LIST OF ACRONYMS Abbreviation ExplanationAcc2 H-bond acceptor projection EPIs Efflux pump inhibitors Hyd ITC Hydrophobic centroidIsothermal titration calorimetryMATE The multidrug and t
LITERATURE REVIEW
Antibiotic resistance
Antibiotics are antimicrobial agents that target bacteria, ushering in the modern antibiotic era after the discovery of penicillin and streptomycin in the early 20th century However, the overuse and overprescribing of antibiotics have accelerated bacterial evolution, leading to the rise of antibiotic‑resistant strains Antibiotic resistance—where bacteria, viruses, fungi or parasites survive drugs that would normally kill them—renders standard treatments ineffective, causing longer illnesses, higher healthcare costs, and increased mortality Common infections such as pneumonia, urinary‑tract infections, tuberculosis and food‑borne illnesses are becoming harder, and sometimes impossible, to treat The World Health Organization (2014) warns that antibiotic resistance is a serious, worldwide threat that is already affecting people of all ages in every country.
Bacteria develop antibiotic resistance through five key mechanisms: they reduce outer‑membrane permeability to block drug entry, alter the molecular targets so antibiotics can no longer bind, enzymatically modify and inactivate the drugs, spread resistance genes via plasmids, and activate efflux pumps that expel antibiotics from the cell (Blair et al., 2015).
Efflux pumps are essential microbial mechanisms that protect cells by actively removing toxic substances and antimicrobial agents, thereby maintaining internal homeostasis Six major families dominate efflux transport: the ATP‑binding cassette (ABC) superfamily, the resistance‑nodulation‑division (RND) family, the small multidrug resistance (SMR) family, the major facilitator superfamily (MFS), and the multidrug and toxic compound extrusion (MATE) system Recognizing the structure and function of each efflux pump family is crucial for developing effective strategies to combat antimicrobial resistance.
The MATE family represents one group of bacterial efflux carriers, but Acinetobacter baumannii also possesses another important efflux protein known as the Proteobacterial Antimicrobial Compound Efflux (PACE) superfamily; PACE is structurally similar to the SMR family, highlighting its role in multidrug resistance (Seukep et al., 2019).
MATE SMR ABC MFS RND Cyloptavn
Figure 1.1 Six superfamilies of efflux pumps found in bacteria: MATE, SMR,
MFS, ABC, RND and PACE (Seukep et al., 2019)
The Major Facilitator Superfamily (MFS) is the largest group of membrane transporters, moving a wide range of substrates such as sugars, oligosaccharides, inositols, drugs, amino acids, nucleosides, organophosphate esters, Krebs‑cycle intermediates, and numerous organic and inorganic anions and cations While some MFS transporter families are exclusive to bacteria or to eukaryotes, many are shared between these domains, reflecting their essential role in cellular metabolism and the transport of diverse molecules (Pao et al., 1998).
The Major Facilitator Superfamily (MFS) proteins are 400‑600‑amino‑acid transporters characterized by 12 or 14 transmembrane α‑helices and are ubiquitous in both Gram‑negative and Gram‑positive bacteria These transporters mediate uniport, antiport, and symport mechanisms for a wide range of substrates, and in Gram‑negative organisms they can form a tripartite pump complex that spans the outer membrane In Gram‑positive bacteria, MFS pumps such as NorA, QacA, QacB (from *Staphylococcus aureus*) and LmrP (from *Lactococcus lactis*) are the best‑studied examples and play a critical role in multidrug resistance (MDR) by expelling antibiotics including chloramphenicol, norfloxacin, tetracycline, ceftazidime, cefepime, and streptomycin This dual presence across bacterial groups underscores the importance of MFS transporters as key vectors of antimicrobial resistance.
