Proposed Allosteric Inhibition of Cyclin-Dependent Kinase 4 by a Proline-Derived Acylsemicarbazide Compound with Antiproliferative Activity in Breast Cancer Cells Authors: Xu Huang, Qingyang Nian, Xizhe Sun, Yuheng Zhou, Yuxin Wang, Jiayin Yue, Fanhao Meng, Jingwei Liang Published: 2026 ABSTRACT Background: Cyclin-dependent kinase 4 (CDK4) is a key regulator of cell-cycle progression and an established therapeutic target for breast cancer. Although the unique architecture of its ATP-binding site has enabled the development of highly selective inhibitors, the emergence of acquired resistance highlights the need for alternative therapeutic strategies targeting protein conformational regulation. Methods: The conformational landscape of CDK4 was investigated in this study using accelerated molecular dynamics (aMD) simulations combined with Markov state model (MSM) analysis to identify cryptic conformational states and potential allosteric binding sites. Particular attention was given to the glycine-rich loop (G-loop), a critical structural element that shapes the ATP-binding pocket, and to the effects of compound 8i (1-(2-(3-chlorobenzoyl)hydrazine-1- carbonyl)-N-(pyridin-3-yl)pyrrolidine-2-carboxamide), previously synthesized in our laboratory, on CDK4 dynamics. Results: A distinct conformational transition was identified in which the G-loop shifted toward the N-terminus, resulting in the exposure of a previously unrecognized allosteric pocket adjacent to the catalytic site. Compound 8i interacted with Leu147 and was associated with stabilization of conformational states that favor exposure of the cryptic pocket. Conclusions: These observations suggest that ligand binding may modulate the conformational landscape of CDK4 and favor formation of a cryptic pocket with potential allosteric characteristics. The identified conformational mechanism provides new insights into the dynamic regulation of CDK4 and suggests that stabilization of transient allosteric states represents a promising strategy for the rational design of next-generation CDK4 inhibitors with the potential to overcome resistance in breast cancer therapy. 1. Introduction Cyclin-dependent kinases (CDKs) constitute a class of serine/threonine protein kinases that collaborate with cyclins to orchestrate and modulate the progression of the cell cycle at different phases. Their pivotal role in the regulation of the cell cycle has been well established [1]. Within tumor cells, inhibiting CDK activity instigates cell cycle arrest, impeding cell proliferation, thereby underscoring their significance as prime anti-tumor targets [2,3,4]. Notably, CDK2 emerges as a cyclin-dependent kinase pivotal for steering cells through the G1 phase and transitioning into the S phase. CDK2’s interaction with and activation by cyclin E sustains the phosphorylation of the retinoblastoma protein (RB) during the latter part of G1, facilitating the seamless traversal from G1 to S phase [5,6]. In the realm of anti-breast cancer drug research and development, inhibitors directed against CDK2 have garnered considerable attention. Several compounds have advanced to clinical studies, revealing promising anti-tumor efficacy [7]. Nevertheless, the potential adverse effects of these inhibitors observed in clinical trials cannot be disregarded, encompassing conditions such as neutropenia, leukopenia, and myelosuppression. These effects primarily stem from CDK2’s extensive substrate spectrum and its involvement in diverse signaling pathways. Excessive inhibition poses a substantial risk of disrupting normal cellular functions, thus challenging the progress of drug development for this specific target [8]. In recent years, CDK4/6 inhibitors represented by Palbociclib have been approved for the treatment of breast cancer. Compared with CDK2 inhibitors, CDK4/6 inhibitors also inhibit the proliferation of breast cancer cells by inducing G1/S phase arrest. But the difference is that CDK4/6 inhibitors do not affect other CDKs involved in the regulation of normal cell cycle, effectively reduce the side effects. It brought these inhibitors become the most valuable and promising anti-breast cancer drugs [9,10]. However, with the clinical application of CDK4/6 inhibitors, the number of drug-resistant cases is increasing, which seriously limits the effectiveness of these inhibitors [11,12]. Research findings have underscored the significant role of compensatory CDK4/6 activation in mediating breast cancer resistance to CDK2 inhibitors. A pivotal requisite for cells transitioning from G1 to S phase is the phosphorylation of the retinoblastoma protein (RB). Within the CDK family, CDK2 and CDK4 act in concert to intricately orchestrate the cell cycle progression from G1 to S phase, thereby collectively facilitating the RB protein’s phosphorylation process. In cases where CDK4 in breast cancer cells faces inhibition, unabated CDK2 activity persists, leading to the continued phosphorylation of the RB protein as a bypass pathway and the uninterrupted propulsion of