Computational identification and mechanistic characterization of natural product binders targeting the PDE6D prenyl binding tunnel

Abstract

RAS oncogenesis remains a significant clinical challenge due to the difficulty of directly targeting RAS proteins. PDE6D, a prenyl-binding chaperone involved in the RAS membrane trafficking pathway, represents an indirect yet tractable target whose modulation has been proposed to influence RAS localization and signaling. This study employed a multiscale structure-based in silico workflow to identify natural compounds with putative binding potential toward the prenyl-binding tunnel of PDE6D. A curated natural product library was screened using molecular docking, followed by density functional theory–based geometry optimization. Additional 500 ns molecular dynamics simulations were then used to analyze the stability of binding and the overall conformational behavior of the most promising complexes. Further PCA and free energy landscape mapping highlighted distinct low-energy conformational states adopting unique structural transitions, where the MolPort-039-052-621 complex presented the most compact and well-defined low-energy basin. In that respect, superimposing free energy minima with the respective initial docked poses demonstrated minimal structural deviation from the binding orientation during dynamic evolution. Finally, QM/MM calculations qualitatively described the electronic stabilization of the ligands within the PDE6D environment. The present study identifies natural compounds with computationally favourable tunnel-binding characteristics and provides mechanistic insights that may guide future experimental validation.

Supplementary Information

The online version contains supplementary material available at 10.1038/s41598-026-37939-4.

Introduction

Cancer remains one of the leading causes of morbidity and mortality worldwide. According to recent pan-cancer profiling of over 10,800 tumor samples, mutations in KRAS occur in approximately 20% of cancers globally, with especially high prevalence in pancreatic ductal adenocarcinoma (PDAC), colorectal cancer (CRC), and lung adenocarcinomas¹.

In PDAC, the overall KRAS mutation rate approaches 82–90%, with G12D being the most frequent KRAS alteration (~ 39–42%), followed by G12V, G12R, and Q61 variants². Importantly, cohort studies from 2024-to 2025 have confirmed that patients carrying KRAS G12D or G12V mutations experience significantly worse outcomes—including shorter overall survival and faster progression—than those with wild-type or less aggressive KRAS subtypes³. In colorectal cancer, KRAS mutations are present in metastatic cases; G12D and G12C are among the most common⁴. Overall, in metastatic CRC, Asia boasts mutation rates similar to or slightly lower than those in Western Europe, but the subtype distributions can differ⁵.

Despite decades of study, direct targeting of mutant RAS has remained elusive. The RAS proteins (KRAS, NRAS, HRAS) are small GTPases that cycle between GDP-bound (inactive) and GTP-bound (active) states, serving as central nodes in signaling pathways (RAF/MEK/ERK, PI3K/AKT, etc.)⁶. Oncogenic RAS mutations drive many cancers but are difficult to target directly, leading to interest in indirect approaches that modulate RAS localization and signaling. Mutations at positions G12, G13, or Q61 lock RAS in an active or activation-prone state, driving uncontrolled proliferation, survival, and metastasis⁶. The high frequency and grave prognosis associated with such mutations make RAS a critical therapeutic target. However, RAS lacks easily druggable pockets, has high affinity for physiological ligands (GTP/GDP), and its membrane localization is essential for oncogenic function⁷.

An alternative strategy to directly targeting RAS is to interfere with its trafficking and membrane localization machinery. PDE6D is a prenyl binding chaperone that facilitates cytosolic transport and membrane delivery of prenylated RAS proteins, enabling proper signaling localization. Disruption of this process can lead to RAS mislocalization and reduced downstream signaling, making PDE6D an attractive indirect therapeutic target⁸. Structurally, this is mediated through a conserved hydrophobic tunnel within PDE6D, which accommodates the farnesylated tail of RAS. This binding pocket, illustrated in Figure S1, is lined by critical residues such as Ser115, Gln114, Leu87, Glu88, and Trp90 that contribute to stabilizing ligand interactions and define the pharmacological target site. Epidemiologically, the need for new therapies against non-G12C RAS mutations is urgent. Approved inhibitors (such as sotorasib and adagrasib) are specific to G12C mutations, which represent only a small fraction of RAS-driven cancers in PDAC or CRC⁶. For the many patients with G12D, G12V, and other RAS mutations, prognosis remains dismal under current standard chemotherapies⁹. Because of its role in RAS membrane localization, PDE6D has gained traction as a potential indirect target to inhibit RAS oncogenic signaling. Preclinical studies have shown that interfering with the PDE6D-RAS interaction results in mislocalization of RAS, reduced activation of downstream effectors (e.g., ERK, AKT), and decreased proliferation in RAS-mutant cells⁸.

Several small molecules have been developed to inhibit the PDE6D prenyl binding pocket^10,11. Among early chemotypes, deltarasin binds PDE6D with moderate affinity and disrupts RAS localization but exhibits steep dose response behavior and off target cytotoxicity at higher concentrations. Subsequent compounds such as deltazinone showed improved selectivity and reduced cytotoxicity while retaining tunnel binding through interactions with key residues including Tyr149 and Arg61¹². More recently, DW0254 was validated in leukemia models: this compound inhibits PDE6D-RAS binding, induces RAS mislocalization, and reduces downstream pathway activation while being less toxic to normal cells compared to older inhibitors⁸. Several PDE6D inhibitors have been reported, establishing the prenyl binding tunnel as a druggable site for modulating RAS trafficking, although early compounds were limited by solubility, cytotoxicity, and modest biological efficacy. More recent inhibitors, including the Deltaflexin series, further support the feasibility of targeting PDE6D but still face challenges related to pharmacokinetics and incomplete suppression of RAS signaling. These findings indicate that while PDE6D is a tractable target, there remains a need for alternative chemical scaffolds with improved physicochemical and functional properties. In this context, the present study extends existing PDE6D focused efforts by exploring new natural product derived scaffolds using a computational framework to support future experimental validation¹³.

Given the high prevalence of RAS mutant cancers and the limitations of current RAS directed therapies, this study aimed to identify natural product derived binders capable of occupying the PDE6D prenyl binding tunnel and potentially interfering with RAS membrane localization. Natural compounds were selected due to their chemical diversity and potential to provide novel scaffolds with improved selectivity and physicochemical properties. A structure based in silico workflow combining molecular docking, density functional theory optimization, molecular dynamics simulations, and post simulation conformational analyses was employed to investigate dynamic and energetic features governing PDE6D ligand interactions. The objective was to computationally prioritize natural scaffolds and gain mechanistic insight into tunnel binding behaviour to support future experimental validation. The overall computational workflow is illustrated in Fig. 1.

Fig. 1.

Computational workflow employed in the present study.

Materials and methods

Virtual screening of the natural product library

As a first effort to screen hit compounds against the PDE6D–RAS interface, a structure-based virtual screening approach was used through deployment of the MTiOpenScreen webserver (http://bioserv.rpbs.)¹⁴. The NP Lib natural product library available on the MTiOpenScreen platform was selected for structure based virtual screening due to its pharmacologically enriched and chemically diverse scaffold composition. This library contains approximately 1,228 purchasable stereoisomers derived from 653 unique natural products and is pre filtered using the FAF Drugs pipeline to remove compounds with unfavorable physicochemical properties, toxicophores, and PAINS patterns. The 3D structure of PDE6D (PDB ID 7PAD) was prepared for docking by adding missing hydrogens, assigning protonation states at physiological pH, removing crystallographic water molecules and irrelevant heteroatoms, correcting bond orders, and assigning Gasteiger charges using UCSF Chimera, followed by brief energy minimization to relieve local steric clashes. All NP Lib compounds were standardized through the MTiOpenScreen pipeline, which performs stereochemical normalization, hydrogen addition, removal of reactive groups, and generation of three dimensional conformers, yielding chemically valid ligands suitable for AutoDock Vina scoring. These preparation steps ensured that both receptor and ligand libraries were structurally consistent and appropriate for reliable docking evaluation^8,15–17.