SMR transport proteins are the smallest members of the membrane transporter family, typically comprising 100–120 amino acids and four transmembrane helices, and are divided into three subclasses: small multidrug pumps, GroEL mutation suppressors, and paired SMR proteins The small multidrug‑pump subclass is notable for conferring multidrug resistance (MDR) in both Gram‑positive and Gram‑negative bacteria through the expression of a single gene The best‑studied model, EmrE from *Escherichia coli*, is a 110‑amino‑acid protein that efficiently extrudes acriflavine and quaternary ammonium compounds Additional key SMR efflux transporters include SepA and QacC from *Staphylococcus aureus* and EbrAB from *Bacillus subtilis*, highlighting the broad relevance of SMR proteins in bacterial drug resistance.
Active multidrug efflux pumps—primarily secondary transporters from the Major Facilitator Superfamily (MFS), Small Multidrug Resistance (SMR) family, and Resistance‑Nodulation‑cell Division (RND) superfamily—play a crucial role in the baseline or intrinsic resistance of many bacteria to antimicrobial agents Recently, the Multidrug and Toxic Compound Extrusion (MATE) family has been identified as a new contributor to drug resistance, although its impact has been documented in only a few isolated cases (Fernandez et al., 2012).
Efflux pumps exist in both Gram‑positive and Gram‑negative bacteria, yet antibiotic resistance in Gram‑negative organisms is particularly complex because they have an extra outer membrane that functions as a strong barrier to the influx of antimicrobial agents, thereby enhancing the effectiveness of their multidrug‑efflux systems.
Efflux pumps of the RND (resistance‑nodulation‑cell division) superfamily—exemplified by AcrB in *Escherichia coli* and MexB in *Pseudomonas aeruginosa*—are key contributors to multidrug resistance in Gram‑negative bacteria These transporters function as part of a tripartite complex that couples the inner‑membrane pump with an outer‑membrane channel (TolC in *E. coli* and OprM in *P. aeruginosa*, members of the OMF family) and a periplasmic adaptor protein (AcrA in *E. coli* and MexA in *P. aeruginosa*) By linking these three components, the RND efflux systems expel a broad range of antibiotics, making them critical targets for combating both intrinsic and acquired drug resistance.
Importantly, each ofthese three component proteins is essential for drug efflux, and the absence of even one component makes the entire complex totally nonfunctional.
The tripartite RND efflux pumps export drugs directly into the external medium, bypassing the periplasm, which gives Gram‑negative bacteria a huge advantage because the outer membrane barrier prevents re‑entry of antibiotics Wild‑type Gram‑negative strains exhibit intrinsic resistance to many lipophilic antibiotics—including penicillin G, oxacillin, macrolides, novobiocin, linezolid, and fusidic acid—and this resistance has traditionally been linked to exclusion by the outer membrane However, disrupting the outer membrane only modestly sensitizes *E. coli* to these drugs, whereas inactivation of the constitutively expressed RND pump AcrB dramatically lowers the MIC of compounds such as cloxacillin—from 512 µg/ml in the wild type to just 2 µg/ml—even with an intact outer membrane Consequently, the intrinsic resistance of Gram‑negative bacteria relies as much on RND efflux pumps as on the outer membrane barrier.
1.2 E.Coli AcrAB-TolC efflux pump
1.2.1 Structure ofE coli AcrAB-TolCefflux pump
The RND (resistance‑nodulation‑cell division) superfamily is exclusive to Gram‑negative bacteria and includes the clinically significant AcrAB‑TolC efflux system of *Escherichia coli* This tripartite complex consists of the inner‑membrane transporter AcrB, the outer‑membrane channel TolC, and the periplasmic adaptor protein AcrA, which stabilizes the AcrB‑TolC interaction and enables efficient drug extrusion.
1.2.2 Substrates and mechanism of effluxpump
AcrB, the active component of the AcrAB‑TolC efflux system, expels a broad spectrum of antibiotics and chemotherapeutic agents—including tetracycline, chloramphenicol, β‑lactams, novobiocin, fusidic acid, nalidixic acid and fluoroquinolones—as well as bile salts, contributing to multidrug resistance in Gram‑negative bacteria.
Figure 1.2 Proposed model of the AcrA-AcrB-TolC complex (Kobyika et al., 2020)
Effluxpumps are prominent in terms ofboth their high efficiencyof drug extrusion and broad substrate specificities, underlying their roles in multidrug resistance.
