IntroductionCerebral cavernous malformations (CCMs) are low-flow vascular lesions of the central nervous system composed of dilated, endothelial-lined caverns that can remain clinically silent or present with seizures, focal neurological deficits, headache, or intracerebral hemorrhage1. Familial CCM is associated with pathogenic variants in KRIT1/CCM1, CCM2, and PDCD10/CCM3, whose protein products function as an endothelial signaling complex required for vascular stability and junctional integrity2. Loss of CCM signaling disrupts endothelial barrier regulation, with prominent involvement of RhoA–ROCK activity and endothelial gain of MEKK3–KLF2/4 signaling, pathways that have been implicated in CCM lesion formation and vascular dysfunction3,4.A central event in this pathway is recruitment of KRIT1 to endothelial junctions by heart of glass 1 (HEG1), a transmembrane protein that couples junctional membrane localization to CCM complex signaling5. Structural studies showed that the KRIT1 FERM domain binds the cytoplasmic tail of HEG1, with the HEG1 C-terminal Tyr-Phe motif occupying a hydrophobic groove formed at the KRIT1 F1/F3 interface4. KRIT1 can also bind Rap1 simultaneously with HEG1, supporting a model in which KRIT1 integrates membrane anchorage and small-GTPase signaling at endothelial cell–cell junctions6. Disruption of this interface therefore provides a focused strategy for probing early signaling events downstream of HEG1–KRIT1 complex remodeling, as summarized schematically in Fig. 17.Although protein–protein interactions are often difficult to modulate with small molecules8,9, the HEG1-binding groove of KRIT1 presents a structurally defined pocket that can be exploited for inhibitor discovery6. The small-molecule HEG1–KRIT1 inhibitor HKi2 was shown to compete orthosterically with HEG1 for binding to the KRIT1 FERM domain, occupy the same binding pocket as the HEG1 cytoplasmic tail, and form a reversible covalent interaction with KRIT1 Lys72010. Acute inhibition of the HEG1–KRIT1 interaction increased KLF4 and KLF2 mRNA and protein levels in endothelial cells and induced endothelial klf2a expression in zebrafish, establishing HKi compounds as pharmacological probes for interrogating HEG1–KRIT1-dependent signaling7. More recently, 2-hydroxy-1-naphthaldehyde (HNA) was characterized as a site-directing reversible covalent ligand of KRIT1 Lys720; HNA displayed selective KRIT1 engagement, prolonged residence time, and inhibition of the HEG1–KRIT1 interaction, while HNA derivatives showed improved target engagement and inhibitory activity11.In this study, we sought to identify new lead candidates that disrupt the HEG1–KRIT1 interaction using a focused virtual screen of 2-hydroxy-1-naphthaldehyde scaffolds against the KRIT1 HEG1-binding pocket. Selected candidates were further evaluated by 1000 ns molecular dynamics simulations to examine binding stability, persistence of key contacts, and conformational dynamics within the KRIT1 interface. Finally, computational predictions were experimentally tested using a bead-based flow-cytometry competition assay that measures disruption of KRIT1 FERM binding to HEG1-coated beads, providing an integrated structure-guided and experimental workflow for prioritizing HEG1–KRIT1 PPI disruptors.Fig. 1Full size imageProposed strategy for small-molecule disruption of the HEG1–KRIT1 protein–protein interaction. Under native endothelial junctional conditions, the cytoplasmic tail of HEG1 interacts with the FERM domain of KRIT1, supporting assembly of the CCM signaling complex with CCM2 and CCM3 and maintaining regulated downstream MEKK3–MEK5–ERK5–MEF2–KLF2/4 signaling. VE-cadherin is shown as a representative endothelial junctional component. Candidate 2-hydroxy-1-naphthaldehyde–derived small molecules identified by virtual screening are proposed to bind at or near the HEG1–KRIT1 interface, thereby competitively disrupting the interaction. Loss of HEG1–KRIT1 binding is predicted to alter CCM complex organization and dysregulate downstream signaling, resulting in increased KLF2/KLF4 transcriptional response. The inset illustrates the predicted binding of a small-molecule disruptor at the HEG1 cytoplasmic tail–KRIT1 FERM-domain interface.Materials and methodsVirtual screeningA focused virtual-screening workflow was applied to identify candidate small-molecule disruptors of the HEG1–KRIT1 protein–protein interaction using a starting library of 1,121 compounds containing a 2-hydroxy-1-naphthaldehyde scaffold. Compounds were first filtered for drug-like physicochemical properties, including molecular weight, calculated logP/logD, topological polar surface area, hydrogen-bond donor and acceptor counts, and rotatable bonds, followed by exclusion of compounds containing PAINS or potentially reactive substructures12,13. The remaining compounds were subjected to a diversity and purchasability gate to prioritize chemically tractable and experimentally accessible candidates. Docking was then performed with AutoDock Vina using multiple independent runs and binding poses, and compounds were ranked based on predicted binding affinity and pose consistency at or near the predicted HEG1 cytoplasmic tail–KRIT1 FERM-domain interface14. Top-ranked compounds were further refined by short molecular dynamics simulations to assess pose stability, ligand RMSD, and contact occupancy, followed by MM/PBSA-based rescoring. This hierarchical workflow yielded three prioritized candidates for experimental evaluation: HN1, HN2, and HN3 (Figure S3).Molecular docking analysisLigand preparationThe three prioritized 2-hydroxy-1-naphthaldehyde–based candidates, HN1–HN3, were built as three-dimensional structures and energy-minimized using Gaussian 16, Revision C.01. Ligand geometries were refined with density functional theory at the B3LYP/6-31G(d, p) level to generate low-energy conformations suitable for docking15,16. Frequency analyses were subsequently performed to verify that the optimized structures corresponded to stable minima, and only structures without imaginary frequencies were advanced15,16. The minimized ligands were then imported into AutoDockTools for docking preparation, during which polar hydrogens were introduced, Gasteiger charges were assigned, and torsional degrees of freedom were specified to permit ligand flexibility17,18. The prepared ligands were saved as PDBQT files and used for AutoDock Vina docking against the HEG1/HKi2-binding pocket within the KRIT1 FERM domain14.Receptor preparationThe docking receptor was generated from the KRIT1-containing crystal structure deposited as PDB ID 6OQ3, a 1.85 Å X-ray structure of KRIT1 in complex with Rap1 GTPase and the HEG1–KRIT1 inhibitor HKi27. Before docking, the KRIT1 structure was processed in AutoDockTools by deleting crystallographic waters and other nonessential heteroatoms or non-receptor components, followed by addition of polar hydrogen atoms and assignment of docking-compatible charge parameters. The final receptor model was exported in PDBQT format14,17,19. Docking calculations were restricted to the HEG1/HKi2-recognition region of the KRIT1 FERM domain by