A target-guided drug repurposing strategy for antibacterial discovery

Abstract

Background

The scarcity of effective antibiotics against drug-resistant pathogens has intensified the need for alternative antibacterial development strategies. Drug repurposing (DR) offers a practical solution; herein, we present detailed experimental workflows and in vitro results to validate the feasibility of a DR-based strategy for new antibacterial development.

Methods

Using Vina-GPU, we performed molecular docking between 125 highly conserved bacterial essential proteins and 2,027 approved non-antibacterials drugs to predict drug-target interactions. As preliminary screening, 14 candidate drugs were tested for inhibitory activity against six bacterial strains (Escherichia coli MC4100, Pseudomonas aeruginosa PAO1, Acinetobacter baumannii ATCC 19606, Salmonella enterica Typhi CMCC 50071, Salmonella enterica Typhimurium ATCC 14028, and Klebsiella oxytoca ATCC 13182) at 100 μg/ml. Subsequently, lower-concentration assays (combined with Polymyxin B nonapeptide, PMBN) were conducted for preliminarily active drugs. The FLUOStar Omega was used to measure bacterial suspension optical density (OD₆₀₀) and calculate drug inhibition rates. Surface plasmon resonance (SPR), MD simulation, and folate rescue experiments were used to further validate the mechanism of the drug.

Results

Among the 2,027 non-antibacterials drugs, several exhibited antibacterial activities. Additionally, 173 existing antibacterials were clustered into two groups based on their binding affinity similarity to the 125 essential proteins, with the two groups showing distinct antibacterial activities. Notably, multiple repurposed drugs inhibited the growth of multiple bacterial species.

Conclusion

Our study revealed that the combination of 8 μg/ml lifitegrast and PMBN can effectively inhibit six gram-negative bacteria. Folate rescue experiments showed that the preliminary antibacterial mechanism of lifitegrast is to inhibit the function of FolA. In addition, future research should explore the structure–activity relationship of lifitegrast and its impact on antibacterial activity.

Graphical abstract

Supplementary Information

The online version contains supplementary material available at 10.1186/s12866-026-04710-8.

Introduction

The rise of drug-resistant bacteria has rendered many antibiotics ineffective, posing a significant challenge in modern medicine [1]. In 2021, an analysis of the World Health Organization's (WHO) antibiotic development pipeline indicated that there has been virtually no progress in developing new and urgently needed antibacterials to combat drug-resistant infections [2]. While the advancement of science has led to the discovery of promising new classes of antibacterials, such as halicin [3], clovibactin [4], cresomycin [5], and abaucin [6], these compounds unfortunately remain in the preclinical research stage, and their path to market remains lengthy.

Drug repurposing (DR) aims to determine the new therapeutic uses of approved drugs. This strategy offers a potential solution to accelerate the discovery pipeline, which is critical in the field of antibacterial research [7]. Compared to de novo discovery, DR candidates possess established safety, toxicity, and pharmacokinetic profiles. This significantly reduces the time and financial investment required for preclinical and early-stage clinical trials, thereby increasing the probability of successfully developing a new antibacterial therapy. However, historically, many DR discoveries have been serendipitous. A rational, target-guided strategy is necessary to improve the efficiency of identifying new antibacterial activities from existing pharmacopeia.

Essential genes, which are indispensable for bacterial survival, represent ideal targets for the design of antibacterial drugs [8, 9]. The mechanisms of action of most clinically successful antibiotics are concentrated on three main categories of these targets. These include components of the protein synthesis machinery, enzymes involved in cell wall synthesis, and proteins regulating nucleic acid metabolism [10–12]. Because inhibition of these core pathways and the proteins encoded by these essential genes can directly lead to bacterial death, they remain the most critical targets for antibacterial drug development.

Molecular docking and molecular dynamics (MD) simulation are fundamental and essential computational methods in drug research, primarily used to predict the interaction between small molecules and biomacromolecules [13]. This efficient screening method accelerates lead compound discovery and provides valuable theoretical guidance for subsequent experimental validation, making it an indispensable part of modern drug development.

Given the complexity of bacterial resistance mechanisms and the diminishing efficacy of existing antibacterials, traditional antibacterials discovery frameworks appear increasingly inadequate [14]. To address this challenge, we propose investigating the interactions between known drugs and key bacterial proteins, particularly those involved in protein synthesis, cell wall synthesis, and nucleic acid metabolism. We hypothesize that a drug capable of simultaneously binding to proteins encoded by these three essential proteins would be very difficult to develop drug resistance and holds promise as a potential antibacterial candidate. To this end, we have developed a DR strategy that leverages molecular docking technology to identify antibacterial candidates from approved non-antibacterials drugs. This approach is further validated through in vitro antibacterial assays and SPR to pinpoint hits with inhibitory activity against gram-negative bacteria.

Methods

Acquisition of bacterial essential proteins and approved drugs

The essential bacterial proteins used in this study were obtained from the CEG2.0 database developed by our research group [15]. Initially, we identified proteins regulating protein synthesis, cell wall synthesis, and nucleic acid metabolism from the literature. These proteins were then used as keywords to search the CEG2.0 database, and the retrieved proteins were defined as essential proteins. All drug information, including antibacterials and non-antibacterials drugs, was sourced from the DrugBank 5.0 database [16], representing a major update since 2018. After acquiring all approved drugs, we categorized them into antibiotics and non-antibacterials based on the literature review and clinical indications. To facilitate the study, we performed preliminary screening to exclude drugs with overly simple structures or impractical properties, such as those lacking SMILES sequences, combination drugs, single-element compounds, and those labeled with "illicit," "withdrawn," or "nutraceutical" in the Groups category.