Substrates are first taken up from the lipid bilayer of the cell membrane and then expelled by the AcrB transporter Structural evidence shows that AcrB moves these compounds from the interior across its transmembrane region into periplasmic vestibules, where they accumulate in a central cavity From this cavity the substrates are actively pumped through the AcrB pore into the TolC tunnel, completing their extrusion out of the cell (Nikaido et al., 2012).
1.3 Inhibition of AcrAB-ToIC efflux pump
The tripartite AcrAB‑TolC efflux pump is a primary driver of antibiotic resistance in *Escherichia coli*, making it a critical drug‑target for novel combination therapies that pair antibiotics with efflux‑pump inhibitors (EPIs) Similar to the classic β‑lactam/β‑lactamase inhibition strategy, the synergistic use of EPIs and antibiotics boosts intracellular drug concentrations, lowers intrinsic bacterial resistance, expands the antibacterial spectrum, and restores the efficacy of antibiotics that have become ineffective against resistant pathogens.
Multi- drug efflux systems
119 compounds from 12 scientific articles (J A Bohnert & Kern, 2005, J A Bohnert et al., 2016, Abdali et al., 2016, Nguyen et al., 2015, Wang et al., 2018, J.
A Bohnert et al., 2011, Haynes et al., 2017, Whalen et al., 2015,Ohene-Agyei etal.,
In 2014 and 2015, Yilmaz et al and Su et al (2007) pioneered the construction of novel pharmacophore models targeting the AcrAB‑TolC efflux pump of *Escherichia coli*, a strategy later refined by Mowla et al (2018) These models identified four compounds with high binding affinity, quantified by their dissociation constant (Kd), and 115 additional agents whose inhibitory potency was measured using the MPC4 metric—the minimal concentration needed to reduce an antibiotic’s MIC by fourfold Detailed chemical structures, supporting research articles, and full references are available in the supplementary material (Appendix 1 and Appendix 2).
The database comprises two set: training and testing set The training set is used for constructing a predictive model whose predictive performance is evaluated on the testing set.
Training set includes 4 compounds which are considered as the most potential efflux pump inhibitors They are FIM_205_4211_D13-9001, BBA_2018_1860_
The compounds 878 NMP, BBA_2018_1860_878_Minocycline, and BBA_2018_1860_878_A2 have been identified as potent potential inhibitors, with Minocycline additionally serving as an antibiotic and a recognized substrate of the AcrB efflux pump Moreover, FIM‑205‑4211‑D13‑9001 belongs to the pyranopyrimidine scaffold class, while BBA_2018_1860_878_NMP is classified as an arylpiperazine‑derived molecule.
BBA_2018_1860_878_Minocycline belongs to tetracyclin, BBA_2018_1860_878_ A2 is 4-isobutoxy-2-naphthamide.
Those compounds determined inhibitory ability based on Kd -the descriptor measures the binding affinity between AcrB and ligands.
The smaller Kd value is, the strongerinhibitory abilityis The Kd (pM) values were converted to pKd = -logKd to develop pharmacophore model Full list ofstructure compounds, research journalswas in Table 2.1.
SUBJECTS - RESEARCH METHOD
Data sets
119 compounds from 12 scientific articles (J A Bohnert & Kern, 2005, J A Bohnert et al., 2016, Abdali et al., 2016, Nguyen et al., 2015, Wang et al., 2018, J.
A Bohnert et al., 2011, Haynes et al., 2017, Whalen et al., 2015,Ohene-Agyei etal.,
Based on the seminal works of Yilmaz et al (2014), Su et al (2015), Mowla et al (2007, 2018), we constructed novel pharmacophore models targeting the AcrAB‑TolC efflux pump of *E. coli* Among the screened molecules, four compounds exhibited high binding affinity, as quantified by their dissociation constant (Kd), while an additional 115 candidates demonstrated inhibitory potency measured by MPC₄—the minimal concentration required to reduce the minimum inhibitory concentration (MIC) of a companion antibiotic by four‑fold A comprehensive list of the identified structures, along with the corresponding research journals and references, is provided in Appendix 1 and Appendix 2.