defining a cubic search space centered at x = 85.08, y = 18.85, and z = 10.88 Å, with box dimensions of 20 × 20 × 20 Å along the x, y, and z axes.Molecular dockingMolecular docking was performed with AutoDock Vina to predict the binding orientation and relative affinity of the selected 2-hydroxy-1-naphthaldehyde derivatives within the HEG1/HKi2-recognition site of the KRIT1 FERM domain14. The prepared KRIT1 receptor and ligand PDBQT files were used as docking inputs, and the search space was defined around the predicted HEG1–KRIT1 interface using the grid parameters described above. For each compound, multiple docking runs were carried out to improve sampling of possible ligand orientations within the binding region. The resulting poses were evaluated according to Vina-predicted binding energy, spatial convergence among repeated docking poses, and the ability of each compound to occupy the HEG1/HKi2-interacting pocket. Docked conformations showing consistent placement at the target interface and favorable predicted binding scores were selected for further inspection. Protein–ligand interactions, including hydrogen bonding, hydrophobic contacts, π-associated interactions, and contacts with residues located near the HEG1-binding surface, were analyzed to identify compounds most likely to interfere with the HEG1–KRIT1 interaction. The best-ranked and most reproducible docking poses of HN1, HN2, and HN3 were subsequently used as starting structures for molecular dynamics simulation.Docking validationThe docking protocol was validated by a self-docking approach using the co-crystallized HEG1–KRIT1 inhibitor HKi2 present in the KRIT1 crystal structure (Figure S1). HKi2 was removed from the receptor model, prepared using the same ligand-preparation procedure described above, and then redocked into the HEG1/HKi2-recognition site of the KRIT1 FERM domain using the same grid-box coordinates and docking parameters applied to HN1–HN3. The top-ranked redocked pose of HKi2 was superimposed onto its crystallographic binding orientation, and the heavy-atom root-mean-square deviation was calculated to assess whether the docking workflow could reproduce the experimentally observed ligand position. The docking setup was considered reliable when the redocked HKi2 pose showed low deviation from the crystal pose and preserved the key binding-site contacts within the HEG1-recognition pocket. This validation step confirmed that the selected grid space and docking parameters were suitable for evaluating the binding modes of the 2-hydroxy-1-naphthaldehyde candidates at the KRIT1 interface.Molecular dynamics simulationsAll-atom molecular dynamics simulations were conducted for the KRIT1–HN1, KRIT1–HN2, and KRIT1–HN3 complexes using GROMACS 2022.4 with the CHARMM36 biomolecular force field and the CHARMM-compatible TIP3P explicit water model20,21,22. The initial coordinates for each simulation system were taken from the most favorable and reproducible docking pose obtained for each 2-hydroxy-1-naphthaldehyde derivative within the HEG1/HKi2-recognition region of the KRIT1 FERM domain. Protein parameters were assigned according to the CHARMM36 force field, whereas ligand parameterization was performed using the CHARMM General Force Field through CGenFF, followed by conversion into GROMACS-readable topology files23. Each KRIT1–ligand complex was centered in a periodic dodecahedral simulation box with a final volume of 930.11 nm³, solvated with CHARMM-modified TIP3P water, and charge-neutralized by the addition of two chloride ions. Prior to equilibration, each system was subjected to steepest-descent energy minimization until the maximum force was reduced below 1000 kJ mol⁻¹ nm⁻¹24. The minimized systems were equilibrated under position restraints in sequential canonical and isothermal–isobaric ensembles, consisting of 100 ps NVT equilibration followed by 100 ps NPT equilibration. The simulation temperature was controlled at 300 K using the velocity-rescaling thermostat, while the pressure was maintained at 1 bar with the Parrinello–Rahman barostat. Electrostatic interactions beyond the short-range cutoff were calculated using the particle mesh Ewald scheme, and both Coulombic and van der Waals short-range interactions were treated with a 1.2 nm cutoff25. Bonds involving hydrogen atoms were constrained with the LINCS algorithm, permitting a 2 fs integration time step throughout the simulations. After equilibration, each KRIT1–ligand system was simulated without restraints for 1000 ns. Trajectories were subsequently processed to remove periodic-boundary artifacts and analyzed to characterize complex stability, ligand retention, and conformational behavior, including PPI disruption metrics, interface analysis, protein backbone RMSD, residue-wise RMSF, radius of gyration, ligand RMSD, hydrogen-bond occupancy, and persistence of contacts between each compound and residues within the HEG1-binding surface of KRIT1.MM/PBSA free-energy calculationsTo complement the docking analysis and long-timescale MD simulations, endpoint binding free-energy analyses were carried out using gmx_MMPBSA v1.6.3 on equilibrated portions of the simulation trajectories26,27,28. The calculations were performed to compare the relative energetic favorability of the selected 2-hydroxy-1-naphthaldehyde candidates, HN1–HN3, within the HEG1-recognition region of the KRIT1 FERM domain and to examine whether ligand binding was predicted to alter the stability of the HEG1–KRIT1 interface. For each system, snapshots were collected from the last 200 ns of the 1000 ns production run, corresponding to the 800–1000 ns trajectory window, and the resulting energy values were expressed as the mean ± standard deviation across the analyzed frames. The binding free energy was estimated according to the MM/PBSA framework:$$\:{\Delta\:}{G}_{\text{b}\text{i}\text{n}\text{d}}={\Delta\:}{E}_{\text{M}\text{M}}+{\Delta\:}{G}_{\text{s}\text{o}\text{l}\text{v}}-T{\Delta\:}S$$where \(\:{\Delta\:}{E}_{\text{M}\text{M}}\)includes the gas-phase molecular mechanics contributions, mainly electrostatic and van der Waals interactions, and \(\:{\Delta\:}{G}_{\text{s}\text{o}\text{l}\text{v}}\)represents the solvation-energy term. The polar component of solvation was determined using the Poisson–Boltzmann continuum model, whereas the nonpolar contribution was approximated from the solvent-accessible surface area29. The entropic term \(\:\left(T{\Delta\:}S\right)\)was not included unless specifically indicated; therefore, the calculated MM/PBSA values were used for relative comparison among the tested compounds rather than as absolute experimental binding affinities.Ligand-dependent modulation of the HEG1–KRIT1 interface \(\:\left(\varvec{\Delta\:}\varvec{\Delta\:}{\varvec{G}}_{\mathbf{P}\mathbf{P}\mathbf{I}}\right)\)To determine whether each compound was energetically associated with disruption or stabilization of the HEG1–KRIT1 complex, the protein–protein binding energy in the ligand-bound state was compared with that of the corresponding ligand-free reference system. The ligand-induced change in interfacial binding energy was calculated