Computational identification of multi-target natural compounds from Sesbania grandiflora as potential therapeutic agents against Klebsiella pneumoniae

Abstract

Klebsiella pneumoniae (K. pneumoniae) is a Gram-negative bacterium that causes severe community- and hospital-acquired infections. Its rising multidrug resistance complicates therapy, highlighting the need for novel drugs with broad-spectrum, multi-target potential. Leveraging the traditional use and therapeutic evidence of Sesbania grandiflora, this study performed structure-based computational screening of its phytochemicals against K. pneumoniae targets. Initially, 93 proteins with high annotation scores and resolved X-ray structures were identified. Six key therapeutic targets, including LpxH, fabG, KPC-2, GlmU, chbG, and ompA, were prioritized for their pathogenic role. Molecular docking revealed that 59 of 73 compounds interacted with all six targets with high affinity, while the remaining 14 compounds interacted with five targets. Network pharmacology indicated KPC-2, fabG, and ompA had the highest connectivity (73 compounds), followed by chbG and LpxH (72), and GlmU (61). ‘3′,6-di-O-feruloylsucrose’ had the strongest affinity for ompA, LpxH, and GlmU, while ‘Acarbose hydrate’ ranked top for chbG, fabG, and KPC-2. Out of 47 drug-like compounds, 9 passed ADMET filters. Sonchuionoside A was selected for molecular dynamics simulations, demonstrating stable binding to all targets. This suggests S. grandiflora phytoconstituents as multi-target regulators against K. pneumoniae and highlights Sonchuionoside A as a promising lead for further validation.

Supplementary Information

The online version contains supplementary material available at 10.1038/s41598-026-37613-9.

Introduction

K. pneumoniae, a clinically significant opportunistic Gram-negative, non-motile, encapsulated bacterium of the Enterobacteriaceae family, is a major cause of both community-acquired and nosocomial infections, ranging from necrotizing pneumonia, urinary tract infections, pharyngitis, periodontitis, sepsis, liver abscesses, to meningitis. The bacterium mainly affects immunocompromised individuals and is one of the leading causes of healthcare-associated infections^1,2. Due to the rapidly evolving drug resistance, K. pneumoniae poses significant challenge on disease management. Carbapenem-resistant K. pneumoniae (CRKP) prevalence is 3–7% in Europe and the United and is around 20% in Asia and Latin America, with mortality rates reaching ~ 50% in bloodstream infections³. K. pneumoniae exhibits antimicrobial resistance through diverse molecular mechanisms causing significant challenges to antibiotic efficacy and treatments. Production of β-lactamases, overexpression of efflux pumps, decreased outer membrane permeability, mutations in RNA polymerase (RpoB), and biofilm formation, collectively reduce the efficacy of its conventional therapeutic approaches^4,5. Carbapenems, including meropenem and imipenem, were long considered the mainstay of treatment, but increasing resistance has led to the global spread of Carbapenems resistant traits. Among the therapeutics, aminoglycosides (e.g., Gentamicin, Amikacin) demonstrated partial efficacy, particularly in combination⁴. Multi-target directed antibiotics could be a promising approach to combat these, as these agents reduce the likelihood of emerging resistance, by simultaneously regulating multiple pathways associated with disease progression. Unlike single-target drugs, resistance to multi-target agents would require concurrent mutations in multiple genes, a statistically less probable event.

Plant-derived compounds are increasingly recognized for their multi-target antimicrobial properties, such as disruption of cell membrane integrity, inhibition of biofilm formation, and interference with resistance pathways. It has to be noted that, phytochemicals including phenolics (e.g., gallates, ferulates), terpenoids (e.g., thymol, carvacrol), organosulfur compounds, and alkaloids have demonstrated broad-spectrum activity against multidrug-resistant pathogens^6,7. Several anti-microbial agents have been identified from medicinal plants, such as allicin from Allium sativum (garlic), and curcumin from Curcuma longa⁸. Given the growing crisis of multidrug resistance in K. pneumoniae, phytochemicals from medicinal plants with multi-target therapeutic profiles may be considered as a promising strategy for the development of novel therapeutics with promising pharmacological profiles. The medicinal plant, Sesbania grandiflora, commonly known as the vegetable hummingbird, have a long history of traditional use, and exhibit diverse medicinal properties, including antimicrobial, antioxidant, anti-inflammatory, and anticancer activities⁹. Different solvent-based extracts from its leaves, flowers, and bark have demonstrated inhibitory effects against different human pathogens, including Staphylococcus aureus, Vibrio cholerae, and Escherichia coli^10,11. In view of its promising antimicrobial potential, the present study employed a comprehensive in-silico approach to identify potential lead compounds with multi-target regulatory effects against K. pneumoniae. Initially, the key molecular targets of K. pneumoniae implicated in multiple pathways of disease pathogenesis were identified using proteome data analysis. Further, a structure-based virtual screening approach was employed to evaluate the binding affinity of phytochemicals from the plant against the identified key molecular targets of the pathogen. A compound-target interaction network was constructed to monitor the multi-targeting potential of the screened compounds. The promising hits were subsequently filtered based on pharmacokinetic properties including oral bioavailability and toxicity. Finally, molecular dynamics (MD) simulations were carried out on the promising candidates to validate its binding stability with the selected targets.

Results

Target identification

Using the defined selection strategy and filtering criteria, the complete proteome of K. pneumoniae from the UniProt database (Taxonomic ID: 573) were retrieved. From this dataset, we identified a total of 93 proteins that met the criteria of having a UniProt annotation score ≥ 3 and possessing experimentally resolved three-dimensional structures, specifically determined through X-ray crystallography. Subsequent literature-based functional analysis was performed to evaluate the relevance of these proteins as potential drug targets, focusing on their essentiality, involvement in critical bacterial pathways, virulence, and previous reports of druggability. This analysis led to the identification of six high-priority therapeutic targets. These include enzymes involved in lipid A biosynthesis (LpxH), fatty acid metabolism (fabG), peptidoglycan and amino sugar biosynthesis (GlmU), antibiotic resistance (KPC-2), chitin degradation (chbG), and a periplasmic domain of the Outer Membrane Protein A (OmpA). Details of these proteins, including their UniProt accession numbers, resolved PDB structures, and crystallographic resolution, are summarized in Table 1.

Table 1.

Selected protein targets of Klebsiella pneumoniae retrieved from UniProt.

NO.	Protein name	Gene symbol	UniProtID	PDB/AlphaFold ID	Resolution (Å)
1.	UDP-2,3-diacylglucosamine pyrophosphate hydrolase	LpxH	A6T5R0	6PJ3	1.85 Å
2.	3-oxoacyl-[acyl-carrier-protein] reductase	fabG	W9B6I8	6T77	1.75 Å
3.	Carbapenem-hydrolyzing beta-lactamase KPC-2	KPC-2	Q9F663	6QWC	1.30 Å
4.	Bifunctional protein GlmU	GlmU	A6TG34	AF-A6TG34-F1-v4
5.	Chitooligosaccharide deacetylase	chbG	A6T7U7	AF-A6T7U7-F1-v4
6.	Outer membrane protein A	OmpA	P24017	5NHX	1.95 Å

Molecular Docking and network Pharmacology

From the docking analysis, it was observed that the majority of the screened compounds were able to interact with all the selected 6 target proteins with favourable binding modes. The target-ligand interaction network for the phytochemicals against selected target is depicted in Fig. 1. Based on the docking results, phytocompounds interacting with each target of K. pneumoniae were included for the network.

Fig. 1.

Interaction network of Sesbania grandiflora phytochemicals with K. pneumoniae targets based on molecular docking. The network consists of 79 nodes, including 6 protein targets and 73 phytochemicals, connected by 424 edges representing predicted interactions. Central magenta-colored nodes represent the protein targets (denoted by gene symbols), while blue-colored nodes on the left (LHS) correspond to phytochemicals interacting with all six targets, and blue-colored nodes on the right (RHS) denote phytochemicals interacting with five targets. Node size is proportional to degree centrality, highlighting highly connected targets and ligands. The network reveals that certain phytochemicals can simultaneously target multiple K. pneumoniae proteins, suggesting potential multi-target therapeutic efficacy. The highly connected phytochemicals could serve as lead compounds for broad-spectrum inhibition.

