Introduction

Type 2 diabetes mellitus is a prevalent metabolic disorder characterized by insufficient insulin secretion, insulin resistance, and persistent hyperglycemia1,2. Postprandial hyperglycemia, a key contributor to diabetic complications, exerts greater detrimental effects on microvascular and macrovascular systems than fasting hyperglycemia, highlighting the necessity of its effective management in diabetes care3. Carbohydrate absorption in the small intestine relies on four key α-glucosidases: maltase (EC 3.2.1.20) and glucoamylase (EC 3.2.1.3), which form the maltase-glucoamylase complex (MGAM); sucrase (EC 3.2.1.48) and isomaltase (EC 3.2.1.10), which form the sucrase-isomaltase complex4,5,6. These enzymes catalyze the hydrolysis of polysaccharides into monosaccharides (such as glucose and fructose), which are then transported into enterocytes via glucose transporters (GLUTs), primarily sodium-glucose transport protein-1 and GLUT2 and GLUT57,8.

AGIs are oral antidiabetic drugs that reduce postprandial blood glucose and glycated hemoglobin by delaying carbohydrate absorption9,10. Consequently, the development of safer and less toxic AGIs, particularly those derived from natural products, has become a research priority. Numerous active components in natural plants and foods, including anthocyanins, flavonoids, alkaloids, fatty acids, and phenolic compounds, have been identified as effective α-glucosidase inhibitors11,12,13,14,15,16. To date, microbial AGIs have predominantly included glucose analogs and structurally diverse non-sugar compounds. Classical glucose analog AGIs, including acarbose, miglitol, voglibose, and 1-deoxynojirimycin (DNJ), are derived from microbial metabolites produced by actinomycetes or streptomycetes17,18,19,20,21. Acarbose significantly lowers blood glucose and HbA1c levels but is associated with side effects like bloating and diarrhea22,23. DNJ, initially isolated from mulberry root bark, is now produced via fermentation by actinomycetes, bacilli, and other microorganisms, highlighting its potential as a drug candidate24,25,26. Additionally, other AGI compounds without sugar structures were also discovered, such as CKD-711 and CKD-711a (IC₅₀ values of 2.5 and 6.5 µg/mL, respectively)27, genistein has been reported as a non-competitive inhibitor28, and HGA (hydroxyglutaric acid), isolated from the fermentation products of Paenibacillus sp. TKU042, which exhibits α-glucosidase inhibitory activity (IC₅₀ = 220 µg/mL) stronger than acarbose29.

Microorganisms capable of producing AGIs include bacteria and fungi, such as Paenibacillus, Streptomyces, Aspergillus, and Penicillium species30,31,32. Paenibacillus is a genus of facultatively anaerobic, endospore-forming Gram-positive bacteria widely distributed in the environment and known for producing diverse bioactive compounds with potential applications33,34. Previous studies have demonstrated that Paenibacillus species can produce a variety of bioactive metabolites, including exopolysaccharides35,36, antimicrobial peptides (e.g., paenibacillin, polymyxin, bacitracin, paenibacilysin, and fusaricidin)37,38,39,40,41, and degradative enzymes like cellulases and xylanases42,43,44. They also possess nitrogen-fixing abilities and plant growth-promoting effects, often being developed as agricultural microbial inoculants45,46. Furthermore, Paenibacillus has strong metabolic activity and can degrade complex polysaccharides such as cellulose and chitin, making it valuable for biomass conversion and bioenergy production47,48. It has also been reported to synthesize various natural products with antibacterial and antitumor activities49,50,51,52.

This study aims to identify and characterize α-glucosidase inhibitors produced by Paenibacillus sp. JNUCC 31. Whole-genome sequencing and annotation were first performed to predict biosynthetic gene clusters. Subsequently, compounds were isolated and identified from microbial cultivation extracts, which were then subjected to bioactivity screening to assess their inhibitory effects. Enzyme kinetics analyses were conducted to elucidate the mechanism of action. Furthermore, molecular docking and MD simulations were employed to reveal the detailed interaction mechanisms between the active compounds and α-glucosidase. Together, these findings provide foundational evidence for the potential of microbial α-glucosidase inhibitors and offer insights for developing new treatments for type 2 diabetes.

Materials and methods

Bacterial isolation and genomic analysis of strain JNUCC 31

Paenibacillus sp. JNUCC 31 was isolated in September 2019 from soil collected at Baengnokdam, the summit crater lake of Mount Halla, Jeju Island, Republic of Korea. For isolation, 0.5 g of soil was suspended in 0.45 mL of 0.1% (w/v) Tris buffer (pH 7.5), shaken at 180 rpm at 30 °C for 1 h, diluted stepwise (10⁻⁵ to 10⁻⁹), and spread onto MRS agar plates. The strain was routinely cultured aerobically at 30 °C for 24 h on Luria–Bertani (LB) agar plates or in LB broth, and preserved at − 80 °C in 20% (v/v) glycerol. Genomic DNA was extracted from colonies using the QIAGEN Genomic-tip Kit (Qiagen Inc., Shenzhen, China). Whole-genome sequencing was performed using PacBio RS II and Illumina platforms (Macrogen Inc., Seoul, Republic of Korea). Functional annotation of the assembled genome was conducted using the Clusters of Orthologous Groups (COG), Gene Ontology (GO), and Kyoto Encyclopedia of Genes and Genomes (KEGG) databases.

GC-MS analysis of cellular fatty acids in strain JNUCC 31

The cellular fatty acid composition of strain JNUCC 31 was determined following the method of Miller53,54. Approximately 40 mg of cultured cells were transferred to a Teflon-lined screw-cap tube and mixed with 1 mL of 15% NaOH in 50% methanol. The mixture was heated at 100 °C for 30 min and then cooled to room temperature. Subsequently, 2 mL of methanolic HCl (a mixture of 6.0 N HCl and methanol, 325:275 mL) was added and heated at 80 °C for 10 min. After rapid cooling, 1.25 mL of hexane/methyl tert-butyl ether (1:1, v/v) was added and shaken for 10 min. After phase separation, the lower aqueous layer was removed. Then, 3 mL of dilute NaOH (10.8 g NaOH in 900 mL deionized water) was added, mixed for 10 min, and the upper organic layer was collected into a screw-cap vial (12 × 32 mm, Agilent Technologies) for analysis. Samples were analyzed using an Agilent 6890 gas chromatograph equipped with a HP-1 column (30 m × 0.320 mm × 0.25 μm). Fatty acid profiles were processed using Sherlock MIS software (version 6.3). Peaks were identified by comparison with a standard calibration mixture, and retention times, peak areas, and relative abundances were recorded.