The database comprises two set: training and testing set The training set is used for constructing a predictive model whose predictive performance is evaluated on the testing set.
Training set includes 4 compounds which are considered as the most potential efflux pump inhibitors They are FIM_205_4211_D13-9001, BBA_2018_1860_
The study identifies three promising inhibitor candidates—878 NMP, BBA_2018_1860_878_Minocycline, and BBA_2018_1860_878_A2—highlighting Minocycline’s dual role as an antibiotic and a substrate of the bacterial efflux pump AcrB Additionally, the compound FIM 205 4211 D13‑9001 is characterized by a pyranopyrimidine scaffold, while BBA_2018_1860_878_NMP belongs to the arylpiperazine derivative class, both offering valuable structural insights for the development of novel antimicrobial agents.
BBA_2018_1860_878_Minocycline belongs to tetracyclin, BBA_2018_1860_878_ A2 is 4-isobutoxy-2-naphthamide.
Those compounds determined inhibitory ability based on Kd -the descriptor measures the binding affinity between AcrB and ligands.
The smaller Kd value is, the strongerinhibitory abilityis The Kd (pM) values were converted to pKd = -logKd to develop pharmacophore model Full list ofstructure compounds, research journalswas in Table 2.1
❖ The compound names are given as follows: Abbreviation of Journal-Publish year_Volume_Page_Name in literature
The testing set comprises 115 compounds with measured MPC4 values against eight key antibiotics—Levofloxacin, Ciprofloxacin, Chloramphenicol, Erythromycin, Clarithromycin, Novobiocin, Piperacillin, and Rifampicin—identified across ten peer‑reviewed research journals Detailed information on the source articles and the corresponding number of compounds for each antibiotic is presented in Table 2.2.
9 MO_2014_3_6_885 2 (Ohene-Agyei et al., 2014)
Testing setincludes 115 compounds divided into two categories:
• Data set 1: 42 active compoundswith MPC4 < lOpM
• Data set 2: 73 inactive compounds with MPC4 > 1O|1M
Using MOE 2008.10, a comprehensive drug‑discovery platform, we constructed both ligand‑ and structure‑based pharmacophore models to identify potent AcrB inhibitors The study employed well‑established in‑silico techniques, focusing on ligand‑based pharmacophore modeling to efficiently screen and pinpoint novel compounds that can block the AcrB efflux pump.
The process for developing a pharmacophore model generally involves the following steps in Figure 2.1
Structural compounds were initially drawn using ChemBioDraw Ultra 12.0 and saved as cdx files; these were then converted to sdf format with Open Babel GUI 2.4.1 and imported into the Molecular Operating Environment (MOE 2008.10) for energy minimization before further calculations.
Kd values are collected and converted to pKd =-logKd
Energy minimization is the process of finding an arrangement in space of a collection of ligands Energy Minimize tool in MOE 2008 is used to optimize structures with the following options:
The dataset conformations were generated using the default settings of the MOE 2008 software; specifically, the Conformation Import Tool in MOE 2008 was employed to produce the minimum‑energy conformations for each active and inactive compound.
A collection of possible molecular conformations was built with the following options:
Conformation Import is a fast‑processing tool that calculates low‑energy conformations for a collection of molecules and stores the results in a molecular database It accepts MOE Molecular Database files or *.cdx drawing files as input, making it compatible with common chemical data formats Designed for speed, the underlying algorithms excel at handling large combinatorial libraries, delivering rapid conformation generation and seamless integration with molecular database workflows.
• Forcefield-Based A molecular force field is used as the basis to determine low energy conformations(as opposed torule-based methods).
• Fragment Approach Molecules are divided into overlapping fragments whose conformations are determined independently prior to assembly For speed, the conformations of these fragments are savedfor futureuse.
• Systematic Search The conformational search is systematic (as opposed to stochastic) The resulting conformations are reproducible.