As illustrated in Fig. 1. the network consists of 79 nodes (6 targets, and 73 phytochemicals) and 424 edges, where the edges encode interaction and the nodes represent potential targets and selected ligands. Degree centrality was prioritized to identify highly connected protein targets and phytochemicals with potential multi-target therapeutic relevance. The degree centrality of the targets and phytochemicals are represented in Table 2.

Table 2.

The degree of centrality of selected targets of K. pneumoniae and phytochemicals of Sesbania grandiflora.

Targets		Degree
	KPC-2, fabG, OmpA	73
	chbG, LpxH	72
	GlmU	61
Ligands	7855, 69,663, 572,621, 556,368, 10,329, 31,337, 14,186,645, 543,960, 10,080,425, 237,332, 136,810, 892, 64,689, 531,884, 10,441,857, 6,421,196, 82,076,714, 531,954, 90,232, 689,043, 91,705,314, 445,858, 91,700,532, 6,426,041, 5,280,462, 570,777, 14,038,380, 91,708,329, 73,062, 51,136,538, 594,771, 64,971, 273,591,684, 336,327, 91,700,443, 22,327, 88,881, 10,494, 91,740,724, 170,570, 5,319,706, 128,861, 5,320,946, 101,729, 286,761, 5,280,863, 472,422,923, 531,404, 5,280,633, 5,280,343, 5,282,102, 5,281,116, 9,986,191, 274,632,687, 1,794,427, 5,318,767, 482,028,337, 41,774, 169,552,964	6
	6560, 3,032,731, 7362, 8369, 10,914, 69,505, 579,280, 100,332, 370, 244,581, 532,328, 615,116, 5,280,805, 6,436,722	5

Degree centralities of the selected K. pneumoniae protein targets and Sesbania grandiflora phytochemicals were calculated using Cytoscape. Protein targets are denoted by their gene symbols, while phytochemicals are denoted by their PubChem IDs.

From the docking results, it was noted that the mean docking scores ranged from 82.25 to 91.55 for the selected targets. The mean docking score for each target and the number of compounds that exhibited docking scores higher than the mean values are represented as bar plot (Fig. 2) and the number of overlapping compounds across the targets are showed in Fig. 3.

Fig. 2.

Docking score analysis of Sesbania grandiflora phytocompounds against six K. pneumoniae protein targets. (a) Bar plot showing the distribution of LibDock scores for all phytochemicals across each protein target. The mean docking score for each target is indicated, along with the number of compounds exceeding the mean, highlighting ligands with above-average predicted binding affinity. (b) Venn diagram illustrating the overlap of top-binding phytochemicals (those exceeding mean docking scores) across all six targets. Twenty-two compounds were shared among all targets, indicating potential broad-spectrum inhibitory activity. This analysis identifies lead phytocompounds with strong and consistent predicted binding across multiple K. pneumoniae targets, supporting their prioritization for multi-target therapeutic evaluation.

Fig. 3.

Heatmap visualization of docking scores for high-affinity Sesbania grandiflora phytocompounds across six K. pneumoniae protein targets. The heatmap displays LibDock scores of the 22 phytochemicals identified as top binders across all six targets (from Fig. 2b). Rows represent phytochemicals (PubChem IDs) and columns represent protein targets (gene symbols). Color intensity corresponds to docking score, with darker shades indicating higher predicted binding affinity. The heatmap highlights distinct binding patterns of individual phytochemicals, allowing identification of compounds with consistently strong interactions across multiple targets, as well as subtle target-specific variations in binding strength. This visualization supports the prioritization of phytocompounds with broad-spectrum and potentially multi-target inhibitory activity.

From the Fig. 2(b), it is evident that 22 compounds were shared across all targets, demonstrating higher binding affinity towards each of them than the mean docking score. Distinct binding trends of these phytochemicals based on docking scores towards the target proteins are shown as heatmap (Fig. 3). The complete docking score distribution used for compound prioritization is provided in Supplementary Table S3. The PubChem IDs along with their corresponding SMILES strings for the selected phytochemicals are listed in Supplementary Table 4.

Drug-likeness assessment

Lipinski filter showed that out of 73, 47 compounds passed the criteria to act as orally bioavailable drugs. These compounds were further screened for their ADMET profiles. The compounds that met ADMET criteria is listed in Table 3.

Table 3.

Selected phytochemicals of Sesbania grandiflora that passed ADMET filter.

Compounds PC IDs	Name	SMILES	Solubility	BBB		CYP2D6	Hepatotoxicity	Absorption
100,332	Loliolide	C[C@@]12 C[C@H](CC(C1 = CC(= O)O2)(C)C)O	3	3		FALSE	FALSE	0
445,858	Ferulic acid	COC1 = C(C = CC(= C1)/C = C/C(= O)O)O	4	3		FALSE	FALSE	0
689,043	Caffeic acid	C1 = CC(= C(C = C1/C = C/C(= O)O)O)O	4	3		FALSE	FALSE	0
5,280,462	Vomifoliol	CC1 = CC(= O)CC([C@]1(/C = C/[C@@H](C)O)O)(C)C	4	3		FALSE	FALSE	0
1\0080425	(2E)−4-Amino-5-methyl-2-hexen-1-ol	CC(C)[C@@H](/C = C/CO)N	4	3		FALSE	FALSE	0
51,136,538	(3 S,5R,6R,7E)−3,5,6-Trihydroxy-7-megastigmen-9-one	CC(= O)/C = C/C1(C(CC(CC1(C)O)O)(C)C)O	4	3		FALSE	FALSE	0
91,708,329	Glutaric acid, 3-oxobut-2-yl propyl ester	CCCOC(= O)CCCC(= O)OC(C)C(= O)C	4	3		FALSE	FALSE	0
274,632,687	Sonchuionoside A	CC1 = C(C(CC(C1O[C@H]2[C@@H]([C@H]([C@@H]([C@H](O2)CO)O)O)O)O)(C)C)/C = C/C(= O)C	4	4		FALSE	FALSE	1
286,761	2-Acetylflexuosin A	CC1CC2C(C(C3(C1C(CC3O)OC(= O)C)C)OC(= O)C)C(= C)C(= O)O2	3	3		FALSE	FALSE	0

Phytocompounds (with corresponding PubChem CIDs) that satisfied key absorption, distribution, metabolism, excretion, and toxicity (ADMET) criteria. Parameters reported include predicted aqueous solubility (0 = extremely low, 1 = low but possible, 2 = possible, 3 = good, 4 = optimal, 5 = overly soluble), blood–brain barrier (BBB) penetration level (0 = very high penetrant, 1 = high penetrant, 2 = medium penetrant, 3 = penetrant, 4 = undefined), CYP2D6 inhibition (TRUE/FALSE), hepatotoxicity (TRUE/FALSE), and human intestinal absorption (0 = good, 1 = moderate, 2 = poor, 3 = very poor). These categorical scales reflect the internal standards of the Discovery Studio ADMET models.

From the overall filtering criteria, it is evident that, 9 compounds emerged as promising candidates in terms of target affinity, oral bioavailability and toxicity. Of these, the compound, Sonchuionoside A that demonstrated higher docking score toward all selected K. pneumoniae targets was further evaluated for binding stability with all the six targets using molecular dynamics simulations. The binding modes of Sonchuionoside A with the targets, along with their interacting residues, are illustrated in Fig. 4.

Fig. 4.

Binding modes of Sonchuionoside A with selected K. pneumoniae protein targets. (a: LpxH, b: fabG, c: KPC-2, d: GlmU, e: chbG, f: OmpA). For each target, the left panel (LHS) shows the overall binding pose of Sonchuionoside A within the active site of the protein (depicted using the respective PDB structure), while the right panel (RHS) highlights key interactions such as hydrogen bonds, hydrophobic contacts, π–π stacking, and electrostatic interactions with specific amino acid residues. This visualization reveals how Sonchuionoside A achieves strong binding affinity across all six targets, identifying critical residues that stabilize the ligand and contributing to its potential as a multi-target therapeutic agent against K. pneumoniae.