The binding pocket was defined around the binding ligand in the crystal structure to show the farnesyl-binding tunnel crucial to the localization of RAS. The interaction between PDE6D and RAS is mediated through the binding of the farnesylated C-terminal tail of RAS to a deep hydrophobic tunnel within the PDE6D protein¹⁸. The prenyl binding tunnel of PDE6D serves as the biologically relevant cavity for accommodating the hydrophobic prenyl group of RAS and was therefore selected as the docking site in this study. Docking grids were centered on the co crystallized ligand coordinates in PDB ID 7PAD to accurately represent the native interaction interface, and grid-based screening was performed using the MTiOpenScreen platform with AutoDock Vina scoring. All NP Lib compounds were docked into this tunnel and ranked based on predicted binding affinity¹⁹. Top ranked hits were further filtered based on ligand efficiency, proper placement within the tunnel, and interactions with functionally important PDE6D residues, followed by visual inspection and drug likeness assessment. Based on these criteria, three ligands were shortlisted for downstream quantum and molecular dynamics analyses. Docking protocol validation was performed by re docking the co crystallized control ligand from PDB ID 7PAD into the PDE6D binding pocket using identical grid parameters, followed by RMSD calculation based on ligand heavy atoms within the binding tunnel. An RMSD of 0.017 Å confirmed accurate reproduction of the experimental binding pose, as shown in Supplementary Figure S2.

Quantum chemical evaluation via density functional theory

Density functional theory calculations were performed to examine electronic features and optimized geometries of the shortlisted ligands. All quantum mechanical calculations were carried out using the B3LYP functional with the cc pVDZ basis set implemented in the PySCF framework, which provides a suitable balance between accuracy and computational efficiency for organic molecules. This level of theory was selected to enable relative comparison of electronic properties and frontier orbital characteristics rather than absolute energy estimation. Ligand structures were imported in SDF format using the RDKit toolkit for subsequent geometry optimization and electronic analysis²⁰. Ligand structures with embedded three dimensional coordinates were imported into PySCF for electronic structure calculations, and hydrogen atoms were retained to preserve correct molecular geometry. Following self consistent field optimization, HOMO and LUMO orbital energies were obtained. Electron density and orbital distributions were visualized using cube files generated on an 80 × 80 × 80 grid. HOMO–LUMO energy gaps were calculated in electron volts after conversion from Hartree units using the relation 1 Hartree = 27.2114 eV²¹.

The entire automated script is developed on a reproducible Python pipeline that combines RDKit to read structures and PySCF to deliver the DFT backend^20,22. The workflow ensured standardized file management and consistent processing across all compounds without manual intervention. A summary output was generated containing HOMO and LUMO energies, energy gaps, and related electronic descriptors used for qualitative assessment of molecular stability and potential electronic compatibility with the protein environment. This pre MD electronic screening step allowed early elimination of chemically unfavorable candidates prior to molecular dynamics simulations. The analysis was implemented using PySCF mean field workflows following standard PySCF computational protocols.(https://pyscf.org/quickstart.html#mean-field-theory).

ADMET analysis

The pharmacokinetic, physicochemical, and toxicity properties of the top three hit compounds, with a reference control, were accessed by using the ADMETlab 2.0 web server at https://admetmesh.scbdd.com/^23,24. This web platform aggregates a large number of predictive models on the basis of machine learning algorithms trained on curated experimental datasets, and calculates more than 300 descriptors relevant to drug discovery. For each compound, the canonical SMILES representation was submitted to generate a full ADMET profile. Predicted parameters included key physicochemical properties such as molecular weight, LogP, topological polar surface area (TPSA), hydrogen bond donors and acceptors, number of rotatable bonds, synthetic accessibility score (SA), and the quantitative estimate of drug-likeness (QED). The absorption and distribution profile encompassed human intestinal absorption (HIA), Caco-2 and MDCK cell permeability, blood-brain barrier (BBB) penetration, plasma protein binding (PPB), volume of distribution (VDss), and fraction unbound (Fu), together with interactions with P-glycoprotein (P-gp). Metabolic liability was assessed from the calculated interaction profile with major cytochrome P450 isoenzymes, including substrate and inhibition probabilities for CYP1A2, CYP2C9, CYP2C19, CYP2D6, and CYP3A4. Clearance-related parameters such as total clearance and biological half-life were also calculated. Moreover, a large panel of in silico toxicology endpoints was predicted, including hERG inhibition, hepatotoxicity (DILI), Ames mutagenicity, skin sensitization, and potential for carcinogenicity. Results were analyzed to evaluate drug likeness, natural product likeness, and developability.

Re-docking of DFT-optimized compounds

An initial round of structure based virtual screening was performed using the MTiOpenScreen server to rapidly prioritize compounds based on predicted binding affinity and pocket compatibility. Top ranked hits were subsequently re docked using AutoDock Vina implemented in UCSF Chimera to refine binding poses within the PDE6D prenyl binding tunnel under explicitly defined grid parameters. After DFT based geometry optimization, ligands were re docked again using the same protocol to confirm pose stability and retention of key interactions prior to molecular dynamics simulations. This stepwise docking strategy was adopted to reduce false positives and ensure reliable candidates for downstream dynamic and energetic analyses.

Gas phase DFT optimization refined intrinsic ligand geometry, and subsequent re docking ensured biologically relevant conformations within the solvated PDE6D prenyl binding tunnel prior to molecular dynamics simulations^15,19. The docking grid covered the farnesyl-binding tunnel with grid parameters of x = −20.37 Å, y = 0.89 Å, z = −11.24 Å, and the grid box size: 20 × 20 × 20 ų. The same grid center and box dimensions were applied during both the initial virtual screening, ensuring a consistent definition of the prenyl-binding tunnel and maintaining uniformity in binding-site evaluation throughout the workflow. The PDE6D structure was pre-processed to attain a conformation ready for docking. The protonation state of the ionizable residues and ligands was set at physiological pH (7.4) using UCSF Chimera in order to ensure that subsequent calculations had correct hydrogen bonding and electrostatic modeling. The optimized ligands were docked using an exhaustiveness of 8. Ranked-1 pose of 0 Å RMSD for each ligand having stable key interactions was selected for complex building. The hydrogen bond and hydrophobic contact interaction profile were investigated to ensure it is compliant with the tunnel architecture of PDE6D. This re-docking confirmed that DFT-refined ligands retained biologically relevant conformers before dynamic simulation.

Molecular dynamics simulation

All MD simulations in multiples of three under physiological conditions were performed using the free academic AMBER suite to determine the dynamic behaviour and structural stability of the PDE6D-ligand complexes. All-atom molecular dynamics (MD) simulations were performed using the free academic AMBER suite to study the dynamic behaviour and structural stability of the PDE6D-ligand complexes under physiological conditions^25–28. The protein–ligand complexes optimized in re-docking were initiated into simulation by the LEaP module²⁹. The protein was described by using the ff14SB force field, while the ligand parameters were obtained through the Antechamber program by using the General Amber Force Field, GAFF2, with AM1-BCC partial charge assignment to guarantee consistent parameterization for subsequent molecular dynamics simulations²⁹.

Every system was immersed in an octahedral box of TIP3P water molecules at a minimum buffer length of 10 Å from the protein surface to the box boundary. The system charge was neutralized by the incorporation of counterions (either Na⁺ or Cl⁻). The solvated complex was energy-minimized in two steps: first by restraining solute atoms and relaxing the water and ions, and then by an unrestrained minimization of the entire system. Then, the systems were heated in steps from 0 K to 300 K in an NVT ensemble with gentle restraints on the complex to prevent distortion of the structure. Following this, a 500 ps density equilibration at 1 atm in the NPT ensemble was done to allow the system to relax. Production runs were 500 ns long and were carried out using the pmemd.cuda module to take advantage of calculation acceleration through support of GPUs. An integration time step of 2 fs was applied throughout the duration of the simulation using the SHAKE method to limit all hydrogen-containing bonds³⁰. The PME method and non-bonded interaction cutoff at 10 Å were used to calculate long-range electrostatics³¹. The trajectories were saved every 50 ps and then post-processed.

