Mechanistic basis of aldose reductase inhibition by phenolics from Achillea fragrantissima: experimental insights and computational validation

Abstract

This study explores the inhibitory potential of phenolic compounds derived from Achillea fragrantissima against aldose reductase (AR), a key enzyme implicated in diabetic complications. Among the six tested compounds, salvianolic acid B exhibited the strongest inhibitory activity, with an IC₅₀ of 1.90 ± 0.1 µM, followed by chlorogenic acid (IC₅₀ = 3.79 ± 0.17 µM) and 3-caffeoyl-4-feruloylquinic acid (IC₅₀ = 4.49 ± 0.38 µM). The remaining three phenolics showed weak AR inhibition and were excluded from further analysis. Enzyme kinetics studies revealed that all three active compounds function as noncompetitive inhibitors, a finding further supported by in silico analyses. Molecular docking demonstrated strong binding affinities, with salvianolic acid B adopting the most favorable binding pose within the AR binding site. Molecular dynamics (MD) simulations confirmed the stability of AR–ligand complexes throughout the simulation period, while MM/PBSA calculations identified salvianolic acid B as having the lowest binding free energy, reinforcing its superior inhibitory potential. Additional validation through free-energy landscape (FEL) analysis and interaction energy profiling further substantiated the stability and efficacy of these inhibitors. Moreover, ADMET analysis indicated that all active compounds exhibit drug-like properties, with favorable bioavailability and gastrointestinal absorption. Overall, the in silico findings align with experimental results, underscoring the therapeutic potential of these phenolic compounds as AR inhibitors. These insights has the potential of the development for therapeutics for diabetic complications.

Supplementary Information

The online version contains supplementary material available at 10.1007/s13205-025-04486-0.

Introduction

Aldose reductase (AR) is an NADPH-dependent enzyme that initiates the polyol pathway by reducing glucose to sorbitol (Gupta 2024). Under hyperglycemic conditions, this pathway is upregulated, leading to intracellular sorbitol accumulation, osmotic and oxidative stress, and diabetic complications including cataract, neuropathy, retinopathy, and nephropathy (Yabe-Nishimura 1998). Inhibiting AR is therefore a validated strategy to blunt polyol pathway-mediated damage; however, several synthetic AR inhibitors have encountered limitations in selectivity, pharmacokinetics, or safety that constrain long-term use (Balestri et al. 2022; Bernardoni et al. 2024). Natural AR inhibitors—particularly plant‐derived phenolic compounds—show promising inhibitory activity along with complementary antioxidant effects that may improve tolerability and multi-target benefit (Kamel et al. 2025c, d, a).

Natural products—especially phenolics—offer a complementary route, combining enzyme-inhibitory activity with antioxidant properties that may counter polyol pathway-associated oxidative stress (Veeresham et al. 2014; Kamel et al. 2025d, f). Prior reports show that selected polyphenols can inhibit AR with potencies comparable to reference inhibitors, highlighting their potential as drug leads (Alqhtani et al. 2025; Demir et al. 2017). Achillea fragrantissima (Forssk.) Sch. Bip. (“Qaisum”) is a traditional Middle Eastern medicinal herb distributed across arid regions of the Middle East and North Africa (El-Shazly et al. 2020). It is rich in flavonoids, phenolic acids, and related secondary metabolites with reported antioxidant, anti-inflammatory, and enzyme-inhibitory activities, making it a promising source of natural AR inhibitors (Sadeq et al. 2021; Awad et al. 2017; Tawfik et al. 2024).

Computational and experimental workflows provide complementary evidence for enzyme inhibition (Alqhtani et al. 2024; Alwaili et al. 2025). Molecular docking, molecular dynamics (MD), MM/PBSA binding free-energy estimates, and principal component/energy landscape analyses elucidate binding modes, stability, and key interactions, while ADMET predictions help anticipate pharmacokinetic liabilities and safety risks (Kamel et al. 2025e, 2024a, 2024b). Complementary in vitro inhibition assays and enzyme kinetic analyses (e.g., Michaelis–Menten/Lineweaver–Burk) quantify potency (IC₅₀), binding affinity (K_i), and mechanism of inhibition, providing empirical benchmarks for the computational predictions (Kamel et al. 2024b, 2025b, g).

Here, we evaluate the phenolic constituents of A. fragrantissima as AR inhibitors using an integrated design. In vitro assays quantify potency and mechanism (IC₅₀, Ki), and targeted in silico studies—docking, MD, MM/PBSA, PCA/FEL—interpret the experimental trends and prioritize chemotypes with favorable ADMET profiles. This combined approach provides a coherent view of how A. fragrantissima phenolics engage AR and highlights candidates suitable for further development toward mitigating diabetes-associated complications.

Materials and methods

Phytochemical investigation

General

The structural characterization of phytochemicals isolated from A. fragrantissima was conducted using a variety of spectroscopic techniques. ¹H and ¹³C NMR spectra were recorded on a Bruker AVANCE III-600 spectrometer, with tetramethylsilane (TMS) serving as the internal reference. Chemical shift values (δ) were expressed in parts per million (ppm), and coupling constants (J) were reported in hertz (Hz). Infrared (IR) spectra were obtained using a Shimadzu FTIR-8400 spectrophotometer, employing potassium bromide (KBr) pellets as the medium. Ultraviolet–visible (UV–Vis) spectra were determined on a Shimadzu UV-2201 spectrophotometer. Optical rotations of the isolated phenolic compounds were measured with a Rudolph Autopol III polarimeter. High-resolution electron impact mass spectrometry (HREIMS) and electron impact mass spectrometry (EIMS) data were acquired using an Agilent 6210 time-of-flight (TOF) mass spectrometer.