3D-pharmacophore model was built by Pharmacophore Elucidator tool in MOE 2008.
The Pharmacophore Elucidator is designed to create a robust set of pharmacophore queries that accurately represent the key molecular features of compounds active against a specific biological target By analyzing a diverse collection of scaffolds, it generates queries that retrieve the majority of active molecules while deliberately excluding inactive compounds, ensuring that all or most active agents satisfy the defined pharmacophore patterns This approach maximizes hit identification and streamlines virtual screening for drug discovery.
Pharmacophore Elucidation operates on 3D conformations of the input molecules The Conformations options control how the Input Database molecules are interpreted The options are:
Integrating the conformational database with the original active database that stores pKd values enables you to automatically retrieve the binding affinity for each entry; by merging these datasets, every conformational record is seamlessly matched with its corresponding pKd value, streamlining data analysis and providing quick, reliable access to essential pKd information.
Pharmacophore queries were imported byPharmacophore Elucidation tool with the following indicators: cover, overlap, accuracy.
The model was evaluated by the following indicators:
(i) Cover: The number of actives that match the query.
(ii) Overlap: Thealignment score that varies between zero and the number of active molecules Higher scores indicate betteralignments.
Accuracy measures how effectively a query distinguishes active conformations from inactive ones An accuracy score of 1.0 represents perfect performance, meaning the query correctly matches every active compound while excluding all inactive compounds This metric is essential for evaluating query precision in differentiating actives versus inactives.
2.2.5 Applying queries on testing set
Pharmacophore queries generated from potent inhibitors were evaluated by screening the testing set (Table 2.2) to identify compounds that satisfy these queries Using the Pharmacophore Search tool in MOE 2008.10, we applied the queries to the 3D conformational database, allowing the model to distinguish active compounds from the remaining entries After execution, MOE produced a results table where each row represents a match of the query to a specific conformation, often showing multiple matching orientations for the same structure This analysis demonstrates the effectiveness of the pharmacophore model in separating active molecules from the broader dataset.
2.2.6 Validation pharmacophore query score of 1, showed the best predictive ability with high selectivity and specificity which are defined by the retrieval of active and inactive compounds, respectively. The performance of the classification models was evaluated by parameters calculated as follow:
TP X ((TP + FN) + (TP + FPỴ)
TN X ((FP + TN) + (TN + FN)
Here, TP: the number of true positives They are active compounds that match the model.
TN: the number oftrue negatives They are inactive compounds that do not match the model.
FP: the number of falsepositives They are active compounds that do not match the model.
FN: the number of false negatives They are inactive compounds that match the model.
Accuracy: The accuracy tells that overall how often the model is making a correct prediction.
Sensitivity, also known as recall, is a key performance metric that measures the fraction of relevant (positive) instances correctly retrieved out of the total number of actual positive cases It quantifies how well a model identifies true positives, indicating the proportion of actual positive outcomes that are accurately classified High sensitivity means the model is effective at detecting positive instances, making it essential for evaluating the model’s ability to correctly identify the positives and ensure reliable results.
Specificity: It is also known as the True Negative Rate It measures the proportion of actualnegatives that are correctly identified as such.
Specificitytells howgood our model is for identifying the negatives correctly.
Goodness-of-hit (GH) score: This parameter takes into account both the yield (the fraction ofactive structures hit) and the percentage of actives thatare retrieved from the database.
Virtual screening
The top‑performing pharmacophore models were employed to virtually screen comprehensive, freely accessible online databases—including DrugBank, Traditional Chinese Medicine (TCM) libraries, and curated natural‑product collections—to pinpoint compounds most likely to inhibit the AcrB efflux pump This high‑throughput in‑silico screening workflow, illustrated in Figure 2.2, leveraged the predictive power of the best‑ranked pharmacophore hypotheses to rapidly identify promising AcrB inhibitors from a diverse chemical space, accelerating the discovery of novel antimicrobial agents.
DrugbankandTraditional Chinese Medicine is a curated collection of commercially available chemical compounds preparedespecially for virtual screening.