After obtaining information on essential proteins, we retrieved the crystal structures of proteins in the three categories (protein synthesis, cell wall synthesis, and nucleic acid metabolism) from the RCSB database [17]. These structures were associated with specific PDB IDs (Supplementary Table S1). Notably, since the crystal structures of E. coli proteins have been extensively resolved, offering accessibility and reliability, we selected the crystal structure information of E. coli K12. The 3D structures of all drugs in SDF format were generated from SMILES sequences provided by DrugBank using the Chem3D software in the ChemOffice 2019 suite.

Molecular docking and visualization

Before molecular docking, the crystal structures of proteins and drugs were preprocessed. For protein structure, the process involves removing water molecules, retaining a single conformation of the multimer, and converting the PDB structure to pdbqt format. For the drug molecule, hydrogen atoms were added, and the structures were saved in mol2 format before being converted to pdbqt format. Discovery Studio 2019 was used to remove water molecules and retain protein conformations, while UCSF Chimera software [18] was employed to add hydrogen atoms to the drug molecules. All conversions to pdbqt format were performed using OpenBabel software [19].

After the docking was completed, we used TBtools software [20] to design an affinity heatmap and performed row scaling using Dist method set to Euclidean, the Cluster method set to Complete, and the Branch Form set to Cladogram. To facilitate the display of results, we used TBtools to scale the affinity values between 0 and 1 (the scale method of affinity is ZeroToOne, and Rows and Cols clustering are used, the smaller the value, the redder the color), indicating a higher affinity between the drug and the protein. Additionally, we classified the marketed antibiotics into two categories based on their affinity values, namely Class I (average affinity ≤ −6 kcal/mol) and Class II (−6 kcal/mol < average affinity < 0 kcal/mol) [21–23]. To demonstrate the difference in affinity between the two class antibiotics, we also conducted T-test statistical analysis on the average affinity between approved antibiotics and key targets.

Additionally, the active site box regions of protein crystals were identified using AutoDockTools-1.5.7 software [24]. If the protein crystal lacked a native ligand, AutoDockTools predicted the active site box. After completing all preparations, batch molecular docking was performed using Vina-GPU 2.0 [25]. Studies have shown that the Vina series of tools exhibits superior performance among traditional molecular docking tools [26]. We use hydrogen bonding to show the interaction between drugs and amino acid residues (yellow dashed line).

ADMET analysis

Following molecular docking results, drugs with favorable binding energies were screened for toxicity using the ADMETlab 3.0 database [27] to identify drugs with minimal adverse effects. The screening criteria included nHA (number of hydrogen bond acceptors) ≥ 2, nHD (number of hydrogen bond donors) ≥ 2, TPSA (topological polar surface area) ≥ 95, nRot (number of rotatable bonds) ≤ 20, nRing (number of rings) < 10, logS < 0, logD ≤ 4.5, cl-plasma (plasma clearance) ≤ 10, hERG (human ether-a-go-go related gene) ≤ 0.9, hERG-10 μm ≤ 0.45, BCRP (breast cancer resistance protein inhibitor) ≤ 0.5, t0.5 (half-life of a drug) ≥ 0.6, BBB (blood–brain barrier) ≤ 0.4, Carcinogenicity ≤ 0.5, Ames (mutagenicity) ≤ 0.93, Hematotoxicity ≤ 0.55, RPMI-8226 (cell immunotoxicity) ≤ 0.7, NR-PPAR-gamma (peroxisome proliferator-activated receptors) ≤ 0.5, and A549 (cytotoxicity) < 1. The drugs selected were ordered.

Ordering information for materials

The bacterial strains used in this study were the E. coli MC4100, P. aeruginosa PAO1, A. baumannii ATCC 19606, S. enterica ser. Typhi CMCC 50071, S. enterica ser. Typhimurium ATCC 14028, and K. oxytoca ATCC 13182. Among them, the MC4100 genotype is a variant of E. coli K12 with minimal genetic engineering, closely resembling wild-type strains, and contains a genome of 4,631,469 bp with over 4,000 proteins [28]. Compared to other E. coli strains, the MC4100 genotype exhibits higher tolerance and adaptability. It can survive and proliferate under diverse environmental conditions and exhibits resistance to several commonly used antibiotics. These traits, combined with its minimal genetic modification and close similarity to wild-type strains, make it a particularly suitable model for investigating resistance mechanisms and advancing antibacterial development [29, 30].

All small-molecule drugs used in this study were ordered (excluding peptides, degarelix, and afamelanotide). For ease of analysis, the 14 drugs were labeled DR1-DR14. Their DrugBank IDs, generic names, part numbers, and other details are presented in Table 1. DR1, DR2, and DR6 were purchased from Aladdin Biochemical Technology Co., Ltd (Shanghai, China); DR3, DR7, DR8, DR10, and DR12 were purchased from Macklin Biochemical Technology Co., Ltd (Shanghai, China); DR4, DR5, DR9, DR11, and DR14 were purchased from Yuanye Bio-Technology Co., Ltd (Shanghai, China); and DR13 was purchased from Bepharm Science & Technology Co., Ltd (Shanghai, China). Additionally, the consumables used in this experiment included standard LB liquid and solid media, DMSO solvent, PBS solvent, 96-well plates (Costar®), disposable sterile Petri dishes, and Oxford cups. DMSO and PBS were purchased from Meilun Bio-Technology Co., Ltd (Dalian, China), and the experimental instruments included the FLUOstar Omega microbial growth curve detection system (Germany) and a centrifuge, among others.