Molecular dynamics

The structural stability, residual fluctuations, compactness, and interactions of six K. pneumoniae protein targets (OmpA, LpxH, KPC-2, FabG, ChbG, and GlmU) were evaluated through 300-ns MD simulations in both the apo (maroon) and Son.A-bound (yellow) states. The RMSD and RMSF plots for the target-ligand complexes are shown in Figs. 5 and 6, respectively. The results of Rg, SASA, and hydrogen-bond analyses are presented in Figs. 7 and 8, and Fig. 9, respectively.

Fig. 5.

RMSD trajectories of apo proteins and protein-Sonchuinoside A complexes. Root Mean Square Deviation (RMSD) profiles of backbone atoms for six Klebsiella pneumoniae protein targets in their apo states (maroon) and Sonchuinoside A (Son.A)-bound forms (yellow) over 300 ns. The plots illustrate that Son.A binding generally maintains or improves structural stability, with complexes showing either similar or reduced RMSD relative to apo structures. (a) OmpA maintains low RMSD (~ 0.10–0.20 nm) in both states, indicating a highly stable fold, (b) LpxH displays broader fluctuations, yet the Son.A complex consistently deviates less than the apo form. (c) KPC-2 shows stable behavior with tight RMSD clustering (~ 0.10–0.40 nm), (d) FabG apo exhibits higher fluctuations, whereas FabG-Son.A achieves improved stability, (e) ChbG-Son.A shows lower RMSD spread, indicating stabilizing ligand interactions, (f) GlmU-Son.A displays localized deviations but maintains low global RMSD, reflecting preservation of its core structural architecture.

Fig. 6.

Residue-level flexibility analysis using RMSF. Root Mean Square Fluctuation (RMSF) of Cα atoms for apo proteins (maroon) and Son.A-bound complexes (yellow), reflecting residue-wise mobility during the simulation. The data reveal that Son.A binding does not induce destabilizing flexibility; instead, several proteins exhibit reduced mobility in functional or catalytic regions. (a) OmpA shows low flexibility with minor loop fluctuations, (b) LpxH displays expected terminal mobility, while mid-region fluctuations remain comparable between states, (c) KPC-2 maintains uniformly low fluctuations across residues, (d) FabG shows reduced flexibility in catalytically relevant segments upon Son.A binding, (e) ChbG maintains similar fluctuation patterns, with slight dynamical modulation in loop regions and near His125 and (f) GlmU–Son.A exhibits slightly higher localized fluctuations, consistent with ligand-induced adjustments in flexible loop domains while preserving global structural integrity.

Fig. 7.

Compactness analysis via radius of gyration (Rg). Radius of gyration trajectories for apo proteins (maroon) and Son.A complexes (yellow) over 300 ns. Rg values (nm) assess overall structural compactness. Across all systems, Rg profiles remain stable, indicating no unfolding events. Son.A binding frequently produces slightly lower or more stable Rg values, reflecting improved packing or tighter structural organization. (a) OmpA shows consistent compactness (~ 1.50–1.55 nm), (b) LpxH exhibits moderate fluctuations but maintains overall structural integrity, (c) KPC-2 remains tightly compact throughout the simulation, (d) FabG-Son.A shows slightly enhanced compactness relative to the apo form, (e) ChbG displays stable Rg with mild reductions upon ligand binding and (f) GlmU, as the largest protein, shows higher Rg (~ 3.0–3.4 nm), yet the Son.A complex demonstrates a narrower and more stable distribution.

Fig. 8.

Solvent Accessible Surface Area (SASA) profiles of apo and Son.A-bound proteins. SASA trajectories (nm²) of proteins in apo form (maroon) and in complex with Son.A (yellow). SASA reflects solvent exposure and surface reorganization during MD. Overall, Son.A induces subtle but meaningful reshaping of solvent-exposed regions without triggering any destabilizing expansion. (a) OmpA maintains stable SASA (~ 70–80 nm²), indicating preserved surface architecture, (b) LpxH shows slight SASA reduction upon binding, consistent with hydrophobic pocket engagement, (c) KPC-2 remains stable with minor fluctuations around 115–130 nm², (d) FabG displays modest SASA elevation early in the simulation before stabilizing, (e) ChbG shows gradual SASA increases followed by equilibration and (f) GlmU exhibits stable SASA despite its size, with the complex showing slightly more consistent surface exposure.

Fig. 9.

Hydrogen-bond dynamics in Sonchuinoside A complexes. Time evolution of hydrogen bonds between Son.A and each protein target over 300 ns (yellow). The number of H-bonds reflects binding persistence and polar complementarity. Interaction patterns show that Son.A forms stable, target-specific hydrogen-bond networks, supporting persistent engagement across diverse protein folds. (a) OmpA-Son.A maintains 1–4 sustained H-bonds, with transient peaks up to six, (b) LpxH-Son.A shows increased H-bonding after 150 ns, forming 2–5 sustained bonds, (c) KPC-2-Son.A exhibits intermittent 0–3 H-bonds, reflecting moderate but persistent interactions, (d) FabG-Son.A forms the highest density of interactions (3–6 H-bonds), with peaks up to eight, (e) ChbG-Son.A displays moderately stable bonding (1–4 H-bonds) and (f) GlmU-Son.A shows fewer but consistent interactions (0–2 H-bonds), consistent with binding largely driven by hydrophobic or shape complementarity rather than polar contacts.

To quantitatively evaluate structural stability, mean backbone RMSD and RMSF values were calculated over the final 200 ns of the 300-ns trajectories, corresponding to the converged phase of each simulation. All systems attained convergence within ~ 75–100 ns based on RMSD plateau behavior. The RMSD (mean ± SD) values for the Son. A–bound and apo forms, respectively, were: OmpA (0.14 ± 0.01 nm) and OmpA apo (0.15 ± 0.02 nm); LpxH (0.50 ± 0.11 nm) and LpxH apo (0.52 ± 0.11 nm); KPC-2 (0.3 ± 0.04 nm) and KPC-2 apo (0.24 ± 0.06 nm); FabG (0.27 ± 0.03 nm) and FabG apo (0.25 ± 0.02 nm); ChbG (0.29 ± 0.05 nm) and ChbG apo (0.23 ± 0.03 nm); GlmU (0.63 ± 0.09 nm) and GlmU apo (0.82 ± 0.02 nm). Across all systems, RMSD trajectories revealed that ligand binding did not induce destabilization but instead contributed to improved or comparable structural stability (Fig. 5). For OmpA, both apo and Son.A-bound forms stabilized between ~ 0.10–0.20 nm. OmpA displayed low RMSD values, with the Son.A complex maintaining a compact conformation throughout the simulation. LpxH displayed higher intrinsic flexibility (0.20–0.80 nm). Although the LpxH-Son.A complex showed wide RMSD deviations without clear convergence, it still demonstrated lower fluctuations than the apo protein. KPC-2 exhibited stable RMSD profiles in both trajectories (~ 0.10–0.40 nm), achieving stability after 100 ns and remaining stable and unaffected by ligand binding. The FabG apo structure showed larger fluctuations after 150 ns (~ 0.30–0.45 nm) and did not achieve convergence throughout the simulation. Whereas the FabG-Son.A complex remained comparatively stable with lower overall deviation, reflecting stabilizing interactions. ChbG also demonstrated reduced fluctuations upon ligand binding, with the complex showing a narrower RMSD distribution. Although GlmU exhibits comparatively higher localized deviations across specific flexible regions, its overall RMSD remains low because the global fold of the protein stays structurally conserved throughout the simulation, indicating that the fluctuations are confined to peripheral or loop segments rather than affecting the core architecture. Overall, Son-A binding either maintained or improved structural stability across all proteins, with no evidence of destabilizing effects.

Residue-wise flexibility (RMSF) was analyzed to assess local perturbations induced by Son.A binding. Overall, no significant differences were observed between the apo and ligand-bound states, indicating that the ligand-bound proteins adopted RMSF patterns similar to their apo counterparts and exhibited only negligible deviations (Fig. 6).