For simulation quality assessment, Root Mean Square Deviation (RMSD) of backbone atoms, Root Mean Square Fluctuation (RMSF), and hydrogen bond behavior were computed utilizing CPPTRAJ³². Moreover, for each complex, independent replicate MD simulations were initiated using different random initial velocity seeds, followed by its protein backbone and ligand RMSD profiles were compared to assess reproducibility and convergence. The stability of the ligand in the PDE6D binding tunnel was also monitored by tracing its center-of-mass distance and conformational drift. The above setup of simulation helped us to deduce the conformational mobility, binding persistence, and overall conformational coherence of the complexes of ligands and PDE6D in nearly physiological conditions.

MMGBSA analysis

The binding free energy was computed for each of the protein-ligand complexes using the MMPBSA.py module of the AMBER suite³³. Representative snapshots were extracted from the equilibrated region of the production molecular dynamics trajectories of each protein-ligand complex. Each snapshot was processed for the complex, receptor, and ligand separately under identical conditions. Binding free energies were estimated by summing gas-phase interaction energies with implicit solvation contributions. Polar solvation energies were computed using the generalized Born model, while the nonpolar component was estimated from the solvent-accessible surface area. Entropic contributions were not included; hence, the values reported here represent relative binding free energies. Additional energy decomposition was carried out on a per-residue basis to identify important residues that contribute to ligand binding.

Principal component analysis and free energy landscape construction

For analysing dominant conformational motions in the PDE6D–ligand complexes in the course of molecular dynamics, Principal Component Analysis (PCA) was carried out using the Bio3D software in the programming environment R^34,35. This was performed on Cα atom positional fluctuations after trajectory alignment to remove rotational and translational motions. A covariance matrix was constructed and diagonalized to obtain eigenvectors describing dominant collective motions. The first two principal components (PC1 and PC2) were selected as collective variables for free energy landscape (FEL) construction, as they captured the major conformational variance sampled during the simulations and enabled visualization of ligand induced dynamic states^36,37. The distribution of conformations in PC1 and PC2 space was examined, and corresponding energy values were extracted by probability-based population mapping.

Quantum mechanics/molecular mechanics (QM/MM) calculations

QM/MM calculations were performed using the PySCF framework to capture interaction energies at the quantum level that cannot be resolved through classical molecular dynamics simulations^21,38,39. In this, only the ligand was treated at the quantum mechanical level, while the surrounding protein and solvent environment were represented as fixed molecular mechanics point charges (electrostatic embedding). This scheme captures polarization of ligand electron density induced by the protein electrostatic field but does not explicitly model quantum mechanical protein–ligand interactions or charge transfer between ligand and binding site residues. QM/MM calculations were performed as single point energy evaluations on FEL minimum structures obtained from MD simulations, without additional geometry optimization at the QM/MM level. This step, toward evaluating electronic stabilization of each ligand in its lowest-energy MD-derived binding conformation, has been included to provide insight beyond the energetics from force-field-based approaches⁴⁰. The MM region, constituting the protein environment, contributed electrostatic embedding into the QM Hamiltonian, thus self-consistently accounting for polarization effects. The thusly streamlined QM/MM protocol allowed for high-resolution assessment of the electronic contributions governing ligand-PDE6D stabilization.

Results

Virtual screening

This first approach relied on the MTiOpenScreen virtual screening platform to screen the NP-LIB natural product library for potential small-molecule binders of the PDE6D–RAS interface. For the study, the researchers targeted the hydrophobic inner tunnel of PDE6D where prenylated RAS peptides are known to reside during their trafficking^8,41.

Screening output was established for the recovery of the best 100 ligands, ranked according to their binding affinities estimated through the AutoDock scoring function (Supplementary Table S1). Among the recovered hits, the docking energies ranged from − 13.3 kcal/mol (most favorable) to − 9.5 kcal/mol and indicate energetically favorable and strong protein-ligand interactions for the PDE6D pocket. All of these shortlisted molecules were visually inspected to confirm proper insertion within the tunnel-like binding site structure.

From docking orientation, tunnel occupancy, and interaction potential, three molecules, viz. MolPort-039–052-621, MolPort-002–507-186, and MolPort-001–768-161, of good binding capability, having docking values of − 13.3 kcal/mol, − 12.1 kcal/mol, and − 11.0 kcal/mol, respectively, and better than the control ligand of (–10.4 kcal/mol), were selected for refinement and deeper level analysis. They were selected not only based on their binding scores, but shape complementarity and consistency of interaction between a few conformers, as well. A co-crystallized calibration ligand of the 7PAD structure was, similarly, left for calibrating the end-of-screen scoring, including energy decompositions based on DFT and molecular dynamics.

Electronic structure analysis via DFT

In order to predict the reactive potential and electronic structure of selected molecules, a DFT-driven investigation of their frontier molecular orbitals was undertaken. It was found that the localization of the Highest Occupied Molecular Orbital (HOMO) and Lowest Unoccupied Molecular Orbital (LUMO), and hence of electron density localization, points of electrophilic/nucleophilic attack, and intramolecular charge transfer ability (Fig. 2).

Fig. 2.

HOMO and LUMO isosurfaces of four ligands: a, b MolPort-039–052-621, c, d MolPort-002–507-186, e, f MolPort-001–768-161, and (g-h) Control. These ligands contain electron-rich (pink) and electron-deficient (blue) regions, which indicate potential regions of charge transfer and interaction beneath the binding tunnel.

Out of the compounds analyzed, MolPort-002–507-186 exhibited the highest energy gap of 7.480 eV, thus ensuring good intrinsic kinetic stability and minimal inherent reactivity under standard conditions. MolPort-001–768-161, on the other hand, exhibited the lowest gap of 4.068 eV, exhibiting enhanced polarizability and susceptibility for active interactions at the protein binding site. MolPort-039–052-621 was the second short-listed compound and exhibited a comparable HOMO–LUMO gap of 4.572 eV, achieving a balance between stability and reactivity.

Control ligand, of PDE6D X-ray structure, had a gap of 6.084 eV, thus positioning it between ligands under investigation w.r.t electronic delocalisation. Importantly, visual observation of orbital distribution confirmed active electron densities of the HOMO and LUMO of each of the three ligands under investigation, localized in the region of functional groups of tunnel binding, in particular, hydroxyl, carbonyl, and aromatic systems. It would suggest here that while structural stability at physiological conditions would be higher for MolPort-002–507-186, smaller intermolecular gaps of MolPort-001–768-161 and MolPort-039–052-621 would give enhanced binding flexibility and intermolecular charge transfer, and hence dynamical stabilization of the PDE6D channel would be favorable.

