How Leucine Locks Sestrin2’s Shape to Recognize Specific Branched-Chain Amino Acids: A Computational Study

Construction of a full-length SESN2 model was essential for studying conformational dynamics involving unresolved regions missing from the 5DJ4 crystal structure. Integration of the AlphaFold-predicted loops into the crystallographic template produced a complete structure that preserved the SESN2 folds while reinstating key flexible regions. Structural alignment of the final minimized hybrid model with the original 5DJ4 coordinates demonstrated excellent conservation of the global architecture, particularly within the C-terminal domain (residues 339–480), which encloses the leucine-binding site. This domain exhibited a backbone root-mean square deviation (RMSD) of only 0.88 Å, indicating that the incorporation and subsequent minimization of the reconstructed segments did not perturb the core structure. Although the N-terminal and internal linker segments displayed greater conformational variability, consistent with their intrinsic flexibility, their incorporation did not disrupt local packing arrangements. These observations confirm that the hybrid structure retains the experimentally supported three-dimensional framework while providing a physically realistic representation of previously unresolved regions. The final model, therefore, provides a robust starting point for molecular docking and MD simulations to characterize the dynamic roles of these flexible segments in SESN2 function.

Molecular docking was performed to generate initial energetically favorable binding poses and to validate the computational setup. The protocol was first validated by re-docking the native ligand, leucine, into the SESN2 binding site. As shown in Figure 1, the resulting docked pose showed excellent agreement with the experimentally determined crystallographic pose, yielding an RMSD of only 0.68 Å. This confirms that the docking protocol can reliably reproduce the correct binding orientation. Analysis of the docked pose revealed extensive hydrogen bonding and polar contacts with key residues, including THR374, TYR375, THR377, GLU451, and HIS454, all of which were conserved in the experimental structure.

Following this validation, the same protocol was used to predict the binding modes of isoleucine and valine (Figure 2). The docking scores for leucine, isoleucine, and valine were −6.15 kcal/mol, −6.15 kcal/mol, and −6.40 kcal/mol, respectively. These scores suggest comparable binding affinities, with valine even appearing slightly more favorable. This highlights a known limitation of docking scoring functions: while excellent for pose prediction, they may lack sufficient sensitivity to rank the binding affinities of closely related ligands accurately. Nevertheless, the docking successfully identified a conserved binding mode for all three BCAAs, anchored by interactions with GLU451 and HIS454, with their side chains occupying the hydrophobic pocket. These validated poses served as ideal initial configurations for the extensive all-atom MD simulations required to dissect the subtle energetic and dynamic differences that govern specificity.

Before dissecting the specific molecular interactions, we first validated the convergence and reproducibility of the extensive MD simulations that underpin this study. For each of the four systems (apo, leucine-bound, isoleucine-bound, and valine-bound), three independent 500 ns simulations were performed to ensure robust sampling. The stability of each system was assessed by monitoring key structural and energetic parameters over time. The RMSD of the protein Cα atoms for all systems stabilized after an initial relaxation period (typically within ~50 ns), indicating that the simulations reached a stable structural equilibrium (Figure 3). Furthermore, the total and potential energies of each system remained stable throughout the simulation runs, confirming thermodynamic equilibration.

Crucially, the trajectories from the three independent replicates for each system spanned a narrow, consistent range of conformational space, as reflected by low variance in both RMSD and Radius of Gyration (Rg) (Figure 3 and Figure 4). This high degree of consistency demonstrates the reproducibility of our results and provides confidence that the observed conformational states are representative. The trajectories from all three independent replicates for each system are presented directly in the subsequent figures to demonstrate the reproducibility of our results and provide a clear view of the conformational sampling.

A quantitative basis for the binding specificity of BCAAs to SESN2 was established by first determining their relative binding affinities. We employed the Molecular Mechanics/Generalized Born Surface Area (MM/GBSA) method, a robust computational technique for estimating the binding free energy (ΔG_bind) from MD simulation trajectories. A more negative ΔG_bind value indicates a more stable, thermodynamically favorable interaction. The calculations were performed on the final, equilibrated portion of three independent 500 ns simulations for each complex, ensuring the results are statistically sound.