Extraction and isolation

The aerial parts of A. fragrantissima were collected from arid regions in Beni-Suef City. The plant material was identified morphologically by a qualified taxonomist in the Department of Botany, using standard regional floras and diagnostic keys, and verified against authenticated herbarium material. The dried and finely powdered plant material (1.5 kg) was extracted through cold maceration at room temperature three times using absolute methanol (3 × 2.0 L; total 6 L), selected to maximize recovery of phenolic acids and flavonoids, with occasional stirring. After 72 h of extraction, the solvent was filtered, and the process was repeated twice. The combined methanolic extracts were concentrated under reduced pressure using a rotary evaporator, yielding a crude extract of 310 g. To further purify the extract, it was dissolved in a water–methanol mixture (7:3 v/v) and successively partitioned with n-hexane, chloroform, ethyl acetate, and n-butanol. Among these, the ethyl acetate fraction (24 g) showed a rich, well-resolved phenolic pattern on TLC/PC—multiple discrete bands visible under UV (254/366 nm), strong responses to FeCl₃ and NP/PEG spray reagents, and minimal streaking—so it was selected for further isolation. The ethyl acetate fraction was subjected to silica gel column chromatography, eluted with a gradient of n-hexane:ethyl acetate (10:1 to 1:10, v/v), to yield 45 subfractions. These subfractions were pooled based on their TLC and PC profiles into seven main fractions (A1–A7). Each fraction was subjected to further chromatographic processes to isolate the target phenolic compounds.

Compound 1 was obtained from fraction A3, which was further fractionated using a silica gel column with an n-hexane:ethyl acetate (3:7, v/v) gradient. The resulting subfraction A3.2 was purified on a Sephadex LH-20 column using methanol as the eluent, yielding 28 mg of compound 1 as a yellow crystalline powder. Similarly, compound 2 was purified from fraction A2 using preparative thin-layer chromatography (PTLC) with ethyl acetate:methanol:water (75:20:5, v/v/v) as the mobile phase, affording 19 mg of the compound as pale yellow crystals. Compound 3 was isolated from fraction A5. The fraction was separated on a silica gel column eluted with n-hexane:ethyl acetate (4:6, v/v), yielding subfraction A5.3. Further purification of this subfraction using Sephadex LH-20 with methanol resulted in 24 mg of the compound as a yellow amorphous solid. Compound 4 was obtained from fraction A4 through PTLC using ethyl acetate:water:methanol (70:20:10, v/v/v), yielding 27 mg of a brown amorphous solid. Compound 5 was purified from fraction A6. The fraction was first subjected to a polyamide column with methanol as the eluent, yielding six subfractions (A6.1–A6.6). Subfraction A6.4 was further purified using Sephadex LH-20 with methanol, producing 18 mg of a greenish-brown powder. Finally, compound 6 was isolated from fraction A7, which was purified using a polyamide column and methanol gradient. Subfraction A7.2 was further processed using Sephadex LH-20 with methanol, yielding 26 mg of compound 6 as a reddish-brown powder. The structures of all isolated compounds were confirmed using ¹H and ¹³C NMR, UV–Vis, FTIR, and mass spectrometry. The spectral data were compared with previously reported values in the literature to ensure structure elucidation accuracy.

In vitro inhibitory activity assays

Chemicals and reagents

Human recombinant AR, DL-glyceraldehyde, NADPH, and other critical reagents were primarily sourced from Sigma-Aldrich. While standard chemicals were widely accessible, specific compounds required for the study were exclusively procured from Sigma-Aldrich. Additional solvents and reagents used in the experiments met reagent-grade quality standards and were obtained from various commercial suppliers.

AR inhibitory activity assay

AR inhibition was quantified spectrophotometrically by following the decrease in absorbance at 340 nm from NADPH oxidation (Kim et al. 2013). Standard reactions (200 µL, 96-well) contained 100 mM sodium phosphate buffer (pH 6.2), 1 mM EDTA, 0.10 mM NADPH, 10 mM DL-glyceraldehyde, and serial dilutions of each phenolic (DMSO ≤ 1% v/v). Mixtures were pre-incubated at 37 °C for 5 min, initiated with 10 µL recombinant AR, and read every 30 s for 5 min. No-inhibitor wells defined 100% activity and quercetin served as a positive control. Initial rates were taken from the linear portion of the traces, and IC₅₀ values were obtained by nonlinear regression (GraphPad Prism). All measurements were performed in triplicate.

Enzyme kinetics analysis

Inhibition mechanisms and kinetic parameters were determined under the same conditions by varying NADPH and DL-glyceraldehyde concentrations at several fixed inhibitor levels (200 µL total volume, 96-well). After 5 min pre-incubation at 37 °C, reactions were started with enzyme and A340 was recorded for 5 min. Initial velocities were calculated from linear segments and fitted by nonlinear Michaelis–Menten analysis to estimate Vmax and Km (GraphPad Prism). Inhibition type (competitive, noncompetitive, uncompetitive, or mixed) was assigned from Lineweaver–Burk plots, and Ki values were derived from secondary replots (slope or intercept versus inhibitor concentration). Experiments were performed in triplicate and reported as mean ± SD; controls without inhibitor established the uninhibited baseline.