ADMET and drug-likeness profiling of top-ranked compounds

Since binding affinity alone is insufficient to evaluate drug candidacy, ADMET profiling was performed to assess potential pharmacokinetic and toxicity liabilities of the shortlisted compounds. To evaluate the biological relevance and preliminary pharmacokinetic suitability of the top-ranking hits identified through virtual screening, I conducted a comprehensive in silico ADMET and cheminformatic analysis. The evaluated compounds include MolPort-039–052-621, MolPort-002–507-186, and MolPort-001–768-161, in comparison to a known reference compound used as control. The results are summarized in Supplementary Table S2. The physicochemical analysis indicated that all three hits complied with Lipinski’s Rule of Five, suggesting acceptable baseline drug-likeness. However, significant variability was observed in solubility (LogS), polarity (TPSA), and lipophilicity (LogP), which directly impact bioavailability and permeability. Notably, MolPort-039–052-621 exhibited high polarity (TPSA = 162.98 Ų), which may restrict its membrane permeability, while MolPort-001–768-161 had a zero TPSA, suggesting extremely poor aqueous solubility. The absorption and distribution parameters revealed limited oral bioavailability for all three compounds, with low predicted human intestinal absorption (HIA < 2%) and unfavorable Caco-2 permeability. Only MolPort-001–768-161 showed moderate blood-brain barrier (BBB) penetration potential, although it also demonstrated strong plasma protein binding (PPB = 99.59%) and low free drug fraction, which could affect systemic exposure. In terms of metabolism, the compounds presented varying interaction probabilities with cytochrome P450 enzymes. MolPort-001–768-161 was predicted to be both an inhibitor and substrate of CYP2D6, raising potential for metabolic liabilities and drug-drug interactions. MolPort-002–507-186, while more metabolically stable, showed a very high probability of hERG inhibition (0.997), indicating a strong risk of cardiotoxicity. Toxicological profiling further highlighted concerns, particularly for MolPort-001–768-161, which demonstrated high mutagenicity (Ames test positive), carcinogenicity potential, and drug-induced liver injury (DILI) risk. MolPort-039–052-621 and MolPort-002–507-186 also showed DILI risks and high skin sensitization probabilities. Alarmingly, both MolPort-039–052-621 and the control compound triggered structural alerts, including toxicophores and genotoxicity flags. Despite these liabilities, all three compounds are classified as natural-product-like, with MolPort-002–507-186 scoring highest in NP-likeness (2.988). However, its synthetic accessibility score (5.33) suggests considerable challenges in chemical synthesis or large-scale procurement. Overall, these findings indicate that all three compounds exhibit significant pharmacokinetic and toxicity liabilities and may not be directly suitable as drug candidates without further chemical optimization. Therefore, the identified ligands are best interpreted as mechanistically relevant binding scaffolds rather than fully optimized therapeutic candidates, and their primary value lies in guiding future structure-based optimization and experimental validation efforts.

Redocking and molecular interaction analysis

In order to verify ligand binding stability following electronic optimization and ADMET analysis, refined structures of the DFT were subjected to redocking. Importantly, while redocking scores were mildly reduced, always positive scores were obtained: MolPort-039–052-621 at − 10.2 kcal/mol, MolPort-002–507-186 at − 9.6 kcal/mol, and MolPort-001–768-161 at − 10.9 kcal/mol. The modest decrease in docking scores after DFT optimization reflects the relaxation of intrinsic ligand geometries into lower-strain conformations that do not necessarily maximize the docking affinity; these refined structures retained their tunnel-compatible orientations upon re-docking, supporting preservation of biologically relevant binding poses rather than indicating loss of stability. Docking of the control compound optimized by DFT yielded − 10.1 kcal/mol, proving that candidate molecules shortlisted exhibited or surpassed comparable binding efficacies even after structural relaxation.

Molecular interaction analysis revealed typical binding signatures of the three ligands. MolPort-039–052-621 (Fig. 3a-b) formed exhaustive hydrogen bonds via Glu88, Thr131, Ser143, and Tyr149, and interacted via an expansive hydrophobic network of Leu17, Leu22, Val49, Leu63, and Phe133, and others. It, in addition, presented π–alkyl and stacking interactions via significant tunnel-lining residues of Trp90, Ile129, and Met20, and hence secured a great initial docking score. MolPort-002–507-186 (Fig. 3c-d), while having a slightly broader HOMO–LUMO gap and a comparatively smaller redocking score, presented a heterogeneous interaction profile. It established a hydrogen bond via Ile53, and hydrophobic contacts via Cys56, Leu54, Gln78, Tyr149, and Fhe133, and aromatic contacts via Trp32, Leu38, and Ile129, anchoring the molecule deep inside the hydrophobic tunnel. MolPort-001–768-161 (Fig. 3e-f), lacking significant hydrogen bonds, was stabilized through a compact aromatic and nonpolar contact network. Some of the residues participating in anchoring of the ligand through hydrophobic and van der Waals forces were Met20, Arg61, Trp90, Tyr149, and Val145. Lack of polar interactions may be countered by the conformational plasticity of the ligand and the favorable shape complementarity of the ligand and the interior of the tunnel. Conversely, the control molecule (Fig. 3g-h) demonstrated similar-in-scope interactions, featuring both hydrogen bonding (e.g., Ile53, Glu88) and a general hydrophobic contact face including Val49, Leu87, Trp90, and Leu147, validating its application as a comparative standard. Overall, these outcomes signify that the selected natural molecules maintain conserved associations of the PDE6D tunnel following geometry refinement at the quantum level, affirming their possibility of use for additional molecular dynamics and free energy investigation.

Fig. 3.

Redocking Interaction Profiles of Ligands with PDE6D Binding Pocket. 3D (a, c, e, g) and 2D (b, d, f, h) interaction diagrams of the binding poses and non-covalent interactions of the selected ligands, a, b MolPort-039–052-621, c, d MolPort-002–507-186, e, f MolPort-001–768-161, and g, h Control, with PDE6D.

Protein–ligand stability and residue flexibility assessment

During validation of the structural integrity of the complexes of PDE6D and ligands and assessing the influence of ligand binding on protein dynamics, a detailed analysis of Root Mean Square Deviation (RMSD) and Root Mean Square Fluctuation (RMSF) of the protein moiety was performed through 500 ns of MD simulation, as shown in the Fig. 4a for the protein and Fig. 4b for the ligand. While the replicates of 2nd and 3rd run is shown in the supplementary file as Figure S3 and S4. Within the MolPort-039–052-621 complex, there was a smooth protein backbone motion with RMSD between 1.2 and 2.3 Å for most of the simulation time, suggesting a relatively consistent conformational ensemble. There was a ligand trajectory exhibiting initial small motions but stabilization, showing that despite inner motion, there were sustained contacts of the ligand within the tunnel. However, MolPort-039–052-621 shows an overall slight increase in ligand RMSD after ~ 400 ns, reflecting minor local rearrangements within the binding tunnel. The protein backbone remained stable throughout the trajectory.

Fig. 4.

Protein backbone (a) and ligand (b) RMSD profiles of all complexes during the 500 ns molecular dynamics simulation.

Protein RMSD of the MolPort-002–507-186 complex was highly consistent, only varying narrowly between 1.4 and 1.9 Å, typical of very good structural preservation throughout the simulation. The ligand exhibited a correspondingly contained RMSD profile, indicative of rigid tunnel interaction and limited position drift. Such a profile is typical of a tightly bound, dynamically compatible interaction with PDE6D. Conversely, the MolPort-001–768-161 complex demonstrated contrasting behavior. While the protein backbone remained stable (~ 1.2–1.8 Å), the ligand RMSD showed significant growth, rising over 9 Å after 100 ns. This is reflected in the pronounced rise in ligand RMSD for MolPort-001–768-161, which indicates reduced stability of binding and partial egress from the tunnel, in concert with its overall weaker interaction profile. Such a curve might reflect an egress from the binding tunnel or entry into a loosely bound surface-bound state, reminiscent of poor anchoring or affinity within the tunnel environment. Protein was stable in the control complex over the course of the simulation, with RMSD between 1.3 and 2.1 Å. Ligand RMSD rose at a constant rate to ~ 5 Å, showing some motion in the pocket but no dissociation at all. That was what would be anticipated for a moderate-affinity ligand, which was, in this instance, used only as a comparator referent.

The RMSF plots listed below gave additional data on the regional motion of the protein. For the four complexes, fluctuation was nearly entirely confined to the surface-exposed termini and loops, and core residues lining up the binding tunnel indicated minimal fluctuation. For the control (Fig. 5d) and MolPort-002–507-186 (Fig. 5b), motion was suppressed across the entirety of the data set, supporting the structural preservation interpretation. On the other hand, MolPort-039–052-621 and MolPort-001–768-161 (Fig. 5a&c), evoked larger fluctuations at some of the loop areas, and poor ligand retention might have led to localized instability.

Fig. 5.

Residue-wise RMSF of PDE6D during 500 ns MD simulation for all ligand-bound systems: a MolPort-039–052-621, b MolPort-002–507-186, c MolPort-001–768-161, and d Control. Lower fluctuations in binding site residues suggest strong stabilizing interactions with the ligands.