The computed binding free energies, presented in Table 1, reveal a clear and distinct hierarchy of binding affinity among the three BCAAs. Leucine binds to SESN2 with the highest affinity, exhibiting a ΔG_bind of −37.60 ± 2.39 kcal/mol. This is followed by isoleucine, with a moderately strong affinity of −34.47 ± 1.98 kcal/mol. Valine demonstrates the weakest interaction, with a ΔG_bind of −30.32 ± 3.18 kcal/mol. Acknowledging the standard deviations observed across the three replicates, we performed independent two-sample t-tests to assess whether the differences in binding free energies were statistically significant. The analysis revealed a ~7.3 kcal/mol advantage of leucine over valine (p = 0.034), providing strong quantitative support for SESN2's specificity. This substantial thermodynamic advantage is the primary driver of specificity, ensuring that, under physiological conditions, leucine preferentially occupies the binding site over other BCAAs to initiate the signaling cascade.

In contrast, the minor difference between leucine and isoleucine (~3.1 kcal/mol) did not reach statistical significance (p = 0.155) with the current sample size of three replicates. This suggests that while a clear binding preference for leucine over isoleucine is observed, a larger number of simulations would be required to establish this difference with high statistical confidence. Nevertheless, the overall computationally derived binding order—with leucine as the most potent binder—is in excellent agreement with experimental evidence that firmly establishes SESN2 as the primary cellular sensor for leucine [10].

The energetic contributions that govern these differential affinities were dissected by decomposing the total binding free energy into its constituent energetic terms (Table 1). This analysis provides insight into the nature of the BCAA-SESN2 interaction. The most significant favorable contribution to binding for all three ligands comes from van der Waals energy (ΔEvdW), which reflects non-polar interactions such as dispersion forces and serves as a proxy for shape complementarity. The magnitude of this term directly correlates with the binding affinity hierarchy: leucine shows the most favorable ΔEvdW (−28.09 ± 1.28 kcal/mol), followed by isoleucine (−25.92 ± 1.38 kcal/mol), and valine (−21.96 ± 1.29 kcal/mol). This trend strongly suggests that the primary determinant of binding strength is the ability of the BCAA's aliphatic side chain to maximize favorable contacts within the hydrophobic binding pocket of SESN2. Leucine's larger isobutyl side chain provides the greatest surface-area contact, leading to the strongest interaction.

Conversely, the electrostatic contribution, represented by the Coulomb energy (ΔE_Coulomb), is unfavorable for all three complexes. This indicates that the binding process likely involves a significant desolvation penalty for the charged amino and carboxyl groups of the BCAAs as they move from the aqueous solvent into the more confined binding pocket. While these groups facilitate favorable interactions within the pocket, they are insufficient to offset the energy cost of removing their hydration shells. The solvation energy terms, particularly the favorable lipophilic energy (ΔG_Lipo), further reinforce the dominance of hydrophobicity in driving the binding event. Finally, the calculated ligand strain energy was minimal for all three BCAAs, indicating that they can adopt their bound conformation without incurring a significant energetic penalty. In summary, the thermodynamic analysis establishes that SESN2 preferentially binds leucine due to stronger hydrophobic and van der Waals interactions, providing the quantitative foundation for a structural and dynamic explanation of this specificity.

Having established that leucine binds to SESN2 with the highest affinity, the next important question is to understand the structural consequences of this binding event. To investigate this, we analyzed the overall structural stability of SESN2 in its unbound (apo) state and in complexes with each of the three BCAAs, using RMSD and Rg as metrics. The RMSD of the protein's Cα atoms measures the deviation of the backbone from its initial structure, providing a quantitative assessment of conformational stability over time. The Rg measures the protein's overall size and shape, indicating whether it undergoes global conformational change [11]. The convergence of the trajectories in independent MD simulation runs was confirmed before proceeding to further analysis. The RMSD analysis in Figure 3 shows ligand-induced stabilization. The apo-SESN2 protein consistently exhibits the highest RMSD values across the simulations, with large fluctuations centered on a mean of ~3.5–4.0 Å. This indicates that the unbound protein is intrinsically flexible and dynamically samples a broad range of conformations, which is consistent with the experimental difficulty of obtaining a crystal structure of the apo state [4]. In contrast, binding any of the three BCAAs reduces RMSD, indicating that ligand engagement stabilizes the protein and restricts its conformational freedom.