as:$$\:{\Delta\:}{\Delta\:}{G}_{\text{P}\text{P}\text{I}}={\Delta\:}{G}_{\text{P}\text{P}\text{I}}^{\text{l}\text{i}\text{g}\text{a}\text{n}\text{d}-\text{b}\text{o}\text{u}\text{n}\text{d}}-{\Delta\:}{G}_{\text{P}\text{P}\text{I}}^{\text{a}\text{p}\text{o}}$$or equivalently:$$\:{\Delta\:}{\Delta\:}{G}_{\text{P}\text{P}\text{I}}={\Delta\:}{G}_{\text{b}\text{i}\text{n}\text{d}}\left(\text{H}\text{E}\text{G}1\text{\--}\text{K}\text{R}\text{I}\text{T}1\:\text{w}\text{i}\text{t}\text{h}\:\text{l}\text{i}\text{g}\text{a}\text{n}\text{d}\right)-{\Delta\:}{G}_{\text{b}\text{i}\text{n}\text{d}}\left(\text{H}\text{E}\text{G}1\text{\--}\text{K}\text{R}\text{I}\text{T}1\:\text{a}\text{p}\text{o}\right)$$For the ligand-bound calculations, trajectory frames were taken from HEG1–KRIT1 complexes containing HN1, HN2, or HN3, whereas the apo reference values were obtained from the matching HEG1–KRIT1 system in the absence of ligand using identical MM/PBSA settings. This approach allowed the energetic consequence of ligand occupancy near the KRIT1 HEG1-binding region to be assessed directly. In this analysis, positive \(\:{\Delta\:}{\Delta\:}{G}_{\text{P}\text{P}\text{I}}\)values were interpreted as a reduction in HEG1–KRIT1 interfacial stability, consistent with a predicted disruptive effect, while negative values indicated relative stabilization of the protein–protein complex compared with the apo state.MM/PBSA analysis of compound binding to KRIT1In parallel, MM/PBSA calculations were performed for each KRIT1–ligand trajectory to estimate the relative binding strength of HN1, HN2, and HN3 toward the KRIT1 FERM domain. Protein–ligand binding free energy was calculated using:$$\:{\Delta\:}{G}_{\text{b}\text{i}\text{n}\text{d}}^{\text{p}\text{r}\text{o}\text{t}\text{e}\text{i}\text{n}-\text{l}\text{i}\text{g}\text{a}\text{n}\text{d}}={G}_{\text{K}\text{R}\text{I}\text{T}1-\text{l}\text{i}\text{g}\text{a}\text{n}\text{d}}-\left({G}_{\text{K}\text{R}\text{I}\text{T}1}+{G}_{\text{l}\text{i}\text{g}\text{a}\text{n}\text{d}}\right)$$The energy values obtained from the 800–1000 ns portion of each trajectory were averaged to provide a comparative estimate of ligand stability within the KRIT1 binding region. These MM/PBSA-derived results were evaluated together with the docking scores and MD trajectory analyses, including protein backbone RMSD, residue flexibility, ligand RMSD, hydrogen-bond persistence, and maintenance of ligand contacts within the HEG1-binding surface. This integrated assessment was used to prioritize compounds that not only remained stably associated with KRIT1 but also showed a predicted capacity to weaken the HEG1–KRIT1 protein–protein interaction.MM/PBSA values were treated as approximate relative estimates rather than quantitatively precise free energies. The reported standard deviations describe frame-to-frame variability within the analyzed trajectory window and do not constitute formal confidence intervals for differences between independently simulated systems. Because trajectory frames are temporally correlated and independent replicate free-energy calculations were not performed, no statistical significance was assigned to the calculated ΔΔGPPI values.Free energy landscape analysisFree energy landscape (FEL) analysis was performed to characterize the dominant conformational states sampled by the KRIT1–HN1, KRIT1–HN2, and KRIT1–HN3 complexes during the 1000 ns production simulations. Principal component analysis (PCA) was first applied to the MD trajectories using the covariance-analysis utilities implemented in GROMACS. Prior to PCA, each trajectory was aligned to a reference structure by least-squares fitting to eliminate overall translational and rotational movement, allowing the analysis to focus on internal structural motions of the KRIT1–ligand complexes. A covariance matrix was then generated from the positional fluctuations of the selected atoms and diagonalized to obtain eigenvectors and eigenvalues. The eigenvectors defined the major collective motions sampled during the simulations, while the eigenvalues reflected the magnitude of motion along each principal direction. Each trajectory was projected onto the first two principal components, PC1 and PC2, which captured the main conformational variations explored by each ligand-bound KRIT1 system. These projections were used to construct two-dimensional conformational maps representing the sampled structural space of each complex.The FELs were calculated from the PC1 and PC2 probability distributions using the SHAM approach in GROMACS. In this method, the relative Gibbs free energy is estimated from the population density of conformational states through Boltzmann inversion, according to:\(\:G(PC1,PC2)=-{k}_{B}T\text{l}\text{n}P(PC1,PC2)\)where \(\:{k}_{B}\)is the Boltzmann constant, \(\:T\)is the simulation temperature, and \(\:P(PC1,PC2)\)represents the probability of observing a given conformational state within the PC1–PC2 subspace. The resulting FEL maps were examined to identify energetically favorable basins, intermediate or metastable states, and transition regions associated with each KRIT1–ligand complex. The GROMACS-generated FEL files were exported in XPM format and converted into numerical data using an in-house Python script, enabling consistent visualization and comparison of the conformational energy profiles of HN1–HN3 bound to the HEG1-recognition region of KRIT1.In silico ADMET profilingThe pharmacokinetic and toxicity-related properties of the prioritized 2-hydroxy-1-naphthaldehyde candidates, HN1–HN3, were predicted using the ADMETlab 3.0 online platform30. The chemical structures of each compound were submitted as canonical SMILES strings, and the resulting outputs were used to evaluate physicochemical features, absorption, distribution, metabolism, excretion, toxicity, and medicinal-chemistry-related parameters. Particular attention was given to drug-likeness, gastrointestinal absorption, blood–brain barrier permeability, cytochrome P450 interaction liability, clearance-related descriptors, hepatotoxicity, mutagenicity, and overall toxicity risk. The ADMET predictions were used as a complementary filtering step together with docking, MD simulation, and MM/PBSA results to assess the suitability of HN1–HN3 as candidate HEG1–KRIT1 disruptors for further experimental testing.In vitro HEG1–KRIT1 competition assayThe inhibitory activities of the prioritized 2-hydroxy-1-naphthaldehyde candidates HN1–HN3 against the HEG1–KRIT1 interaction were evaluated using a bead-based flow-cytometry competition assay based on previously described HEG1–KRIT1 inhibitor assays10,11. SPHERO NeutrAvidin polystyrene beads (6–8 μm) were washed twice with wash buffer containing 20 mM Tris, 150 mM NaCl, 0.01% NP-40, and 1 mM EDTA at pH 7.4. The beads were subsequently blocked for 30 min at room temperature in reaction buffer containing 20 mM Tris, 150 mM NaCl, 0.01% NP-40, 1 mM EDTA, 1 mM DTT, and 0.1% BSA at pH 7.4. After centrifugation, the beads were resuspended in reaction buffer to a density of 3,600 beads/µL and coated overnight at 4 °C with N-terminally biotinylated HEG1 