Table 1.

Ordering information of 14 approved drugs

Number	DrugBank ID	Generic name	Part numbers	Molecular weight	Solvent
DR1	DB11577	Indigotindisulfonic acid	I105011	422.39	H2O
DR2	DB09297	Paritaprevir	P413284	765.89	DMSO
DR3	DB11942	Selinexor	K860909	443.31	DMSO
DR4	DB11611	lifitegrast	S81722	615.48	DMSO
DR5	DB09280	Lumacaftor	S81128	452.41	DMSO/Ethanol
DR6	DB08995	Diosmin	D111391	608.54	DMSO
DR7	DB11575	Grazoprevir	M872316	766.9	DMSO/Ethanol
DR8	DB00559	Bosentan	B882229	551.61	DMSO/Ethanol
DR9	DB06290	Simeprevir	S56440	749.94	DMSO
DR10	DB13879	Glecaprevir	G873456	838.87	DMSO
DR11	DB00706	Tamsulosin	S80063	408.51	DMSO
DR12	DB13751	Glycyrrhizic acid	G810520	822.94	DMSO/Ethanol
DR13	DB14895	Vibegron	BD00846990	444.54	DMSO
DR14	DB04845	Ixabepilone	S81436	506.7	DMSO

Preliminary assessment of antibacterial activity

Bacterial cultures were streaked onto LB agar plates and incubated overnight at 37 °C. Single colonies were washed with physiological saline and adjusted to a McFarland turbidity of 0.5, followed by a 150-fold dilution in Mueller–Hinton Broth (MHB). To evaluate the antibacterial activity of 14 drugs, we measured the optical density at 600 nm (OD₆₀₀). First, bacterial cells at the mid-log phase were diluted 1:50 into 200 μL LB (Luria–Bertani) medium containing each compound (final concentration of 100 μg/ml) in 96-well plates. Subsequently, the 96-well plate was incubated at 37 ℃ without shaking. After 24 h, the absorbance at 600 nm was measured using a microplate spectrophotometer. Percent inhibition was calculated as Eq. (1), where OD_X is the end point OD₆₀₀ for a culture treated with compound X, OD_P, and OD_N are the end point OD₆₀₀ for bacterial cells (50 μL) with 150 μL LB medium and the blank control (200 μL LB medium), respectively. All wells in the microplate underwent three biological replicates. Among the 14 FDA-approved drugs, those showing > 50% inhibition against E. coli MC4100, P. aeruginosa PAO1, A. baumannii ATCC 19606, S. enterica ser. Typhi CMCC 50071, S. enterica ser. Typhimurium ATCC 14028, or K. oxytoca ATCC 13182, in preliminary screening were selected for this combined activity assessment. All wells in the microplate underwent two biological replicates.

1

OD: Optical Density at 600 nm; OD_P: The absorbance of the positive control at 600 nm; OD_N: The absorbance of the positive control at 600 nm; OD_X: The absorbance of the drug at 600 nm.

Assessment of combined activity with PMBN

Test compounds were prepared at two concentrations (8 μg/ml and 4 μg/ml) in 96-well plates, and each well was supplemented with 100 μL of the bacterial dilution plus 100 μL of the compound solution (to achieve final drug concentrations of 8 μg/ml and 4 μg/ml) with the addition of Polymyxin B nonapeptide (PMBN) at a fixed concentration of 8 μg/ml. Wells containing only PMBN (8 μg/ml) or bacterial suspension served as controls. After 24 h of incubation at 37 °C, bacterial growth was assessed by measuring optical density at 600 nm (OD₆₀₀) to calculate inhibition rates.

The possible antibacterial mechanism of candidate drug

To preliminarily investigate the antibacterial mechanism of drug, we conducted a folate (5-formyl-tetrahydrofolate, 5-F-THF) rescue experiment [31]. Not only that, the interaction between drug and the essential bacterial protein was analyzed through molecular docking, molecular dynamic simulations, and surface plasmon resonance (SPR) assays. Firstly, the crystal structure of protein was obtained from the PDB database and prepared by removing water molecules, adding hydrogen atoms, and optimizing side-chain conformations. The drug was optimized by energy minimization, and molecular docking was performed to evaluate the potential competitive inhibition of protein by drug. Subsequently, a ligand–residue interaction map was generated to identify key amino acids involved in binding and to compare drug with known protein ligands.

In addition, MD simulation program (GROMACS) was used to study the drug-target interactions [32]. We have exported RMSD (root mean square deviation), RMSF (root mean square fluctuation), Rg (radius of gyration), SASA (solvent accessible surface area), hydrogen-bonding interactions, and ΔG (free energy landscape) as the basis for drug target binding stability. As is well known, the Rg of protein reflects the volume and structural state of the protein macromolecule. A larger Rg value of the same system indicates that the system undergoes expansion during the simulation. SASA is an important means to describe the hydrophobicity of proteins, and the hydrophobicity of amino acid residues is a key factor affecting protein folding. RMSD and RMSF area Eqs. (2) and (3). Additionally, we use the hydrogen bonding interaction force of MD simulation throughout the entire process as the binding strength between the drug and its target, that is, the stable existence of hydrogen bonds can indirectly represent the stable binding between the drug and the target.

2

3

Among them, R_t-R_ref represents the position of the t-th atom in a certain frame minus its position ref (position offset) in the reference conformation, N is the number of atoms, and T is each sampling time.