A clear hierarchy of stabilization emerges, mirroring the thermodynamic affinities determined by MM/GBSA. The leucine-bound complex is the most stable, settling into a low and very stable RMSD plateau around ~2.5–3.0 Å. The valine-bound complex is also highly stable, with a similar RMSD profile, whereas the isoleucine-bound complex is intermediate in stability, with a slightly higher, more fluctuating RMSD. This robust correlation between binding energy and structural rigidity suggests that the more tightly a ligand binds, the more effectively it dampens the protein's intrinsic motions and locks it into a stable fold.

The Rg analysis reveals a key consequence of ligand binding: a significant stabilization of the protein's overall structure, as shown in Figure 4. The apo-protein exhibits substantial fluctuations in its Rg, with an average of ~21.4–21.5 Å, indicating that it exists in a dynamic, conformationally flexible state. Upon binding leucine or valine, these fluctuations are markedly dampened, and the protein settles into a more stable state with a consistent Rg of ~21.3 Å. While the change in the average size is subtle, the reduction in dynamic variability is significant. This effect is characteristic of a 'conformational selection' mechanism, where the ligand does not induce a large-scale compaction but rather stabilizes a pre-existing, signaling-competent conformation from the protein's dynamic ensemble. This stabilization is the hallmark of an allosteric mechanism, providing a plausible explanation for how a local binding event can remodel and stabilize distant functional interfaces, such as the one required for GATOR2 interaction. Notably, this stabilizing effect is less pronounced for isoleucine, whose Rg profile remains more flexible and similar to the apo state, providing a structural rationale for SESN2's ligand specificity.

The observed conformational change upon binding of leucine and valine raises the question of how this signal is propagated throughout the protein structure. To map these pathways of intramolecular communication, we analyzed residue-level flexibility and correlated motions within the protein using root-mean-square fluctuations (RMSF) and a dynamic cross-correlation matrix (DCCM).

The RMSF analysis, which measures the fluctuation of each residue, provides a high-resolution map of protein dynamics. As expected, binding of all three BCAAs dramatically reduces the flexibility of residues within the binding pocket, physically constraining their motion (Figure 5). However, the most significant insight comes from examining regions distant from the binding site. The apo-protein displays several prominent peaks of high flexibility, corresponding to mobile loop regions that connect the structural domains of SESN2 (e.g., residues ~240–260 and ~320–340). Upon leucine binding, the fluctuations in these distant loops are markedly dampened. This reduction in flexibility in regions far removed from the ligand provides direct, compelling evidence of an allosteric communication network. The binding event in the Sesn-C domain does not remain localized; instead, it transmits a stabilizing signal that propagates throughout the protein's architecture, rigidifying its overall structure. This allosteric stabilization is the likely mechanism by which leucine binding remodels the conformation of the GATOR2 binding interface, which is known to be spatially distinct from the leucine pocket, to cause its dissociation.

To map the pathways of intramolecular communication and understand how the ligand-binding signal is propagated, we analyzed correlated motions using DCCM analysis. To distill complex residue-level data into a clear picture of allosteric regulation, we constructed simplified network models that capture the average correlations among key functional domains (Figure 6). This analysis reveals a clear ligand-dependent activation gradient (Leucine > Isoleucine >> Valine ≈ Apo), where only leucine optimally organizes the protein's dynamic network for full allosteric communication.

The unbound apo-protein (Figure 6, far right) exhibits a baseline level of internal communication, reflecting its native flexibility. However, its global network is suboptimal, characterized by diffuse correlations and a lack of organized long-range communication. Similarly, valine binding provides minimal stabilization, resulting in a fragmented, sparsely connected network in which the leucine-binding microdomains and the CTD are largely disconnected from other functional regions.