RMSF (mean ± SD) values followed a similar trend, with RMSF means (± SD) of 0.0839 ± 0.0445 nm (OmpA), 0.0857 ± 0.0524 nm (OmpA apo); 0.2009 ± 0.2413 nm (LpxH), 0.2088 ± 0.2177 nm (LpxH apo); 0.0947 ± 0.1153 nm (KPC-2), 0.1061 ± 0.1423 nm (KPC-2 apo); 0.1374 ± 0.1108 nm (FabG), 0.1774 ± 0.1391 nm (FabG apo); 0.1232 ± 0.0800 nm (ChbG), 0.1272 ± 0.0724 nm (ChbG apo); and 0.3015 ± 0.1920 nm (GlmU) compared with 0.2520 ± 0.1758 nm (GlmU apo). Fluctuations were predominantly localized to loop and terminal regions, consistent with stable, well-equilibrated protein conformations across all systems. OmpA showed low residue fluctuations (< 0.20 nm) with only slight increases in surface loop regions upon ligand binding, while KPC-2 exhibited minimal flexibility changes and remained largely stable. LpxH exhibited modest flexibility across residues 50–150 and 175–230, while C-terminal residues showed sharp fluctuations (~ 2.0) in both apo and complex states, indicating intrinsic terminal mobility. This high terminal mobility can distort the global RMSD metric, making RMSD less reliable for assessing overall structural stability in this system. For FabG, Son.A binding markedly reduced fluctuations in catalytically relevant regions compared to apo, indicating localized stabilization. ChbG showed similar RMSF patterns in both the apo and complex forms, with slight reductions in loop mobility and modest increases in fluctuations in some regions, including the C-terminal, upon ligand binding. The GlmU-Son.A complex displayed slightly elevated RMSF values at a residue level, suggesting that ligand engagement can induce localized mobility without compromising the global structural stability. Overall, RMSF results indicate that, Son.A binding does not induce destabilizing flexibility in any target and, in several cases, reduces mobility in functionally significant regions.

The radius of gyration (Rg) profiles of all six targets was monitored to evaluate the effect of SonA binding on global structural compactness (Fig. 7). The Rg trajectories for all proteins remained stable throughout the simulation, reflecting maintained compactness in both apo and ligand-bound forms. OmpA and KPC-2 exhibited tightly clustered Rg values around ~ 1.50–1.55 nm and ~ 1.78–1.88 nm respectively, with negligible differences between apo and complex states. LpxH, FabG, and ChbG displayed moderate fluctuations (~ 1.75–2.00 nm), with SonA-bound complexes consistently showing slightly lower or more stable Rg values than apo proteins. GlmU, the largest target, showed expected higher Rg values (~ 3.0–3.4 nm), yet the SonA-bound complex maintained a narrower distribution. No system exhibited signs of unfolding or expansion. SASA analysis revealed protein-specific changes upon SonA binding (Fig. 8). OmpA and KPC-2 exhibited minimal SASA variation between apo and complex forms (~ 70–80 nm² and ~ 115–130 nm², respectively), indicating stable surface exposure. LpxH showed slight reductions in SASA in the SonA-bound forms, suggesting increased burial of hydrophobic or catalytic residues. FabG, ChbG and GlmU complexes displayed modest increases in SASA during early simulation but stabilized over time, converging near apo profiles. No protein demonstrated drastic SASA fluctuations, indicating stable solvation and compactness across trajectories.

The hydrogen-bond profiles of the six K. pneumoniae–Son.A complexes revealed target-specific interaction patterns over the 300-ns simulation (Fig. 9). For OmpA-Son.A, the complex maintained 1–4 H-bonds with intermittent peaks reaching six bonds, demonstrating a stable but dynamically shifting interaction network. LpxH-Son.A showed pronounced H-bond enrichment after ~ 150 ns, frequently sustaining 2–5 bonds and reaching a maximum of six, suggesting enhanced stabilization during the later stages of the trajectory. KPC-2-Son.A exhibited more transient H-bonding, oscillating between 0 and 3 bonds with occasional four-bond peaks, indicating moderate but persistent interactions. The FabG-Son.A complex demonstrated the highest H-bond density among all targets, with frequent periods of 3–6 bonds and peaks up to eight, reflecting strong polar complementarity within the active-site environment. ChbG-Son.A displayed moderate hydrogen-bond formation (1–4 bonds), with transient peaks up to five, whereas GlmU-Son.A showed sparse interactions, predominantly 0–2 H-bonds with infrequent single spikes, indicating weaker polar engagement compared to other systems. Overall, the hydrogen-bond analysis reveals that, Son.A maintains continuous binding across all six proteins, with FabG, LpxH, and OmpA exhibiting the strongest and most persistent interaction profiles.

MM/PBSA free energy calculation

The binding free energies (ΔG_binding) of Sonchuinoside A (Son.A) toward the six K. pneumoniae protein targets were computed using gmx_MMPBSA (Fig. 10; Table 4). All complexes demonstrated distinct thermodynamic signatures, reflecting the unique structural and chemical environments of their respective binding pockets. Among the evaluated systems, the LpxH–Son.A complex exhibited the most favorable binding free energy (ΔG = −30.69 ± 2.18 kcal/mol), indicating a highly stabilizing interaction dominated by strong enthalpic contributions. FabG-Son.A also displayed robust affinity (ΔG = −21.46 ± 4.76 kcal/mol), which is consistent with the dense and persistent hydrogen-bond network observed in MD simulations. OmpA-Son.A showed moderate but clearly favorable affinity (ΔG = −11.89 ± 3.73 kcal/mol), while KPC-2-Son.A retained a stable interaction profile (ΔG = −8.77 ± 4.46 kcal/mol), underscoring sustained contact within the active-site region throughout the simulation. For ChbG–Son.A, the binding free energy (ΔG = −2.08 ± 5.37 kcal/mol) suggests a ligand interaction that is more dynamic in nature, consistent with a shallow or flexible binding region that accommodates transient contacts. The GlmU-Son.A complex displayed a positive ΔG value (+ 10.81 ± 7.99 kcal/mol), indicative of a pocket environment that favors weaker or short-lived interactions, yet MD analyses confirm that the global protein architecture remained stable, demonstrating that Son.A binding does not perturb the structural integrity of the enzyme. Across all targets, binding was predominantly enthalpy-driven, particularly for LpxH, FabG, and OmpA, reflecting strong polar and van der Waals complementarity. Entropic factors contributed variably, with some complexes showing a modest entropy penalty commonly associated with tightly bound ligands.

Fig. 10.

MM/PBSA binding free energy profiles of Sonchuinoside A with six K. pneumoniae target proteins. MM/PBSA-derived energetic components, including enthalpy (ΔH), entropy contribution (-TΔS), and binding free energy (ΔG), for Sonchuinoside A (Son.A) in complex with six Klebsiella pneumoniae proteins: (a) OmpA-Son.A, (b) LpxH-Son.A, (c) KPC-2-Son.A, (d) FabG-Son.A, (e) ChbG-Son.A, and (f) GlmU-Son.A. Energies are reported in kcal/mol. Bars represent mean values calculated from the 300-ns trajectory, with error bars denoting standard deviation. LpxH and FabG show strongly favorable binding energies, indicating deep and stable ligand engagement, whereas OmpA and KPC-2 demonstrate moderate but meaningful binding. ChbG exhibits marginal stability, and GlmU reflects weaker complementarity. These trends correlate with the interaction persistence observed in the MD trajectories and identify LpxH and FabG as the most promising Son.A targets.

Table 4.

MM/PBSA binding free energies (ΔG_binding) of Sonchuinoside A with K. pneumoniae protein targets.

Sl. No.	Complex	ΔG_binding (kcal/mol)	SD
1	OmpA-Son.A	−11.89	3.73
2	LpxH-Son.A	−30.69	2.18
3	KPC-2-Son.A	−8.77	4.46
4	FabG-Son.A	−21.46	4.76
5	chbG-Son.A	−2.08	5.37
6	GlmU-Son.A	10.81	7.99

Summary of calculated MM/PBSA binding free energies (ΔG_binding, kcal/mol) with associated standard deviations for Sonchuinoside A in complex with six K. pneumoniae targets. Values reflect the averaged energetics of snapshots extracted from the equilibrated region of the 300-ns MD trajectories. The data underscore the strong and thermodynamically favorable binding of Son.A toward LpxH, FabG, and OmpA, supporting their identification as high-priority inhibitory targets.