cells into the S phase. Consequently, this phenomenon engenders drug resistance in breast cancer cells. To circumvent this challenge, concurrent inhibition of CDK4 and CDK2 has emerged as a potential solution to surmount drug resistance associated with CDK4/6 inhibitors [13]. However, the central hurdle in the development of dual-target inhibitors targeting CDK2 and CDK4 resides in the pronounced spatial divergence characterizing the active pockets of these two kinases [14]. This divergence predominantly stems from the conformational disparities exhibited by the G-loop domains implicated in the constitution of these active pockets. In recent years, with the rapid development of computer technology, the combination of molecular dynamics (MD) simulation [15] and Markov state model (MSM) analysis [16] has been successfully applied to the study of protein folding [17], molecular recognition [18], and conformational dynamics [19,20]. Inspired by our previous work, we synthesized a novel CDK4 inhibitor designated as 8i ((S)-1-(2-(3-chlorobenzoyl)hydrazine-1-carbonyl)-N-(pyridin-3-yl)pyrrolidine-2-carboxamide) in this study [21]. Combined with molecular dynamics and the MSM, we found a new allosteric mechanism in CDK4 and found an amino acid residue that is in the active pocket that contributes greatly to the stability of protein-ligand binding. This work provides a theoretical basis and new ideas for solving the problem of drug resistance of CDK4 inhibitors and designing new CDK2/4 dual-target inhibitors. 2. Result 2.1. Putative Allosteric Role of Leu147 in CDK4 Inspired by our previous work [21], we synthesized several chiral enantiomers to further explore this skeleton. One of them is 8i, which is an enantiomer of the previously reported compound 7i. We found that its selectivity was different from the compounds we previously reported. Its inhibitory activity against CDK4 was similar to that against CDK2, while it had almost no inhibitory activity against CDK6 (Table 1). This indicates that it has unique combining characteristics. To further elucidate its interaction profile, we conducted a 300 ns accelerated molecular dynamics simulation of the protein–ligand complex (Figure 1A,B). Post-simulation analysis revealed that both the ligand and protein backbone exhibited RMSD (Supplementary Figure S2 for RMSD profiles) fluctuations of approximately 0.2 nm, indicating a stable system. The RMSD profiles over time from accelerated molecular dynamics simulations of the wild-type protein and the wild-type protein–ligand complex suggests that the small molecule 8i can markedly accelerate conformational changes of the protein. Furthermore, analysis of the RMSD of 8i in complex with both the wild-type and mutant proteins indicates that Leu147 may play an important role in the ligand-associated conformational regulation of the protein. In parallel, RMSF analysis was conducted for both the wild-type protein–ligand complex and the mutant protein–ligand complex (Figure 1C,D). RMSF analysis shows that binding of the small-molecule ligand is associated with conformational stabilization of wild-type CDK4, stabilizing key regions and reducing residue flexibility. The Leu147 to Ala147 mutation significantly alters local dynamics, changing fluctuations in specific areas. These results suggest that ligand binding alters CDK4 conformational dynamics and that Leu147 contributes to maintaining these ligand-associated structural changes. A representative 50 ns trajectory was extracted near the end of the accelerated molecular dynamics, and the binding free energy was calculated using the single-trajectory method in gmx MMPBSA [22], yielding a value of −5.67 kcal/mol (Figure 2A). The CDK4L147A mutant was constructed using Modeller 10.2. Following energy minimization, the protein model was successfully validated using PROCHECK and a 300 ns molecular dynamics simulation of the mutant–small molecule complex was performed. Comparative analysis with the CDK4 wild-type revealed that the protein backbone RMSD of the CDK4L147A mutant fluctuated around 0.3 nm within a 0.1 nm range. The RMSD of the wild-type protein–ligand complex changed more rapidly over time than that of the mutant complex, suggesting that the small-molecule ligand 8i significantly accelerates the conformational dynamics of the protein, while Leu147 appears to contribute to ligand-associated conformational regulation. To corroborate the findings above, we conducted Drug Affinity Target Stability (DARTS) experiments on both wild type and mutant CDK4 proteins, analyzing the samples through Western blot (WB). When bound with the small molecule 8i, the wild-type CDK4’s stability significantly increased, it resisted to enzymatic hydrolysis by Pronase E. Conversely, due to the Leu147 mutation, the CDK4L147A protein failed to bind with 8i. It exhibited reduced binding stability with small molecules, resulting in a weakened resistance to enzymatic hydrolysis (Figure 2B). Furthermore, to elucidate the rational design concept that led to the discovery