The surface plasmon resonance (SPR) is a biophysical technique developed based on the principles of physical optics for detecting molecular interactions. This method can detect intermolecular interactions in real-time and with high sensitivity without the need for labeling. SPR utilizes light to generate evanescent waves in different media, which then resonate with plasma waves to construct a biosensing analysis technique for detecting the interactions between ligands and analytes on biosensing chips. In the experiment, one of the molecules to be measured is fixed on the surface of the chip, while the other molecule flows through the surface of the chip in a flowing manner. The analyte interacts with the ligand, and when the molecule binds and dissociates, the surface mass of the chip changes, causing a change in the resonance angle. The sensor image will record the signal values under the interaction in real time, and based on this change curve, information about intermolecular interactions can be obtained. The experimental materials required for SPR are listed (Table 2), for information on protein purification, please refer to Supplementary Material S13.

Table 2.

The partial experimental materials required for SPR

Reagent/Squipment	Item number/specification	Company
CM5 Sensor Chip	BR-1005–30	Cytiva
Amino coupling reagent kit	BR100050	Cytiva
Sodium acetate buffer solution (pH 4.0/4.5/5.0)	BR100349	Cytiva
Biacore	8 K	GE HEALTHCARE
Recombinant Escherichia coli Dihydrofolate reductase (FolA)	Molecular Weight: 18.8 kDa Expression host: E. coli Purity: 98%	CUSABIO
Lifitegrast	615.48 g/mol	MCE

Results

Bacterial essential proteins and approved drug information

This study identified 125 essential proteins from the CEG 2.0 database [15]. This includes 55 proteins associated with protein synthesis, 24 proteins associated with cell wall synthesis, and 46 proteins that regulate nucleic acid metabolism (Supplementary Table S1). These 125 essential proteins showed a cluster size of ≥ 2, indicating that each protein is present in two or more strains of bacteria. Subsequently, we retrieved 2,200 drugs classified as "approved" from the DrugBank database (v2025) [16]. After excluding entries that did not meet our criteria, we retained 173 antibiotics and 2,027 other drugs (Supplementary Tables S2 and S3).

Clustering of binding energies of existing 173 antibiotics

The binding-energy matrix of 173 antibiotics against 125 essential bacterial proteins revealed distinct clustering patterns (Fig. 1). More than half of the drug–target pairs showed weak affinities, and only a smaller subset exhibited strong binding, indicating heterogeneous target engagement across compounds. In Fig. 1, the horizontal axis represents the 173 antibiotics clustered according to their binding-energy profiles, while the vertical axis shows the 125 essential bacterial proteins clustered by their interaction patterns. These proteins can be classified into three major functional categories, are protein synthesis, cell-wall synthesis, and nucleic-acid synthesis respectively.

Fig. 1.

Affinity clustering of 173 antibacterials and 125 essential proteins. This heatmap shows the binding energy profiles between 173 approved antibiotics (horizontal axis) and 125 essential bacterial proteins (vertical axis). The color gradient from white to dark red indicates low to high affinity, respectively. Dark red regions suggest antibiotics with strong, multi-target binding

The heatmap (full matrix in Supplementary Table S4) displays a clear energy gradient from white (low affinity) to dark red (high affinity), reflecting variability in drug–target interactions. Overall, most clinical antibiotics exhibited weak binding toward a large portion of the essential proteins, which aligns with their generally selective mechanisms of action. In contrast, concentrated dark-red regions indicate antibiotics with strong binding to multiple essential targets, suggesting potential broad-spectrum capabilities.

This multi-target affinity-based clustering effectively differentiated antibiotics into distinct mechanistic groups and supports the reliability of binding energy as a quantitative indicator for prioritizing antibacterial candidates. These results provide a benchmark for subsequent screening of the 2,027 non-antibacterial compounds.

Based on average binding energy values (Supplementary Table S5), the 173 antibiotics were divided into two groups (Fig. 2B). Class I (≤ − 6 kcal/mol) contained 82 antibiotics (47.4%), including 15 cephalosporins, 12 penicillins, 14 quinolones, 5 macrolides, 1 sulfonamide, 1 aminoglycoside, 12 tetracyclines, and 22 other types. Class II (− 6 to 0 kcal/mol) contained 91 antibiotics (52.6%), including 22 cephalosporins, 14 penicillins, 14 sulfonamides, 11 aminoglycosides, 3 quinolones, 2 macrolides, and 25 other types. The two classes showed a statistically significant difference in their overall affinity patterns with the 125 targets (p = 2.483 × 10⁻³¹, Fig. 2C).

Fig. 2.

The information on 173 antibiotics and 245 approved drugs. A Overview of the number of indications for 245 repurposed drugs shows the primary therapeutic uses of these repurposed drugs, including anti-tumor, cardiovascular disease, and pain. B Two classifications of 173 antibiotics illustrate the division of 173 antibiotics into Class I and Class II based on their average binding affinity. C Statistical difference of average affinity between 173 antibiotics and 125 essential proteins confirm a statistically significant difference between the affinity patterns of Class I and Class II antibiotics (T-test, p = 2.483e-31)

Notably, all latest-generation antibiotics were enriched in Class I, including all fifth-generation cephalosporins and newly developed cephalosporin derivatives, whereas Class II mainly contained first- and second-generation members of the same categories. The clinical usage status of both classes is summarized in Supplementary Table S5. These affinity-based classifications provide an indication of potential broad-spectrum target engagement. Although not a direct surrogate for clinical efficacy, the trend supports the utility of large-scale docking-driven antibiotic profiling and highlights candidates with higher exploration value for antibacterial potential.