Discussion

In an effort to identify promising natural compounds from the medicinal plant Sesbania grandiflora with multi-regulatory effects on molecular targets of K. pneumoniae, computational screening was performed, revealing several phytochemicals with desirable pharmacological profiles.

Initially, using the defined selection strategy and filtering criteria, the complete proteome of K. pneumoniae was retrieved from the UniProt database. From this dataset, a total of 93 proteins that met the specific filtering criteria including annotation scores and structural data was selected for further analysis. Subsequent literature-based functional analysis was performed to assess the relevance of the selected proteins as potential drug targets, focusing on their essentiality, involvement in critical bacterial pathways, virulence, and previous reports of draggability, which led to the identification of six high-priority therapeutic targets. These include enzymes involved in lipid A biosynthesis (LpxH), fatty acid metabolism (fabG), peptidoglycan and amino sugar biosynthesis (GlmU), antibiotic resistance (KPC-2), chitin degradation (chbG), and a major structural component of the bacterial outer membrane (OmpA). UDP-2,3-diacylglucosamine pyrophosphate hydrolase (LpxH) is a critical enzyme in K. pneumoniae, responsible for a key step in the biosynthesis of lipid A, an essential component of the outer membrane of Gram-negative bacteria. LpxH catalyzes the hydrolytic cleavage of the pyrophosphate bond in UDP-2,3-diacylglucosamine to produce lipid X and uridine monophosphate (UMP). Given its indispensable role in the Raetz pathway and the absence of homologous enzymes in mammalian systems, LpxH represents a highly attractive target for the development of novel antibacterial agents, particularly against multidrug-resistant Gram-negative pathogens^12,13. Recent investigations have identified sulfonyl piperazine compounds, such as AZ1 and its derivatives, as potent inhibitors of LpxH activity in K. pneumoniae, exhibiting promising antibacterial efficacy in vitro. Notably, unlike other enzymes involved in lipid A biosynthesis, such as LpxC and LpxH has not yet been targeted by clinically approved antibiotics, thereby offering a novel therapeutic avenue that may circumvent current mechanisms of antibiotic resistance¹³.

The enzyme 3-oxoacyl-[acyl-carrier-protein] reductase, encoded by the fabG gene, plays an essential role in the type II fatty acid synthesis (FAS II) pathway in bacteria. Also referred to as β-ketoacyl-acyl carrier protein (ACP) reductase, FabG catalyzes the first reductive step in the fatty acid elongation cycle, wherein β-ketoacyl-ACP intermediates are converted to β-hydroxyacyl-ACP products using NADPH as a reducing cofactor. This reaction is a key regulatory step in the FAS II pathway, which is indispensable for bacterial membrane biogenesis, growth, and survival. Owing to its critical function and the divergence of FAS II from the mammalian type I fatty acid synthesis pathway, FabG has emerged as a compelling target for the development of novel antibacterial agents^14–16. Recent studies have identified allosteric inhibitors that bind to a previously uncharacterized site on FabG in Pseudomonas aeruginosa, inducing conformational changes that disrupt catalytic activity and hinder NADPH binding. These findings highlight the potential of structure-based drug design strategies to target FabG allosterically¹⁷. Given its essential role in bacterial viability and the absence of closely related human homologues, FabG remains a promising target for the development of next-generation antibiotics. The carbapenem-hydrolyzing β-lactamase KPC-2 is a clinically significant enzyme in K. pneumoniae and other Gram-negative pathogens, primarily due to its capacity to confer resistance to carbapenem antibiotics, agents frequently regarded as the last line of defence against multidrug-resistant bacterial infections. KPC-2 belongs to the Ambler Class A β-lactamases and exhibits an extended substrate profile, efficiently hydrolyzing a wide range of β-lactam antibiotics, including carbapenems, penicillins, cephalosporins, and monobactams such as aztreonam^18,19. The dissemination of KPC-2-producing strains represents a critical public health threat, as it severely limits available therapeutic options and is associated with increased morbidity, mortality, and healthcare burden. Given the rising prevalence of KPC-2-mediated resistance in K. pneumoniae and the paucity of effective treatment alternatives, the development of novel and potent KPC-2 inhibitors is urgently needed to combat these life-threatening infections and restore the efficacy of existing β-lactam antibiotics. GlmU is an essential bifunctional enzyme in K. pneumoniae that plays a critical role in the biosynthesis of UDP-N-acetylglucosamine (UDP-GlcNAc), a key precursor in the formation of bacterial cell wall components, including peptidoglycan and lipopolysaccharide. This enzyme catalyzes two sequential enzymatic reactions: the C-terminal domain mediates the acetylation of glucosamine-1-phosphate (GlcN-1-P) to generate N-acetylglucosamine-1-phosphate (GlcNAc-1-P), while the N-terminal domain facilitates the uridylation of GlcNAc-1-P to produce UDP-GlcNAc^20,21. Given its indispensable role in bacterial viability and its involvement in cell wall biosynthesis, GlmU has emerged as a promising target for the development of novel antimicrobial agents. A variety of GlmU inhibitors spanning multiple chemical scaffolds have been identified and characterized, underscoring the potential for structure-based drug design approaches to develop effective therapeutics against K. pneumoniae and other Gram-negative pathogens^21,22. Chitooligosaccharide deacetylase (ChbG) in K. pneumoniae is a recently characterized enzyme involved in the degradation of chitin-derived oligosaccharides. It catalyzes the deacetylation of N, N’-diacetylchitobiose, removing a single acetyl group, a critical step for the bacterium’s ability to utilize acetylated chitooligosaccharides as a carbon source. This enzymatic activity also contributes to the induction of the chb operon, underlining ChbG’s functional importance in nutrients. Uniquely, ChbG lacks significant sequence homology with other known chitooligosaccharide deacetylases and is therefore categorized as a non-classified carbohydrate esterase, reflecting its structural and functional divergence from classical members of this enzyme family acquisition^23,24. In addition to its metabolic role, ChbG plays a critical part in modulating hypermucoviscosity (HMV) in K. pneumoniae, a phenotype closely associated with enhanced virulence. Although not a canonical virulence factor, ChbG’s activity in deacetylating capsular polysaccharides (CPS) directly influences the expression of HMV, which facilitates resistance to phagocytosis and enhances the pathogen’s ability to invade host tissues and evade immune clearance^25,26. These findings underscore ChbG as a functionally significant enzyme at the intersection of metabolism and virulence regulation in K. pneumoniae. Given its dual role in nutrient metabolism and virulence regulation, ChbG represents a promising target for the development of novel anti-virulence therapies aimed at attenuating K. pneumoniae pathogenicity. Outer membrane protein A (OmpA) in K. pneumoniae is a highly conserved and essential outer membrane protein that plays a pivotal role in bacterial virulence, pathogenesis, and modulation of host immune responses. As a multifunctional virulence factor, OmpA contributes to K. pneumoniae pathogenicity through various mechanisms, including facilitating penetration of the blood-brain barrier (BBB) during meningitis and inducing host cell damage via pyroptotic and apoptotic pathways^27,28. OmpA plays a dual role in host-pathogen interactions. In airway epithelial cells, it can suppress inflammatory responses, potentially promoting bacterial persistence in the respiratory tract. Conversely, it is capable of activating antigen-presenting cells (APCs), such as macrophages and dendritic cells, through engagement with pattern recognition receptors including Toll-like receptor 2 (TLR2) and scavenger receptors (SRs)²⁹. OmpA is also known to interact with a variety of host immune and epithelial cells, including Langerhans cells, where it enhances allostimulatory function. Furthermore, it has been shown to induce cytokine production in innate immune cells and promote cell-cycle arrest and pyroptosis in epithelial cells²⁷.