intracellular-tail protein at a final concentration of 150 nM. Excess unbound HEG1 was removed by three washes with ice-cold reaction buffer, and the HEG1-functionalized beads were adjusted to 2,000 beads/µL.Test compounds were prepared as stock solutions in DMSO and serially diluted, with HKi2 included as a positive inhibitor control. For each condition, His₆–EGFP–KRIT1 FERM domain (residues 417–736) was combined with vehicle or test compound in a 100-µL preincubation mixture containing 140 nM EGFP–KRIT1 FERM and 10% DMSO. The mixtures were rotated for 15 min at room temperature in the dark. An equal volume of HEG1-coated beads was subsequently added to produce a final assay volume of 200 µL containing 70 nM EGFP–KRIT1 FERM and 1,000 beads/µL. The reactions were incubated for an additional 15 min at room temperature with rotation. Final compound concentrations were calculated after bead addition, and the final DMSO concentration was maintained consistently across all conditions. Background fluorescence was determined using HEG1-coated beads without EGFP–KRIT1 FERM, whereas maximum binding was defined using HEG1-coated beads incubated with EGFP–KRIT1 FERM and DMSO vehicle. Competitive displacement was verified using 2 µM non-biotinylated HEG1 7-mer peptide corresponding to residues 1375–1381. Bead-associated EGFP fluorescence was quantified using a BD Accuri flow cytometer. The geometric mean fluorescence intensity (GMFI) of the gated bead population was used as the assay readout. Percentage inhibition was calculated using the following equation:$$ \% ~inhibition = 100 \times \frac{{GMFI_{{DMSO}} - GMFI_{{compound}} }}{{GMFI_{{DMSO}} - GMFI_{{no~KRIT1}} }} $$Dose–response curves were generated by plotting percentage inhibition against the log₁₀-transformed compound concentration and fitting the data using a four-parameter logistic model with variable slope. IC₅₀ was defined as the concentration required to reduce EGFP–KRIT1 FERM binding to HEG1-coated beads by 50%. The assay was performed in three independent experiments, with each compound concentration tested in technical triplicate. Technical replicates were averaged before curve fitting and were not considered independent experimental replicates. A separate concentration–response curve and IC₅₀ estimate were obtained for each independent experiment. Because IC₅₀ values are distributed on a logarithmic concentration scale, replicate IC₅₀ estimates were log₁₀-transformed before statistical analysis. Results are reported as geometric mean IC₅₀ values with 95% confidence intervals obtained by back-transforming the mean log₁₀(IC₅₀) and its corresponding t-based confidence limits. Differences among HKi2, HN1, HN2, and HN3 were evaluated using one-way analysis of variance applied to the independently determined log₁₀(IC₅₀) values. Statistical analysis and nonlinear regression were performed using GraphPad Prism version 10.0 (GraphPad Software, Boston, MA, USA).Results and discussionA stepwise virtual-screening funnel prioritized three 2-hydroxy-1-naphthaldehyde candidates as putative HEG1–KRIT1 disruptorsThe virtual-screening workflow summarized in Fig. 2 began with a focused library of 1,121 compounds containing the 2-hydroxy-1-naphthaldehyde scaffold and applied a sequential, multicriteria selection process to prioritize candidates for computational and experimental evaluation. Candidate advancement was based on the combined physicochemical, structural, computational, and practical characteristics of each compound rather than on a single docking-score cutoff. In the first stage, compounds were assessed for molecular weight, calculated logP/logD, topological polar surface area, hydrogen-bond donor and acceptor counts, and rotatable-bond number. Compounds with clearly unfavorable drug-likeness profiles, PAINS alerts, or additional potentially reactive functionalities were deprioritized. The 2-hydroxy-1-naphthaldehyde group was intentionally retained because it represents the established reversible-covalent recognition scaffold associated with KRIT1 Lys720; therefore, the reactivity filter was used to exclude additional nonspecific reactive groups rather than the intended aldehyde functionality itself.The remaining compounds were then assessed for commercial availability, chemical tractability, and structural diversity. Commercial availability was required to permit experimental evaluation without additional synthesis, while structural-diversity assessment was used to avoid selecting numerous nearly identical analogues. Preference was given to compounds representing distinct substitution patterns around the shared 2-hydroxy-1-naphthaldehyde core. The purchasable candidates were subsequently docked into the HEG1/HKi2-recognition pocket of the KRIT1 FERM domain using AutoDock Vina. Multiple docking runs and poses were evaluated for each compound. Candidate ranking considered the predicted docking score, reproducibility of the ligand orientation across docking runs, placement of the shared naphthaldehyde core within the target pocket, contacts with residues surrounding the HEG1-recognition region, and absence of severe steric clashes. Docking scores were treated as comparative ranking parameters and not as experimental binding affinities. The calculations modeled the noncovalent pre-reaction complexes, and no covalent bond with Lys720 was imposed.The highest-priority docking candidates underwent short molecular-dynamics refinement. Candidates were favored when they remained localized within the KRIT1 pocket, showed stabilization of ligand RMSD following initial structural relaxation, and retained contacts with the HEG1/HKi2-recognition region. MM/PBSA estimates were used only as an additional qualitative ranking descriptor because endpoint free-energy calculations are subject to uncertainties associated with conformational sampling, solvation treatment, and omission of entropic contributions. Candidate selection therefore reflected the overall consistency of the physicochemical, docking, trajectory-stability, contact-persistence, chemical-diversity, and commercial-availability criteria.This integrated evaluation produced three final candidates: HN1, 6-chloro-2-hydroxy-1-naphthaldehyde; HN2, 2-hydroxy-6-[5-(4-methylpiperazine-1-carbonyl)thiophen-2-yl]-1-naphthaldehyde; and HN3, 2-hydroxy-6-[5-(4-morpholinylcarbonyl)-2-thienyl]-1- naphthaldehyde. HN1 represented a compact halogen-substituted analogue, whereas HN2 and HN3 represented extended C6-substituted analogues containing piperazine- and morpholine-derived groups, respectively. All three compounds were commercially available and were selected before experimental testing; their subsequent IC₅₀ values were not used retrospectively during prioritization. Accordingly, the workflow provided a transparent rationale for advancing three chemically differentiated analogues to long-timescale molecular-dynamics simulations and biochemical HEG1–KRIT1 competition testing.Fig. 2Full size imageVirtual-screening workflow used to identify candidate HEG1–KRIT1 disruptors.Docking analysis supports HN1–HN3 binding within the HEG1-recognition pocket of