In silico studies

System preparation

Phenolic ligands from A. fragrantissima were geometry optimized at the B3LYP/6-311G(d,p) level in Gaussian 16; harmonic frequency analysis confirmed true minima (no imaginary frequencies) (Frisch et al. 2016; Lee et al. 1988; Becke 1988; Hehre et al. 1986). The X-ray structure of human aldose reductase (PDB ID: 3DN5; 1.45 Å) was retrieved and inspected for missing atoms/residues and alternate conformations; if present, gaps were rebuilt in Swiss-PdbViewer, non-essential heteroatoms were removed, and hydrogens were added/validated in UCSF Chimera (Pettersen et al. 2004). Ligands and receptor were converted to PDBQT with MGLTools/AutoDockTools 1.5.6. In line with the AutoDock Vina workflow, Gasteiger charges were assigned to ligands and nonpolar hydrogens were merged (Trott and Olson 2010; Eberhardt et al. 2021). The binding site was defined around the catalytic pocket identified from the crystal structure; any bound inhibitors were removed prior to docking. Grid-box center and dimensions, together with other docking parameters, are reported below.

Molecular docking analysis

The AR structure was prepared in AutoDockTools (ADT 1.5.6): nonstandard residues/waters and bound inhibitors were removed, polar hydrogens were added, protonation states were inspected, and receptor/ligands were converted to PDBQT (Eberhardt et al. 2021; Trott and Olson 2010). The binding site was defined to cover the catalytic pocket with a cubic grid (center: x = 15.468, y = 15.468, z = − 12.088; size: 30 × 30 × 30 Å). Docking was performed with AutoDock Vina (Trott & Olson 2010), keeping the grid fixed across ligands; exhaustiveness and a fixed random seed were specified for reproducibility (other settings at default). Predicted poses were ranked by Vina score and visually inspected for chemically reasonable contacts within the active site; the top pose for each ligand was advanced to subsequent analyses.

Molecular dynamics simulation

Docking against AR (PDB 3DN5) identified the best-scoring enzyme–phenolic complexes, which were advanced to 200-ns molecular dynamics simulations in GROMACS 2022.4 to assess stability, key interactions, and conformational changes under near-physiological conditions. (Bauer et al. 2022; Abraham et al. 2015). Before running the simulations, docked inhibitors were carefully extracted from their binding conformations and parameterized to ensure accurate molecular representations. The CHARMM36m force field was utilized to generate interaction parameters, while the modified TIP3P water model from CHARMM simulated the aqueous environment (Bauer et al. 2022; Abraham et al. 2015). Geometries and topologies for the inhibitors were constructed using the CGenFF server and integrated into the topology files of the enzyme–inhibitor complexes. These systems were placed in a dodecahedral simulation box, solvated with TIP3P water molecules, and adjusted to a simulation volume of 378.25 nm³, ensuring adequate space for free molecular motion. Sodium ions were introduced to neutralize the system, ensuring proper electrostatic balance (MacKerell et al. 1998).

Energy minimization was performed using the steepest descent algorithm for 10 picoseconds (ps), effectively resolving unfavorable interactions and removing steric clashes (Hess et al. 2008). This was followed by a two-phase equilibration process. The first phase, conducted under the NVT ensemble, focused on stabilizing temperature and molecular configurations, while the second phase, under the NPT ensemble, equilibrated pressure (Parrinello and Rahman 1981). Both steps were carried out over 100 ps each at a constant temperature of 300 K. After equilibration, a 200-ns production run was performed under constant pressure (1 bar) and temperature (300 K), enabling detailed observation of structural dynamics, binding interactions, and complex stability. Insights derived from these simulations shed light on the molecular mechanisms governing AR inhibition by the phenolics, providing critical information on their potential therapeutic applications.

Free-energy landscape (FEL)

To derive the free-energy landscape (FEL) from the MD trajectory, post-equilibration frames were least square fitted to a reference structure to remove overall translation/rotation, and positional fluctuations of backbone Cα atoms (or all heavy atoms, as specified) were used to build the covariance matrix. Principal component analysis (PCA) was performed to obtain eigenvectors/eigenvalues, and the trajectory was projected onto the first two principal components (PC1, PC2). A 2D probability density P(PC1,PC2) was obtained from a binned histogram (uniform bins; bin width chosen to balance resolution and sampling), and the FEL was computed by Boltzmann inversion, G(PC1,PC2) = − k_B​TlnP(PC1,PC2), referenced to the global minimum and reported in kJ·mol⁻¹ at the simulation temperature. Energy basins, barriers, and putative transition pathways were identified from contour maps, and representative structures were extracted from basin centroids. Convergence/robustness were assessed by cumulative variance of the first PCs and block averaging. All analyses were carried out with GROMACS utilities (e.g., gmx covar, gmx anaeig, gmx sham) and corroborated with in-house Python scripts when needed.