of 8i, a structural comparison among the general structure from our previous study, and compound 8i is presented in Figure 2C [21]. In the wild-type CDK4, the ligand 8i adopts a deeply inserted binding conformation, stably embedded within the protein’s binding pocket (Figure 3A). The surrounding residues form a well-defined enclosure, tightly wrapping the ligand and resulting in a compact and stable structure. In contrast, in the L147A mutant, the 8i ligand exhibits a noticeable positional shift, with incomplete embedding and partial exposure to the solvent, leading to a looser and less confined binding pocket (Figure 3B). Leucine 147, a moderately sized, hydrophobic residue, is located at the edge or semi-enclosed surface of the binding pocket. It plays a crucial role by providing steric support or compression to constrain the ligand’s position, forming a hydrophobic platform that promotes π–hydrophobic interactions or clustering, and maintaining the geometric closure of the pocket to prevent ligand slippage. When mutated to alanine, which has a smaller side chain and lower hydrophobicity, the pocket becomes enlarged and lacks effective spatial filling, resulting in greater conformational freedom but reduced stability of the ligand (Figure 3C,D). Ligand interaction diagrams further reveal interaction changes. In the wild-type CDK4, 8i forms hydrophobic interactions (indicated by yellow/green lines) near Leu147, potentially involving van der Waals contact surfaces (Figure 4A). Additionally, it forms one to three hydrogen bonds or polar interactions (blue lines) with surrounding residues, creating a multipoint cooperative interaction network that stabilizes ligand positioning. In contrast, in the L147A mutant, the loss of hydrophobic packing with the original Leu147 residue leads to a local misalignment of the ligand, resulting in the loss of other hydrogen bonding interactions (Figure 4B). The overall interaction network is reduced and more unevenly distributed, leading to significantly weakened binding affinity. Although Leu147 may not directly form hydrogen bonds, its spatial conformation and hydrophobic nature contribute to a favorable “hydrophobic platform” and help maintain the positioning of nearby polar residues, thereby indirectly stabilizing other hydrogen bonds. This site thus plays both a structural support and an indirect enhancing role in ligand binding. Structural analysis showed that Leu147 does not form identifiable hydrogen bonds, salt bridges, or π-stacking interactions with 8i. However, its proximity to the ligand and the observed conformational changes in surrounding residues during the simulations suggest that Leu147 may contribute to local structural rearrangements associated with ligand binding. Based on these computational observations, we propose that Leu147 could function as a conformational modulator within the binding pocket. Further experimental studies will be required to validate this proposed role. One possible explanation consistent with our simulation results is that ligand binding may induce subtle steric effects around Leu147, leading to local side-chain rearrangements and changes in the geometry of neighboring residues. Such structural adjustments could contribute to formation of a binding environment more favorable for ligand accommodation. We emphasize that this model represents a computational hypothesis derived from molecular dynamics analyses and has not yet been experimentally verified. A potential interpretation of the simulation data is that ligand binding may alter the conformational preferences of the local structural environment surrounding Leu147. Such local changes could, in principle, influence more distant regions of the protein through coupled structural motions. Within this computational framework, Leu147 may represent one component of a broader network of residues associated with conformational regulation. However, the existence of specific signal propagation pathways, dynamic coupling networks, or long-range allosteric communication was not directly tested in the present study and therefore should be regarded as a hypothesis requiring further experimental validation. To validate the binding affinity of 8i to wild type CDK4 and mutant CDK4 respectively, we conducted Surface Plasmon Resonance (SPR) experiments (Figure 5). The proteins were immobilized on a Sensor Chip teem with -COOH, 8i was tested at concentrations ranging from 31.25 to 1000 μmol/L for wild type and 62.5 to 1000 μmol/L for the mutant respectively. The sensorgram fitting curve indicated a Kd value of 0.289 mmol/L for CDK4 wild type with 8i and 0.774 mmol/L for the CDK4 mutant with 8i. These results underscored the higher affinity of CDK4 wild type for 8i compared to the mutant, highlighting the impact of the Leu147 mutation on the binding capability of proteins and small molecules. 