KRIT1The docking analysis shown in Figs. 3 and 4 indicates that the three prioritized 2-hydroxy-1-naphthaldehyde derivatives occupy the HEG1/HKi2-recognition region of the KRIT1 FERM domain, supporting their selection as candidate HEG1–KRIT1 disruptors. In Fig. 3, the docked poses of HN1, HN2, and HN3 are positioned within the same surface-accessible pocket used by the reference inhibitor HKi2, suggesting that these compounds may interfere with HEG1 engagement by sterically or energetically competing for the KRIT1 interface. HN1 adopts a compact orientation within the pocket, consistent with its smaller scaffold, whereas HN2 and HN3 extend more deeply across the binding groove through their thiophene-linked substituents, allowing broader contact with the KRIT1 surface. The 2D interaction maps in Fig. 4 further show that all three candidates contact residues located near the HEG1-binding site, including recurring interactions with residues such as Gln473, Lys475, Val512, Ala638, Phe640, Leu717, and Leu721. These residues overlap with, or lie close to, the interaction environment observed for HKi2, supporting the idea that the selected compounds target a relevant KRIT1 interface rather than a distant nonspecific pocket. HN1 appears to preserve the core naphthaldehyde-associated contacts while maintaining a relatively simple binding mode, whereas HN2 and HN3 show expanded hydrophobic and polar interaction networks involving additional residues such as His478, His483, Ile490, and Leu494. This broader interaction pattern may improve pocket occupancy and pose stabilization, particularly for HN2 and HN3. Overall, the docking results suggest that HN1–HN3 can engage the KRIT1 HEG1-recognition surface in poses compatible with competitive disruption of the HEG1–KRIT1 interaction, with the extended analogues HN2 and HN3 providing more extensive interface coverage for subsequent MD-based stability evaluation.Fig. 3Full size imagePredicted docking poses of HN1–HN3 within the HEG1-recognition pocket of KRIT1. Three-dimensional docking models showing the binding positions of the reference inhibitor HKi2 and the selected 2-hydroxy-1-naphthaldehyde candidates HN1, HN2, and HN3 at the HEG1-binding region of the KRIT1 FERM domain. Each panel displays the protein–ligand complex in both cartoon and surface representations to illustrate ligand placement within the KRIT1 interfacial pocket. The compounds are shown in ball-and-stick representation, highlighting their predicted occupancy of the HEG1/HKi2-recognition site and their potential to interfere with HEG1–KRIT1 binding.Fig. 4Full size imageTwo-dimensional interaction profiles and surface binding views of HKi2 and HN1–HN3 docked to KRIT1. Docking interaction maps and corresponding surface representations showing the predicted contacts formed by HKi2 and the candidate compounds HN1, HN2, and HN3 within the KRIT1 HEG1-recognition region. The 2D diagrams summarize key ligand–residue interactions, including hydrogen-bonding, hydrophobic, and other noncovalent contacts with residues surrounding the binding pocket. The surface views illustrate the position of each ligand within the KRIT1 pocket, supporting the predicted binding of HN1–HN3 at or near the HEG1–KRIT1 interface.MM/PBSA PPI-energy analysis identifies HN1 as the only candidate with a predicted weakening effect on the HEG1–KRIT1 interfaceThe MM/PBSA decomposition analysis in Table 1 was used to estimate how ligand binding influenced the energetic stability of the HEG1–KRIT1 interface during the final 200 ns of the 1000 ns MD simulations. The apo HEG1–KRIT1 complex showed a calculated interfacial free energy of − 59.86 ± 9.49 kcal/mol and was used as the reference state for ΔΔGPPI calculations. HKi2 produced a nearly unchanged PPI energy relative to apo, with ΔGPPI = − 60.19 ± 7.06 kcal/mol and ΔΔGPPI = − 0.33 kcal/mol, indicating that it did not weaken the HEG1–KRIT1 interface under this energetic criterion. Among the three HN candidates, HN1 was the only compound associated with a positive ΔΔGPPI value, shifting the interfacial energy to − 57.20 ± 9.45 kcal/mol and giving ΔΔGPPI = + 2.66 kcal/mol, consistent with a modest predicted weakening of the HEG1–KRIT1 interaction. In contrast, HN2 and HN3 gave more negative ΔGPPI values than the apo complex, with ΔΔGPPI values of − 17.00 and − 7.91 kcal/mol, respectively, suggesting relative stabilization rather than disruption of the protein–protein interface. However, the HN2 and HN3 calculations were characterized by very large opposing electrostatic and solvation terms, as well as high standard deviations in the total PPI energies, indicating substantial energetic compensation and greater uncertainty in these estimates. Therefore, while docking suggested that HN1–HN3 can occupy the KRIT1 HEG1-recognition region, the MM/PBSA PPI-disruption analysis supports HN1 as the most promising candidate for ligand-mediated weakening of the HEG1–KRIT1 interaction, whereas HN2 and HN3 appear more consistent with stable KRIT1 binding without predicted PPI destabilization.Table 1 MM/PBSA energy decomposition and predicted ligand-mediated effects on the HEG1–KRIT1 interface. Binding free-energy components were calculated from the final 200 ns of the 1000 ns MD trajectories and are reported as mean ± SD in kcal/mol. The apo HEG1–KRIT1 complex was used as the reference state, and ΔΔGPPI was calculated as ΔΔGPPI (ligand-bound) − ΔΔGPPI (apo), using − 59.86 kcal/mol as the apo reference value. Positive ΔΔGPPI values indicate predicted weakening of the HEG1–KRIT1 protein–protein interaction, whereas negative values indicate relative stabilization of the interface.Full size tableHN1–HN3 perturb the native HEG1–KRIT1 interface despite maintaining overall complex associationThe PPI-disruption metrics in Figs. 5 and S2 and Table 2 indicate that ligand binding altered the native HEG1–KRIT1 interface without causing complete dissociation of the protein complex. In the apo simulation, the overall HEG1–KRIT1 center-of-mass distance remained stable at 3.631 ± 0.097 nm, while the native interface COM distance was maintained at 0.627 ± 0.048 nm, consistent with a compact and persistent interfacial arrangement. HKi2 produced a similar native interface distance of 0.627 ± 0.045 nm and a comparable contact profile to the apo system, suggesting that, under these simulation conditions, HKi2 did not strongly separate the native interface. In contrast, all three HN compounds increased the native interface COM distance, with HN1, HN2, and HN3 reaching 0.748 ± 0.042, 0.763 ± 0.060, and 0.694 ± 0.043 nm, respectively. This shift supports partial opening or rearrangement of the native HEG1–KRIT1 contact region, with HN2 producing the largest separation and HN1 showing a similarly strong destabilizing trend. Although the overall HEG1–KRIT1 COM distance was lower for the HN-bound systems than for apo, the increased native-interface distance indicates that the compounds did not simply pull the two proteins apart globally, but instead promoted local remodeling of the native contact surface.Additional contact-based metrics support this interpretation. Compared