Figure 2.2 The virtual screening process
The database from these libraries was screened using Lipinski’s Rule of 5, a widely accepted filter that distinguishes drug‑like molecules from non‑drug‑like compounds By applying this rule, researchers can quickly assess the drug‑likeness of each compound and predict a high probability of success—or potential failure—in later development stages Molecules that meet two or more of Lipinski’s criteria—appropriate molecular weight, lipophilicity, hydrogen‑bond donors, and hydrogen‑bond acceptors—are flagged as promising candidates for drug discovery, helping streamline the selection process and improve overall screening efficiency.
• Molecular mass less than 500 Dalton
• High lipophilicity (expressed as LogP less than 5)
Using open‑source software, we screened extensive compound libraries—including DrugBank and Traditional Chinese Medicine databases—against the Lipinski Rule of 5 to identify drug‑like molecules The highest‑scoring pharmacophore model was then selected to drive a virtual‑screening workflow, focusing on potential AcrB efflux‑pump inhibitors Compounds that matched the pharmacophore and displayed the lowest root‑mean‑square deviation (RMSD) conformations were highlighted as the most promising AcrB inhibitors for further drug‑discovery studies.
RESULT AND DISCUSSION
Building 3D- Pharmacophore Models
Based on the activity values of four training‑set compounds, fifteen pharmacophore models were generated using Pharmacophore Elucidation—one 5‑point model and fourteen 4‑point models The eleven highest‑scoring models, ranked in descending order of overlap scores, are presented in Table 3.1 with detailed descriptions.
Table 3.1 Pharmacophore model with descriptions
Accuracy scoring Factors Aro/PiR Hyd Acc2
Pharmacophore (PH4) models of AcrB efflux pump inhibitors are characterized by key features such as an aromatic center (Aro) or Pi‑ring center (PiR), a hydrophobic centroid (H), and an H‑bond acceptor projection (Acc2) In most models, the hydrophobic centroid dominates, accounting for 50 % or more of the total features, which highlights the lipophilic nature of the AcrB binding pocket and underscores the importance of hydrophobic interactions in designing effective AcrB inhibitors This insight guides the development of potent AcrB efflux pump inhibitors with optimized aromatic, Pi‑ring, and hydrogen‑bond acceptor groups to enhance binding affinity and overcome bacterial resistance.
Model PH4_1 consisted of 3 hydrophobic centroids and 1 aromatic center.
Model PH4 2 consisted of3 hydrophobic centroids and 1 aromatic center and 1 El- bond acceptor.
Three models PH4_4, PH4_5, PH4 6 consisted of 3 hydrophobic centroids and 1 Id- bond acceptor.
Model PH4_7 consisted of 2 hydrophobic centroids and 2H-bond acceptors.
Two models PH4_8, PH4_9 consisted of 3 hydrophobic centroids and IH-bond acceptor.
Two models PH4_10, PH411 consisted of2 hydrophobic centroids and 2H-bond acceptors.
Validation models
11 models were built from 4 compounds:by MOE 2008.10 Theywere evaluated by the testing set including 115 compounds divided into two categories:
• Data set 1: 42 active compounds with MPC4 < lOpM
• Data set 2: 73 inactive compounds with MPC4> lOpM.
The results were shown in Table 3.2
All pharmacophore models demonstrate promising performance, with 10 of 11 achieving high sensitivity, specificity and accuracy—exceeding 50%—and eight models surpassing 80% across all metrics Notably, four models attain GH scores approaching 100%, confirming them as top‑ranked, reliable pharmacophore models.
There are two kinds of pharmacophore model: aromatic ring group and non aromatic ring group.
The aromatic ring group includes the RHHH, RHHHa, and RHHa models, as shown in Figure 3.1 Both the RHHH and RHHa models achieve the same high sensitivity of 90.48 % but differ slightly in specificity and overall accuracy Among the 73 inactive compounds, 57 are not matched by the RHHH model, whereas 62 are not matched by the RHHa model.