MM/PBSA analysis

MM/PBSA binding free energies (ΔG_total) were computed with gmx_MMPBSA using a single-trajectory protocol: snapshots were extracted every 100 ps from the last 50 ns of each production run (≈500 frames) (Valdés-Tresanco et al. 2021). Complex, receptor, and ligand energies were evaluated with the same force field/topologies as MD. Polar solvation was obtained from the Poisson–Boltzmann model (ε_solvent = 80, ε_solute = 2, ionic strength 0.15 M), and nonpolar solvation from an SASA-based term; unless noted, other parameters were kept at gmx_MMPBSA defaults. When indicated, per-residue energy decomposition was performed. Frame-wise energies were averaged to yield mean ΔG_total (kcal·mol⁻¹), and uncertainties are reported as mean ± SEM (bootstrap resampling). Entropic contributions were not included.

ADME analysis

ADME and drug-likeness properties were predicted with the SwissADME web server using default settings (Daina et al. 2017). For each compound, canonical SMILES were provided (tautomer/ionization standardized to the neutral form), and the following outputs were recorded: Lipinski/Veber/Egan/Muegge rules (and violation counts), bioavailability score, topological polar surface area (TPSA), H-bond donors/acceptors, rotatable bonds, consensus logP (from iLOGP/XlogP3, etc.), ESOL water solubility class, GI absorption and BBB permeability (BOILED-Egg), P-gp substrate status, predicted inhibition of major CYP isoforms (1A2, 2C19, 2C9, 2D6, 3A4), PAINS/Brenk structural alerts, and synthetic accessibility. Results are reported as numerical values with pass/fail annotations, and compounds with multiple red flags were noted as lower-priority leads.

Results and discussion

Phytochemical studies

The ethyl acetate fraction obtained from the methanolic extract of A. fragrantissima underwent multiple sequential chromatographic purification processes utilizing a range of stationary phases. These meticulous separation techniques culminated in the isolation of six distinct phenolic compounds. The structural characterization of these compounds was accomplished through comprehensive spectroscopic analyses, cross-referencing their TLC profiles against authenticated standards, and corroborating the findings with previously published data. Consequently, the identified phenolic compounds (Fig. 1) were determined to be myricetin (1) (Kamel et al.; Kamel et al. 2022), 3,5-dihydroxy-7-methoxyflavone (2) (Mancipe et al. 2023), 5-hydroxy-3,6,7,4′-tetramethoxyflavone (3'-O-methyl-eupatorin) (3) (Sen et al. 2017), 3-caffeoyl-4-feruloylquinic acid (4) (Chen et al. 2022), chlorogenic acid (5) (Alruhaimi et al. 2024), and salvianolic acid B (6) (Ma et al. 2021).

Fig. 1.

Structures of isolated phenolics

Enzyme assays identify salvianolic acid B, chlorogenic acid, and 3-caffeoyl-4-feruloylquinic acid as potent noncompetitive AR inhibitors

The AR inhibitory activity of six A. fragrantissima phenolics was quantified with quercetin as a positive control. Dose–response curves (Fig. 2A) and IC₅₀ values (Fig. 2B) ranked salvianolic acid B as most potent (1.90 ± 0.10 µM), followed by chlorogenic acid (3.79 ± 0.17 µM) and 3-caffeoyl-4-feruloylquinic acid (4.49 ± 0.38 µM); all three were comparable to or better than quercetin (2.24 ± 0.25 µM). Myricetin, 3′-O-methyl-eupatorin, and 3,5-dihydroxy-7-methoxyflavone were weaker (73.33 ± 2.66, 92.27 ± 15.22, and 146.47 ± 27.18 µM, respectively).

Fig. 2.

Inhibitory activity and kinetics of isolated phenolic compounds against AR. A Dose-dependent inhibition curves for phenolic compounds isolated from A. fragrantissima against AR. B IC₅₀ values (µM) of the tested compounds. C–E Lineweaver–Burk plots of C 3-caffeoyl-4-feruloylquinic acid, D chlorogenic acid, and E salvianolic acid B. F–H Michaelis–Menten plots for F 3-caffeoyl-4-feruloylquinic acid, G chlorogenic acid, and H salvianolic acid B. Data are presented as mean ± standard deviation (SD) from triplicate experiments

Given their low IC₅₀ values, the top three compounds were advanced to enzyme kinetic analysis. Lineweaver–Burk plots (Figs. 2C–E) showed families of lines intersecting at the x-axis, and Michaelis–Menten plots (Figs. 2F–H) showed decreased Vmax with essentially unchanged Km—consistent with noncompetitive inhibition. The inhibition constants (Ki) were 2.32 µM (salvianolic acid B), 3.85 µM (chlorogenic acid), and 4.10 µM (3-caffeoyl-4-feruloylquinic acid). Together, these data identify salvianolic acid B as the strongest AR inhibitor, with chlorogenic acid and 3-caffeoyl-4-feruloylquinic acid as additional promising leads for follow-up.

Docking reveals active-site engagement and highest affinity for salvianolic acid B, followed by chlorogenic acid and 3-caffeoyl-4-feruloylquinic acid

Docking placed all three phenolics in the AR active-site cleft (PDB 3DN5), with poses stabilized by an extensive H-bond network to catalytic/recognition residues (e.g., Tyr48, His110, Trp111) and by hydrophobic contacts within the specificity pocket (e.g., Trp20/Trp219, Phe122, Leu300, Cys298) (Figs. 3 and 4). The predicted binding affinities (kcal·mol⁻¹) ranked salvianolic acid B as strongest (− 9.6), followed by chlorogenic acid (− 8.9) and 3-caffeoyl-4-feruloylquinic acid (− 7.8), mirroring the IC₅₀ trend observed experimentally.