2.2. Putative Allosteric Process of CDK4 To explore the conformational changes during protein movement, we then carried out large-scale aMD simulations of CDK4 before and after small molecule 8i binding. Then we use PyEMMA to identify the key structural changes in the binding process [23]. Firstly, the backbone torsions were characterized to describe the overall topological structure of the protein, and using TICA to reduce the dimension of molecular simulation data. 100 steps were selected as the delay time and retained the first two components IC1 and IC2, which made the greatest contribution to the kinetic variance. We use IC1 and IC2 to construct two-dimensional energy landscape maps of combined and unintegrated systems respectively (Figure 6A,B). It shows that the protein will show more conformational changes when binding ligands. Using a k-means clustering algorithm for clustering, the whole simulation trajectory is clustered into 200 microstates. Then, the implicit time scale is used to calculate the lag time of the Markov model, and the implicit time scale curve converges when the lag time is about 40 ps. Using the PCCA algorithm, the micro-states are further clustered into six steady-state S1–S6 of the bound state and two steady-state S1 and S2 of the unbound state, and then the Chapman-Kolmogorov test verifies that the constructed MSM is Markov (Table 2). To identify the key structural features in the process of pocket change, we extracted representative conformations from each metastable state and superimposed them with CDK4 structures for further study (Figure 6C,D). To identify the structural change process of CDK4, we use protein root mean square deviation (RMSD is the coordinate deviation of a specific part of atoms relative to the reference structure, the smaller the difference, the smaller the RMSD) to evaluate and screen the initial and final states in MSM results. The RMSD values of each metastable conformation and CDK4 G-loop region were calculated, and the initial and final states of the structural changes in the bound state were determined to be S1 and S6, respectively. In the unbound state, the initial state was S1 (1.80 Å) and the final state was S2 (4.86 Å). By aligning the final state of bound and unbound states with the inactive conformation of CDK2, it is found that the conformational similarity of the superimposed conformation is very high. Then we also calculate the RMSD of the final state and CDK2 inactive state of the G-loop in bound and unbound states, which are 2.10Å and 2.30Å, respectively. Whether it binds to small molecules or not, after continuous movement, the G-loop of CDK4 tends to the shape of the inactive state of CDK2, which makes the active pocket similar (Figure 7B). Comparing the MFPT of the two, the time from the initial state S1 to the final state S6 in the bound state is 21.7 ns, while the time from the initial state S1 to the final state S2 in the unbound state is 90.8 ns. The distance between N atom (8i) and H atom (Leu147) in the trajectory of the CDK4 binding state is calculated by using Amber’s CPPTRAJ packet. The contact distance between two atoms usually occurs at 250–300 ns. It is suggested that the interaction between small molecules and the residue Leu147 may accelerate the conformational change of the protein and shorten the timescale over which this change occurs. 3. Discussion The crystal structure of CDK4 (PDB ID: 3G33) was selected as the initial model. We deliberately chose the apo form of this structure, as it represents the natural conformation of CDK4 before ligand binding, allowing us to study the conformational changes induced by ligands and the formation of potential hidden binding sites in the unbound state. Since the 3G33 structure does not contain the co-crystallized small molecule ligand, the complex conformation was obtained through the molecular docking method we reported previously [21]. The present study integrates accelerated molecular dynamics (aMD) simulations with Markov state model (MSM) analysis to investigate a previously unrecognized, putative allosteric mechanism in CDK4 that may be facilitated by the small molecule 8i. Our findings, supported by complementary experimental validation, suggest that Leu147 may serve as an important regulatory residue in the proposed CDK4 allostery, and that 8i binding is associated with a conformational transition in which the G-loop shifts toward the N-terminus, exposing a cryptic pocket with potential allosteric character and structural similarity to the inactive conformation of CDK2. These results carry significant implications for understanding CDK4 regulation and for the rational design of next-generation dual-target inhibitors. 