Evaluation of the potential of marketed drugs as antibiotics

We obtained a binding energy matrix for 2,027 drugs against 125 bacterial essential proteins (Supplementary Table S6). Given that the average affinity of the 173 antibiotics to the 125 targets was −5.9 kcal/mol, we calculated the average affinity of the 2,027 drugs to the 125 proteins and set affinity ≤ −5.9 kcal/mol as the preliminary screening criterion. This resulted in the identification of 245 drugs from the 2,027 approved drugs. Analysis of these 245 drugs revealed that their primary therapeutic applications span anti-tumor, cardiovascular disease, pain, multiple inflammations, gynecological diseases, and hepatitis C, among others (Fig. 2A, the ordinate is the number of drugs).

To prioritize the selection of drug candidates with favorable toxicity profiles, we conducted ADMET analysis on 245 drugs using the ADMETlab 3.0 database (Supplementary Table S7). Based on the optimized parameters, 15 drugs with low predicted toxicity were identified (Supplementary Table S8). Because we only focus on small-molecule compounds, we ordered 14 of these drugs for in vitro experiments. Furthermore, we designated the 14 candidate drugs as DR1, DR2, …, DR14, and obtained the final list of them (Supplementary Table S8).

Analysis of preliminary in vitro antibacterial results

The preliminary antibacterial results of 14 drugs at a dose of 100 μg/ml are presented in Fig. 3. Compared with the positive control and other drugs, lifitegrast and glecaprevir showed lower average OD₆₀₀ values after 24 h (Fig. 3A). To more intuitively distinguish the antibacterial effects of the drugs, we converted the OD values into the percentage inhibition rate against bacteria (retaining two decimal places) according to Eq. (1). It is obvious (Fig. 3B) that among the 6 g-negative bacteria tested, lifitegrast stood out. Its inhibition rates against K. oxytoca ATCC 13182 and S. enterica ser. Typhi CMCC 50071 were 67% and 61%, respectively, followed by 58% against E. coli MC4100, 56% against S. enterica ser. Typhimurium ATCC 14028, and 50% against both P. aeruginosa PAO1 and A. baumannii ATCC 19606. This indicates that lifitegrast may have broad-spectrum potential against gram-negative bacteria. Another drug with relatively good efficacy was glecaprevir, whose inhibition rates against K. oxytoca ATCC 13182 and S. enterica ser. Typhi 50,071 were 61% and 52%, respectively. It is worth noting that these results indicate that lifitegrast and glecaprevir have preliminary antibacterial potential, and their original data can be found in Supplementary Table S9.

Fig. 3.

Preliminary antibacterial effects of 14 drugs. A The average OD₆₀₀ values of bacteria treated with 14 drugs (100 μg/ml) after 24 h, where lower values indicate greater antibacterial inhibition against the six tested gram-negative bacteria. B Percentage inhibition rates of the 14 drugs against 6 g-negative bacteria (calculated via Eq. (1)), highlighting DR4 (lifitegrast) and DR10 (glecaprevir) as having the best preliminary antibacterial potential

It is evident that although lifitegrast and glecaprevir exhibit inhibitory effects against a variety of gram-negative bacteria at 100 μg/ml, their efficacy is not ideal. We hypothesize that this may be attributed to the drugs' inability to effectively penetrate the outer membrane barrier of gram-negative bacteria, thereby preventing them from exerting their intended antibacterial activity. Therefore, we chose to combine lifitegrast and glecaprevir with PMBN to enhance their antibacterial activity.

The results of combining with PMBN

We selected to reduce the doses of lifitegrast and glecaprevir to 8 μg/ml and 4 μg/ml, respectively, and then combined them with PMBN at a fixed concentration of 8 μg/ml to further evaluate their antibacterial activity (Supplementary Table S10).

The results (Fig. 4) showed that lifitegrast at 8 μg/ml or 4 μg/ml exhibited high inhibitory activity against K. oxytoca ATCC 13182 (90% and 88%, respectively). Additionally, at 8 μg/ml, lifitegrast showed inhibition rates of 87%, 84%, 84%, 80%, and 77% against S. enterica ser. Typhi CMCC 50071, E. coli MC4100, A. baumannii ATCC 19606, P. aeruginosa PAO1, and S. enterica ser. Typhimurium ATCC 14028, respectively. At 4 μg/ml, the inhibition rates of lifitegrast against the above-mentioned strains were 85%, 78%, 72%, 74%, and 70%, respectively. In contrast, after combination with PMBN, glecaprevir showed lower inhibitory activity against all 6 bacterial strains than lifitegrast at the same doses (Fig. 4). Meanwhile, PMBN alone at 8 μg/ml had no inhibitory effect on the 6 bacterial strains (Fig. 4). This phenomenon directly indicates that lifitegrast can exhibit a dose-dependent relationship with PMBN, when PMBN is maintained at 8 μg/ml, increasing the dose of lifitegrast can effectively inhibit the growth of various gram-negative bacteria.

Fig. 4.

Antibacterial activity of lifitegrast and glecaprevir combined with PMBN. Displays the enhanced inhibitory activity of lifitegrast and glecaprevir when combined with a fixed dose of Polymyxin B nonapeptide (PMBN) (8 μg/ml) at reduced drug concentrations (8 μg/ml and 4 μg/ml). Lifitegrast exhibited high inhibition rates, showing a synergistic, dose-dependent effect. PMBN alone had no inhibitory effect

Preliminary study on the antibacterial mechanism of lifitegrast

To preliminarily explore the antibacterial mechanism of lifitegrast (DB11611) through bacterial protein binding, we first extracted the raw docking data of lifitegrast with 125 bacterial proteins (Supplementary Table S6). As shown in Fig. 5A, lifitegrast exhibited the lowest affinity value with FolA (dihydrofolate reductase), suggesting that lifitegrast may exert antibacterial activity by targeting bacterial FolA (PDB ID: 4pss). To further elucidate the mechanism by which lifitegrast interacts with FolA, we compared its binding profile with that of 4pss’s ligand (folic acid), FolA’s physiological substrate (dihydrofolate, DHF), and an FDA-approved FolA-targeting antibacterial (trimethoprim). Each compound was docked against FolA with three independent replicates to reduce experimental variability.