Fig. 3.

Molecular docking of 3-caffeoyl-4-feruloylquinic acid, chlorogenic acid, and salvianolic acid B into the binding site of AR. A Ligand positioning within the binding pocket of AR, highlighting distinct binding conformations of the studied phenolic compounds. B Surface representation of AR showing the accommodation of each ligand in the binding site

Fig. 4.

Interaction maps of 3-caffeoyl-4-feruloylquinic acid, chlorogenic acid, and salvianolic acid B with residues in the binding site of AR. Key polar and hydrophobic interactions stabilizing the ligand–protein complexes are depicted

Salvianolic acid B spans the pocket, anchoring through multiple H-bonds to Lys21, Tyr48, His110, Gln49 and contacts with Trp20, Phe122, Leu300, Cys298 (Fig. 4), consistent with its highest score. Chlorogenic acid forms several directional H-bonds with His110, Ile260, Ser210, and Tyr48 and engages nearby aromatics (Trp20/Trp111, Phe122), supporting its intermediate affinity. 3-Caffeoyl-4-feruloylquinic acid also establishes a dense H-bond network (notably with Lys21, Tyr48, Gln49) and hydrophobic contacts (Trp20/Trp111/Trp219, Phe122, Leu300, Cys298), but with a slightly shallower burial, consistent with the modestly weaker score. Overall, the docking poses rationalize inhibition by showing complementary polar and π–π/van der Waals interactions in the catalytic cleft; the stronger-scoring ligands achieve broader pocket coverage and more extensive H-bonding, providing a structural basis for their relative potencies.

Free-energy landscapes (FEL) from PCA of MD trajectories: ligand binding narrows conformational space and stabilizes AR, with the deepest basin for salvianolic acid B

Projection of the 200-ns MD trajectories onto the first two principal components (PC1, PC2) revealed marked differences between free AR and the ligand-bound states (Fig. 5). The free enzyme sampled a broad, rugged landscape with multiple shallow basins, indicating substantial conformational heterogeneity of the active-site cleft. Upon ligand binding, the landscapes collapsed into fewer, deeper minima, signifying restrained motions and stabilization of specific conformers. Among the complexes, salvianolic acid B/AR displayed the deepest and most localized global minimum, chlorogenic acid/AR exhibited an intermediate basin depth/width, and 3-caffeoyl-4-feruloylquinic acid/AR showed a somewhat broader, shallower minimum. This minima depth order mirrors both docking scores and experimental potencies, supporting a consistent structure–activity relationship.

Fig. 5.

Free-energy landscapes of aldose reductase and aldose reductase–ligand complexes. A Two-dimensional FEL maps obtained by projecting MD trajectories onto PC1 and PC2; colors denote relative Gibbs free energy (kJ·mol⁻¹) referenced to the global minimum (cool colors = low energy). From top to bottom: free aldose reductase; 3-caffeoyl-4-feruloylquinic acid-bound complex; chlorogenic acid-bound complex; salvianolic acid B-bound complex. B Corresponding three-dimensional FEL surfaces highlighting basin depth and breadth. Ligand binding reduces conformational heterogeneity and deepens the dominant basin, most prominently with salvianolic acid B

Representative lowest-energy snapshots (Fig. 6) illustrate how each ligand anchors the catalytic pocket while allowing modest breathing motions of loops bordering the site. SalB spans the cleft and forms an extended H-bond/π-stacking network with catalytic/recognition residues (e.g., Lys21, Gln49, Tyr48, His110, Trp111, Phe122, Trp20/Trp219, Cys298, Leu300), consistent with its narrow, deep FEL basin. Chlorogenic acid engages a similar subset of contacts but with slightly reduced burial, while 3-caffeoyl-4-feruloylquinic acid adopts a comparably stable pose with greater side chain plasticity at the pocket edge—features that align with their respective FEL basin widths. Overall, the FEL analysis indicates ligand-induced conformational selection toward stable sub-states, with SalB conferring the greatest dynamic stabilization of AR.

Fig. 6.

Lowest-energy conformations extracted from the free-energy landscapes. Representative snapshots at the global minimum for free aldose reductase and for complexes with 3-caffeoyl-4-feruloylquinic acid, chlorogenic acid, and salvianolic acid B

MM/PBSA binding free energies: salvianolic acid B shows the most favorable ΔG, followed by chlorogenic acid and 3-caffeoyl-4-feruloylquinic acid