Fermentation, extraction, and purification procedures

Paenibacillus sp. JNUCC 31 was initially cultured in 125 mL of LB broth in a 500 mL Erlenmeyer flask at 30 °C for 48 h with shaking at 180 rpm. A 5% (v/v) inoculum was then transferred into four 5 L Erlenmeyer flasks, each containing 1.0 L of LB broth, and incubated at 30 °C for 3 days under aerobic conditions. The culture broth (4 L) was filtered, and the filtrate extracted with methanol to afford a crude extract (800 mg). The extract was partitioned with n-hexane (3 × 4 L) and ethyl acetate (EtOAc) (3 × 4 L), yielding n-hexane (100 mg) and EtOAc (600 mg) fractions. The EtOAc fraction was subjected to vacuum liquid chromatography (VLC) on silica gel, eluted with a CHCl₃–MeOH gradient (300 mL per step), to yield five fractions (V1–V5) (Table S1, Fig. S1). Fractions V1, V2, V4, and V5 were further purified by silica gel column chromatography using CHCl₃–MeOH solvent systems, yielding adenosine (11.8 mg, from V1, 100:1, v/v), uridine (9.5 mg, from V4, 50:1, v/v), 4-hydroxybenzaldehyde (9.6 mg, from V2, 50:1, v/v; RT: 18.068 min), dibutyl phthalate (DBP, 6.8 mg, from V2, 50:1, v/v; RT: 47.438 min), and 1-acetyl-β-carboline (8.0 mg, from V5, 50:1, v/v; RT: 39.287 min).

Adenosine, (Fig. S2, 3)55: white solid; 1H NMR (400 MHz, DMSO-D6) δ 8.35 (s, 1H, H-2), 8.13 (s, 1H, H-8), 7.38 (s, 2H, NH2), 5.87 (d, J = 6.2 Hz, 1H, H-1’), 5.51–5.41 (m, 2 H, 6’−8’-OH), 5.22 (d, J = 4.5 Hz, 1H, 7’-OH), 4.61 (td, J = 6.2, 4.9 Hz, 1H, H-2’), 4.14 (td, J = 4.7, 2.9 Hz, 1H, H-3’), 3.96 (q, J = 3.4 Hz, 1H, H-4’), 3.67 (dt, J = 12.0, 4.0 Hz, 1H, H-5’a), 3.55 (ddd, J = 12.1, 7.2, 3.6 Hz, 1H, H-5’b); 13C NMR (101 MHz, DMSO-D6) δ 156.72 (C-6), 152.93 (C-2), 149.58 (C-4), 140.50 (C-8), 119.90 (C-5), 88.43 (C-1’), 86.44 (C-4’), 73.95 (C-2’), 71.22 (C-3’), 62.22 (C-5’).

Uridine, (Fig. S4, 5)56: white solid; 1H NMR (400 MHz, DMSO-D6) δ 11.32 (s, 1H, NH), 7.89 (d, J = 8.1 Hz, 1H, H-6), 5.78 (d, J = 5.4 Hz, 1H, H-1’), 5.65 (d, J = 8.0 Hz, 1H, H-5), 5.39 (d, J = 5.7 Hz, 1H, OH), 5.10 (t, J = 5.0 Hz, 2 H, OH), 4.02 (q, J = 5.3 Hz, 1H, H-2’), 3.96 (q, J = 4.1 Hz, 1H, H-3’), 3.84 (q, J = 3.4 Hz, 1H, H-4’),3.62 (ddd, J = 12.0, 5.0, 3.3 Hz, 1H, H-5’), 3.54 (ddd, J = 12.0, 4.9, 3.3 Hz, 1H, H-5); 13C NMR (101 MHz, DMSO-D6) δ 163.04(C-4), 150.65(C-2), 140.62(C-6), 101.64(C-5), 87.52(C-1’), 84.70(C-4’), 73.42(C-3’), 69.76(C-2’), 60.71(C-5’).

4-hydroxybenzaldehyde, (Fig. S6, 7, 8)57: light yellow solid; 1H NMR (400 MHz, CHLOROFORM-D) δ 9.86 (s, 1H, CHO), 7.85–7.79 (m, 2 H, 2–6-OH), 7.02–6.94 (m, 2 H, 3–5-OH); 13C NMR (101 MHz, CHLOROFORM-D) δ 191.45 (C-7), 161.70 (C-4), 132.70 (C-2, C-6), 129.98 (C-1), 116.16 (C-3, C-5).

Dibutyl phthalate, (Fig. S9, 10, 11)58: white solid; 1H NMR (400 MHz, CHLOROFORM-D) δ 7.76–7.67 (m, 2H, H-3 and H-4), 7.57–7.48 (m, 2H, H-2 and H-5), 4.30 (t, J = 6.7 Hz, 4H, CH2−2’ and CH2−2’’), 1.77–1.65 (m, 4 H, CH2−3’ and CH2−3’’), 1.51–1.37 (m, 4 H, CH2−4’ and CH2−4’’), 0.96 (t, J = 7.4 Hz, 6 H, CH3−5’ and CH3−5’’); 13C NMR (101 MHz, CHLOROFORM-D) δ 167.87 (C-1’ and C-1’’), 132.41 (C-1 and C-6), 131.07 (C-3 and C-4), 128.97 (C-2 and C-5), 65.71 (C-2’ and C-2’’), 30.68 (C-3’ and C-3’’), 19.31 (C-4’ and C-4’’), 13.87 (C-5’ and C-5’’).