In general, the dynamic analysis shows that MolPort-002–507-186 has the best dynamic profile, in ligand accommodation and protein rigidity, of the three ligands under investigation here. MolPort-039–052-621showed controlled flexibility, and MolPort-001–768-161 did not maintain a good binding mode at longer time scales and is potentially compromising lead compound status.

Intermolecular interaction analysis of final MD conformations

To identify the impact of dynamic relaxation on ligand placement and interaction profiles, end frames of the 500 ns MD trajectories were screened for intermolecular contacts. It was then possible to identify persistent interactions, stable over the length of the simulation and most likely supporting binding retention. The final snapshot of the 500 ns MD simulation was used for the analysis of protein-ligand interactions. However, I recognize this conformation may not represent the global minimum of the free energy surface. Interaction profiles were prepared both before and after MD simulations; thus, I could present interaction profiles both for the redocked complexes and post-MD snapshots. This allows an assessment of the persistence of interactions and stability of the binding mode throughout the simulation course.

The MolPort-039–052-621 (Fig. 6a) complex had a strongly conserved and complete binding network at the end of the simulation. Most importantly, the Glu88, Thr131, and Ser143 of the complex maintained direct hydrogen bonds with the ligand, holding it within the binding tunnel. There was a broad range of hydrophobic contacts, including residues of the entrance and core of the PDE6D pocket, including Trp90, Val49, Leu87, Ile109, Leu123, Gln78, and Val127. There were additional non-covalent contributions, including π-alkyl and π–π stacking interactions, maintained by residues of the type of Met20, Leu22, Ile129, and Val145, which reflect the compound’s strong anchoring effect.

Fig. 6.

2D interaction profiles of PDE6D with ligands at the last MD snapshot: a MolPort-039–052-621, b MolPort-002–507-186, c MolPort-001–768-161, and d Control. Key hydrogen bonds and hydrophobic contacts highlight sustained interactions post-equilibration.

In MolPort-002–507-186 (Fig. 6b), the ligand was observed forming hydrogen bonds via Gln116 and Ala112, and the ligand went on to form multiple hydrophobic contacts over a wide range of residues, including Leu54, Val59, Trp90, Ser115, Thr131, and Tyr149. Of particular significance was the fact that the interaction profile included both deeply buried residues (e.g., Ile129, Ala111) and more shallow ones (e.g., Trp32, Arg61), consistent with the ligand adopting a stabilized pose in an orientation available to contact both the tunnel floor and laterally located walls. Such interactions corresponded to the measured low RMSD of the complex, confirming a steady and specific binding orientation.

Whereas the MolPort-001–768-161 (Fig. 6c) complex had fewer persistent contacts at the last frame, no hydrogen bonds were discovered, and the hydrophobic interactions proved relatively localized at the tunnel exit, i.e., at Glu88, Leu123, Tyr149, and Val127. There were interior residues, including Val80, Ile129, and Trp90, with contacts, but these were fewer in quantity than in the remaining complexes. Such an interaction map, so sparing in character, most likely explains the ligand’s previous displacement or motion within the binding site, considering its greater tendency toward RMSD and RMSF during simulation.

End pose analysis of the control ligand showed relatively stable hydrogen bonding to Arg61 and Tyr149, and a large hydrophobic network of residues from around the tunnel, including Gln78, Val80, Phe82, Met118, and Leu147. Contact was present between the control (Fig. 6d) and both surface residues at the entrances (e.g., Leu17, Met20) and deeply buried hydrophobic residues (e.g., Ala111, Ile129), validating its position as a moderately stable binder of broad distribution throughout the PDE6D cavity.

In general, this analysis confirms that MolPort-039–052-621 and MolPort-002–507-186 retained a robust and durable interaction profile through long-term dynamic evolution, confirming their status as top candidate molecules. By contrast, MolPort-001–768-161 demonstrated lower contact persistence, suggesting possible egress or destabilization over the course of the simulation time course.

MMGBSA analysis

The various components of the MM/GBSA binding free energy as calculated by MMPBSA.py for the three MolPort compounds and the control ligand are summarized in Table 1. Complex stabilization was strong for MolPort-039–052-621, driven by favorable van der Waals and electrostatic interactions, thus returning an overall binding free energy, ΔG_total = − 75.93 ± 16.53 kcal/mol, comparable with the control. For MolPort-002–507-186, a balanced contribution from hydrophobic interactions and nonpolar solvation yielded a stable binding free energy, ΔG_total = − 75.27 ± 9.37 kcal/mol, despite its relatively weaker electrostatic interactions. On the other hand, MolPort-001–768-161 demonstrated lower gas-phase interaction energies, which translated into a reduced binding affinity, ΔG_total = − 50.69 ± 7.41 kcal/mol. The control ligand demonstrated the most favorable binding free energy, ΔG_total = − 82.30 ± 12.16 kcal/mol, which was supported by strong van der Waals and electrostatic contributions. Overall, from Table 1, van der Waal interactions emerge as the dominant force that governs complex stabilization, with MolPort-039–052-621 and MolPort-002–507-186 demonstrating binding affinities comparable to the control.

Table 1.

MMGBSA analysis of top selected compounds.

Energy components/Complexes	MolPort-039–052-621	MolPort-002–507-186	MolPort-001–768-161	Control
Van der Waal energy (ΔVDWAALS)	−69.73 ± 3.79	−52.85 ± 2.52	−36.52 ± 2.80	−63.78 ± 2.75
Electrostatic energy(ΔEEL)	−15.32 ± 6.80	−8.31 ± 3.19	−4.30 ± 1.50	−20.16 ± 5.06
Polar solvation energy (ΔEGB)	29.81 ± 4.09	16.79 ± 1.75	14.55 ± 1.26	30.39 ± 2.91
Non-polar solvation energy (ΔESURF)	−20.69 ± 1.84	−30.90 ± 1.90	−24.42 ± 1.83	−28.75 ± 1.43
Net gas phase energy (ΔGGAS)	−85.05 ± 10.59	−61.17 ± 5.71	−40.83 ± 4.30	−83.95 ± 7.81
Net solvation energy (ΔGSOLV)	9.12 ± 5.94	−14.10 ± 3.66	−9.86 ± 3.10	1.64 ± 4.34
ΔGtotal	−75.93 ± 16.53	−75.27 ± 9.37	−50.69 ± 7.41	−82.30 ± 12.16

Although docking identified MolPort 039 052 621 as the top candidate based on static binding poses, MMGBSA analysis performed on MD relaxed complexes indicated that the control ligand exhibited a more favorable binding free energy. This difference reflects the distinct nature of the two approaches, as docking primarily evaluates geometric fit using largely rigid receptor models, whereas MMGBSA accounts for protein flexibility, solvent contributions, and time averaged interaction stability derived from molecular dynamics simulations. Consequently, MMGBSA values are interpreted as indicators of post binding energetic stability rather than initial docking affinity. A comparative summary of docking scores, MMGBSA binding free energies, and qualitative QM/MM stabilization trends for all complexes is provided in Supplementary Table S3 to facilitate direct comparison of all energetic metrics. MMGBSA calculations were carried out using equilibrated frames from the production trajectory, and the reported mean ± standard deviation represents frame wise statistical variation. While independent replicate simulations were used to assess overall dynamic stability of the complexes, MMGBSA analysis was performed on representative trajectories for comparative evaluation of binding free energy trends.