The RHHHa predictive model exhibits the highest specificity, sensitivity, and accuracy among the three evaluated models, correctly identifying 65 of the 73 inactive compounds As a result, RHHHa outperforms the other models in forecasting inactive activity, making it the most reliable tool for distinguishing inactive compounds in drug discovery and related research.
Table 3.2 11 best pharmacophore models of AcrB inhibitors with descriptions
According to GH‑score analysis, the RHHHa model demonstrates superior performance in the aromatic ring group, with an active‑state ratio of 85.61% and an inactive‑state ratio of 90.95%, confirming RHHHa as the best model for this chemical series.
Blue sphere is H-bond acceptor projection (Acc2), green sphere is hydrophobic centroid (Hyd) and orange sphere is aromatic center (R)
The non -aromatic models are two types: HHHa and HHaa ThePH4 4, PH4_5, PH48, PH4_9 models are HHHa and the PH4 7, PH4_8 and PH4_9 model are
The HHHA pharmacophore models incorporate two key feature types—three hydrophobic centroids (H) and one hydrogen‑bond acceptor (a) Among these models, PH4‑4, PH4‑5, and PH4‑9 exhibit high predictive performance, each achieving a sensitivity exceeding 90%, whereas the PH4‑8 model displays a markedly lower sensitivity of only 23.81% This discrepancy is primarily due to differences in the spatial arrangement of the features, as illustrated by the varying radii (distances) shown in Figure 3.2.
The PH4_8 model shows the largest radius (8.18 Å) between hydrophobic centroids and H‑bond acceptors among the three models, while PH4_5 and PH4_9 have very similar radii and comparable predictive abilities Both PH4_5 and PH4_9 correctly identify 38 of 42 active compounds—the highest success rate—yielding a specificity of 90.48 % and an overall accuracy of 90.43 % Their GH scores are also strong, with 87.46 % for active compounds and 92.35 % for inactive compounds, making PH4_5 and PH4_9 the best-performing models in the HHHa group.
HHaa group includes PH4_7, PH4_8, PH410 models They have pharmacophore features: 2 Hydrophobic centroids (H) and IH-bond acceptor(a).
PH4_11 model shows the best result when compared to its counterparts Its specificity and accuracy are 90.41% and 90.43%, respectively It matches 38 out of
The PH4_11 model outperforms other variants by achieving the highest GH score—87.46 % for active compounds and 92.35 % for inactive compounds—demonstrating superior predictive power In contrast, the PH4_7 and PH4_10 models show lower performance, with sensitivities of 92.76 % and 76.19 % and specificities of 76.71 % and 87.67 %, respectively Notably, the dataset includes 42 active compounds that match the model, while 66 out of 73 inactive compounds are correctly identified, highlighting the robustness of the PH4_11 model in distinguishing active versus inactive compounds.
The PH4_5 and PH4_9 pharmacophore models from the HHHa group achieve GH scores comparable to the top‑performing PH411 model of the HHaa group, placing them among the three highest‑ranking pharmacophore models used for database screening These leading models—highlighted in Figure 3.3—demonstrate robust predictive power and are essential tools for efficient virtual screening in drug discovery.
Figure 3.3 Top three pharmacophore models of inhibitors of AcrB with
In molecular modeling, the blue sphere represents the hydrogen‑bond acceptor projection (F4:Acc2) while the green sphere denotes the hydrophobic centroid (F:Hyd), and the distance between their centers is calculated using the Ả method The alignment of active molecules that satisfy the query is visualized with color‑coded atoms: oxygen appears in red, nitrogen in blue, fluorine in green, chlorine in dark green, bromine in dark red, iodine in purple, sulfur in yellow, and carbon in grey, providing a clear and SEO‑optimized overview of chemical element representation and molecular interactions.
Discussion
Doxorubicin functions as a substrate of the bacterial efflux pump AcrB, binding within a distinct hydrophobic pocket identified by Thomas Eicher et al in 2012 This pocket is exclusive to one monomer of the asymmetric AcrB trimer (the T‑binding monomer) and is formed by residues such as Phe666 and Leu828 Hydrophobic interactions between the doxorubicin centroid and the Phe666‑Leu828 pocket facilitate its coordination inside the AcrB access channel, explaining the drug’s transport mechanism.