Fig. 5.

Docking results of lifitegrast, folic acid, DHF, and trimethoprim with 125 proteins. A The affinity of lifitegrast with 125 proteins, shows that lifitegrast had the lowest affinity value with FolA (dihydrofolate reductase), suggesting it is the primary target. B Comparison of affinity between 4 drugs and 125 proteins, with colors representing affinity values (kcal/mol), compares lifitegrast's affinity to FolA with that of folic acid, DHF, and trimethoprim, showing lifitegrast has a higher mean affinity toward FolA

The results (Fig. 5B) showed that the mean affinity of lifitegrast toward FolA was higher than that of DHF, folic acid, and trimethoprim, indicating that lifitegrast may form a more stable interaction with FolA. An alternative explanation is that lifitegrast could compete with DHF for FolA binding, thereby interrupting the reduction of DHF to tetrahydrofolate (THF) and leading to rapid bacterial cell death.

Furthermore, to analyze the amino acid sequence identity of FolA across the six bacterial strains, we used the FolA sequence from E. coli MC4100 as the reference to compare its similarity with the FolA sequences of the other five bacterial strains. The results showed (Fig. 6A) that the FolA sequences of the three Enterobacteriaceae strains (K. oxytoca ATCC 13182, S. enterica Typhi 50,071, and S. Typhimurium ATCC 14028) shared more than 90% similarity with that of E. coli MC4100. As an essential metabolic enzyme in bacteria, FolA is highly conserved among strains within the same family, a phenomenon consistent with the evolutionary law. In addition, the FolA sequences of the other two strains (P. aeruginosa PAO1 and A. baumannii ATCC 19606) exhibited less than 60% similarity to that of E. coli MC4100. A potential explanation for this observation is that the families Pseudomonadaceae and Moraxellaceae diverged from Enterobacteriaceae at an early evolutionary stage, leading to significant overall sequence differences.

Fig. 6.

The similarity of FolA sequences and the interaction between liftegrast and DHF with FolA. A The similarity between FolA sequences of 5 bacteria and FolA sequences of E. coli MC4100. B The optimal conformation of lifitegrast bound to FolA, showing the interaction between lifitegrast and FolA at the binding site, and D showing the interaction between DHF and FolA at the binding site. C Align the sequence (35aa) containing the 8 residues of FolA bound by liftegrass with the FolA sequences of 5 different bacteria. Yellow dashed lines represent hydrogen bond interactions between the drug and residues

However, through analyzing the interaction between lifitegrast and the FolA protein of E. coli MC4100 (Fig. 6B), we identified 8 amino acid residues involved in the binding with lifitegrast, including 5 residues (ASN18, ALA19, THR46, SER49, and ARG52) that form hydrogen bonds and 3 residues (MET20, LEU28, and ILE50) that participate in hydrophobic interactions. Notably, these 8 residues are located within a 35-amino acid segment of FolA, and this specific segment is present and relatively conserved in all the six bacterial strains (Fig. 6C). This alignment result incorporates an analysis of the physicochemical property similarity of the 35 amino acids (with similarity expressed as a percentage), where residues with identical polarity or acid–base properties are colored uniformly. In particular, except for P. aeruginosa, the 5 residues that form hydrogen bonds with lifitegrast exhibit nearly identical physicochemical properties across the six bacterial strains, indicating that lifitegrast is also capable of forming hydrogen bond interactions with the FolA proteins of these strains.

The docking results showed that lifitegrast can compete with DHF for the residues ASN-18 and THR-46 of FolA (such as hydrogen bond with ASN18, THR46, hydrophobic interaction with LEU28), thereby preventing the effective binding of DHF to FolA (Fig. 6D), reducing bacterial metabolic efficiency, and consequently inducing rapid bacterial death (Supplementary Table S11).

To explore the potential of FolA as a key target, we conducted 100 ns molecular dynamics (MD) simulations using GROMACS to systematically compare the binding stability and dynamic behavior of lifitegrast-FolA and DHF-FolA complexes (DHF serves as the physiological substrate of FolA, with the crystal structure derived from PDB ID: 4pss). This extended simulation timeframe was selected to ensure sufficient sampling of conformational space, enabling reliable assessment of long-term binding dynamics between the ligand and target protein.

Following the MD simulations, four critical structural parameters were analyzed to characterize complex stability: root mean square deviation (RMSD), root mean square fluctuation (RMSF), radius of gyration (Rg), and solvent-accessible surface area (SASA). As shown in Fig. 7A-D, the RMSD values of all simulated systems remained within a narrow range (0.15–0.30 nm) throughout the simulation, indicating that both lifitegrast and DHF maintained stable binding to FolA without significant conformational drift. The RMSF profiles further revealed that the flexible loops adjacent to the active site (residues 45–55 and 80–90) exhibited similar fluctuation amplitudes in both complexes, suggesting that neither ligand induced abnormal local flexibility in FolA. Additionally, the Rg values (1.54–1.60 nm) and SASA measurements of FolA remained consistent between the two groups, confirming that the overall folding state and hydrophobic exposure of the protein were not substantially altered by ligand binding. Collectively, these four parameters failed to distinguish differences in binding stability between lifitegrast and DHF.