Principal component analysis (PCA) of protein–ligand dynamics

To define the dominant motions governing protein-ligand motion within the 500 ns simulation, principal component analysis (PCA) was undertaken through Bio3D within R. PC1 and PC2, the principal components, were compressed and utilized for characterizing the collective displacements of atoms of each of the complexes, and eigenvalue contribution plots were generated to assess the distribution of variance. As shown in Fig. 6, PCA was performed on Cα atoms of the entire protein to assess global conformational stability, while binding pocket dynamics were evaluated separately using RMSF and interaction persistence analyses of tunnel lining residues. PCA projections revealed typical motion clusters for every ligand-bound system, which would suggest each compound triggered different conformational rearrangements of the PDE6D protein. The use of PCA to interpret dominant collective motions and ligand-induced conformational changes in protein-ligand systems is well established, and similar PCA-driven analyses have been successfully applied for long-timescale molecular dynamics simulations to elucidate functionally relevant dynamics and conformational transitions^42,43. For the MolPort-039–052-621 complex (Fig. 7a), the conformation space was clustered predominantly with moderate spreading, which would suggest a distribution of global protein motion symmetrical around the protein’s structure. In addition, projections along PC2 and PC3 showed relatively limited conformational spread, indicating restricted large-scale transitions beyond the dominant PC1 motion. This suggests that the MolPort-039–052-621 complex exhibits comparatively rigid dynamic behaviour with constrained collective fluctuations. Red and blue cluster distance refers to the ligand locking the protein into two dominant conformational basins, and it corresponds to an intermediate level of plasticity with sharp transitions between conformational substates. MolPort-002–507-186 complex (Fig. 7b) exhibited a broader distribution over PC1 and PC2, suggesting broader sampling of the conformation landscape. Interestingly, the discovery of two extremely populated but disconnected basins would imply significant ligand-induced flexibility, arguably linked to allosteric motion or binding pocket rearrangement. It aligns with its relatively stable RMSD behaviour and preservation of intermolecular interactions. In the MolPort-001–768-161 complex (Fig. 7c), the PCA was characterized by a more linear and restrictive trajectory path, which suggests the system was constrained into a smaller conformational channel. Such restriction can be accounted for by decreased protein or ligand slippage, and limited protein plasticity, both of which relate to its relatively smaller interaction number and larger fluctuations. Such a confined energy valley points towards small global transitions and limited degrees of freedom throughout the simulation. Control complex (Fig. 7d) exhibited moderate variance distribution between PC1 and PC2, producing overlapping groups characteristic of smoothly varying dynamical behavior with occasional jumps. It implies a relatively stable but moderately flexible complex structure, consistent with the binding energy and hydrogen bond data. Scree plots of eigenvalues of each system show that the first three components accounted for together a large percentage of total variance (between ~ 65% and ~ 73%), suggesting that the major global motions of the proteins are sufficiently captured by the first principal components. These data suggest that MolPort-039–052-621 and MolPort-002–507-186 exhibit favourable modulation of the dynamic behavior of PDE6D, stabilizing productive but distinct conformational states. Conversely, MolPort-001–768-161 exhibited limited exploration, which may be associated with suboptimally stable interaction or lack of conformational plasticity.

Fig. 7.

PCA plots depict time-evolved conformational sampling of protein–ligand complexes, with color gradients indicating structural transitions. Scree plots confirm that dominant motions are captured by the first few principal components across all systems. Panels: a MolPort-039–052-621, b MolPort-002–507-186, c MolPort-001–768-161, and d Control.

Dynamic cross-correlation analysis of residue motions

To explore the influence of ligand binding on the overall motion of the protein PDE6D, dynamic cross-correlation matrix (DCCM) analysis for each of the complexes was done at the level of the 500 ns MD simulation. The correlation coefficients of the residue pair atomic displacements were computed and presented, so positive (tending toward + 1.0) values indicate cooperative motion, and those taking negative values (tending toward − 1.0) indicate anti-correlated motion. Corresponding heatmaps of the above analysis are provided in Fig. 8. The MolPort-039–052-621 complex (Fig. 7a) had large positive correlation areas across the diagonal and off-diagonal sections and strongly coordinated intra-domain motions. There were several residue clusters of strong cooperative motion, particularly between residues 25–45 and 110–130, and thus suggest long-range communication between domains is enabled through this ligand. The MolPort-002–507-186 system (Fig. 8b) showed more dispersed but significant correlation signals, demonstrating dynamic couplings between otherwise distant components. Of note was the increase of crosstalk between helical components of the N-terminal domain and loop segments around the C-terminal segment. Such group fluctuations would be beneficial for reshaping the binding site or structural transitions of the protein, which would be ligand driven. For the MolPort-001–768-161 structure (Fig. 8c), the matrix showed moderate correlation patterns, smaller strong positive areas, and rising uncorrelated or neutral regions, revealing a relatively rigid complex conformation, and reduced cooperative plasticity. It can be a sign of a rigid binding mode, which can compromise allosteric modulation or adaptive binding efficiency. Control complex (Fig. 8d) had an even split between positive and negatively correlated domains. While some of the localized residue pairs exhibit correlations of motion typical of ligand-bound structures, overall intensity and correlation ranges were decreased. It would thus indicate ligand binding places a higher level of coordinated motion between residues, which may lock into place functionally important dynamic networks. These DCCM analyses collectively verify that each ligand differentially regulates the motion patterns of internal motion of PDE6D. MolPort-039–052-621 and MolPort-002–507-186 displayed notable coordinated dynamics, most likely supporting favorable inhibition conformational states, and the remaining molecules showed reduced dynamic coupling.

Fig. 8.

Dynamic cross-correlation matrices display correlated and anti-correlated motions within PDE6D for each ligand: a MolPort-039–052-621, b MolPort-002–507-186, c MolPort-001–768-161, and d Control.

Principal component analysis-based free energy landscape (FEL)

To uncover the conformational heterogeneity and thermodynamic stability of the complexes of ligands with PDE6D, principal component analysis (PCA) was utilized to extract the principal motions of the protein backbone, and then the free energy landscape (FEL) was constructed based on the first two principal components (PC1 and PC2). FEL plots were generated by PyMOL’s Geo Measures plugin, and the energy basins correspond to the most probable conformational states sampled during the 500 ns simulation.

As described in Figure S5, the FEL of the MolPort-039–052-621 complex (Figure S5a) was characterized by a sharp and narrow energy minimum and steep neighboring ridges of energy. This would thus suggest that the system overwhelmingly resides in a single, ultra-stable conformation and shows few transitions among the alternate substates. The basin is narrow and deep, and this hints at tight conformational confinement, most likely due to favorable protein-ligand contacts. In contrast, the MolPort-002–507-186 complex (Figure S5b) had a shallower and broader energy basin and a more diffuse landscape. Such a profile is typical of enhanced plasticity and sampling of a number of low-energy forms. Such motions may give adaptive binding benefits but may instead reflect lower overall bound-state rigidity. For the MolPort-001–768-161 system (Figure S5c), two seeming minima within the FEL existed, and there might be at least two energetically favorable conformational substates. Interconversion of metastable forms on the time scale of simulation would be implied by the distance between these wells, and this may be connected with dynamic ligand-driven rearrangements. Control complex (Figure S5d) exhibited a relatively shallower free energy well of moderate curvature and gradient. There was no significant obstacle nor localized minima, which indicated perhaps a more plastic and less stabilized binding structure than lead candidate ligands. It is consistent with the general reduced stability and persistence of interaction observed in other analyses. At the same time, these FEL maps define the varied dynamic profiles of ligand-attached complexes. MolPort-039–052-621 confines the protein in a predominant low-energy state, in accordance with tight and strong binding. MolPort-002–507-186 and MolPort-001–768-161 possess broader or multi-basin topographies, consistent with flexible motion of interaction. By contrast, the control exhibits weaker conformational bias, and this highlights the enthalpic advantages of the natural ligands selected in binding PDE6D against conformational flux associated with oncogenic trafficking.

Comparison of conformations between minima structures and initial redocked positions

To calculate the extent of the conformation deviation between the MD-refined lowest energy conformations (Figure S6) and their original docked poses, structural superimposition analysis was carried out through PyMOL. By interpreting this analysis, the extent of structural rearrangement taking place over the course of the simulation, most particularly in the area of the binding pocket, can be deduced. For MolPort-039–052-621 (Fig. 9a), the RMSD value of the re-dock and post-MD minima pose was 1.239 Å, validating moderate but persistent ligand alignment throughout the simulation. For MolPort-002–507-186 (Fig. 9b), the superposed structures were at an RMSD of 1.202 Å, portraying almost continuous binding orientation and minimal deviation, hence suggesting persistent ligand-protein interactions. Of particular interest was MolPort-001–768-161 (Fig. 9c), which showed a slightly increased RMSD of 1.375 Å, and this could reflect increased intrinsic flexibility or minute displacements of the binding tunnel of PDE6D under dynamic sampling conditions. It was observed that the lead compound had an RMSD of 1.250 Å, which was in an analogous range, and hence confirmed the dynamic similarity of the obtained hits. Graphical overlay of each of the poses verifies the fact that all of the ligands maintained their principal binding geometries with small positional changes, supporting their binding stability and applicability as future PDE6D inhibitors.