Given its essential role in mediating host-pathogen interactions and contributing to immune evasion, OmpA has garnered interest as a promising target for vaccine development. Recombinant OmpA has demonstrated immunogenic potential in preclinical models, eliciting protective immune responses and, when conjugated with tumor antigens, inducing anti-tumor cytotoxic activity²⁹. These findings underscore the therapeutic potential of targeting OmpA in the development of both antimicrobial and immunomodulatory strategies against K. pneumoniae infections.

A structure-based molecular docking study was carried out to assess the binding affinity and multi-targeting potential of phytocompounds from Sesbania grandiflora against six identified targets of K. pneumoniae. From the constructed network based on docking scores, it is evident that that the targets KPC-2, fabG, and OmpA exhibited the highest number of interacting phytochemicals (73), followed by chbG and LpxH (72), and GlmU (61). Among the phytochemicals screened, 60 interacted with all the 6 selected targets and the remaining 13 showed interaction with 5 targets. The compound, ‘3′,6-di-O-feruloylsucrose’ showed highest affinity for ompA, LpxH, and GlmU, while the compound, ‘Acarbose hydrate’ demonstrated highest docking score against chbG, fabG, and KPC-2. Further the docking scores were quantitatively evaluated to identify compounds exhibiting higher affinities than the mean dock score. The mean docking scores for each target highlighted the diverse interaction potential of the phytocompounds. A significant number of compounds exhibited binding scores above the respective mean values for each target. Specifically, 32 compounds exceeded the mean docking score (84.49) for OmpA; 33 compounds outperformed the mean of 88.96 for LpxH; 30 compounds surpassed the mean score (82.25) for GlmU; 32 compounds were above the mean value (83.15) for ChbG; 34 compounds exceeded the mean of 90.52 for FabG; and 33 compounds had scores above the mean of 91.55 for KPC-2. The heterogeneity in docking scores reflects the structural and chemical diversity of both protein binding pockets and phytocompounds. A total of 22 phytocompounds were consistently identified as top binders across all six targets, indicating strong and potentially selective interactions with the binding pockets. Heatmap analysis of docking scores revealed distinct binding trends among the proteins. These findings underscore the structural compatibility of Sesbania grandiflora’s phytocompounds with the active sites of diverse K. pneumonia protein targets. The presence of shared high-affinity compounds across multiple proteins suggests potential for multi-target drugs. This is particularly desirable in the context of K. pneumoniae, which requires multi-target antibiotics due to its complex resistance mechanisms.

Next, all the identified phytocompounds were screened for their biological activity prediction based on oral bioavailability and toxicity. The result revealed that, 9 compounds out of 73, emerged as promising candidates in terms of target affinity, oral bioavailability and toxicity.

It has to be note that, many of these shortlisted phytochemicals exhibited recurring structural features that likely contribute to their multi-target interaction potential. A consistent pattern was observed across compounds such as loliolide, vomifoliol, and (3 S,5R,6R,7E)−3,5,6-trihydroxy-7-megastigmen-9-one, all of which belong to the apocarotenoid/lactone family, characterized by a cyclic lactone ring and multiple hydroxyl groups. These features facilitate hydrogen bonding and hydrophobic anchoring within protein active sites. Likewise, ferulic acid and caffeic acid share a hydroxycinnamic acid backbone, defined by a phenolic ring with electron-rich hydroxyl groups and a conjugated side chain, enabling π–π stacking with aromatic residues and stabilizing polar interactions. In addition, Sonchuinoside A and 2-acetylflexuosin A represent glycosylated or polyoxygenated aglycone structures, providing a combination of rigid hydrophobic cores and multiple hydrogen-bond donors/acceptors, which supported stable binding across multiple enzyme targets. Together, these observations indicate that hydroxylated aromatic systems, lactone-containing apocarotenoid scaffolds, and glycosylated polyoxygenated frameworks constitute recurring chemotypes among the highest-performing ligands. This provides a rational basis for future structure-guided optimization and the development of multi-target antibacterial leads.

From the above list of 9 compounds, 2-Acetylflexuosin A, and Sonchuionoside A overlapped with the 22 compounds previously found to interact with all six targets. This overlap highlights them as particularly strong and safe candidates with natural origin for further investigation to combat emerging drug resistance in the pathogen. Of these, Sonchuionoside A, which achieved the highest relative score, and favourable binding mode was chosen for further detailed analysis of molecular interaction. For LpxH, the compound established strong hydrogen bonding with residues such as Glu44 Asn79, and Arg80, along with hydrophobic contacts that stabilize the ligand within the binding pocket. Since the residues, Asn79, and Arg80, participate in substrate binding, interactions specifically to these residues could reduce the protein’s substrate affinity and highlights potential of the identified compound as a probable competitive regulator for lipid A biosynthesis. In the case of FabG, the compound showed H-bond interaction with key residues, including Gly12, Gly18, Asn86, Ala87, and Gly88, that involved in NADP+ binding, highlighting its possible role in modulating the fatty acid synthesis pathway. For KPC-2 β-lactamase, the ligand is positioned deep within the active site, forming stable hydrogen bonds with Ser70, Arg104, Asn170, Thr216, Thr235, and Thr237. Of these, the residues Thr216, Thr235, and Thr237 are shown to be engaged with the co-crystalized drug, Relebactam, a beta-lactamase inhibitor which highlight the compound’s possible inhibitory potential. For target GlmU, Sonchuionoside A interacted with residues His46, Pro115, Arg227, Leu116, Ile117, Ile316, Lys267, and Asn164, primarily through hydrogen bonding and van der Waals forces. The exact binding site features for this target is unavailable. However, the residue, Arg227 is known to be mediating interactions with UDP-N-acetyl-alpha-D-glucosamine binding site and the cofactor, Mg²⁺. The vast hydrogen bonding engagement with the receptor through multiple residues including Arg227 suggest the compound’s ability to interfere with cell wall biosynthesis. In the case of ChbG, hydrogen bonding with residues, Asp10, Ser123, His124, His125, His126, Met203, and His205 and hydrophobic interactions with surrounding residues were observed that stabilized the ligand within the catalytic pocket, possibly inhibiting chitooligosaccharide deacetylation and impairing virulence regulation. For OmpA, the compound occupied within the binding pocket, interacting with residues Ser12, Asp13, Phe16, Asn17, phe18, Asp55, and Arg70 via hydrogen bonds. These interactions may disrupt OmpA’s structural role in host-pathogen interactions and virulence modulation.

Collectively, the interaction profiles demonstrate that Sonchuionoside A could mediate strong interaction through multiple hydrogen bonds, hydrophobic contacts, and π interactions possibly stabilizing its binding mode across all six targets.

Given these promising interaction profiles, we next evaluated the dynamic behavior and stability of the selected protein-Son.A complexes through molecular dynamics simulations. Accordingly, 300-ns MD simulations were performed for all six targets in both their apo forms and in complex with Sonchuionoside A to evaluate their structural stability and dynamic behavior. The RMSD analysis provides insight into the global structural behavior of each K. pneumoniae protein and the influence of Son-A binding on their conformational dynamics. The analyses collectively demonstrate that Son.A forms stable complexes with all six K. pneumoniae proteins. In most systems, particularly OmpA, FabG, GlmU, and chbG, the ligand-bound complexes show reduced structural deviation relative to the apo forms, indicating strong and persistent interactions with the binding pockets. Stabilization over the 300-ns trajectory suggests that Son.A is well accommodated within the protein architectures and does not disrupt global structural integrity. This behavior is characteristic of ligands that form favorable hydrogen-bonding networks and hydrophobic interactions, promoting conformational stability of the receptor-ligand complex.