Fig. 7.

MD simulation fluctuations data of lifitegrast and DHF bound to FolA. A Root mean square deviation (RMSD); B Root mean square fluctuation (RMSF); C Radius of gyration (Rg); D Area per residue over the trajectory/Solvent accessible surface area (SASA)

To gain deeper insights into ligand-FolA interactions, we further analyzed hydrogen bonding (H-bond) dynamics throughout the 100 ns simulation, as H-bonds are critical for mediating specific and stable protein–ligand binding. The time-series analysis of H-bond formation (Fig. 8A-B) demonstrated that both lifitegrast and DHF formed H-bonds with FolA over the entire simulation period. However, distinct patterns emerged in H-bond stability and quantity: lifitegrast consistently maintained 2–4 H-bonds with key residues of FolA (primarily ASN18, THR46, and SER49), with a stable H-bond occupancy rate exceeding 85% after 65 ns. In contrast, DHF formed 1–3 H-bonds with FolA (predominantly ASN18 and THR46), and its H-bond occupancy fluctuated significantly (40–70%) throughout the simulation, with frequent dissociation and reformation events.

Fig. 8.

Hydrogen bonding interactions and ΔG landscape of lifitegrast and DHF bound to FolA. A Hydrogen bonding interactions between lifitegrast and FolA; B Hydrogen bonding interactions between DHF and FolA; C ΔG landscape of lifitegrast; D ΔG landscape of DHF

Complementary free energy landscape (ΔG) analysis based on RMSD and Rg further validated the superior binding stability of lifitegrast (Fig. 8C-D). The ΔG landscape of the lifitegrast-FolA complex exhibited a single deep energy minimum (ΔG = −12.0 kJ/mol), indicating a highly stable and conformationally restricted binding state. In contrast, the DHF-FolA complex displayed multiple shallow energy minima (ΔG ranging from −7.5 to −9.0 kJ/mol) and broader conformational distribution, reflecting higher dynamic flexibility and lower binding specificity. These findings collectively demonstrate that despite similar overall structural stability inferred from RMSD, RMSF, Rg, and SASA, lifitegrast forms more stable and specific H-bond interactions with FolA compared to the native substrate DHF, supporting its role as a potent competitive inhibitor of FolA.

Results of 5-F-THF rescue assay and SPR

Moreover, we hypothesize that DHF in the bacterial metabolic pathway cannot be converted into THF, resulting in the inability of the cells to obtain the tetrahydrofolate required for survival. This subsequently blocks nucleotide metabolism and leads to bacterial death. To validate this hypothesis, we performed a folate rescue experiment. The results show that 5-F-THF can restore 60% of the growth inhibited by lifitegrast combined with PMBN in E. coli MC4100, indicating that 5-F-THF is able to rescue E. coli from lifitegrast-induced killing (Table 3, Supplementary Table S12).

Table 3.

Impact of 5-F-THF on lifitegrast and PMBN synergistic inhibition of E. coli MC4100

Dose	Group	E. coli MC4100			Average
8 μg/ml + 8 μg/ml	Lifitegrast + PMBN	0.84	0.843	0.838	0.84
8 μg/ml + 50 μM + 8 μg/ml	Lifitegrast + 5-F-THF + PMBN	0.193	0.179	0.179	0.18
50 μM + 8 μg/ml	5-F-THF + PMBN	0.059	0.058	0.054	0.057
8 μg/ml	Lifitegrast	0.492	0.49	0.487	0.49
50 μM	5-F-THF	0.0013	0.004	0.002	0.002

The value in the average column represents the average inhibition rate of the current group against E. coli MC4100 (range 0–1)

To further investigate whether lifitegrast can tightly bind to FolA, we conducted surface plasmon resonance (SPR) assays. First, FolA protein was immobilized on the sensor chip with a coupling level of 3245 RU. Next, six concentrations of lifitegrast were tested (0.3125 μM, 0.625 μM, 1.25 μM, 2.5 μM, 5 μM, and 10 μM). The results showed that lifitegrast exhibited strong binding affinity to FolA, with a dissociation constant (KD) of 5.08 μM (Fig. 9A, Table 4). Additionally, to check whether the FolA used in SPR is consistent with the FolA used in molecular docking and molecular dynamics simulations, we conducted sequence analysis, and the results showed (Fig. 9B) that the FolA used in this study is consistent.

Fig. 9.

SPR and sequence alignment. A Measurement of the interaction between FolA protein and lifitegrast, the dissociation constant K_D was determined to be 5.08 μM. B Amino acid sequence alignment of FolA, confirms that the FolA protein sequence used in the SPR is consistent with that used in the computational studies

Table 4.

The experimental results for SPR

Receptor	Analyte	KD(M)	Ka(1/Ms)	Kd(1/s)
Dihydrofolate reductase (FolA)	lifitegrast	5.08E-06	1.95E + 04	9.89E-02

K_D: Dissociation constant, which reflects the affinity of the analyte to the target. The smaller the value, the stronger the affinity; Ka: Association rate constant, representing the speed of intermolecular binding, with larger values indicating faster binding; Kd: Dissociation rate constant, representing the speed of intermolecular dissociation, with larger values indicating faster dissociation

Discussion

This study establishes a target-guided drug repurposing (DR) framework centered on bacterial essential proteins, aiming to overcome current limitations in antibacterial discovery caused by slow innovation and increasing antibacterial resistance. The choice of bacterial essential proteins as core targets is well-supported by existing literature, for instance, a study by Thulasi et al. elucidated importance that essential protein-focused antibacterial discovery [33], moreover, another study by our research group has shown that drug reuse strategies based on essential genes are feasible [34]. By integrating large-scale molecular docking with in vitro validation, we demonstrated that this computational-experimental workflow is feasible, efficient, and suitable for prioritizing clinically available drug candidates before extensive pharmacological evaluation. Compared with traditional and AI-driven antibiotic development, this strategy has advantages in terms of cost, time, and risk, and does not rely on high-quality training data [35], which supports it as a practical supplement to existing antibiotic discovery pipelines.