Fig. 9.

Superimposition of Initial Docked Structures with First Free Energy Landscape (FEL) Minima. Structural alignment of the initial docked poses with their corresponding FEL-derived first minima for a MolPort-039–052-621, b MolPort-002–507-186, c MolPort-001–768-161, and d the control ligand.

QM/MM energy assessment of first minima structures

Following rigorous 500 ns molecular dynamics simulations and identification of thermodynamically favourable conformations through application of the free energy landscape (FEL) analysis, the initial minimum structure of each protein–ligand complex was addressed by hybrid quantum mechanics/molecular mechanics (QM/MM) calculations. It was performed to evaluate ligand electronic stabilization within the electrostatic field of the binding pocket, while the surrounding protein and solvent were treated classically to provide environmental context rather than explicit quantum mechanical protein–ligand interaction energies. Under the QM/MM setup, ligands were positioned in the quantum mechanical (QM) region, and protein residues and the solvent environment were computed under molecular mechanics (MM) calculations. Three-dimensional division of the QM and MM spaces (Fig. 10a–d) shows spatially encapsulated positioning of ligands at the protein binding pocket, where red dots indicate the QM space and blue dots indicate the MM space. This partitioning allows assessment of ligand electronic structure in the presence of the surrounding protein environment.

Fig. 10.

3D representation of QM/MM regions showing ligands a MolPort-039–052-621, b MolPort-002–507-186, c MolPort-001–768-161, and d the control (QM, red) embedded within PDE6D (MM, blue). Accurate boundary selection ensures valid quantum treatment of active sites.

The QM/MM energies obtained in this study are interpreted only as qualitative indicators of electronic stabilization within the PDE6D environment rather than as quantitative protein–ligand interaction or binding energies, since absolute electronic energies inherently include large baseline system contributions. Since binding site residues were not included in the QM region, QM/MM results were not used to infer residue level electronic interactions or binding energies, and were instead employed only to assess ligand electronic compatibility within the tunnel environment. Within this qualitative framework, MolPort-039–052-621 (–1910.39 Ha), MolPort-002–507-186 (–1280.11 Ha), and the control ligand (–1339.98 Ha) exhibited comparable electronic stabilization trends, whereas MolPort-001–768-161 (–769.22 Ha) showed relatively weaker stabilization. These observations are therefore used only to support electronic compatibility with the tunnel environment and not to establish binding strength rankings (Fig. 10).

As a supplementary analysis, Density Functional Theory (DFT) calculations were performed to approximate frontier orbital energies. As Fig. 11 demonstrates, there were narrowly spaced HOMO–LUMO levels for each ligand, characteristic of desirable electronic structures and a tendency toward electron delocalization within the tunnel-like PDE6D pocket. These electronic descriptors are therefore used only as supportive qualitative indicators and not as direct measures of interaction strength within the protein pocket.

Fig. 11.

HOMO and LUMO levels of the four ligands a MolPort-039–052-621, b MolPort-002–507-186, c MolPort-001–768-161, and ) the control ligand, showing their electronic stability and potential reactivity in the PDE6D binding tunnel. All four molecules exhibit orbitals.

All these hybrid energy and quantum accounts conclude that MolPort-039–052-621 possesses the most energetically and electronically favourable inhibition profile of PDE6D and is, therefore, the most promising candidate for disrupting RAS trafficking through occupancy of the tunnel.

Discussion

The present study was designed as a comprehensive computational investigation aimed at identifying and mechanistically characterizing natural product derived binders of PDE6D, rather than establishing biological inhibition of the PDE6D RAS interaction. Through the integration of molecular docking, density functional theory-based ligand optimization, extended molecular dynamics simulations, PCA based free energy landscape analysis, MMGBSA binding free energy estimation, and QM/MM evaluation, this work provides a multiscale in silico framework for prioritizing compounds capable of occupying the prenyl binding tunnel of PDE6D.

In recent computational drug discovery studies, the combination of docking with long timescale molecular dynamics simulations and post docking energetic refinement has been widely adopted to better account for protein flexibility and solvent effects, particularly for targets involving dynamic binding pockets. Very recently, medicinal chemistry efforts targeting PDE6D have emphasized the importance of a precise mode of engagement of the hydrophobic prenyl-binding tunnel to disrupt RAS trafficking while maintaining selectivity and minimizing cytotoxicity. Thus, next-generation PDE6D inhibitor development has been described to underscore that ligand stability within the tunnel, balanced hydrophobic-polar interactions, and conformational persistence during dynamic simulations are critical determinants of efficacy. These features consistently emerge for the top-ranked natural compounds identified herein to support their relevance as promising starting points for further optimization and experimental validation^44,45.

One of the most closely related recent works is by Majrashi et al. (2024), who evaluated a series of K-Ras inhibitors (e.g., deltarazine, deltaflexin-1, deltaflexin-2) against PDEδ by docking, molecular dynamics, and other methods. Their docking of a few of the compounds was in ranges similar to our pre-DFT docking (≈ − 10 to − 12 kcal/mol), and their MD simulations of lead hits confirmed stability⁴⁶. Another closely related example is the work (2025) on natural product inhibitors of PDE6D derived from marine fungi, computationally screening marine natural products for inhibitors of PDE6D; many screened natural product hits had good binding in docking and good dynamics⁴⁷. Also, Kaya et al.‘s work of generating the “Deltaflexin3 + Sildenafil” PDE6D inhibitor of good intracellular potency and solubility advanced beyond purely computer predictions by adding biochemical assays. Their design overcame prenyl binding tunnel accommodation issues and ligand solubility issues⁴⁸. By comparison, several fronts were advanced in this work:

The HOMO and LUMO isosurfaces describe the electronic distribution and potential reactivity of each ligand within the PDE6D binding tunnel. Particularly, the positioning of the HOMO on electron-donating groups, such as hydroxyl and the LUMO on electron-deficient regions supports the formation of important hydrogen bonds and π-π stacking interactions in docking and MD simulations. For instance, MolPort-039–052-621 displayed HOMO density near polar substituents involved in the formation of stable hydrogen bonds with Glu88 and Ser143, while its LUMO distribution was found in good superposition with tunnel-lining residues, suggesting preferred electron-accepting capacity. In fact, such quantum mechanical features, besides explaining the observed binding modes, support the ability for charge transfer interactions that stabilize protein-ligand complexes. Data of this type guides drug design by highlighting electron-rich or deficient regions to be modified for enhancing the binding affinity or specificity toward the PDE6D pocket. Better docking energies before refinement: Our lead compound (MolPort 039 052 621) had a docking energy of 13.3 kcal/mol, which was better than most of the marine natural product leads, and comparable to or slightly better than some of the deltaflexin/deltarazine leads before refinement. It thus follows that our virtual screening library and prefiltering steps were good at sampling the chemical space of high affinities.

Post-quantum refinement consistency and dynamics: Most previous reports talk of reduced docking score or pose following MD or energy minimisation; a few hit compounds suffer loss of significant interactions. Here, though the redocking scores after optimisation at the level of DFT were generally poorer than initial screening (e.g., MolPort 002 507 186 dropped to 9.6 kcal/mol), they were robust, and structural analyses (RMSD, interaction networks) proved retention of significant contacts.