3.1. Leu147 as a Potential Conformational Regulator of CDK4 Our computational and experimental results consistently suggest that Leu147 contributes to ligand binding and conformational stabilization of CDK4. The aMD simulations revealed that the wild-type CDK4–8i complex exhibited faster RMSD convergence and lower fluctuations compared to the L147A mutant complex, suggesting that 8i binding accelerates conformational stabilization in a Leu147-dependent manner. RMSF analysis further demonstrated that ligand binding reduces residue flexibility in key structural regions—particularly around the G-loop—while the L147A mutation disrupts this stabilizing effect, leading to altered fluctuation patterns across multiple domains. The MM/GBSA binding free energy calculations provided quantitative evidence that the Leu147Ala substitution substantially diminishes the binding affinity of 8i for CDK4. This was corroborated by SPR experiments, which showed an approximately 2.7-fold increase in Kd (from 0.289 mmol/L to 0.774 mmol/L) upon mutation. Although these Kd values indicate moderate binding affinity overall—consistent with the micromolar-range IC50 values—the consistent trend across independent methods strongly supports the functional importance of Leu147. Notably, the DARTS assay provided direct biochemical evidence that 8i confers proteolytic resistance to wild-type CDK4 but fails to protect the L147A mutant, confirming that Leu147 is essential for ligand-induced structural stabilization. Structural analysis offered mechanistic insight into Leu147’s role. Although Leu147 does not form direct hydrogen bonds or π-stacking interactions with 8i, its hydrophobic side chain provides steric support and maintains the geometric closure of the binding pocket. In the wild-type complex, 8i is deeply embedded within a compact, well-defined pocket, with Leu147 contributing to a hydrophobic platform that constrains the ligand’s position. Upon mutation to alanine, the reduced side-chain volume creates excess cavity space, leading to ligand misalignment, partial solvent exposure, and disruption of the cooperative interaction network—including loss of key hydrogen bonds with neighboring residues. This finding is consistent with the concept of “hydrophobic platform” residues that indirectly stabilize ligand binding by maintaining the spatial organization of nearby polar residues, an analogy previously invoked for G protein-coupled receptors (GPCRs) and other regulatory kinases. Based on the simulation results, we propose a working model in which Leu147 may contribute to ligand-associated conformational regulation. Upon 8i entry into the binding pocket, minor steric repulsion with Leu147 may trigger side-chain displacement, which could propagate through a dynamic coupling network to reshape the binding site into a conformation more favorable for ligand accommodation. This proposed pre-organization mechanism could reduce the entropic penalty of binding and enhance affinity—a paradigm well-documented in allosterically regulated systems. The analogy to thermostabilized GPCRs, where key residues serve as structural anchors that prolong ligand residence time, further supports this interpretation. 3.2. G-Loop Dynamics and a Putative Cryptic Pocket with Potential Allosteric Character The MSM analysis revealed a marked expansion of conformational space upon 8i binding, with six metastable states (S1–S6) identified in the bound system compared to only two (S1–S2) in the apo system. This diversification indicates that 8i does not merely stabilize a single conformation but rather facilitates sampling of multiple conformational substates, broadly analogous to the role ascribed to allosteric modulators in reshaping the free-energy landscape [24,25]. Transition path theory (TPT) analysis identified the dominant pathway S1→S3→S6 (54.30% of total flux) and the secondary pathway S1→S3→S2→S6 (27.15%), together accounting for over 81% of conformational transitions. Along this pathway, the G-loop undergoes a progressive shift toward the N-terminus, resulting in gradual enlargement of the active-site pocket. This directional movement is of particular significance because it represents a structural transition from the CDK4-typical active-site geometry toward a conformation resembling the inactive state of CDK2. The RMSD values of the G-loop region between the CDK4 final states and the CDK2 inactive conformation (PDB: 5A14) were 2.10 Å (bound) and 2.30 Å (unbound), indicating substantial structural convergence regardless of ligand presence. The observation that CDK4 can adopt a CDK2-like inactive geometry has profound implications for the design of dual-target CDK2/CDK4 inhibitors. The pronounced spatial divergence between the active pockets of CDK2 and CDK4, which primarily originates from conformational differences in their respective G-loop domains, has long been recognized as a central challenge in developing such dual-target agents [14]. Our findings suggest that this divergence is not static but rather dynamic: the G-loop of CDK4 can transition to a conformation that narrows the structural gap between the two kinases, thereby creating a transient binding pocket capable of accommodating ligands designed for CDK2. This dynamic convergence provides a structural rationale for the moderate CDK2 inhibitory activity of 8i (IC50 = 9409.8 nM) alongside its CDK4 activity (IC50 = 17,677.3 nM), as well as its negligible CDK6 inhibition—underscoring the selectivity profile unique to this compound. 3.3. 