −25.6 kcal/mol. Table lists all binding free energy differences. The binding free energies of systems 3h409 and 3h416 are significantly higher than those of 3h389, namely at −19.7 and −14.4 kcal/mol, respectively. But the latter also emerge as markedly strong inhibitors. The result obtained in this part not only emphasizes three promising compounds but also proves that our strategy, including MD and Umbrella Sampling MD computations, is effective as a realistic scenario for computationally aided drug design.

7.

Free energy profile obtained by the gmx wham package of gromacs. Data are collected from more than 120 windows. Related diagrams of each US performance are inserted in Figure S4 (SI file). Bootstrapping errors of 3h389 and EAI045 are also computed and plotted in Figure S5 (SI file).

4. Absolute Binding Free Energies of 3h389, 3h409, 3h416, and EAI045 Systems .

system	binding free energy (kcal/mol)
3h389	–25.6
3h406	–19.7
3h416	–14.4
EAI045	–12.1

^aRepresentative structures obtained from 1.1 us MD steps are employed to perform the Umbrella Sampling MD methods.

4. Concluding Remarks

When a molecular dynamics simulation on a ligand–protein interaction is long enough, a ligand can undergo one of the two possible scenarios, either progress toward a more favorable equilibrium state of the resulting complex or be released from its initial binding position as we have statistically replicated in Figures S7–S9 (SI file). While the limitations of the docking algorithm may cause some potential candidates to be overlooked, it appears that a compound which successfully completes two independent 500 ns MD simulations has a high probability of being a strong inhibitor.

On the initial configurations suggested by docking computations, 100 ns simulations of first-round molecular dynamics (MD) simulations are both necessary and effective for screening the binding abilities of 15 novel imidazole­[1,5-a]­pyridine compounds with the EGFR kinase protein from which 8 compounds were ejected. The following simulation round, using longer (500 ns) simulation times, eliminates 4 of the rest.

The present theoretical study also recommends that assessment of the activity of a compound requires consideration of various parameters rather than solely relying on conventional metrics such as the RMSD, hydrogen bonds, or contact numbers. These metrics alone are insufficient to draw comprehensive conclusions. In practice, the distances from the compound considered to the K745 and S797 residues can serve as indicators as to whether a ligand remains in the binding pocket or is dissociated. The computed RMSD values, contact numbers, or hydrogen bond numbers cannot provide us with a reliable determination. Finally, from a computational viewpoint, although evidence from the umbrella sampling MD method is reliable enough for a good prediction in drug design, application of this method always needs a strategic approach in the context of low computing resources. The high binding free energy computed, −25.6 kcal/mol, should be regarded with caution, because it is a strictly computational prediction and does not represent a real experimental binding energy.

Our present investigation into the allosteric binding capacity of the imidazole-pyridine compound assigned as 3h389 is particularly encouraging. This opens new research avenues, such as an exploration of the synergistic dual-inhibiting effects of 3h389 in combination with known ATP-competitive TKIs, including the erlotinib, gefitinib, or osimertinib drugs. An effective combination of 3h389 with the first-, second-, and third-generation TKIs offers a valuable therapeutic strategy for an EGFR inhibition using small organic molecules. In future studies, detailed analyses of the interactions between fragments within the 3h389 structure (cf. Figure ) and specific residues of the protein can provide us with deeper insights that are needed for the evaluation of this compound as a new drug.

The Supporting Information is available free of charge at https://pubs.acs.org/doi/10.1021/acsomega.5c06704.

• Minimum distances between the ligand and residues surrounding the allosteric region of the EGFR mutant; minimum distances between the ligand and specific residues, indicating compound dissociation from the allosteric pocket; averaged values calculated from the last 200 ns of the second-round MD simulations; minimum distance between 15 ligands and two specific residues, K745 (red) and S797 (black), as a function of time and root mean square (PDF)

The authors declare no competing financial interest.