In contrast, isoleucine binding elicits a partial activation, maintaining some core network connections (e.g., between Cys125 and the CTD) but failing to propagate dynamic communication as efficiently as leucine, particularly to the NTD and the GATOR2 docking domain.

The binding of leucine (Figure 6, far left) induces the highest degree of coordinated dynamics, transforming the protein into a fully activated allosteric network. The DCCM shows the strongest and most structured positive correlations, with extensive long-range communication. The network model reveals that leucine establishes a strong allosteric axis centered around the Cys125, CTD, Leu_Lid, and GATOR2-interacting domains. This dense web of communication provides a physical pathway for the binding signal to be transmitted to distal functional sites. These results highlight that, although SESN2 possesses an intrinsic capacity for dynamic coupling, binding only its primary ligand, leucine, can fully organize and stabilize the specific allosteric network required for maximal biological activity.

Taken together, these analyses reveal that leucine binding does not merely stabilize SESN2 but actively organizes its internal dynamics into a coherent, allosterically coupled network. This network provides the physical means to transmit the signal of leucine's presence from the sensory pocket to distal functional sites. This allosteric communication is fundamental to understanding SESN2's role as a multi-functional signaling hub. The same network that likely modulates the GATOR2 interface may also interact with other functional regions, such as the Sesn-A domain, which contains the Cys125 residue critical for its antioxidant activity [12]. This provides a structural framework for hypothesizing how the binding of a single amino acid could simultaneously and differentially regulate SESN2's distinct roles in metabolic signaling and redox homeostasis.

We performed a detailed analysis of the atomic-level interactions within the binding pocket and assessed the stability of the ligands themselves to uncover the ultimate origin of SESN2's binding specificity. This provides the foundational, chemical explanation for the thermodynamic and dynamic differences observed among the three BCAAs. Our analysis of the protein–ligand interaction networks across the simulations reveals a set of conserved contacts that anchor the common backbone of all three amino acids. The charged amino and carboxyl groups of the BCAAs are consistently stabilized by a network of hydrogen bonds and water-mediated bridges, primarily involving the side chains of residues Glu451 and His454, and a cluster of threonine residues (THR374, THR375, THR377) (Figure 7). This common anchoring motif ensures that each BCAA is appropriately positioned within the binding cleft, presenting its unique side chain for recognition. The critical determinants of specificity and affinity, however, lie in the differential interactions between the hydrophobic side chains and the pocket's nonpolar lining. The nature and extent of these contacts vary significantly for each amino acid. Leucine forms the most diverse and robust interaction network, which accounts for its superior binding affinity. Its binding is characterized by multiple persistent hydrogen bonds with a cluster of threonine residues (THR374, THR375, THR377) and ASN376. This polar network is complemented by crucial water bridges and electrostatic interactions with HIS454 and GLU451, which further anchor the ligand. Finally, its isobutyl side chain forms extensive hydrophobic contacts with residues like VAL453 and, critically, ILE378. This dense, multivalent network of polar, hydrophobic, and water-mediated interactions provides a powerful enthalpic driving force for binding. Isoleucine retains some key interactions, including hydrogen bonds with the threonine cluster and contacts with GLU451 and HIS454. However, its interaction profile is less cooperative. It forms fewer stabilizing water bridges, and its sec-butyl side chain makes fewer hydrophobic contacts than leucine. This results in a moderate, but significantly weaker, interaction network. Valine's smaller isopropyl side chain severely restricts its interaction footprint. While it maintains strong hydrogen bonding with the threonine cluster, it forms significantly fewer interactions with HIS454 and with key hydrophobic residues, such as ILE378, VAL455, and LEU458. This results in a more straightforward, less robust binding interface, as evidenced by the lowest binding free energy.