The affinity-based clustering analysis of 173 approved antibiotics revealed two distinct groups with significantly different binding affinities toward 125 essential proteins, where newer broad-spectrum antibiotics were mainly enriched in the high-affinity group. This observation supports the utility of affinity-based profiling for evaluating potential antibacterial target breadth, thus serving as a reference for prioritizing existing drugs or guiding derivative design.

Lifitegrast is originally used to treat dry eye syndrome [32] and functions via blocking the T-cell integrin LFA-1 [36]. Although its clinical toxicity is manageable [37], no previous studies have reported gram-negative antibacterial activity. This work provides the first evidence-based indication that lifitegrast may exhibit antibacterial potential when membrane permeability is enhanced. Through docking-based screening of 2,027 non-antibacterial drugs, followed by toxicity profiling, 14 compounds were selected for experimental testing. Among them, lifitegrast showed the most promising inhibitory profile against six gram-negative pathogens, particularly when applied together with PMBN, suggesting a permeability-dependent antibacterial effect.

FolA, also known as dihydrofolate reductase (DHFR), is an essential enzyme in the folate biosynthesis and regeneration pathway of bacteria. It catalyzes the reduction of DHF to THF using NADPH as an electron donor. THF is a critical one-carbon carrier required for nucleotide synthesis (purines, thymidylate) and several amino-acid biosynthesis reactions. Due to this central metabolic role, FolA is indispensable for bacterial growth and survival. Mechanistic exploration indicated that lifitegrast targets FolA, and docking comparison suggested a stronger binding tendency than physiological substrate DHF, folic acid, and trimethoprim. SPR analysis further validated the interaction, reinforcing FolA as a plausible mechanistic target [38]. Folate rescue experiments showed that the preliminary antibacterial mechanism of lifitegrast is to inhibit the function of FolA. These results imply that lifitegrast may function as a competitive FolA inhibitor, interfering with folate metabolism and RNA/DNA biosynthesis, ultimately resulting in bacterial growth inhibition.

However, several limitations should be noted. Firstly, antibacterial evaluation was limited to in silico and in vitro [34, 39, 40], and in vivo pharmacokinetics remain unknown. Secondly, its potency depends on permeability enhancement by PMBN, implying that structural optimization, formulation modification, or nanoparticle-based delivery may be required before clinical translation [41, 42].

Conclusion

We successfully developed and validated a bacterial essential protein-guided drug repurposing strategy for antibacterial discovery, and clarified computational predictions and experimentally validated findings. Computational work included docking 2,027 non-antibacterial drugs with 125 essential proteins, clustering 173 antibiotics by affinity, and screening 14 candidate drugs via ADMET analysis. Experimental results confirmed: 8 μg/ml lifitegrast + PMBN inhibited six gram-negative bacteria; folate rescue, MD simulations, and SPR (K_D = 5.08 μM) verified lifitegrast targets FolA to block folate synthesis. Overall, this study highlights that integrating drug repurposing with molecular docking and essential protein targeting represents a valuable approach for accelerating antibacterial discovery, and future work should emphasize expanding the screening scope to larger drug libraries, optimizing the chemical structure of lifitegrast to improve membrane permeability and potency.

Abbreviation

DR

Drug repurposing

OD₆₀₀

Optical Density at 600 nm

SPR

Surface plasmon resonance

MDs

Molecular dynamics simulation

CEG

Cluster of essential genes

PDB

Protein Data Bank

KD

Dissociation constant

LB

Luria–Bertani

MHB

Mueller–Hinton Broth

SMILES

Simplified Molecular-Input Line-Entry System

DS

Discovery Studio

FDA

Food and Drug Administration

PMBN

Polymyxin B nonapeptide

FolA

Dihydrofolate reductase

DHF

Dihydrofolate

DHFR

Dihydrofolate reductase

THF

Tetrahydrofolate

ADMET

Absorption, Distribution, Metabolism, Excretion, Toxicity

RU

Response Unit

BIO

Biotechnology Innovation Organization

WHO

World Health Organization

Authors’ contributions

Dongdong Zhang and Haotian Li: Writing – original draft, Methodology, Investigation, Formal analysis. Anqiang Ye, Xiang Lian and Hongtu Cui: Investigation, Formal analysis. Feng-Biao Guo and Zhenshun Cheng: Resources, Supervision, Methodology, Investigation, Formal analysis, Foundation.

Funding

This work was supported by the National Natural Science Foundation of China (Grant NO. 32370696) and the Climbing Project for Medical Talent of Zhongnan Hospital of Wuhan University (Grant NO. PDJH202406), and Noncommunicable Chronic Diseases-National Science and Technology Major Project (grant no. 2023ZD0506200/2023ZD0506202).

Data availability

All the supporting information reported in this study is available for free, and you need to access 18182075 (The Zenodo link). These data include crystal files and docking parameters of 125 essential proteins, *.pdbqt and *.sdf files of 173 antibiotics, and *.pdbqt and *.sdf files of 2,027 approved drugs.

Declarations

Ethics approval and consent to participate

Not applicable.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.