with the apo complex, the HN-bound systems showed reduced native interface contact fractions and increased native contact loss, with contact loss rising from 0.158 ± 0.131 in apo to 0.229 ± 0.086, 0.222 ± 0.090, and 0.223 ± 0.091 for HN1, HN2, and HN3, respectively. Interface hydrogen bonding was also markedly reduced relative to apo, decreasing from 1.893 ± 0.619 hydrogen bonds in the ligand-free complex to 0.364 ± 0.692 for HN1, 0.561 ± 0.586 for HN2, and 0.557 ± 0.649 for HN3, indicating weakening of specific polar interactions that help stabilize the native HEG1–KRIT1 interface. At the same time, the HN compounds maintained substantial ligand–protein contacts, particularly HN1 and HN2, with 94.247 ± 16.089 and 102.016 ± 31.331 contacts within 0.45 nm, respectively. This suggests that the ligands remained engaged with KRIT1 while the native interface was partially reorganized.The SASA/BSA analysis in Table S1 further refines this interpretation by showing that the HN-bound systems did not simply separate the two chains, but instead formed a larger, remodeled chain-to-chain contact surface. The apo and HKi2 systems showed low Chain A–Chain B total BSA values of 3.225 ± 0.393 and 3.227 ± 0.358 nm², respectively, whereas HN1, HN2, and HN3 increased total BSA to 20.350 ± 1.988, 20.344 ± 2.821, and 20.428 ± 2.220 nm². The corresponding BSA per partner increased from approximately 1.61 nm² in the apo/HKi2 systems to approximately 10.17–10.21 nm² in the HN-bound complexes. This increase in buried surface area, together with the loss of native interface contacts, suggests that HN1–HN3 promote non-native or expanded interfacial packing rather than complete HEG1–KRIT1 dissociation. Therefore, the combined distance, native-contact, hydrogen-bond, SASA, and BSA metrics support a model in which HN1–HN3 do not fully dissociate the HEG1–KRIT1 complex, but instead remodel the native binding interface. This remodeling is characterized by increased native-interface separation, reduced retention of native contacts, loss of interfacial hydrogen bonding, and expansion of buried surface area, suggesting a shift from the original HEG1–KRIT1 contact geometry toward an altered, non-native protein–protein association.Fig. 5Full size imageTime-dependent HEG1–KRIT1 chain-to-chain and native-interface distance metrics during 1000 ns MD simulations. Center-of-mass distance analyses were used to monitor global HEG1–KRIT1 association and local native-interface separation in the apo complex and in complexes containing HKi2, HN1, HN2, or HN3. For each system, the upper trace shows the overall HEG1–KRIT1 chain-to-chain COM distance, whereas the lower trace shows the COM distance across the native HEG1–KRIT1 interface. Representative surface snapshots illustrate the relative orientation of the two binding partners and ligand position during the simulations. Increased native-interface COM distance in the HN-bound systems indicates local interfacial rearrangement consistent with partial disruption of the HEG1–KRIT1 contact region.Table 2 MD-derived PPI-disruption metrics for apo and ligand-bound HEG1–KRIT1 complexes. Quantitative comparison of global complex association, native-interface separation, buried surface area, native contact retention or loss, interface hydrogen bonding, close-contact pairs, and ligand–protein contacts calculated from the 1000 ns MD simulations. Values are reported as mean ± SD. The apo HEG1–KRIT1 complex was used as the reference state for evaluating ligand-induced changes in interface stability. Increased native-interface COM distance, greater native contact loss, and reduced interfacial hydrogen bonding were interpreted as indicators of HEG1–KRIT1 interface perturbation.Full size tableHN1–HN3 preserve global HEG1–KRIT1 structural stability while inducing modest ligand-dependent conformational changesThe global MD stability analyses in Fig. 6 and Table S2 show that all simulated HEG1–KRIT1 systems remained structurally stable over the 1000 ns trajectories, although ligand binding produced modest increases in conformational deviation relative to the apo complex. The apo HEG1–KRIT1 system displayed the lowest backbone RMSD, with an average value of 0.231 ± 0.035 nm, indicating a stable reference trajectory. Ligand-bound systems showed slightly higher RMSD values, increasing to 0.288 ± 0.083 nm for HKi2, 0.280 ± 0.044 nm for HN1, 0.295 ± 0.066 nm for HN2, and 0.319 ± 0.053 nm for HN3, suggesting that compound binding increased structural mobility but did not cause large-scale destabilization. The radius of gyration remained narrowly distributed across all systems, ranging from 2.515 ± 0.016 nm for HKi2 to 2.533 ± 0.020 nm for HN2, supporting preservation of overall KRIT1 compactness during the simulations. RMSF analysis further showed similar residue-flexibility profiles among the apo and ligand-bound complexes, with localized fluctuations but no widespread destabilization of the protein backbone. HN2 had the lowest average Chain A RMSF among the HN compounds, 0.163 ± 0.100 nm, whereas HN1 and HN3 showed slightly higher flexibility at 0.181 ± 0.099 and 0.185 ± 0.100 nm, respectively. The SASA values reported in Table S2 were higher for HN1–HN3 than for the apo and HKi2 systems, consistent with the broader interface remodeling observed in the PPI-disruption analysis. Together, these data indicate that HN1–HN3 do not globally unfold or destabilize HEG1–KRIT1, but instead preserve overall complex stability while promoting localized conformational and surface-accessibility changes compatible with native-interface remodeling.Fig. 6Full size imageGlobal structural stability and flexibility of apo and ligand-bound HEG1–KRIT1 complexes during 1000 ns MD simulations. Time-dependent MD analyses of the apo HEG1–KRIT1 complex and complexes containing HKi2, HN1, HN2, or HN3. (A) Backbone RMSD was used to monitor global structural deviation during the simulations. (B) Radius of gyration was calculated to assess overall compactness of the protein complex. (C) Solvent-accessible surface area was used to evaluate changes in surface exposure. (D) Residue-wise RMSF was calculated to compare local flexibility patterns across the simulated systems.FEL analysis reveals ligand-specific conformational basins that support stable KRIT1 engagement by HN1–HN3The free energy landscape analysis in Fig. 7 shows that the ligand-bound HEG1–KRIT1 systems sampled distinct conformational energy profiles during the 1000 ns MD simulations, indicating that each compound influenced the conformational behavior of the complex in a different manner. The HKi2 reference system displayed a broad low-energy region with multiple connected minima, consistent with conformational flexibility within the KRIT1 interface. HN1 produced a more elongated landscape along PC1 and PC2, with a defined low-energy basin separated from higher-energy regions, suggesting that HN1 binding favored a restricted subset of conformations rather than broad conformational sampling. HN2 showed a wider and more continuous low-energy surface, indicating that this compound allowed multiple related conformational