1-acetyl-β-carboline, (Fig. S12, 13, 14)59: light yellow solid; 1H NMR (400 MHz, PYRIDINE-D5) δ 8.65 (dd, J = 4.9, 1.4 Hz, H-10), 8.34–8.32 (m, H-9), 8.31–8.28 (m, H-6), 7.89 (dt, J = 8.3, 1.1 Hz, H-3), 7.67 (tt, J = 8.3, 1.3 Hz, H-2), 7.40 (ddd, J = 8.1, 7.0, 1.2 Hz, H-1), 2.92 (d, J = 1.5 Hz, CH3−13); 13C NMR (101 MHz, PYRIDINE-D5) δ 201.86 (C-12), 142.58 (C-11), 137.88 (C-10), 136.99 (C-4), 133.32 (C-7), 131.70 (C-8), 129.27 (C-5), 122.06 (C-2), 121.00 (C-6), 120.59 (C-1), 119.25 (C-9), 112.76 (C-3), 25.83 (C-13).

α-glucosidase inhibitory assay and kinetic analysis

The α-glucosidase inhibitory activities of the crude extracts, isolated compounds, and acarbose were evaluated using a p-nitrophenyl-α-D-glucopyranoside (p-NPG)-based colorimetric assay60. A substrate solution (1.5 mM p-NPG for inhibition assay) and α-glucosidase solution (750 mU/mL, from Saccharomyces cerevisiae) were prepared in 0.1 M phosphate-buffered saline (PBS, pH 7.4). The crude extract and isolated compounds were dissolved in DMSO, acarbose in distilled water, and all diluted to a serial of concentrations for assays. In 96-well plates, 20 µL of sample or acarbose was mixed with 100 µL of p-NPG solution. The reaction was initiated by adding 20 µL of α-glucosidase, followed by incubation at 37 °C for 10 min. The reaction was stopped by adding 60 µL of 1 M Na₂CO₃. Absorbance was measured at 405 nm using a microplate reader (Tecan, Switzerland). Acarbose served as the positive control. Inhibition (%) was calculated as:

$$\% {\rm{ inhibition }} = [1-({\rm{O}}{{\rm{D}}_{{\rm{compound}}}}{\rm{-O}}{{\rm{D}}_{{\rm{blank}}}})/({\rm{O}}{{\rm{D}}_{{\rm{control}}}}{\rm{-O}}{{\rm{D}}_{{\rm{blank}}}})] \times 100$$

To elucidate the inhibitory mechanisms of the active compounds against α-glucosidase, enzyme kinetic assays were conducted using varying concentrations of the substrate p-NPG (0.3125, 0.625, 1.25, and 2.5 mM) in the presence of different concentrations of inhibitors: adenosine (0, 31.2, 62.5, and 125 µM), uridine (0, 125, 250, and 500 µM), and DBP (0, 500, 1000, and 2000 µM), acarbose (0, 11.62, 23.24, and 46.47 mM). Kinetic parameters were analyzed by constructing Lineweaver–Burk double-reciprocal plots. The inhibition constants (Ki and Kis) were determined from secondary plots of the slopes and y-intercepts of the Lineweaver–Burk plots versus inhibitor concentration, respectively.

Pharmacokinetics, drug-likeness, and toxicity assessment

Pharmacokinetic parameters were evaluated by predicting the pharmacokinetics and drug-likeness properties using the compounds’ SMILES information. This analysis was conducted comprehensively using several models, including ADMETlab 3.061, SwissADME62, pkCSM63.

Molecular docking simulation

The crystal structures of human maltase-glucoamylase (MAGM; PDB ID: 2QMJ) and microbial isomaltase from Saccharomyces cerevisiae (PDB ID: 3A4A) were downloaded from the RCSB Protein Data Bank (https://www.rcsb.org/). Protein preparation was conducted using PyMOL v3.0.3, involving removal of water molecules and heteroatoms, followed by addition of hydrogens. The ligand structures were retrieved from the PubChem database (https://pubchem.ncbi.nlm.nih.gov/) and energy-minimized using the MMFF94 force field implemented in Open Babel v2.4.1. Ligand preparation including protonation state assignment and rotatable bond definition was performed using AutoDock Tools v1.5.6. Docking grids were centered on the binding sites defined by the co-crystallized ligands, with grid box dimensions set to 28.0 × 28.0 × 28.0 Å. The grid centers were located at coordinates (−20.4, −6.2, −2.6) for MAGM and (21.3, −5.3, 22.2) for isomaltase. Molecular docking was performed using AutoDock Vina v1.2.0 in semi-flexible mode with an exhaustiveness of 25, employing the Lamarckian Genetic Algorithm. The docking protocol was validated by re-docking the co-crystallized ligands into their respective binding sites, with RMSD values below 2.0 Å, indicating acceptable accuracy64.

Molecular dynamics simulations

All MD simulations were performed using GROMACS 2021. AMBER14SB and GAFF2 were used as force fields for the protein and ligand, respectively. The solvated system was built with TIP3P water in a cubic box with at least 1.2 nm between the solute and the nearest box edge, and Na⁺/Cl⁻ ions were added to neutralize the system. Energy minimization was conducted using the steepest descent algorithm for a maximum of 50,000 steps, with a tolerance set at 100 kJ·mol⁻¹·nm⁻¹. The minimized system was subjected to a two-stage equilibration process. The first stage involved a 1 ns NVT ensemble simulation using the V-rescale (T = 310 K, τ = 0.1 ps)65 with separate temperature coupling groups for (i) the protein and ligand, and (ii) the water and ions. This was followed by a 1 ns NPT equilibration using the V-rescale (T = 310 K, τ = 0.5 ps) and C-rescale (P = 1 atm, τ = 1 ps)66, retaining the same temperature coupling groups. During both equilibration phases, harmonic restraints (force constant: 1000 kJ·mol⁻¹·nm⁻²) were applied to all heavy atoms of the protein and ligand to maintain structural stability. Following equilibration, all systems were simulated in the NPT ensemble with periodic boundary conditions, using identical parameters and without applying harmonic restraints. Long-range electrostatic interactions were treated using the Particle Mesh Ewald (PME) method67. A 1.2 nm cutoff was applied for short-range electrostatics and van der Waals interactions. Covalent bonds involving hydrogen atoms were constrained using LINCS68, allowing a 2 fs time step. For each protein-ligand complex, simulations were performed in triplicate to ensure reproducibility and reliability of the results.