This narrative of differential fit is in accordance with the analysis of the per-atom RMSF of the bound ligands themselves. This metric quantifies the positional stability of each atom within the ligand, where lower RMSF values indicate a more rigidly held atom. The backbone atoms exhibit relatively low, comparable fluctuations across all three ligands, confirming stable anchoring, whereas the key distinction, once again, lies in the side chains (Figure 8). Leucine exhibits the lowest overall RMSF, with its central backbone atoms being exceptionally stable (<0.6 Å), confirming that its extensive interaction network tightly anchors it. In contrast, valine shows the highest average RMSF, with its side chain atoms fluctuating up to ~1.9 Å, a direct consequence of its sparse contacts and weaker anchoring. Isoleucine displays intermediate flexibility, with its terminal side chain being notably less restrained than leucine's. This hierarchy of ligand stability (Leucine < Isoleucine < Valine) is a direct physical manifestation of the quality of their respective interaction networks. It provides a compelling, atomic-level explanation for the observed binding affinities and the differential ability of these ligands to stabilize the entire protein structure.

To synthesize our structural, dynamic, and energetic findings into a unified thermodynamic picture, we analyzed the conformational ensembles using PCA and density-based clustering. This approach enabled a direct, quantitative comparison of the energy landscapes of the apo-protein and ligand-bound forms.

Our analysis reveals profound, ligand-specific differences in conformational heterogeneity, as summarized in Table 2 and visualized in Figure 9. The unbound apo-protein explores a vast, rugged energy landscape, sampling 35 distinct conformational states. This is characteristic of a flexible, high-entropy ensemble, consistent with its high RMSD and the experimental difficulty in obtaining an apo-state crystal structure. The isoleucine-bound form remains similarly heterogeneous, occupying 27 distinct states, indicating that its binding is insufficient to restrict the protein's intrinsic flexibility substantially.

In contrast, the binding of leucine or valine induces a dramatic collapse of the conformational landscape. Both ligands reduce the number of accessible stable states by approximately four-fold to just 9 (Table 2). This is visually represented in Figure 9, where the sprawling, scattered distribution of the apo and isoleucine systems is funneled into a few, tightly packed conformational islands for the leucine and valine complexes. This demonstrates that only specific ligands can drive the large-scale conformational selection required to stabilize the protein. However, this reduction in conformational states is not, by itself, sufficient for biological activity. The key differentiator lies in the quality of the resulting stabilized ensemble. As revealed by our dynamic network analysis (Figure 6), only leucine binding establishes a strong, fully connected allosteric communication network capable of propagating the binding signal to distal functional sites. While valine can stabilize the protein, its resulting allosteric network is fragmented and suboptimal, failing to organize the protein for downstream signaling efficiently. Furthermore, leucine is more effective at kinetically trapping the protein, reducing the time spent in transitional states to just 2.6% compared to 4.2% for valine (Table 2). Therefore, SESN2's specificity for leucine arises not just from its ability to bind tightly, but from its unique capacity to lock the protein into a thermodynamically stable and allosterically competent signaling state.

Beyond reducing the number of states, leucine and valine also fundamentally alter the protein's dynamics. The apo and isoleucine forms spend more than 11% of their simulation time in low-density "noise" regions that correspond to high-energy transition states between stable conformations. Upon leucine binding, this value plummets to a mere 2.6% (Table 2). This indicates that leucine not only reduces the static conformational entropy but also imposes a kinetic barrier, effectively trapping or "locking" the protein in its stable basins and preventing exploration of the broader landscape.

The distribution of state populations further underscores this principle of conformational funneling (Figure 10). The apo and isoleucine systems exhibit a characteristic long-tail distribution, with dozens of lowly populated microstates, the fingerprint of a high-entropy system. Conversely, the distributions for the leucine and valine systems are dominated by only three highly populated macrostates, which together account for more than 95% of the total population. This provides clear, quantitative evidence that specific ligand binding overcomes the protein's intrinsic entropy to select for and stabilize a small, well-defined ensemble of functional conformations.