states while remaining associated with the KRIT1 binding region. In contrast, HN3 displayed a relatively compact low-energy basin, suggesting a more localized conformational ensemble. Overall, the FEL profiles indicate that HN1–HN3 do not behave identically after binding to KRIT1: HN1 and HN3 appear to stabilize more defined conformational states, whereas HN2 supports a broader but still energetically favorable conformational distribution.The representative lowest-energy conformations shown in Fig. 8 further support the conclusion that the selected compounds remain positioned within or near the HEG1-recognition region of KRIT1 at energetically favorable states. In these minimum-energy snapshots, HN1 remains localized in the interfacial pocket near residues previously identified in the docking analysis, supporting stable occupation of the HEG1/HKi2-associated binding region. HN2 and HN3 show more extended binding arrangements, with their larger substituents occupying a broader surface area and engaging additional nearby residues and solvent-exposed contacts. This behavior is consistent with the earlier MD and PPI-disruption metrics, in which HN2 and HN3 maintained ligand–protein contacts while promoting remodeling rather than complete separation of the HEG1–KRIT1 complex. Together, the FEL maps and minimum-energy structural snapshots indicate that HN1–HN3 form stable KRIT1-bound conformational states, with ligand-specific differences in conformational restriction, surface coverage, and interfacial remodeling potential.Fig. 7Full size imageFree energy landscape analysis of ligand-bound HEG1–KRIT1 complexes. Two-dimensional and three-dimensional free energy landscapes were generated from the first two principal components, PC1 and PC2, extracted from the 1000 ns MD trajectories of HEG1–KRIT1 complexes containing HKi2, HN1, HN2, or HN3. Color gradients represent relative Gibbs free energy, with lower-energy regions corresponding to more populated and energetically favorable conformational states. The FEL maps show ligand-specific conformational sampling patterns, indicating that each compound drives a distinct energy landscape within the HEG1–KRIT1 complex.Fig. 8Full size imageRepresentative lowest-energy conformations from FEL analysis. Lowest-energy structural snapshots extracted from the dominant free energy basins of the HEG1–KRIT1 complexes containing HKi2, HN1, HN2, or HN3. Each panel shows the ligand position within or near the KRIT1 HEG1-recognition region in cartoon and surface representations, with nearby interacting residues labeled.Ligand–KRIT1 energy analysis identifies HN2 as the strongest KRIT1-binding candidate, driven mainly by short-range van der Waals contactsThe ligand–protein interaction energy analysis in Fig. 9; Table 3 shows that all three HN candidates maintained favorable binding energetics with KRIT1, with total MM/PBSA-derived binding energies broadly comparable to the HKi2 reference. HKi2 showed a ΔTOTAL of − 18.51 ± 2.37 kcal/mol, while HN1, HN2, and HN3 produced values of − 17.31 ± 3.30, − 18.96 ± 2.97, and − 16.36 ± 3.71 kcal/mol, respectively. Among the tested compounds, HN2 had the most favorable overall binding energy, slightly exceeding HKi2, whereas HN3 showed the weakest total binding estimate. The decomposition terms indicate that binding was primarily supported by favorable gas-phase interactions, particularly van der Waals contributions, with HN1–HN3 showing stronger ΔVDWAALS values than HKi2. HN2 gave the most favorable ΔVDWAALS term at − 26.61 ± 2.80 kcal/mol, consistent with its extended structure and broader hydrophobic surface engagement within the KRIT1 pocket. Electrostatic contributions were favorable for all ligands, although HN2 had a weaker ΔEEL value than HKi2, HN1, and HN3, suggesting that its binding advantage was driven more by shape complementarity and dispersion contacts than by electrostatics. As expected for endpoint free-energy calculations, favorable gas-phase interactions were partially offset by positive polar solvation penalties, but the final ΔTOTAL values remained negative for all compounds, supporting stable KRIT1 association.The short-range interaction profiles further support this ranking. The LJ-SR traces in Fig. 9B and the values in Table 3 show that HN2 had the strongest short-range Lennard–Jones interaction energy, − 123.18 ± 29.61 kJ/mol, followed by HN1 at − 95.90 ± 16.68 kJ/mol, HN3 at − 90.03 ± 30.61 kJ/mol, and HKi2 at − 64.49 ± 26.80 kJ/mol. Coulombic short-range interactions were also favorable across the ligand series, with HN2 again showing the strongest average Coul-SR value, − 33.01 ± 19.43 kJ/mol. The per-residue energy decomposition plots in Fig. 9C–F reveal that ligand binding was not dominated by a single residue, but instead distributed across multiple favorable residue-level contacts within and around the KRIT1 HEG1-recognition region. This distributed interaction pattern is consistent with the docking, MD contact, and FEL analyses, which showed that the HN compounds remain engaged within the KRIT1 interface while promoting local interfacial remodeling. Overall, these results identify HN2 as the most energetically favorable KRIT1 binder by total binding energy and short-range contact strength, while HN1 remains notable because earlier PPI-energy and native-interface analyses suggested a stronger disruptive effect on HEG1–KRIT1 contact organization.Fig. 9Full size imageLigand–KRIT1 interaction-energy profiles and per-residue energy decomposition. Time-resolved short-range interaction energies and residue-level energetic contributions for HKi2, HN1, HN2, and HN3 bound to KRIT1 during the MD simulations. (A) Coulombic short-range interaction energies show electrostatic contributions to ligand–protein binding. (B) Lennard–Jones short-range interaction energies show nonpolar and van der Waals contact contributions. (C–F) Per-residue energy decomposition maps identify KRIT1 residues contributing favorably or unfavorably to ligand binding for HKi2, HN1, HN2, and HN3, respectively. More negative values indicate stronger favorable energetic contributions.Table 3 Ligand–KRIT1 binding free-energy components and short-range interaction energies. Summary of MM/PBSA-derived binding-energy components and GROMACS short-range interaction energies for HKi2, HN1, HN2, and HN3 bound to KRIT1. Values are reported as mean ± SD.Full size tableADMET profiling supports HN1–HN3 as drug-like screening hits but identifies liabilities requiring optimizationThe in silico ADMET profiles in Table S3 indicate that HN1, HN2, and HN3 possess generally acceptable physicochemical and drug-likeness properties, but each compound also carries predicted developability liabilities that should be addressed before further lead advancement. All three candidates showed zero Lipinski violations, favorable QED values ranging from 0.707 to 0.729, and low synthetic-accessibility scores, supporting their suitability as tractable early-stage chemical matter. HN1 was the smallest and least flexible compound, with a molecular weight of 206.01 g/mol, TPSA of 37.30 Ų, one rotatable bond, and logP of 