Energetic and electronic profiling (QM/MM): While most studies conclude with docking and MD, this work extended the analysis to include QM/MM first-minima structure energies, providing additional mechanistic insight. While Majrashi et al. included MM/GBSA values, most studies have not reported hybrid QM/MM energetics. In this work, MolPort-039–052-621 was distinguished from the control and other competitors by its strongly negative QM/MM total energy (≈ 1910.39 Hartree), providing fine-grained energetic resolution. Free energy landscape observations: From the FEL analysis, a glimpse of the confinement of MolPort-039–052-621 into a narrow energy basin, typical of rigid binding, was obtained, whereas some of the other hits tried multiple basins. This is seen in the marine natural product work, wherein multiple minima corresponded to weaker binding⁴⁷. However, extension of the QM region to include key tunnel lining residues may provide additional insight into residue specific electronic contributions and could be explored in future studies.

Differences in ligand ranking observed across docking, MMGBSA, and QM/MM analyses highlight the complementary nature of these computational approaches. Docking primarily evaluates static geometric fit and interaction potential, MMGBSA estimates binding stability from dynamically equilibrated complexes, and QM/MM provides qualitative insight into electronic compatibility within the binding environment. In the present study, MolPort 039 052 621 exhibited strong docking scores, persistent intermolecular interactions, and compact free energy basins, whereas the control ligand showed more favorable MMGBSA free energy values. These observations suggest that while initial docking affinity favors MolPort 039 052 621, long term energetic stabilization in solution may favor the control ligand, emphasizing the importance of integrating multiple metrics rather than relying on a single scoring method.

However, our work, as with others before, also shares some of the same flaws. Although docking energies hold promise, there were a few of the earlier studies (e.g. Deltaflexin3) that demonstrated computationally foresaid hits in in vitro or cell-based tests. Our predictions by computer, though intensive, continue to await experimental validation. A few of the compounds (e.g., MolPort-001–768-161), though there was proper docking pre-refinement, had higher ligand RMSD, weaker MD interactions, and worse QM/MM energetics, in line with the deltaflexin analogues losing potency or having poor in-cell activity⁴⁸. Overall, compared to the recent PDE6D computational inhibitors literature, these findings position MolPort 039 052 621 and MolPort 002 507 186 as very promising leads: favorable binding scores, retention of key interactions by MD, favorable free energy basins, and favorable QM/MM stabilization. Besides MolPort‑039‑052‑621, comparison of the other shortlisted ligands yields SAR relevant for the inhibition of PDE6D. MolPort‑002‑507‑186 was slightly higher in QM/MM energy while still maintaining robust hydrogen bonding and a compact tunnel-bound conformation, which would thus be expected to have favorable enthalpic contributions despite its broader conformational flexibility. In contrast, MolPort‑001‑768‑161 had lost persistent hydrogen bonds and showed higher RMSD and reduced tunnel retention, possibly related to a less polar surface and higher rotational freedom. These findings highlight the key role played by a judicious balance of rigidity, tunnel-fit, and polar contacts in yielding optimal PDE6D binding, where MolPort‑039‑052‑621 serves as a structurally optimized template within this particular SAR framework.

Natural products offer deep structural diversity and tend to possess favorable biological profiles; thus, these molecules emerge as promising candidates for the identification of new therapeutic templates. On the other hand, it should not be forgotten that there could exist certain issues regarding the potential of these natural products regarding the possibility of ‘off-target effects’ or ‘toxicity.’ To address the relevance of the results of this study, I investigated the available structural and chemical details of the shortlisted natural products. None of the shortlisted natural products, including MolPort-039–052-621, possess potentially toxic entities like reactive groups or motifs associated with ‘toxicity,’ and ‘non-specific cytotoxicities’ or common patterns of ‘substructural elements’ linked with these properties. Though the specific details regarding ‘in vitro’ cytotoxic profiles of these natural products do not lie under the purview of this computational analysis and theoretical simulation study, the lack of structural ‘toxic entities’ provides a well-grounded initial clue toward the possibility of their favorable ‘safety’ profiles that could redirect these candidates toward successful ‘biological evaluations’; hence, these details may generate more convincing arguments toward their therapeutic potential. ‘Specificity’ toward the target enzyme stands as yet another area of critical significance within the development of therapeutic agents; thus, certain prenyl-binding proteins share partially overlapping structural similarity with the target enzyme ‘PDE6D’s’ binding tunnel. Our results illustrate that the shortlisted ligands tend to target specific amino-acid residues that are significantly responsible toward defining the distinct shape and chemical character of ‘PDE6D’s’ binding cleft like tyrosyl-149, Trp-90, Ile-129, and Leu-87; thus, these observations reveal considerable ‘structural similarity’ toward the target enzyme that could distinguish them clearly from other ‘prenyl binding chaperones’; hence, despite the lack of clear-cut direct computational proofs regarding the ‘cross-reactivity’ of these candidates, these observations could well serve as an initial logical step toward developing well-grounded theoretical justification regarding the potential of these natural candidates toward therapeutic exploration that could reinterpret the discussion. Now that these aspects have been taken into account, the discussion part of this study could well offer an intermediate direct explanation for presenting the therapeutic potential of the selected natural products. Although the compounds identified in the present study have not been experimentally tested in vitro, integrating several computational layers increases the confidence in their prioritization. With due respect to the limitations of in silico drug discovery, I emphasize that experimental follow-up studies, including PDE6D binding assays and RAS trafficking disruption assays, must be executed to confirm the biological activity of these compounds. Nevertheless, the predictive power and mechanistic insight provided by this study represent a valuable hypothesis-generation platform for guiding such future experimental investigations.

Although this study describes a strong in silico strategy that integrates docking, DFT optimization, molecular dynamics, and QM/MM analysis, some limitations remain. First, predictive computational screening cannot yet fully capture complex biological behaviors such as intracellular bioavailability, metabolic stability, or off-target interactions. Binding affinities from both docking and QM/MM approximations, although rigorous, need biochemical validation through in vitro binding assays and cellular models to confirm functional impact on RAS trafficking. Further, this study focused on PDE6D but did not investigate selectivity across other prenyl-binding chaperones (e.g., RhoGDI, REP1), which often share structural homology and may impact specificity. Finally, pharmacokinetic and toxicological profiles of the shortlisted natural products remain uncharacterized and further experimental investigation will be required. These limitations are a reminder that the computational results obtained here have to be experimentally validated further downstream in order to become viable therapeutic leads.

Conclusion

This study employed an end-to-end in silico pipeline of structure-based virtual screening, a disciplinary approach of density functional theory (DFT) optimization, molecular dynamics simulations, PCA-based free energy landscape projection, and hybrid QM/MM energy calculations for computational prioritization of natural product scaffolds capable of occupying the prenyl binding tunnel of the phosphodiesterase delta (PDE6D) protein from natural products. Three ligands, MolPort-039–052-621, MolPort-002–507-186, and MolPort-001–768-161, emerged among screened natural product molecules as computationally prioritized binders of the PDE6D prenyl binding tunnel involved in the RAS trafficking pathway. Among these, MolPort-039–052-621 showed comparatively favorable stability and energetic trends across multiple computational analyses. MolPort-002–507-186 was additionally associated with a good energetic profile and structural convergence, suggestive of stable tunnel engagement and favorable noncovalent interaction patterns within the PDE6D binding environment. Superimposition analyses, moreover, further showed limited deviation of the poses from initial poses, confirming conformational fidelity after simulation. Of particular interest, HOMO–LUMO profiles of each ligand indicated electronic features consistent with potential stabilization within the binding environment, supporting qualitative assessment of ligand compatibility with the tunnel region. Moreover, the analysis was taken further by hybrid QM/MM energy calculations and frontier orbital analysis, thereby enhancing the strength of these in silico predictions. While experimental validation is required to establish biological activity, the present findings provide mechanistically informed hypotheses and structural templates that may guide future biochemical evaluation and medicinal chemistry optimization of PDE6D tunnel binders.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

The author confirms sole responsibility for the conception, design, analysing, drafting, and finalization of the manuscript. All aspects of the work were performed independently by the author.

Funding

The author declares that no funds, grants, or other support were received during the preparation of this manuscript.

Data availability

Data is provided within the manuscript or supplementary information files.

Declarations

Competing interests

The author declares no competing interests.