The experimentally determined X-ray crystal structure of the human SESN2 core domain bound to leucine (PDB ID: 5DJ4) served as the primary template for this study. Although 5DJ4 successfully captures the global fold and ligand-binding architecture, several flexible regions were unresolved, including the N-terminal 65 residues and three internal linker loops spanning residues 242–255, 272–280, and 296–309. These segments are likely to influence inter-domain communication, conformational flexibility, and potential allosteric behavior; therefore, a complete model was required to enable meaningful downstream simulations. A full-length hybrid structure was generated by reconstructing the missing segments using high-accuracy structural predictions from AlphaFold2 [13]. The predicted coordinates for the absent residues were grafted into the 5DJ4 template, while preserving the experimentally observed atomic positions in all resolved regions. Loop insertion was guided by sequence alignment and local structural context to minimize backbone discontinuities, ensuring that the composite model retained the experimental fold while restoring the disordered and flexible motifs essential for dynamic analyses. The resulting chimeric structure was prepared using the Protein Preparation Wizard in Maestro (version 14.5) within the Schrödinger Suite (Schrödinger Release 2025-1) [14]. Hydrogen atoms and proper bond orders were assigned, protonation states were set at physiological pH, disulfide linkages were verified, and local steric clashes at the graft junctions were corrected. A restrained energy minimization with the OPLS4 force field was then performed to relieve local strain and optimize the hydrogen-bonding network, particularly around the reconstructed regions [15]. To further refine the model, the minimized hybrid structure was subjected to a best-practice relaxation protocol in Desmond. The system first underwent vacuum relaxation via steepest-descent minimization, with heavy-atom restraints applied to the crystallographically defined core to relieve short-range steric clashes without distorting the native fold. The model was then solvated in a TIP3P water box with 10 Å padding and neutralized; a second minimization stage was performed with gradually reduced positional restraints to allow proper relaxation of solvent–loop interactions. Finally, the system was subjected to a short, restrained equilibration under NVT conditions, followed by NPT conditions, allowing the AlphaFold-derived regions to equilibrate while maintaining the original 5DJ4 structural framework.

The three-dimensional structures of the branched-chain amino acids L-leucine, L-isoleucine, and L-valine were prepared using the LigPrep tool within Maestro (version 14.5), Schrödinger Suite (Schrödinger Release 2025-1: LigPrep, Schrödinger, LLC, New York, NY, USA, 2025). This step is crucial for generating accurate, low-energy 3D ligand conformations. The process included the generation of possible ionization states at a physiological pH of 7.4 ± 0.2 using Epik (version 7.3) (Schrödinger Release 2025-1: Epik, Schrödinger, LLC, New York, NY, USA, 2025), followed by the generation of tautomers and stereoisomers [16]. The geometry of each resulting ligand structure was then optimized using the OPLS4 force field to ensure it represented a stable, low-energy conformation suitable for docking and simulation studies [15].

To predict the initial binding poses of the BCAAs, a targeted molecular docking study was performed using the Glide (version 10.8) program in extra precision (XP) mode from the Schrödinger Suite (Schrödinger Release 2025-1: Glide, Schrödinger, LLC, New York, NY, USA, 2025) [17,18,19]. A receptor grid was generated by defining a bounding box centered on the coordinates of the co-crystallized leucine ligand within the known binding site of the 5DJ4 crystal structure. This targeted approach was selected over a blind docking study because the specific leucine-binding pocket is well established experimentally. The primary goal was not to discover novel binding sites, but rather to accurately generate initial binding poses for isoleucine and valine within this known active site and to validate the protocol by re-docking leucine, thereby creating reliable starting structures for the subsequent MD simulations. The docking protocol employed a rigid receptor and flexible ligand approach. Glide systematically searches for favorable ligand conformations and orientations within the defined active site through a series of hierarchical filters, including a grid-based search followed by energy minimization and scoring. The final poses were ranked using GlideScore, a proprietary empirical scoring function that approximates ligand-binding free energy and includes energetic terms for van der Waals and electrostatic interactions, hydrogen bonding, and desolvation penalties. The top-scoring pose for each BCAA was selected as the starting conformation for the MD simulations. The two-dimensional (2D) and three-dimensional (3D) ligand interaction diagrams shown in Figure 1 and Figure 2 were generated using the Maestro graphical interface.

All-atom MD simulations were performed using Desmond (version 8.3) (Schrödinger Release 2025-1: Desmond Molecular Dynamics System, D. E. Shaw Research, New York, NY, USA, 2024; Maestro-Desmond Interoperability Tools, Schrödinger, New York, NY, USA, 2025) from the Schrödinger Suite to investigate the structural dynamics and binding mechanisms of SESN2 with its ligands [20].