3.793, whereas HN2 and HN3 were larger and more polar, with molecular weights of 380.12 and 367.09 g/mol and TPSA values of 60.85 and 66.84 Ų, respectively. The absorption and distribution predictions suggest limited intestinal absorption scores for all three compounds, high plasma protein binding, and variable BBB permeability, with HN2 showing the highest predicted BBB permeability. Metabolism predictions identified HN1 as a likely CYP1A2 inhibitor, while HN2 and HN3 showed lower predicted CYP inhibition across most isoforms. However, the toxicity-related outputs raise important concerns: HN2 and HN3 showed high predicted risks for Ames mutagenicity, carcinogenicity, DILI, nephrotoxicity, and neurotoxicity, whereas HN1 showed a strong reactive-compound score and high skin-sensitization risk. These findings suggest that although HN1–HN3 are useful computationally and experimentally testable candidates for probing HEG1–KRIT1 disruption, they should be considered preliminary hit compounds rather than optimized leads. Overall, HN1 appears more favorable from the standpoint of molecular simplicity and PPI-disruption behavior but may require structural modification to reduce reactivity, while HN2 and HN3 may need optimization to reduce predicted toxicity and improve absorption-related properties.HN1 shows HKi2-comparable in vitro disruption of the HEG1–KRIT1 interactionThe bead-based flow-cytometry competition assay in Fig. 10 confirmed that all three selected 2-hydroxy-1-naphthaldehyde candidates inhibited EGFP–KRIT1 FERM binding to HEG1-coated beads in a concentration-dependent manner, validating the computational screening workflow with experimental evidence of HEG1–KRIT1 disruption. The positive-control inhibitor HKi2 produced an IC₅₀ of 3.90 ± 0.13 µM, consistent with strong assay performance and providing a benchmark for comparison. Notably, HN1 showed lower IC₅₀ than HKi2, with an IC₅₀ of 3.51 ± 0.14 µM, identifying HN1 as the most active compound in the series. This result agrees with the computational PPI-disruption analyses, where HN1 was the only candidate associated with a positive ΔΔGPPI and showed clear perturbation of the native HEG1–KRIT1 interface. In contrast, HN2 and HN3 inhibited binding less potently, with IC₅₀ values of 24.12 ± 1.59 and 31.63 ± 1.99 µM, respectively, despite their favorable KRIT1-binding and ligand-contact profiles in MD simulations. This difference suggests that stable KRIT1 binding alone is not sufficient for efficient competitive disruption; rather, ligand orientation, occupancy of the native HEG1-recognition surface, and the ability to weaken productive HEG1–KRIT1 contacts appear to be more important determinants of functional inhibition. Overall, the in vitro data identify HN1 as the strongest experimental HEG1–KRIT1 disruptor among the tested candidates and support its prioritization as the lead scaffold for further optimization.Fig. 10Full size imageIn vitro inhibition of the HEG1–KRIT1 interaction by HN1–HN3. Dose–response curves from the bead-based flow-cytometry competition assay measuring inhibition of KRIT1 FERM binding to HEG1-coated beads by HKi2 and the selected 2-hydroxy-1-naphthaldehyde candidates. (A) HKi2 was included as the positive-control inhibitor. (B–D) HN1, HN2, and HN3 produced concentration-dependent reductions in KRIT1 FERM binding to HEG1-coated beads. Each concentration was evaluated in technical triplicate in each of three independent assay runs. Data points represent the mean geometric fluorescence intensity ± SD and were fitted using a four-parameter logistic model with variable slope. The resulting IC₅₀ values are presented as descriptive potency estimates. No formal statistical test of superiority or equivalence between compounds was performed; therefore, numerical differences among IC₅₀ values should be interpreted descriptively.Previous structure-guided studies identified HKi1 as a low-micromolar HEG1–KRIT1 inhibitor (IC₅₀ approximately 10 µM), whereas its smaller derivatives HKi2 and HKi3 showed improved potency of approximately 3.5 µM; the subsequently optimized HKi6 also demonstrated improved KRIT1 affinity and inhibited KRIT1 recruitment to HEG1 in a cellular colocalization assay7. More recently, 2-hydroxy-1-naphthaldehyde was characterized as a selective, reversible covalent ligand of KRIT1 Lys720 with prolonged residence time, and substituted HNA analogues showed improved target engagement and cellular activity11. Within this context, the observed IC₅₀ of HN1 (3.51 ± 0.14 µM) lies within the low-micromolar range reported for established HKi inhibitors. Nevertheless, differences in assay design and the absence of direct target-engagement and cellular studies for HN1 preclude claims of superiority or equivalence; HN1 should therefore be considered a preliminary analogue for further validation.ConclusionIn this study, a focused library of 2-hydroxy-1-naphthaldehyde derivatives was screened to identify candidate small-molecule modulators of the HEG1–KRIT1 protein–protein interaction, a regulatory interface in CCM signaling. A stepwise computational workflow incorporating drug-likeness filtering, docking, MD refinement, MM/PBSA-based energetic analysis, free energy landscape evaluation, and ADMET profiling prioritized HN1–HN3 for experimental testing. Docking and MD analyses suggested that all three compounds could occupy the HEG1-recognition region of the KRIT1 FERM domain and maintain interactions with KRIT1 over the simulated trajectories. The calculated PPI metrics showed ligand-dependent differences: HN1 yielded a modest positive ΔΔGPPI estimate and changes in several native-interface measures, whereas HN2 and HN3 yielded negative ΔΔGPPI estimates and interaction patterns more consistent with altered or expanded interfacial packing. These trajectory-derived findings should be interpreted descriptively as approximate indicators of ligand-dependent interface modulation rather than definitive evidence of PPI weakening or stabilization. In the bead-based flow-cytometry competition assay, all three compounds produced concentration-dependent reductions in EGFP–KRIT1 FERM binding to HEG1-coated beads. HN1 yielded an IC₅₀ estimate of 3.51 ± 0.14 µM, which was numerically similar to the 3.90 ± 0.13 µM estimate obtained for HKi2 under the same assay conditions, whereas HN2 and HN3 yielded higher IC₅₀ estimates. These numerical differences are descriptive and do not establish superiority or equivalence among the compounds. ADMET predictions suggested generally acceptable early drug-likeness characteristics but also identified potential toxicity and reactivity liabilities requiring further investigation. Collectively, the computational and experimental results support HN1 as a preliminary screening hit and the 2-hydroxy-1-naphthaldehyde scaffold as a potential starting point for further optimization. Additional studies are required to confirm direct KRIT1 engagement, evaluate selectivity and nonspecific reactivity, establish activity using orthogonal biochemical and cellular assays, and determine the effects of HN1 on downstream CCM signaling markers such as KLF2 and KLF4.