8i as a Potential Modulator of CDK4 Conformational Dynamics A particularly notable finding is the kinetic disparity between the bound and unbound systems. The mean first passage time (MFPT) from the initial to the final state was 21.7 ns in the bound system versus 90.8 ns in the unbound system, representing an approximately 4.2-fold acceleration of conformational transition upon 8i binding. CPPTRAJ analysis further revealed that the contact distance between 8i and Leu147 typically forms at 250–300 ns, temporally coinciding with the accelerated phase of conformational change. This suggests that the 8i–Leu147 interaction may act as a kinetic catalyst that lowers the energy barrier for G-loop displacement, potentially channeling the conformational transition along the productive pathway. Our simulations suggest that 8i may influence the conformational landscape of CDK4 beyond simple orthosteric binding; however, additional studies are required to determine whether this represents a bona fide allosteric mechanism. By stabilizing intermediate states along the S1→S3→S6 pathway and promoting the G-loop shift, 8i appears to facilitate the formation of a cryptic pocket that is not present in the ground-state structure. This mode of action is conceptually analogous to the cryptic allosteric site identified in SIRT6, where small-molecule binding induces the formation of a previously inaccessible pocket [25], and to the allosteric modulators of CDK4 described by Morris et al. [4], which regulate kinase activity through conformational mechanisms distinct from direct active-site blockade. 3.4. Implications for Overcoming CDK4/6 Inhibitor Resistance CDK4/6 inhibitors, including palbociclib, ribociclib, and abemaciclib, have significantly improved therapeutic outcomes for hormone receptor-positive breast cancer. However, the development of acquired resistance remains a major limitation of this therapeutic strategy. Multiple resistance mechanisms have been reported, including activation of compensatory CDK2 signaling, Cyclin E amplification, loss or alteration of RB pathway regulation, and adaptive changes in cell-cycle control networks. Among these mechanisms, CDK2 activation has received considerable attention because it can bypass CDK4/6 inhibition and restore RB phosphorylation, allowing tumor cells to continue proliferating despite CDK4/6 blockade [11,12,13]. Current strategies to overcome resistance largely focus on combination therapies, particularly approaches targeting both CDK4/6 and CDK2. However, broad inhibition of multiple CDK family members may increase toxicity and limit therapeutic windows. Therefore, the development of selective CDK4 inhibitors represents an attractive alternative strategy. Selective suppression of CDK4 activity may preserve the antitumor effects associated with CDK4 pathway inhibition while potentially avoiding some adverse effects related to CDK6 inhibition, particularly hematological toxicity. Furthermore, selective CDK4 inhibitors may provide opportunities for rational combination with CDK2 inhibitors to simultaneously block complementary cell-cycle escape mechanisms [14]. Our study offers a potential solution to this challenge. The pocket exposed by the G-loop shift in CDK4 (a putative allosteric pocket) adopts a geometry similar to the CDK2 inactive conformation, suggesting that rationally designed ligands could target both pockets simultaneously or sequentially. The identification of Leu147 as a potential regulatory node further provides a specific target for structure-based drug design: compounds that engage Leu147 and promote the conformational transition could function as allosteric facilitators that sensitize CDK4 to co-administered CDK2 inhibitors, or as bifunctional agents that bridge both active sites. This concept of allosteric sensitization may represent a paradigm shift from traditional orthosteric inhibition strategies and may open new avenues for circumventing resistance. Although our structures reveal how ligands regulate CDK4 conformation, cellular studies are needed to link these changes to signaling, cell-cycle dynamics, and clinical outcomes in breast cancer. 3.5. Limitations and Future Perspectives While our integrated computational and experimental approach provides supportive evidence for the proposed allosteric mechanism, several limitations should be acknowledged. First, the binding affinity of 8i for CDK4 is moderate (kd = 0.289 mmol/L, IC50 = 17,677.3 nM), which limits its direct therapeutic utility and suggests that structural optimization will be necessary to achieve pharmacologically relevant potency. Second, although DARTS and SPR experiments confirm the Leu147-dependent binding behavior, the mechanistic interpretations of signaling pathways and conformational propagation are derived primarily from molecular dynamics simulations, which inherently involve approximations in force-field parameters and sampling efficiency. An additional limitation of the present study is that the proposed ligand-associated conformational