Endpoint MM/PBSA over the last 50 ns (Table 1) ranked the complexes salvianolic acid B/aldose reductase (ΔG_total = − 34.44 ± 2.47 kcal mol⁻¹) > chlorogenic acid/aldose reductase (ΔG_total = − 23.90 ± 3.07 kcal mol⁻¹) > 3-caffeoyl-4-feruloylquinic acid/aldose reductase (ΔG_total = − 18.22 ± 2.23 kcal mol⁻¹) (Table 1). The separation between salvianolic acid B and the other ligands exceeds the combined uncertainties, indicating a robust preference for salvianolic acid B binding. In all cases, favorable gas-phase interactions (ΔG_gas = ΔE_vdw + ΔE_ele) drove binding and were partly offset by desolvation penalties (ΔG_solv > 0). Salvianolic acid B exhibited the strongest van der Waals term (− 50.09 ± 1.72) and a sizable electrostatic contribution (− 27.26 ± 1.07), yielding the most favorable ΔG_gas (− 77.35 ± 2.02) despite the largest solvation cost (+ 42.91 ± 1.42). Chlorogenic acid showed intermediate stabilization (ΔE_vdw = − 38.35 ± 1.91; ΔE_ele = − 22.21 ± 5.80), whereas 3-caffeoyl-4-feruloylquinic acid displayed the weakest noncovalent terms and thus the least favorable ΔG_total. These component trends support a binding mechanism dominated by hydrophobic packing and π/dispersion contacts, with specific electrostatics/hydrogen bonding providing additional stabilization—consistent with the docking poses and the deeper FEL basin observed for salvianolic acid B. Because configurational entropy was not included, absolute values should be interpreted cautiously, but the relative ranking is internally consistent with docking affinities and enzyme inhibition data.

Table 1.

MM/PBSA results for aldose reductase complexes over the last 50 ns of MD (kcal mol⁻¹; mean ± SEM). ΔE_vdw and ΔE_ele are van der Waals and electrostatic gas-phase terms; ΔG_gas = ΔE_vdw + ΔE_ele; ΔG_solv is the polar + nonpolar solvation free energy; ΔG_total = ΔG_gas + ΔG_solv. Snapshots were taken every 100 ps using a single-trajectory protocol. Negative ΔG_total indicates favorable binding

Complex	ΔEvdw	ΔEele	ΔGsolv	ΔGgas	ΔGtotal
3-Caffeoyl-4-feruloylquinic acid/AR	− 25.61 ± 0.71	− 27.59 ± 1.67	34.98 ± 0.71	− 53.20 ± 2.12	− 18.22 ± 2.23
Chlorogenic acid/AR	− 38.35 ± 1.91	− 22.21 ± 5.80	36.66 ± 3.92	− 60.56 ± 5.57	− 23.90 ± 3.07
Salvianolic acid B/AR	− 50.09 ± 1.72	− 27.26 ± 1.07	42.91 ± 1.42	− 77.35 ± 2.02	− 34.44 ± 2.47

200-ns classical MD reveals persistent binding for chlorogenic acid and salvianolic acid B, and a mid-trajectory pose transition for 3-caffeoyl-4-feruloylquinic acid

In the ligand-centric RMSD (Fig. 7A), chlorogenic acid and salvianolic acid B fluctuate within a narrow band (< ~ 1 nm) over 200 ns, indicating persistent, pose-stable binding in the catalytic cleft. In contrast, 3-caffeoyl-4-feruloylquinic acid shows a sharp RMSD rise at ~ 50 ns followed by a new plateau, consistent with a major reorientation from the docked pose to an alternative, long-lived configuration rather than progressive drift.

Fig. 7.

Two-hundred–nanosecond MD analysis of aldose reductase and complexes. A Ligand-centric RMSD (nm) for 3-caffeoyl-4-feruloylquinic acid (red), chlorogenic acid (green), and salvianolic acid B (purple) relative to the docked pose. B Enzyme backbone RMSD (nm) for free aldose reductase (black) and each complex (colors as in A). C Radius of gyration, Rg (nm), of aldose reductase in the free state and in complexes. D Solvent-accessible surface area, SASA (nm²), of the protein over time. Curves show stable backbones for all simulations; chlorogenic acid and salvianolic acid B preserve low ligand RMSD throughout, while 3-caffeoyl-4-feruloylquinic acid undergoes a mid-trajectory pose change and then stabilizes

In the enzyme backbone RMSD (Fig. 7B), all systems equilibrate quickly and remain stable (≈0.10–0.25 nm) throughout, demonstrating that ligand binding does not compromise the global fold. The chlorogenic acid complex and the free enzyme exhibit the lowest long-timescale drift, whereas salvianolic acid B shows a modestly higher yet steady RMSD—compatible with localized loop “breathing” around the pocket rather than global instability.

Radius of gyration (Rg) (Fig. 7C) varies within ~ 1.90–1.96 nm for all trajectories, supporting compact, well-folded proteins. The 3-caffeoyl-4-feruloylquinic acid complex tends toward slightly greater compaction, while chlorogenic acid shows the largest average Rg, and salvianolic acid B lies in between. Solvent-accessible surface area (Fig. 7D), SASA, trends parallel Rg: chlorogenic acid and salvianolic acid B increase solvent exposure relative to free AR, consistent with subtle pocket/loop opening upon ligand accommodation; the 3-caffeoyl-4-feruloylquinic acid complex presents the smallest SASA increase. Taken together, the MD metrics show that all complexes are globally stable on the 200-ns timescale. Chlorogenic acid and salvianolic acid B maintain tight, persistent binding, while 3-caffeoyl-4-feruloylquinic acid undergoes an early conformational transition to a new, stable pose. Small, ligand-dependent differences in Rg and SASA reflect local pocket remodeling rather than loss of structural integrity.

3.7 Hydrogen bonding, residue flexibility, and short-range interaction energies from 200-ns MD: salvianolic acid B maintains the strongest, most persistent contacts

In the hydrogen bonding profile (Fig. 8A), the salvianolic acid B complex sustains the highest hydrogen bond occupancy over time (typically 2–6 concurrent bonds), chlorogenic acid sustains intermediate occupancy (≈1–4), and 3-caffeoyl-4-feruloylquinic acid sustains lowest and more intermittent occupancy (often 0–3). The latter also shows a change in pattern around the midpoint of the simulation, consistent with the pose transition seen in the ligand-centric RMSD.