Protein dynamics analysis

Protein dynamics were analyzed using principal component analysis (PCA) and dynamic cross-correlation matrix (DCCM) calculations with the Bio3D package69 in R Studio. These analyses aimed to identify the dominant modes of motion within the simulations, providing insights into the system’s biological function.

Calculations of MM/GBSA and residue energy analysis

Binding free energies were calculated using the MM/GBSA method via gmx_MM/PBSA70, based on the last 20 ns of stable MD trajectories. The total binding energy was decomposed into gas-phase (van der Waals and electrostatic) and solvation contributions, including polar solvation energy (EGB) calculated by the Generalized Born model and nonpolar solvation energy (ESURF) estimated from solvent-accessible surface area (SASA). Residue energy decomposition was performed for residues within 4 Å of the ligand to identify key contributors to binding.

Results and discussion

General features of strain JNUCC 31 genome

Whole-genome sequencing of strain JNUCC 31 yielded 7,551,121 bp of high-quality sequences with a GC content of 46.4%. The sequences were assembled into a complete circular chromosome. A total of 103 tRNA genes and 33 rRNA genes (11 copies each of 5 S, 16 S, and 23 S) were identified in the genome. The complete genome sequence has been deposited in the NCBI RefSeq database under accession number NZ_CP062165.1. Phylogenomic analysis based on single-copy orthologous genes revealed that strain JNUCC 31 is closely related to Paenibacillus gallinarum, indicating a close evolutionary relationship within the genus Paenibacillus (Fig. 1).

Fig. 1
figure 1

Phylogenomic tree of Paenibacillus sp. JNUCC 31 and related strains. The tree was constructed based on the concatenated sequences of 670 single-copy orthologous genes identified from the whole-genome sequences of 20 Paenibacillus strains. Phylogenetic inference was performed using the maximum likelihood method implemented in IQ-TREE v2.2.2.7 with 1,000 bootstrap replicates. Bootstrap support values are shown at the nodes.

Full size image

Genome annotation revealed a total of 6,700 genes, including 6,458 protein-coding sequences (CDSs), covering 96.4% of the total genome. Functional annotation against the COGs, GO, and KEGG databases identified 5,407, 2,172, and 2,327 genes, respectively. Twenty-three functional classes were revealed by genome functional annotation of strain JNUCC 31 against the COGs database (Fig. 2a). Among the Clusters of Orthologous Groups (COG) categories in JNUCC 31, nine had the largest proportions (each with ≥ 5% of the total COG classifications): E (amino acid transport and metabolism, 420 ORFs, 7.77%), G (carbohydrate transport and metabolism, 894 ORFs, 16.53%), H (coenzyme transport and metabolism, 286 ORFs, 5.29%), J (translation, ribosomal structure, and biogenesis, 314 ORFs, 5.81%), K (transcription, 743 ORFs, 13.74%), M (cell wall/membrane/envelope biogenesis, 288 ORFs, 5.33%), P (inorganic ion transport and metabolism, 498 ORFs, 9.21%), R(general function prediction only, 540 ORFs, 9.99%), T(signal transduction mechanisms, 498 ORFs, 9.21%).

GO analysis suggested that the biological process related genes (gene number: 1110) were the most abundant in strain JNUCC 31, followed by molecular function (gene number: 614) and cellular component (gene number: 483) (Fig. 2b). Among the sub-functions annotated by GO analysis, metabolic process (gene number: 376) and cellular process (gene number: 426) were dominant in biological process category, while catalytic activity (gene number: 318) and binding (gene number: 152) were the core functions in molecular function category, cellular anatomical entity (gene number: 404) in cellular component category.

According to the KEGG database (Fig. 2c), the genome of strain JNUCC 31 harbors a diverse set of functional genes involved in various metabolic pathways, including amino acid metabolism (215 genes), carbohydrate metabolism (299 genes), cofactor and vitamin metabolism (184 genes), and energy metabolism (127 genes). Additionally, 59 genes were annotated under the category biosynthesis of other secondary metabolites, covering critical metabolic pathways related to the synthesis of various secondary metabolites (Table S2). These pathways include the biosynthesis of phenylpropanoids (ko00940), flavonoids (ko00941), anthocyanins (ko00942), isoflavonoids (ko00943), multiple indole alkaloids (ko00960), and several antibiotics such as penicillin (ko00330), cephalosporins (ko00331), and aminoglycosides (ko00524). Furthermore, the annotation revealed that the strain possesses pathways for caffeine metabolism (ko00232), benzylisoquinoline alkaloids (ko00960), and the biosynthesis of other secondary metabolites like acarbose (ko00647) and statosporine (ko01042). Among the annotated biosynthetic pathways, flavonoids, isoflavonoids, anthocyanins, phenylpropanoids, and acarbose represent compound classes previously reported to possess α-glucosidase inhibitory activity71,72, suggesting that strain JNUCC 31 harbors the genetic potential to produce bioactive metabolites with antidiabetic properties. These KEGG annotation results provide valuable insights into the metabolic potential of this strain and its prospective applications in natural product synthesis.

Fig. 2
figure 2

Functional annotation of the genome of Paenibacillus sp. JNUCC 31. (a) COG classification; (b) GO categories; (c) KEGG pathways.

Full size image

Identification and analysis of secondary metabolites

Fatty acids are known for their α-glucosidase inhibitory activity and potential antidiabetic properties. Both saturated fatty acids (e.g., palmitic acid and stearic acid) and unsaturated fatty acids (e.g., oleic acid, linoleic acid, linolenic acid, arachidonic acid, and eicosapentaenoic acid), as well as hydroxylated derivatives such as 10-hydroxystearic acid and 10-hydroxy-cis-12-octadecenoic acid, have demonstrated varying inhibitory effects73,74. To explore microbial sources of such bioactive compounds, the fatty acid profile of strain JNUCC 31 cultured on LB agar plates was analyzed using GC-MS. The major components identified were lauric acid (C12:0, 3.73%, RT 1.65 min), anteiso-pentadecanoic acid (anteiso-C15:0, 57.32%, RT 2.40 min), iso-palmitic acid (iso-C16:0, 9.85%, RT 2.70 min), palmitic acid (C16:0, 7.94%, RT 2.80 min), and anteiso-heptadecanoic acid (anteiso-C17:0, 9.85%, RT 3.00 min). (Figure S15, Table 1). For further secondary metabolite analysis, the EtOAc fraction was purified, affording five known compounds: adenosine, uridine, 4-hydroxybenzaldehyde, DBP, and 1-acetyl-β-carboline.