RMSF patterns further validate the stabilizing behavior inferred from the RMSD analysis. Across all six proteins, residues located in catalytic or ligand-binding regions exhibit either unchanged or decreased flexibility upon Son.A engagement. Reduced RMSF values in OmpA, LpxH, FabG, and KPC-2 specifically highlight ligand-induced local stabilization, which often correlates with productive binding and potential inhibitory activity. Regions displaying higher fluctuations (typically termini and surface loops) behave similarly in both apo and complex forms, indicating that Son.A binding does not perturb natural dynamic properties. Apo OmpA and the OmpA-Son.A complex showed similar RMSF profiles, with slightly elevated fluctuations at the N-terminus and in loop/turn regions, particularly at residues 16–24, 37–46, 55–59, and 96–109. For FabG, both apo and complex forms displayed comparable fluctuation patterns, with increased flexibility in loop and turn regions at residues 88–95, 137–148, and 182–193. Apo ChbG and the ChbG-Son.A complex also showed higher fluctuations in loop regions, notably at residues 75–96, 150–169, and 208–219. The complex exhibited a distinct peak at residue His125, within the binding pocket and crucial for Mg²⁺ coordination, indicating that ligand binding may alter local dynamics at the active site. Importantly, no protein showed widespread increases in flexibility upon ligand binding, which would indicate destabilization or weak interaction. Instead, the consistent pattern of either maintained or reduced mobility demonstrates a stabilizing effect of Son.A across diverse protein folds.

The radius of gyration results demonstrate that SonA binding does not destabilize the global fold of any of the six K. pneumoniae targets. The consistent Rg values across all systems confirm that Son.A binding maintains or enhances the structural compactness of K. pneumoniae proteins. For membrane-associated proteins such as OmpA and enzymatic targets like LpxH, KPC-2, chbG amd GlmU the nearly identical Rg traces highlight that SonA binding preserves their native tertiary organization throughout the simulation. Collectively, Rg analysis strengthens the conclusion that SonA enhances global structural integrity of multiple bacterial targets. SASA behavior indicates that SonA binding induces subtle but meaningful changes in surface accessibility, particularly within catalytic and hydrophobic pockets. The reduced SASA in LpxH complexe suggests deeper ligand burial and enhanced packing, consistent with stable ligand-protein interactions. In contrast, the mild SASA elevation in the OmpA, KPC-2, FabG, chbG and GlmU trajectories reflects binding-induced reorientation, followed by stabilization as the complex equilibrates. The absence of significant SASA increases confirms that Son.A does not disrupt protein solvation shells or induce unfavorable exposure of hydrophobic cores. Overall, SASA trends support compact, well-solvated complexes with SonA contributing to stabilizing interactions across multiple essential targets. Across all six K. pneumoniae targets, Son.A consistently maintained stable interactions, primarily through sustained hydrogen bonding. FabG, LpxH, and OmpA showed the strongest and most stable binding profiles, while the remaining complexes demonstrated moderate but persistent interactions. The interaction patterns indicate that Son.A can stably engage multiple essential or virulence-associated proteins, supporting its potential as a broad-spectrum inhibitor candidate. Collectively, these findings support the hypothesis that Son.A can effectively associate with multiple essential bacterial targets, a potentially valuable feature for phytochemical-based antibacterial strategies.

The free-energy landscape derived from MM/PBSA strongly supports the multi-target binding potential of Sonchuinoside A and aligns closely with the structural stability and interaction patterns observed throughout MD simulations. The exceptionally favorable binding energies for LpxH and FabG emphasize these proteins as primary high-affinity targets. Their strong enthalpic contributions and rich hydrogen-bond landscapes reflect well-defined catalytic pockets that effectively accommodate Son.A, suggesting strong inhibitory potential. OmpA and KPC-2 demonstrate intermediate but consistently favorable ΔG values, which, combined with their stable RMSD, compactness profile, and consistent ligand contact, indicate that Son.A can effectively utilize these proteins.

ChbG-Son.A, despite its milder binding free energy, presents a meaningful interaction signature: the ligand samples a flexible or solvent-exposed region, producing dynamic but recurrent contacts. For GlmU, the positive ΔG reflects a binding environment that is less complementary to Son.A; however, MD analyses confirm that ligand presence does not destabilize the protein.

Collectively, the MM/PBSA and MD data converge on a clear interpretation that the Sonchuinoside A demonstrates strong and selective affinity toward LpxH, FabG, and OmpA, while maintaining stable, non-disruptive interactions with other essential proteins of K. pneumoniae. These findings highlight Son.A as a promising phytochemical candidate capable of engaging multiple bacterial pathways, with particularly compelling energetic support for LpxH and FabG inhibition.

Limitations of the study

Although the study highlights promising results, several limitations must be acknowledged. The phytochemical library used in this work was compiled from previously reported literature and may not fully capture the complete chemical diversity of Sesbania grandiflora. Variations in extraction methods, solvents, and analytical platforms could result in an incomplete metabolite inventory, thereby introducing potential selection bias. Furthermore, the toxicity predictions presented here rely exclusively on computational ADMET models, which depend on predefined training datasets and statistical assumptions. These models do not fully account for factors such as concentration-dependent effects, metabolic pathway variability, or organism-specific physiological conditions. Therefore, the predicted parameters of the identified compounds should be validated through experimental pharmacokinetic and toxicological studies. In addition, the conclusions regarding target-ligand interactions are based solely on computational predictive analyses without experimental confirmation. To substantiate therapeutic potential of identified compounds, experimental assays, including enzymatic assays, bacterial growth inhibition studies, and in vivo infection models are required. Collectively, these limitations indicate that while the computational results provide a strong foundation for hypothesis generation, comprehensive experimental, biochemical, and pharmacological validation is necessary to determine the true translational relevance of these findings. Future studies should incorporate metabolomic profiling, multi-level biological assays, and pharmacological evaluations to further substantiate and refine the proposed therapeutic candidates.

Conclusion

This study adopted a comprehensive target-based and structure-guided strategy to identify novel multi-target therapeutic candidates from the phytocompounds of Sesbania grandiflora against multidrug-resistant K. pneumoniae. Six high-priority bacterial proteins-LpxH, FabG, KPC-2, GlmU, ChbG, and OmpA were shortlisted as promising druggable targets based on their critical roles in disease pathogenesis. Molecular docking revealed several phytochemicals with strong affinities toward multiple targets, while pharmacokinetic and toxicity evaluations identified nine compounds with favorable therapeutic profiles. Among them, Sonchuinoside A emerged as the most promising candidate, displaying potent multi-target affinity, favorable drug-likeness, and desirable safety attributes. Molecular dynamics simulations further supported its stability, compactness, and sustained hydrogen bonding across multiple targets.

Overall, these findings suggest that Sesbania grandiflora is a great source of phytochemicals capable of counteracting multidrug resistance in K. pneumoniae through simultaneous inhibition of multiple therapeutic targets. Future experimental studies are needed to validate its antibacterial efficacy and assess its translational potential. This study highlights the value of medicinal plants as a source of broad-spectrum, multi-target agents to address the growing challenge of antibiotic resistance.

Methods and materials

Target identification

To identify potential therapeutic targets for K. pneumoniae, the complete proteome data of the organism was retrieved from the UniProt database (https://www.uniprot.org/) using the ‘Advanced Search’ option. Specifically, the Taxonomic field was used to filter entries, and the Taxonomic ID: 573 corresponding to K. pneumoniae was identified using the keyword “Klebsiella pneumoniae”. To ensure the selection of biologically and structurally relevant proteins, stringent filtering criteria was applied, retaining only well-characterized proteins with a UniProt annotation score of ≥ 3 (out of 5) and for which high-resolution structures determined by X-ray crystallography were available. A cutoff of annotation score ≥ 3 ensures inclusion of proteins with moderate-to-high curation and experimental support, balancing annotation reliability with adequate dataset coverage. For the proteins with suboptimal resolution, predicted structures were retrieved from AlphaFold. AlphaFold is an AI based protein modelling platform that predicts the 3D structure of proteins³⁰. The shortlisted proteins were then subjected to an extensive literature review to assess their functional importance as druggable targets. Cross-validating with experimental evidence, a subset of the most promising protein targets was selected for subsequent drug discovery studies.

Ligand identification

Phytochemicals of Sesbania grandiflora were retrieved from our previously curated phytochemical library reported in earlier work⁹. All reported phytochemicals, irrespective of plant parts or solvent types, were considered for the present study. The complete list of compounds included in the analysis is provided in Supplementary Table S1.