regulation was validated primarily through computational analyses and biochemical binding assays. Although mutagenesis, DARTS, and SPR experiments support the contribution of Leu147 to ligand binding and protein stabilization, the downstream biological consequences of these conformational changes have not yet been investigated in cellular systems. Future studies will therefore evaluate whether the proposed conformational regulation influences CDK4 kinase activity, RB phosphorylation, cell-cycle progression, and breast cancer cell proliferation. Such functional experiments will be essential for establishing the physiological relevance of the computational model. Future work should employ site-directed mutagenesis in cellular contexts, kinase activity assays with purified allosteric-state trapped conformations, and cryo-EM or X-ray crystallography to capture the cryptic pocket in the ligand-bound state. Additionally, structure-based optimization of 8i derivatives, guided by the allosteric pocket geometry and the Leu147 interaction hotspot, may yield compounds with improved potency and selectivity profiles suitable for preclinical evaluation. In summary, this study supports the proposed Leu147-mediated allosteric pathway in CDK4 that may be facilitated by the small molecule 8i, potentially leading to G-loop displacement and cryptic pocket formation. The convergence of this pocket toward a CDK2-like geometry provides a structural foundation for dual-target inhibitor design and offers a potential strategy to overcome CDK4/6 inhibitor resistance in breast cancer. These findings provide a basis for allosteric modulator discovery that may be applicable to other kinases with structurally divergent active sites. 4. Materials and Methods 4.1. Molecular Docking AutoDock 4 (version 4.2.6) was used for molecular docking. Firstly, the CDK4 (PDB ID: 3G33) structure was imported into SWISS-MODEL web server for residue repair [26]. This specific ligand-free (apo) structure was selected as the starting model because it represents the baseline conformational state of the kinase, and previous studies have successfully employed this structure in docking simulations to identify novel cryptic binding pockets and allosteric sites. The CDK4 and ligand structures were first converted to PDBQT format using AutoDockTools (version 1.5.7) and then loaded into AutoGrid 4 (version 4.2.6). 4.2. Classical Molecular Dynamics (MD) Simulation Classical MD simulations predict the physical movements of atoms and molecules over time by numerically solving Newton’s equations of motion. Empirical force fields are used to calculate the potential energy of the system, allowing us to evaluate the dynamic stability, conformational changes, and binding mechanisms of the protein-ligand complexes under simulated physiological conditions. 4.2.1. GROMACS GROMACS 2021.4 [27] was used to simulate MD. The Amber99SB-ILDN force field was used to prepare protein topology and parameter files, and the GAFF2 force field was used to describe small molecules. These force fields define the mathematical parameters for atomic interactions (such as bond stretching, angle bending, and non-bonded interactions), which are crucial for accurately depicting the dynamic behavior of the system. The RESP2 (Restrained Electrostatic Potential) charge of small molecules was calculated by Multiwfn 3.8 combined with ORCA [28], providing a highly accurate electrostatic representation of the ligand necessary for evaluating binding affinities. The protein-ligand complex was placed in a dodecahedral box and solvated using the TIP3P water model and SPC216 water molecules to mimic the physiological aqueous environment. The distance between the protein edge and the box was at least 10 Å. Enough Na+ and Cl- ions were added to neutralize the system to prevent artifactual electrostatic interactions across periodic boundaries. The steepest descent algorithm eliminated torsional strain and spatial conflicts by setting Fmax to 1000.0 kJ·mol−1·nm−1 to minimize the energy of the system, ensuring a stable starting structure without unnatural atomic overlaps that could crash the simulation. Then the NVT (constant Number, Volume, Temperature) and NPT (constant Number, Pressure, Temperature) equilibrations of 100 ps were performed respectively. These phases allow the solvent and ions to relax around the complex. The heavy atoms of the protein skeleton and ligand were restricted by using the Berendsen thermostat (V-rescale) and the Berendsen barostat. The balance of NVT and NPT gradually heated the system to 300 K and reached the pressure of 1 bar, establishing realistic thermodynamic conditions. After the position constraint was removed, the leap-frog integrator was used to run the MD of 200 ns. The particle mesh Ewald (PME) scheme was used to estimate the long-range electrostatic interaction, which accurately calculates Coulombic forces across the periodic box. The cutoff distance of van der Waals force and Coulomb interaction was set to 12.0 Å, balancing computational efficiency and physical accuracy. The hydrogen bonds were constrained by the LINCS algorithm, which removes high-frequency bond vibrations, allowing the minimum time step to be set to 2.0 fs to accelerate the simulation without losing thermodynamic stability. 