Table 1 Identification of cellular fatty acids in strain JNUCC 31 by GC-MS.
Full size table

α-glucosidase inhibitory activity assay

Previous studies have reported α-glucosidase inhibitory activity in fatty acid-rich extracts from Streptomyces, largely attributed to branched-chain fatty acids. Among these, anteiso-C15:0 was identified as the predominant active component, along with iso-C14:0, iso-C15:0, iso-C16:0, palmitic acid (C16:0), iso-C17:0, anteiso-C17:0, and stearic acid (C17:0)75.

Lauric acid, myristic acid, palmitic acid, and stearic acid, identified in strain JNUCC31, have been reported to possess α-glucosidase inhibitory activity, with studies indicating that certain fatty acids strongly inhibit α-glucosidase but weakly affect α-amylase, suggesting they may reduce colonic fermentation-related side effects compared to acarbose, which inhibits both enzymes73,76.

In this study, the methanol extract of strain JNUCC 31 and its n-hexane and ethyl acetate fractions were tested for α-glucosidase inhibitory activity at 10 mg/mL, acarbose (10 mg/mL) was used as the positive control, and LB medium (25 mg/mL), which corresponds to the strain cultivation concentration, was used as the blank control. As shown in Table 2, the ethyl acetate extract exhibited the strongest inhibition (52.4 ± 0.7%), followed by the n-hexane (30.0 ± 1.8%) and methanol (28.8 ± 1.7%) extracts. All were less active than acarbose (80.2 ± 0.7%). The EtOAc extract, enriched in α-glucosidase inhibitory constituents, was further subjected to isolation, yielding three compounds (adenosine, uridine, and DBP), all of which exhibited α-glucosidase inhibitory activity, consistent with previous reports77,78,79,80. Adenosine and uridine have been reported to modulate glucose metabolism through insulin-related pathways, supporting their potential roles in diabetes and metabolic regulation81,82. In contrast, compounds 4-hydroxybenzaldehyde and 1-acetyl-β-carboline exhibited no inhibitory activity.

Table 2 α-Glucosidase inhibitory effects of strain JNUCC 31 extracts (n = 3, mean ± SD).
Full size table

Kinetic analysis of α-glucosidase Inhibition

Several fatty acids, including palmitic acid, oleic acid, linoleic acid, α-linolenic acid, eicosapentaenoic acid, n-hexadecanoic acid, and arachidonic acid, have been reported to inhibit α-glucosidase via mixed-type mechanisms83,84,85. In contrast, acarbose is a competitive inhibitor of α-glucosidase86. Enzyme kinetic analysis using Lineweaver–Burk plots was performed to clarify the inhibition mechanisms87,88,89 of the isolated compounds. For adenosine (Fig. 3a), the enzyme displayed Km and Vmax values of 1.477 mM and 0.1 µM/min, respectively, under conditions without inhibitors. Similarly, for DBP (Fig. 3c), the corresponding Km and Vmax were 1.359 mM and 0.128 µM/min. With increasing inhibitor concentrations, both compounds induced a decrease in Vmax accompanied by an increase in Km, consistent with a mixed-type inhibition mechanism. Secondary plot analysis revealed that Ki values were lower than Kis values (adenosine: Ki = 90.88 µM, Kis = 116 µM; DBP: Ki = 516.22 µM, Kis = 2533.45 µM), indicating that both inhibitors preferentially bind to the free enzyme rather than the enzyme–substrate complex. In the absence of inhibitors, uridine and acarbose (Fig. 3b, d) exhibited Km values of 4.86 mM and 1.478 mM, respectively. Enzyme kinetic analysis demonstrated that increasing inhibitor concentrations led to Lineweaver–Burk plots intersecting near the y-axis, with Vmax remaining essentially constant and Km significantly increasing. These results are characteristic of competitive inhibition, indicating that both compounds compete with the substrate for the enzyme’s active site, thereby reducing catalytic activity. Secondary plot analysis yielded Ki values of 153.35 µM for uridine and 15.78 mM for acarbose, confirming their competitive binding mechanism.

Fig. 3
figure 3

Lineweaver–Burk plots illustrating the inhibition kinetics of isolated compounds against α-glucosidase. (a) adenosine; (b) uridine; (c) DBP; (d) acarbose.

Full size image

Pharmacokinetics, drug-likeness, and toxicity assessment

In this study, the nucleoside compounds adenosine and uridine exhibited favorable toxicological profiles, with no predicted Ames mutagenicity, hepatotoxicity, hERG inhibition, or major CYP450 inhibition, suggesting a low risk of drug–drug interactions (Table S3). However, both compounds demonstrated limited intestinal absorption, with predicted human intestinal absorption (HIA) values of 61.24% and 47.45%, respectively, and low Caco-2 permeability (adenosine: − 2.346 cm/s; uridine: − 0.161 cm/s), indicating restricted oral bioavailability. Their low plasma protein binding (PPB) rates (adenosine: 18.8%; uridine: 13.8%) suggest a higher proportion of free drug in systemic circulation, which may facilitate rapid distribution and onset of action. In contrast, the fatty acids palmitic acid (C16:0) and anteiso-C15:0 demonstrated excellent absorption potential (HIA > 92%; Caco-2 permeability ~ 1.56–1.58 cm/s) and were predicted to cross the blood–brain barrier (BBB), indicating potential central nervous system (CNS) activity. Both compounds exhibited very high PPB (> 98%), which may prolong their systemic residence time. Metabolic predictions indicated potential CYP-mediated drug interactions, as C16:0 inhibited CYP1A2 and CYP2C9, while anteiso-C15:0 inhibited CYP1A2. Their clearance rates were relatively rapid (1.54–1.76 mL/min/kg) with moderate half-lives (0.61–0.73 h), suggesting suitability for development as short-acting CNS or metabolic modulators. DBP showed favorable pharmacokinetics predictions, including high HIA (95.04%), good Caco-2 permeability (1.622 cm/s), BBB permeability, and high PPB (94.1%), indicating promising oral pharmacokinetic characteristics.