Molecular Docking

A total of 73 compounds were screened against the selected molecular targets of K. pneumoniae using drug docking suite ‘Discovery studio’(DS) client version. The 3D structures of selected targets were retrieved from Protein databank (PDB (https://www.rcsb.org/)). As indicated, proteins with suboptimal structural files and missing residues were retrieved from AlphaFold (https://alphafold.ebi.ac.uk/). Structures with an average predicted Local Distance Difference Test (pLDDT) score of ≥ 95% were considered, ensuring a high level of structural confidence³¹. The 3D structural files of 73 phytochemicals were downloaded from PubChem database (https://pubchem.ncbi.nlm.nih.gov/). Before docking, all target proteins and phytochemicals were prepared following the standard preparation protocols available in Discovery Studio. Protein structures were processed using the Protein Preparation module, which involved the addition of missing hydrogens, correction of bond orders, removal of alternate conformations, optimization of the protonation states, and energy minimization using the CHARMm force field.Ligands were prepared using the Prepare Ligands protocol, which included generation of 3D structures, assignment of appropriate ionization states at physiological pH, enumeration of relevant tautomers, correction of bond orders, and geometry optimization followed by energy minimization to obtain the lowest-energy conformers. Binding sites were defined based on critical amino acid residues known to mediate ligand interactions. The prepared ligands were then docked into the defined binding pockets using the site specific docking algorithm, LibDock.

To validate the docking workflow, redocking was performed for all targets with available co-crystallized ligands. Each ligand was extracted from the crystal structure and redocked into its respective binding pocket using the same docking algorithm and parameters applied in the screening workflow. The resulting docked poses were compared with the experimentally observed conformations using RMSD analysis (values ≤ 2.0 Å) and inspection of key interacting residues, confirming the reliability of the docking protocol. A detailed summary of each target, co-crystallized ligands (where available), and annotated binding-site residues is provided in Supplementary Table S2.

Network Pharmacology

Following docking, a target-ligand interaction network was generated to assess the multi-target interactions of the phytochemicals of Sesbania grandiflora against selected targets of K. pneumoniae. The network was constructed using Cytoscape (https://cytoscape.org/). The degree centrality of the targets and ligands were calculated using cytoNCA plugin in cytoscape to estimate the number of interactions between targets and phytochemicals and to identify the comounds with multi-target interactions.

Drug-likeness assessment

The top scored candidates were screened for their pharmacological parameters such as oral bioavailability and toxicity. The Lipinski filter protocol of DS was used to assess the oral bioavailability. According to this, an orally bioavailable drug should have number of hydrogen bond acceptors < = 10; hydrogen bond donors < = 5; molecular weight < = 500 daltons; octanol-water partition coefficient < = 5. Next, the pharmakokinetic parameters, such as absorption, distribution, metabolism, excretion, and toxicity (ADMET) were predicted using ‘ADMET descriptors’ of DS. The parameters include intestinal absorption, Blood-Brain Barrier penetration (BBB), aqueous solubility, plasma protein binding, hepato-toxicity and cytochrome P4502D6 inhibition.

Molecular dynamics simulation

MD simulation of protein-ligand complexes and apo proteins was carried out using GROMACS (Version: 2024.1) on a system with an Intel^® Core™ i9-14900KF processor and 64 GiB of RAM (https://www.gromacs.org/). A total of 6 protein-ligand complexes and 6 apo proteins were considered for MD simulation. For protein-ligand complexes, the docked structure files were separated into protein and ligand structure files for generating topology and parameter files. The pdb2gmx tool of GROMACS and the CGenFF server were used, respectively, for protein and ligand topology generation. The structure files of protein and ligand were combined to create the complex structure file. The topology files of protein and ligand were combined in order to create the complex topology file. The system was solvated using TIP3P explicit solvent water molecules in a triclinic box with a 1.5 nm distance to the box edge. Sodium and Choline ions were used to neutralize the system before energy minimization. Energy minimization was performed using the steepest-descent algorithm until the maximum force was reduced to < 1000 kJ mol⁻¹ nm⁻¹, with periodic boundary conditions applied in all dimensions. The ligand restraining and thermostat coupling were performed. Equilibration ensembles like NVT and NPT were performed for 200 ps to stabilize the system’s temperature, pressure, and density before the production run. For NVT equilibration, the system was maintained at 300 K using the V-rescale thermostat (modified Berendsen), employing a coupling constant of 0.1 ps. Initial velocities were generated at the target temperature. NPT equilibration was conducted using the Parrinello-Rahman barostat set to 1 bar, with a coupling constant of 2.0 ps and semi-isotropic pressure coupling. Temperature control was preserved using the same V-rescale thermostat at 300 K. The production MD simulations were performed for 300 ns with 15,00,00,000 steps in the NPT ensemble at 300 K and 1 bar, maintaining the thermostat and barostat settings applied during equilibration. Bond constraints were enforced using LINCS, enabling a 2 fs integration timestep. Long-range electrostatics were treated with the Particle Mesh Ewald (PME) method to ensure accurate representation of Coulombic interactions.

For apo proteins, the protein files were prepared, topology was generated using the pdb2gmx tool, and solvated in a triclinic box with a 1.5 nm distance to the box edge. Monovalent ions like Sodium and Chlorine were used to neutralise the system before energy minimization. The temperature, pressure-coupling scheme, and force-field parameters used for the apo systems were identical to those applied to the corresponding protein–ligand complexes.

NVT and NPT ensembles were performed for equilibration for 200 ps, followed by a production MD run of 300 ns with 15,00,00,000 steps with 2 femtoseconds as the integration timestep. The trajectory and structures were analysed for Root Mean Square Deviation (RMSD), Root Mean Square Fluctuation (RMSF), Radius of Gyration (Rg), and Solvent Accessible Surface Area (SASA). In addition, H-bond analysis was carried out for protein-ligand complexes.

Trajectory convergence was evaluated using backbone RMSD and residue-wise RMSF profiles. Convergence was defined as: (i) attainment of a stable RMSD plateau without systematic drift, and (ii) RMSF profiles that did not exhibit progressive increases in residue flexibility over time. Once the equilibrated portion of each trajectory was identified based on these criteria, quantitative structural descriptors were computed. For this analysis, mean RMSD and RMSF values and their corresponding standard deviations (SD) were calculated over the selected equilibrated segment of each 300-ns production run. All statistical calculations were performed using the time-averaged backbone coordinates extracted from the converged region of each trajectory.

MM/PBSA free energy calculation

The binding free energies of the six K. pneumoniae protein-Sonchuinoside A complexes were estimated using the Molecular Mechanics/Poisson-Boltzmann Surface Area (MM/PBSA) approach as implemented in gmx_MMPBSA v1.6.3³². MD trajectories from the 300-ns production phase were used as input for post-simulation free energy calculations. For each complex, snapshots were extracted from the production phase to compute the molecular mechanics energy terms, solvation energies, and entropic contributions. The total binding free energy (ΔG_bind) was calculated using the standard thermodynamic equation:

where.

• ΔG denotes change in binding free energy,

• ΔH denotes change in enthalpy,

• T denotes absolute temperature, and.

• ΔS denotes change in entropy.

The enthalpy (H) consists of contributions from van der Waals interactions, electrostatic interactions, polar solvation energy (Poisson-Boltzmann), and non-polar solvation energy (SASA-based). Entropy (S) was estimated via normal-mode analysis using the built-in gmx_MMPBSA workflow.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

Conceptualization: Anuroopa G. Nadh; Data Curation: Harshit Sajal, Anjali K; Formal analysis and investigation: Harshit Sajal, Anuroopa G. Nadh, Aswin Mohan, Vishal Ravi; Methodology: Anuroopa G. Nadh; Project Administration: Rajesh Raju; Writing - original draft preparation: Harshit Sajal; Visualization: Aswin Mohan; Validation: Anuroopa G. Nadh; Writing - review and editing: Anuroopa G. Nadh; Resources: Rajesh Raju; Supervision: Anuroopa G. Nadh, Niyas Rehman.

Funding

This research did not receive funding.

Data availability

The datasets generated during and/or analysed during the current study are available from the corresponding author on reasonable request.

Declarations

Competing interests

The authors declare no competing interests.