Fig. 8.

Time-resolved interaction analysis over 200-ns MD. A Number of protein–ligand hydrogen bonds vs time for 3-caffeoyl-4-feruloylquinic acid (red), chlorogenic acid (green), and salvianolic acid B (purple). B Per-residue RMSF of aldose reductase for the free enzyme (black) and each complex (colors as in A). C Short-range Coulombic interaction energy (kJ·mol⁻¹; more negative is stronger). D Short-range Lennard–Jones interaction energy (kJ·mol⁻¹). Salvianolic acid B maintains the most favorable electrostatic and van der Waals contacts and the highest hydrogen bond occupancy; 3-caffeoyl-4-feruloylquinic acid shows weaker, more variable interactions with a mid-trajectory reorientation

For the backbone flexibility (Fig. 8B), residue-wise fluctuations of aldose reductase are damped upon ligand binding relative to the free enzyme, particularly across loops bordering the catalytic cleft (e.g., segments around Tyr48/His110–Trp111 and the Cys298–Leu300 region). Salvianolic acid B and chlorogenic acid both reduce peak motions, indicating tighter local stabilization of the active site.

Short-range interaction energies (Fig. 8C–D). Time series for Coulombic (Coul-SR) and Lennard–Jones (LJ-SR) terms show the most negative (favorable) values for salvianolic acid B (≈ − 200 to − 250 kJ·mol⁻¹ for each term), intermediate values for chlorogenic acid (≈ − 100 to − 150 kJ·mol⁻¹), and the weakest, most fluctuating interactions for 3-caffeoyl-4-feruloylquinic acid. The step- change in its energies coincides with the RMSD shift, supporting reorientation to a less strongly packed pose. Together, these metrics indicate that salvianolic acid B forms the most persistent hydrogen bond/dispersion network and confers the greatest local rigidification of aldose reductase, with chlorogenic acid next and 3-caffeoyl-4-feruloylquinic acid weakest. The trends align with the docking scores, free-energy landscape basins, and MM/PBSA ranking.

ADME/drug-likeness profiling indicates chlorogenic acid as the most favorable developability; 3-caffeoyl-4-feruloylquinic acid is borderline; salvianolic acid B faces oral-absorption hurdles

The 3-caffeoyl-4-feruloylquinic acid profile combines high predicted GI absorption with a reasonable bioavailability score (0.55), but it sits just beyond the Lipinski mass cutoff (MW 530.48 g/mol) and is highly polar (TPSA 200.28 Ų) (Table 2). It is non-BBB permeant, not a P-gp substrate, and shows no predicted inhibition of the major CYPs. Its lipophilicity varies by descriptor (consensus Log Pₒ/ 1.703; XLOGP3 6.54) and solubility is moderate (ESOL logS − 3.06). Together, these features suggest oral feasibility with formulation attention and size/polarity constraints. Chlorogenic acid shows the most balanced developability: lower mass (354.31 g/mol), high GI absorption, no BBB permeability, no P-gp or CYP liabilities, and only one Lipinski violation (donors > 5). Its polarity (TPSA 164.75 Ų) and low lipophilicity (Log Pₒ/ − 0.39; XLOGP3 − 0.42) align with its higher predicted solubility (ESOL logS − 1.62). Although its SwissADME bioavailability score is 0.11, the overall profile still favors oral formulation among the three phenolics. Salvianolic acid B presents several oral ADME liabilities: large size (718.61 g/mol), very high polarity (TPSA 278.04 Ų), low predicted GI absorption, low bioavailability (0.11), and poor solubility (ESOL logS − 6.22). It shows no BBB permeability, no P-gp substrate behavior, and no predicted CYP inhibition, but it fails Lipinski on three counts (MW > 500, H-bond acceptors > 10, donors > 5). These features indicate a need for delivery/optimization (e.g., prodrugs or advanced formulations) if oral dosing is intended (Table 2).

Table 2.

Predicted ADME and drug-likeness properties (SwissADME) of phenolics from Achillea fragrantissima