Drug-likeness profiling (Table S4) revealed that adenosine and uridine possess high topological polar surface area (TPSA: 140.3 and 124.8 Ų, respectively), abundant hydrogen bond donors and acceptors (adenosine: HBD 5, HBA 9; uridine: HBD 4, HBA 8), and low LogP values (–2.028 and − 1.709), reflecting their high hydrophilicity. While both met most of Lipinski’s criteria, they violated Ghose and Muegge filters due to excessive polarity and hydrogen bonding, which likely limit membrane permeability and oral bioavailability. In contrast, DBP and the fatty acids had moderate molecular weights (228–278 g/mol), low TPSA (37.3–52.6 Ų), and high LogP values (4.3–6.7), indicating strong lipophilicity. DBP fully satisfied multiple drug-likeness filters, suggesting a favorable lead-like structure. The fatty acids, while highly lipophilic, did not meet all rules (e.g., due to a higher number of rotatable bonds), which may affect conformational stability and absorption. Acarbose, a known non-systemic intestinal α-glucosidase inhibitor, deviated substantially from drug-likeness criteria (MW: 645 g/mol; TPSA: 329 Ų; HBD 14; HBA 19; LogP: − 3.109), consistent with its poor oral absorption and local gut-targeted action.

In summary, nucleoside compounds exhibit limited oral bioavailability due to high polarity and hydrogen bonding capacity. DBP and fatty acid derivatives possess favorable drug-likeness and oral pharmacokinetic properties; However, due to the potential toxicological concerns associated with DBP, particularly its effects on endocrine function and development, further pharmaceutical investigation should be guided by careful risk–benefit evaluation90.

Molecular docking analysis

To further rationalize the inhibitory mechanisms indicated by enzyme kinetics studies, molecular docking simulations were conducted. The binding affinities and interaction modes of acarbose, adenosine, uridine, DBP, and major fatty acids were evaluated against MGAM and isomaltase (Fig. 4, S16). The molecular docking methodology applied to MGAM has been validated in prior studies60. For isomaltase, re-docking was performed using the co-crystallized ligand α-D-glucopyranose, yielding an RMSD of 0.84 Å (Figure S17), which is below the generally accepted threshold of 2.0 Å, indicating the reliability of the docking strategy.

All compounds were successfully docked into the catalytic pockets of the α-glucosidases. Adenosine exhibited slightly higher binding affinity toward isomaltase (− 7.5 kcal/mol) compared to MGAM (− 6.8 kcal/mol). Uridine showed binding energies of − 7.0 kcal/mol and − 7.2 kcal/mol for MGAM and isomaltase, respectively, forming stable hydrogen bonding (green dash line) networks. The ribose moiety in both adenosine and uridine appeared to play a crucial role in enhancing ligand–target binding affinity and specificity. DBP interacted with isomaltase predominantly through hydrophobic contacts (− 7.0 kcal/mol), whereas its binding to MGAM was weaker (− 5.8 kcal/mol). Although fatty acids (e.g., anteiso-C15:0 and C16:0) formed hydrogen bonds with key residues, their simple, hydrophobic structures with few polar or functional groups likely resulted in a limited number of interactions within the active site, leading to reduced binding affinity.

In this molecular docking study of MGAM91, several key residues involved in ligand binding were identified. Asp327, located in the (β/α)₈ barrel, frequently formed hydrogen bonds at the − 1 subsite, while Trp406 (Insert 1) contributed to ligand recognition at the + 1 subsite through hydrogen bonding and hydrophobic interactions, including π–π stacking. Additionally, the catalytic nucleophile Asp443, the putative acid/base catalyst Asp542, and Arg526 formed hydrogen bonds with the ligand. Tyr299 and Phe575 further enhanced binding through hydrophobic and aromatic interactions. Isomaltase features a typical (β/α)₈ barrel with catalytic residues Asp215, Glu277, and Asp352 at the core of an active site formed by domains A and B, with a narrow entrance defined by Tyr158, His280, and the 310–315 loop92. Docking results showed that ligands frequently interact with Glu277, Asp352, Gln279, and Arg442 via hydrogen bonding, supporting their roles in catalysis and substrate stabilization. Tyr158, Phe303, and Phe178 contribute to ligand binding via hydrophobic interactions and π–π stacking.

Integrating molecular docking with enzymatic kinetics, the results indicate that competitive inhibitors (e.g., acarbose and uridine) predominantly bind to the catalytic active site of α-glucosidases. In MGAM, they interact notably with hydrophilic catalytic residues Asp327 and Asp542, while in isomaltase, binding is concentrated on polar residues such as Arg442, Glu277, and Asp352. These inhibitors exert their effect by occupying the substrate-binding pocket, thereby blocking substrate access to the active site. In contrast, mixed-type inhibitors (e.g., adenosine, DBP, and palmitic acid) exhibit a preference for hydrophobic regions surrounding the catalytic core. In MGAM, they commonly engage residues such as His600, Phe575, whereas in isomaltase, frequent interactions are observed with Phe303, Tyr158. These ligands likely modulate enzyme activity by inducing conformational changes or interfering with substrate entry pathways. Notably, residues such as Trp406 in MGAM and Arg442 and Phe303 in isomaltase are recurrently involved in ligand interactions across various inhibitor types, suggesting structural conservation and functional versatility in substrate recognition and inhibitor binding. Collectively, the study reveals both overlapping and distinct binding patterns between competitive and mixed-type inhibitors, offering critical structural insights into the inhibitory mechanisms of α-glucosidases and informing structure-based inhibitor design.