4.2.2. Assisted Model Building with Energy Refinement (Amber) Amber(version 2024) [28] was used to simulate MD. The Amber99SB-ILDN force field was used to parameterize CDK4 and the GAFF2 force field was used to describe small molecules. The RESP2 charge of small molecules was calculated by Multiwfn 3.8 combined with ORCA [29]. The protein-ligand complex was placed in a dodecahedral box and solvated using the TIP3P water model. The distance between the protein edge and the box was at least 10 Å. Counter ions were added to neutralize the protein. Two rounds of energy minimization were carried out after preliminary preparation to remove steric clashes. First, the energy of water and ions was minimized by 5000 steps (the steepest descent in 2000 steps, and then the conjugate gradient minimization in 3000 steps), and the protein skeletons and ligands were constrained by 500 kcal/(mol·Å2) position constraints. Then, without any restriction, the 4000 steepest descent and 6000 conjugate gradient minimization cycles were carried out for the whole composite system. After that, the simulation system was heated to 300 K in 500 ps, then the 700 ps was balanced under the condition of NVT, and the 10 kcal/(mol·Å2) constraint was imposed on the protein and ligand structure, then the global equilibrium of 5 ns was carried out. Finally, the cMD (classical MD) mode of 50 ns was run at the time step of 2 fs at 1 atmospheric pressure and 300 K. After the completion of the simulation, the potential energy and the dihedral energy were calculated to prepare for further accelerated MD (aMD) simulation, which adds a boost potential to reduce energy barriers and enhance conformational sampling. To adjust the temperature of the system, the Langevin dynamics with 1 ps−1 collision frequency was applied, and the Berendsen thermostat with isotropic position scaling was used to adjust the pressure of the system. In the cMD process, the long-range electrostatic interaction was modeled by the particle mesh Ewald (PME) method, and the cutoff values of short-range electrostatic interaction and van der Waals forces were set to 10 Å. In addition, the SHAKE algorithm was used to constrain the covalent bonds involving hydrogen atoms, permitting a larger integration time step without system instability. 4.3. Molecular Mechanics Poisson–Boltzmann Surface Area (MMGBSA) Binding Free Energy Gmx_MMGBSA(2021.6) package was used to calculate the free energy of the single trajectory and Amber99SB-ILDN force field based on GROMACS [30,31]. The tool allows the calculation of free energy using MM/PBSA or GBSA (molecular mechanics/Poisson-Boltzmann or generalized born surface area) methods. The free energy of MM/PB (GB) SA binding between protein and ligand in the solvent can be expressed as follows:(1)∆Gbind=∆Gcomplex−∆Gprotein+∆Gligand The free energies of proteins, ligands, or complex components can be calculated as follows:(2)∆Gbind=∆EMM+∆Gsolv−T∆S(3)∆EMM=∆Ebond+∆EvdW+∆Eelec(4)∆Gsolv=∆GPB/GB+∆GSA In the above Equations, ∆EMM, ∆Gsolv and T∆S are the changes of molecular mechanical energy, solvation free energy and conformational entropy of gas phase when ligands are combined. ∆Ebond is the energy of bonding interactions, which was calculated to be zero in the dynamic simulation, and ∆Evdw and ∆Eelec are the van der Waals and electrostatic interaction energies, respectively. The ∆Gsolv is calculated from the polar ∆G(PB⁄GB) (electrostatic solvation energy) and non-polar ∆GSA between the solute and the continuous solvent. The polar contribution was estimated by GB-OBC1 model, while non-polar energy was estimated by solvent accessible surface area (SASA). In this work, snapshots sampled from the MD trajectories of each protein-ligand complex were used to calculate the binding free energy using the MM/GBSA method because it is computationally efficient and requires fewer resources. In the case of comparing the binding state of ligands to the same protein, the entropy (∆S) effect has the least effect on the total energy, so it was ignored. 4.4. Accelerated Molecular Dynamics (aMD) aMD was performed to enhance sampling and cover a larger conformational space after 50 ns cMD simulation. Throughout the aMD process, “lifting energy” was added to the simulation system to reduce energy barriers and achieved easier conversion between various conformations in the free energy landscape. In Equations (5) and (6), the potential energy of the system is calculated as Vr, and the corresponding threshold is determined by the size of the system and its normal energy value as Ethresh. In the aMD process, when the instantaneous potential energy is less than Ethresh, it is increased by ∆Vr to reach Vr*. Otherwise, the original value Vr is used.(5)Vr*=Vr; Vr≥Ethresh(6)Vr*=Vr+∆Vr; Vr