Post-simulation analysis was performed on the concatenated trajectories from the three replicates for each system. Trajectory visualization and molecular graphics were generated using the Maestro graphical interface (Schrödinger, LLC, New York, NY, USA, 2025). Analysis was conducted using tools within the Schrödinger Suite (Schrödinger Suite 2025-1) and custom Python scripts (version 3.9). Specifically, these scripts used the MDAnalysis library (version 2.7.0) to read and process trajectory data [21]. The quantitative plots of RMSD, Rg, and RMSF were generated using Matplotlib (version 3.8.0). The DCCM plots were also calculated from the Cα atom coordinates and plotted using these custom Python scripts.

The Cα-RMSF was calculated for each residue to identify regions of high and low flexibility, providing insights into the local dynamic changes upon ligand binding. The Rg was computed over time to measure the overall conformation of the protein, indicating large-scale changes such as expansion or contraction. The interactions between SESN2 and each BCAA, including hydrogen bonds, hydrophobic contacts, and water bridges, were monitored throughout the simulations to identify key residues and quantify the stability of specific contacts. DCCM analysis was performed to quantify the correlated motions between pairs of Cα atoms. The resulting matrices highlight regions of the protein that move together (positively correlated) or in opposition (negatively correlated), revealing pathways of allosteric communication.

The binding free energies (ΔG_bind) for the SESN2–BCAA complexes were estimated using the MM/GBSA method as implemented in the Prime module (mmshare version 7.1) of the Schrödinger Suite (Schrödinger Release 2025-1: Prime, Schrödinger, LLC, New York, NY, USA, 2025). This method calculates the binding free energy as the difference between the free energy of the complex and the sum of the free energies of the unbound protein and ligand. MM/GBSA calculations were performed on 100 evenly spaced frames extracted from the final 100 ns of each MD trajectory using the thermal_mmgbsa.py script provided by Schrödinger. The free energy for each species is estimated as the sum of the molecular mechanics energy (E_MM), a polar solvation term (G_GB) calculated using the Generalized Born model, and a non-polar solvation term (G_SA) estimated from the solvent-accessible surface area (SASA). The entropic contribution (−TΔS) was not calculated due to its high computational cost and potential for introducing inaccuracies, a common practice in comparative MM/GBSA studies. The final reported ΔG_bind values and their components are averages over the three independent replicates for each system.

The conformational space explored by each system was characterized through Principal Component Analysis (PCA) and density-based clustering. PCA was performed on the Cα atoms of the protein from the concatenated trajectories using Schrödinger's trj_essential_dynamics.py utility. PCA is a dimensionality reduction technique that identifies the principal modes of collective atomic motion. The analysis was carried out using Schrödinger's trj_essential_dynamics.py utility with atom selection limited to the receptor and ligand. The trajectories were projected onto the first two principal components (PC1 and PC2), which accounted for the dominant conformational variations.

To provide a quantitative, objective comparison of the conformational landscapes, the projected data for each system were clustered using the Density-Based Spatial Clustering of Applications with Noise (DBSCAN) algorithm, implemented in the scikit-learn library. This method was chosen for its ability to identify arbitrarily shaped clusters and to classify low-density transition-state regions as noise, thereby providing a more physically realistic partitioning of the conformational states. A single, consistent eps parameter of 1.2 and a min_samples value of 4 were used across all four systems to ensure a direct and unbiased comparison. The eps value was determined by analyzing the point of maximum curvature on a k-distance graph (k = 4) generated from the apo-protein trajectory, ensuring that the chosen parameter was appropriate for the most complex landscape.

The Gibbs Free Energy Landscape (FEL) was then constructed using the GROMACS gmx sham module (T = 310 K, 100 energy levels), which estimates the Gibbs free energy ΔG = −kBTln(P), where P is the probability density of each conformation in the PC1–PC2 space. The resulting 2D and 3D landscapes depict stable low-energy conformational states as deep basins and transition states as higher-energy barriers, providing a comprehensive picture of the protein's conformational dynamics and thermodynamics.