Fig. 4
figure 4

Molecular docking analysis of bioactive compounds with α-glucosidase. Visualized binding interactions of (a, f) adenosine, (b, g) uridine, (c, h) DBP, (d, i) C16:0, and (e, j) anteiso-C15:0 with MGAM (a–e) and isomaltase (f–j), respectively.

Full size image

Molecular dynamics simulations

RMSD analysis

Root mean square deviation (RMSD) analysis was conducted to assess the structural stability of the ligand–protein complexes involving MAGM and isomaltase throughout MD simulations (Table S5). The average RMSD values calculated for the backbone atoms of the unbound proteins were 1.774 ± 0.086 Å for MAGM and 1.526 ± 0.103 Å for isomaltase, indicating that both proteins maintained stable conformations throughout the simulation. Upon ligand binding, the average RMSD values for the MAGM–ligand complexes ranged from 1.6 to 1.9 Å (Fig. 5a), whereas those for the isomaltase–ligand complexes displayed a broader range, from 1.8 to 2.6 Å (Fig. 5b), These results indicate that the MAGM–ligand complexes exhibited slightly greater conformational stability than the isomaltase–ligand systems. Consistently low RMSD values and relatively small structural fluctuations observed across all systems indicate the structural stability of the protein–ligand complexes throughout the simulations.

Fig. 5
figure 5

RMSD analysis of protein–ligand complexes. (a) MGAM–ligand complexes; (b) Isomaltase–ligand complexes.

Full size image

RMSF analysis

Root mean square fluctuation (RMSF) values indicate protein backbone flexibility during simulations, with higher values reflecting increased residue mobility and potentially reduced stability at the ligand–protein interface. RMSF analysis of the MAGM–ligand complexes revealed that the catalytic region insert 1 loop (red dashed line) exhibited two moderate fluctuations, with amplitudes of approximately 7 Å and 4 Å, respectively (Fig. 6a). In contrast, the insert 2 loop (green dashed line) exhibited a minor fluctuation of around 2 Å, indicating lower flexibility. These values were comparable to those observed in the MAGM and the acarbose-bound complex. This suggests that, despite the inherent flexibility of the catalytic loops, ligand binding did not significantly increase the structural dynamics of these regions. For the isomaltase–ligand complexes, the majority of residues showed RMSF values below 5 Å, indicating generally low conformational flexibility (Fig. 6b). Ligand binding induced fluctuations in specific regions within domain A (green dashed line) and domain B (red dashed line), particularly at the active site entrance formed by Tyr158, His280, and loop 310–315, which were comparable to those observed in the ligand-free system. Due to the narrow conformation of the isomaltase active site entrance, ligand-induced increased rigidity in this region is likely to further restrict substrate access, especially for longer oligosaccharides, thereby potentially reducing substrate affinity and catalytic efficiency92,93. Overall, for both systems, the RMSF values of all backbone residues remained below 10 Å, indicating maintained conformational stability throughout the simulations and supporting the potential inhibitory mechanisms of the studied ligands.

Fig. 6
figure 6

RMSF analysis of protein–ligand complexes. (a) MGAM–ligand complexes; (b) Isomaltase–ligand complexes.

Full size image

Radius of gyration (Rg) analysis

To evaluate the global structural compactness and stability of the protein–ligand complexes during MD simulations, the average Rg values were calculated based on the backbone atoms. As shown in Fig. 7 and Table S6, The average Rg values of all MGAM–ligand complexes ranged from 28.42 Å to 28.53 Å, while isomaltase–ligand complexes showed slightly lower values, ranging from 24.25 Å to 24.70 Å. Minor fluctuations (within 1 Å) in the Rg values were observed throughout the 100 ns simulations for both systems, indicating that each maintained a compact structure with only modest opening and closing motions occurring in the N- and C-terminal regions.

Fig. 7
figure 7

Rg analysis of protein–ligand complexes. (a) MGAM–ligand complexes; (b) Isomaltase–ligand complexes.

Full size image

Protein-ligand H-bonds analysis

To elucidate the contribution of hydrogen bonds to complex stability, we analyzed the number of hydrogen bonds formed between each ligand and the proteins throughout the MD simulations (Figure S18). For the MAGM–ligand complexes, adenosine, uridine, and anteiso-C15:0 were each observed to form two stable hydrogen bonds with the target protein during the simulation trajectory. In contrast, DBP and C16:0 each established a single hydrogen bond, while acarbose exhibited stronger interaction, forming four stable hydrogen bonds with the protein. In the case of the isomaltase–ligand complexes, adenosine, uridine, and C16:0 each formed two hydrogen bonds, while anteiso-C15:0 was involved in the formation of one hydrogen bond. DBP showed weak hydrogen bonding with isomaltase, indicating minimal contribution of hydrogen bonds to complex stability. Collectively, MD simulations showed that adenosine, uridine, C16:0, and anteiso-C15:0 formed stable hydrogen bonds with both proteins, whereas DBP formed weak bonds only with isomaltase.

Average center-of-mass (COM) distance

To evaluate ligand binding stability, the COM distances between each ligand and the proteins were calculated throughout the MD simulations (Table S7). For MGAM–ligand complexes (Fig. 8a), the distances ranged from 21.6 to 32.4 Å, while for isomaltase–ligand complexes (Fig. 8b), they ranged from 9.8 to 12.0 Å. Both systems showed low COM fluctuations, indicating stable ligand–protein binding during the simulations. To further assess positional stability, five representative snapshots were extracted at different time points (Figure S19). All ligands remained consistently near the catalytic pocket with low COM fluctuation (< 10 Å). Among the ligands, adenosine exhibited comparatively greater flexibility, shifting from catalytic domain insert 1 (before 40 ns) to insert 2 over time in the MGAM complex, and displaying notable internal movement within the active site cavity of isomaltase. These observations collectively confirm the stable binding behavior of the ligands and support the reliability of the predicted binding modes derived from docking and MD simulations.

Fig. 8
figure 8

Center-of-mass distance analysis of protein–ligand complexes. (a) MGAM–ligand complexes; (b) Isomaltase–ligand complexes.