	Bioavailability radar	ADMET properties
3-Caffeoyl-4-feruloylquinic acid		• Size: MW = 530.48 g/mol • Num. H-bond donors:6 • Num. H-bond acceptors: 12 • Flexibility: number of rotatable bonds = 10 • Saturation: fraction Csp3 = 0.27 • TPSA (polarity): 200.28 Å2 • Molar refractivity: 131.37 • Log Po/w = 1.703 • Lipophilicity: XLOGP3 = 6.54 • Solubility: logS (ESOL) = -3.06 • Bioavailability score = 0.55 • BBB permeant: no • CYP1A2(No), CYP2C19(No), CYP2C9 (No), CYP2D6(No), and CYP3A4(No) • GI absorption: high • P-gp substrate: no • Lipinski (druglikeness): yes; 1 violation: MW > 500
Chlorogenic acid		• Size: MW = 354.31 g/mol • Num. H-bond donors:6 • Num. H-bond acceptors: 9 • Flexibility: number of rotatable bonds = 5 • Saturation: fraction Csp3 = 0.38 • TPSA (polarity): 164.75 Å2 • Molar refractivity: 83.50 • Log Po/w = -0.39 • Lipophilicity: XLOGP3 = -0.42 • Solubility: logS (ESOL) = -1.62 • Bioavailability score = 0.11 • BBB permeant: no • Log Kp (skin permeation) = -8.76 cm/s • CYP1A2(No), CYP2C19(No), CYP2C9 (No), CYP2D6(No), and CYP3A4(No) • GI absorption: high • P-gp substrate: no • Lipinski (druglikeness): Yes; 1 violation: NH or OH > 5
Salvianolic acid B		• Size: MW = 718.61 g/mol • Num. H-bond donors:9 • Num. H-bond acceptors: 16 • Flexibility: number of rotatable bonds = 14 • Saturation: fraction Csp3 = 0.17 • TPSA (polarity): 278.04 Å2 • Molar refractivity: 178.07 • Log Po/w = 2.23 • Lipophilicity: XLOGP3 = 3.98 • Solubility: logS (ESOL) = -6.22 • Bioavailability score = 0.11 • BBB permeant: no • Log Kp (skin permeation) = -7.86 cm/s • CYP1A2(No), CYP2C19(No), CYP2C9 (No), CYP2D6(No), and CYP3A4(No) • GI absorption: low • P-gp substrate: no • Lipinski (druglikeness): no; 3 violations: MW > 500, N or O > 10, NH or OH > 5

Overall, chlorogenic acid emerges as the most developable lead (high GI absorption, small size, high solubility), 3-caffeoyl-4-feruloylquinic acid remains a plausible oral candidate with a single mass-related violation, and salvianolic acid B—despite its strongest binding—faces substantial oral bioavailability barriers warranting medicinal chemistry or delivery strategies.

Conclusion

Phenolic constituents from Achillea fragrantissima emerged as bona fide aldose reductase inhibitors with complementary experimental and computational support. In vitro, salvianolic acid B, chlorogenic acid, and 3-caffeoyl-4-feruloylquinic acid inhibited the enzyme with low micromolar IC₅₀ values and displayed noncompetitive kinetics (IC₅₀, Ki), indicating modulation consistent with binding outside the catalytic event—i.e., an allosteric or allosteric-like effect—even though docking located stable poses within the canonical active-site cleft. Molecular docking ranked salvianolic acid B as the strongest binder, followed by chlorogenic acid and 3-caffeoyl-4-feruloylquinic acid; 200-ns molecular dynamics simulations confirmed persistent binding, ligand-dependent pocket remodeling, and reduced backbone flexibility. Free-energy landscapes derived from principal component analysis showed ligand-induced narrowing of conformational space, with the deepest, most localized basin for salvianolic acid B. MM/PBSA over the last 50 ns converged on the same hierarchy of binding free energies, dominated by favorable van der Waals and electrostatic terms offset by desolvation penalties. ADME predictions highlighted a practical trade-off: salvianolic acid B achieves the strongest engagement but faces oral absorption liabilities, whereas chlorogenic acid offers the most balanced developability profile, and 3-caffeoyl-4-feruloylquinic acid an intermediate profile.

Taken together, these data nominate salvianolic acid B as a high-affinity chemical probe for aldose reductase and chlorogenic acid as a tractable lead for oral development, with 3-caffeoyl-4-feruloylquinic acid as an additional candidate. Future work should (i) map the binding determinants experimentally (mutagenesis, crystallography or cryo-EM, and detailed kinetics) to reconcile active-site docking with noncompetitive behavior; (ii) assess selectivity across aldo–keto reductase family members; and (iii) advance pharmacology through permeability-enhancing formulations or prodrug/derivatization strategies—particularly for salvianolic acid B—followed by in vivo pharmacokinetics and efficacy in models of diabetic complications. Overall, Achillea fragrantissima phenolics constitute a credible starting point for developing safer, effective aldose reductase inhibitors.

Supplementary Information

Below is the link to the electronic supplementary material.

Authors contributions

Emadeldin M. Kamel: conceptualization, methodology, project administration, formal analysis, data curation, molecular docking and dynamics analysis, experimental studies, enzyme assay development, validation, visualization, preparation of figures and tables, writing—original draft, writing—review and editing, and supervision. Faris F. Aba Alkhayl and Saleh Alkhedhairi: experimental studies, enzyme assay development, and data interpretation. May Bin-Jumah: experimental studies, enzyme assay development, and data interpretation. Doaa A. Abdelrheem: validation, visualization, and preparation of figures and tables. Ahmed A. Allam: supervision, funding acquisition, critical review of results. Al Mokhtar Lamsabhi: conceptualization, methodology, project administration, formal analysis, funding acquisition, writing—original draft, and supervision.

Funding

This work was carried out with the support from the project PID2023-150717NB-I00 from Ministerio de Ciencia, Innovacion y Universidades in Spain. The authors also express their gratitude to Centro de Computación Científica of the UAM (CCC-UAM) for providing computing time. The authors acknowledge Princess Nourah bint Abdulrahman University Researchers Supporting Project number (PNURSP2025R737), Princess Nourah bint Abdulrahman University, Riyadh, Saudi Arabia.

Data availability

The data supporting the findings in this article are available within this article and its supporting information.

Declarations

Conflict of interests

The authors declare no competing interests.

Ethical approval

Not applicable.

Consent to participate

This research does not contain any studies with human participants performed by any of the authors.