Full size image

Principal component analysis (PCA)

To assess the impact of ligand binding on protein conformational dynamics, PCA was performed on the simulation trajectories. The cumulative variance explained by the first three principal components (PC1–PC3) was used to evaluate the dominant collective motions in each system (Figure S20, Table 3). For the MAGM–ligand complexes, the cumulative variance ranged from 23.5% to 33.7%, indicating relatively stable conformational behavior among different ligands. In contrast, the isomaltase–ligand complexes showed higher conformational diversity. The uridine- and DBP-bound isomaltase systems exhibited significantly higher conformational flexibility (both exceeding 56%), suggesting enhanced global motions. In contrast, systems with other ligands showed moderate cumulative variances (28.8–34.9%), while the ligand-free system displayed the lowest value (25.1%), indicating reduced conformational flexibility. PCA revealed distinct conformational dynamics between MGAM and isomaltase, with ligand-specific effects on flexibility, supporting further binding mechanism studies.

Table 3 Cumulative variance (%) of the first three principal components (PC1–PC3) from PCA of MGAM and isomaltase complexes.
Full size table

DCCM analysis

To elucidate the impact of ligand binding on the correlated motions of internal residues, DCCM analysis was performed on the complexes (Figure S21). For the MAGM–ligand complexes, compared to the ligand-free complex, all complexes exhibited significant positive and negative correlated motions within the active site region (residues 300–500), indicating enhanced cooperative dynamics among residues in this region. For the isomaltase–ligand complexes, overall residue motion correlations were markedly higher than those observed in MGAM complexes. Positive correlated motions were predominantly concentrated within the active site region (residues 100–250), with fatty acid ligands inducing especially strong inter-residue correlations. Overall, DCCM analysis shows that ligand binding enhances correlated motions within the active sites of both MGAM and isomaltase.

MM/GBSA binding free energy analysis

MD simulations showed that all complexes maintained high conformational stability. MM/GBSA calculations evaluated ligand–protein binding affinities (Table 4). Binding free energies of adenosine and uridine ranged from − 14.18 to − 28.63 kcal/mol. DBP exhibited average binding energies of − 31.00 kcal/mol (MGAM) and − 27.74 kcal/mol (isomaltase). The two fatty acids also showed strong affinities in both systems, with values between − 28.00 and − 36.50 kcal/mol. Detailed energy components for each complex are listed in Table S8 (MGAM) and Table S9 (isomaltase), covering van der Waals, electrostatic, and solvation energies. Overall, the compounds demonstrated strong binding stability and inhibitory potential, suggesting their promise as AGIs.

Table 4 Calculated MM/GBSA binding free energies of potential α-glucosidase inhibitors.
Full size table

Residue energy decomposition analysis further identified key residues contributing to complex stability. In MGAM (Table 5), the aromatic residue Trp406 located on the insert 1 loop consistently exhibited significant energy contributions for all high-affinity ligands, indicating its role as a core anchoring site for ligand binding. Trp406 cooperates with residues Arg334, Met444, Ser448, Phe450, Phe575, and others to form a stable interaction network, probably through hydrogen bonding or hydrophobic stacking. Our previous study showed that Trp406 contributed relatively weak energy in the acarbose–MGAM complex60, likely due to acarbose’s bulky and highly polar structure (e.g., multiple hydroxyl groups) limiting interactions with Trp406’s aromatic side chain. Moreover, acarbose binding may not induce significant conformational rearrangement of the insert 1 loop, reducing Trp406’s involvement in ligand stabilization. These results suggest that Trp406 preferentially stabilizes smaller or more hydrophobic ligands. Fatty acids possess a simple and highly flexible structure, primarily composed of a long hydrophobic chain and a single carboxyl group, allowing for multiple conformational changes within the binding pocket. Their interactions with proteins largely rely on hydrophobic contacts and limited polar interactions. Compared to semi-flexible docking, MD simulations combined with MM/GBSA account for molecular flexibility and solvent effects, providing a more realistic assessment of binding.

In isomaltase (Table 6), Tyr158 and Gln279 were identified as key residues with significant energy contributions. Tyr158, located at the active site entrance, plays a particularly critical role by collaborating with Phe178, Gln279, His280, Phe303, Phe314, Arg315, and others to form a stable network of interactions that enhances binding stability. In docking poses, DBP tends to remain near the active site entrance, allowing direct interactions with Tyr158 and Gln279, while other compounds may bind deeper or in shifted positions, lacking direct interaction. As docking reflects only a static conformation, MD simulations further revealed consistently high energy contributions from Tyr158 and Gln279, underscoring their roles in complex stability and their relevance for inhibitor design.

Table 5 Residue energy decomposition of MGAM–ligand complexes.
Full size table
Table 6 Residue energy decomposition of isomaltase–ligand complexes.
Full size table

Conclusions

In this study, we conducted a comprehensive genomic and molecular characterization of Paenibacillus sp. JNUCC 31, highlighting its biosynthetic capabilities and α-glucosidase inhibitory activity. Whole-genome sequencing identified a circular chromosome with 6,458 protein-coding genes and diverse metabolic functions. Functional annotation (COG, GO, KEGG) revealed capacities in carbohydrate and amino acid metabolism, secondary metabolism, and enzyme regulation. Importantly, GC-MS analysis and compound isolation identified several bioactive molecules, including adenosine, uridine, DBP, various fatty acids, and others, many of which demonstrated α-glucosidase inhibitory activity in vitro. Mechanistic investigations using enzyme kinetics and molecular docking demonstrated that the active compounds bound stably to α-glucosidases (MGAM and isomaltase), exhibiting favorable affinities and consistent interactions with key catalytic residues. MD simulations further confirmed complex stability and revealed enhanced cooperative motions upon ligand binding, particularly involving Trp406 in MGAM and Tyr158/Gln279 in isomaltase, which function as critical anchoring residues within the active sites. Although DBP demonstrated α-glucosidase inhibitory activity and potential applications in this study, its known endocrine and developmental toxicity risks suggest it may serve as a lead scaffold for mechanistic studies or structural optimization. This study presents genomic and molecular insights into the α-glucosidase inhibitory potential of strain JNUfCC 31, laying a foundation for its application in antidiabetic drug discovery.