In addition, ligand interaction Tool in MOE 2008.10 shows ligand interactions between residue inside AcrB access pocket and doxorubicin (as shown in Figure 3.4.)
• H-bond acceptor forms hydrogenbond with G1U130
• Aromatic center interacts with hydrophobic binding pocket forming by
Phel78 o polar * sidechain acceptor o acidic * sidechain donor o basic • ằbackbone acceptor o greasy •* backbone donor o solvent residue ị 'ziarene-arene o metal complex @+arene-cation solvent contact metal contact
O receptor exposure proximity contour ligand exposure
F igure 3.5 Ligand interaction of 4DX7 (AcrB in complex with doxorubicin)
Pharmacophore features shown as above, which is consistent with Thomas Eicher et al Hence, compounds with hydrophobic centroid, H-bond acceptor and aromatic center features are potentialinhibitors of AcrB.
Lanatoside C was identified for its synergistic potential and validated as a potent efflux‑pump inhibitor through an ethidium bromide accumulation assay, which showed a marked increase in intracellular fluorescence when the compound was present This strong correlation between in silico screening results and positive in vitro efflux inhibition suggests that Lanatoside C can effectively block bacterial efflux mechanisms Consequently, Lanatoside C emerges as a promising candidate for combination therapy aimed at combating drug‑resistant *E. coli* strains.
Lanatoside C was identified as a top‑scoring ligand in virtual screening of the most predictive pharmacophore models (HHHa and HHaa) These models contain a hydrophobic centroid (H) and an H‑bond acceptor projection (Acc2), which match the E‑pharmacophore features described for the AcrB efflux pump by Aparna Vasude et al (2014) Molecular docking confirmed that Lanatoside C binds within the AcrB binding pocket, forming close contacts with key residues surrounding the domain The interaction profile of Lanatoside C therefore satisfies the pharmacophore criteria and highlights its potential as an AcrB inhibitor.
• The hydroxyl groups of the glucose moiety of lanatoside c interacted with
Arg468 of AcrB The digitoxose sugar moiety of lanatoside c formed H-bond
Ser389, Gly296, Asn298 of AcrB In addition, there is a conformational change in the compound that led to the formationof another H-bond with Arg468 ofAcrB.
• Hydrophobic centroid (H) can be aromatic ring, interact with hydrophobic binding pocket forming by Phe386, Gly2387, Phe388, Gln469, Asn391, Ala384, Leu30
• The glucose moiety of lanatoside c formed two H-bonds with AIa831 and
GIn830 of AcrB The digitoxose sugar moiety formed H-bond with Gln657 of AcrB
• Hydrophobic centroid (H) can be aromatic ring, interact with hydrophobic
Figure 3.6 Pharmacophore model with the best Lanatoside c conformations (having the lowest RMSD) (A)Lanatoside c structure (B) HHHa model with Lanatoside c (C) HHaa model with Lanatoside c
MBX2319 is a novel antibiotic potentiator (EPI) featuring a pyranopyridine core with five substituents, designed to enhance the efficacy of a broad spectrum of antibiotics against E. coli without exhibiting membrane‑disrupting or intrinsic antibacterial activity Remarkably, MBX2319 fully potentiates the actions of levofloxacin and piperacillin at ultra‑low concentrations as low as 3 pM—approximately ten times lower than the doses required by earlier inhibitors—making it a highly potent enhancer for existing antibiotic therapies.
Using MBX2319 as a probe, we screened and identified the most favorable pharmacophore models (illustrated in Figure 3.6), thereby confirming that these models accurately predict the critical features that drive AcrB binding This result underscores the relevance of pharmacophore modeling in pinpointing the key molecular determinants and major contributors to the binding properties of the AcrB efflux pump, offering valuable insights for drug‑discovery and computational screening efforts.
Pharmacophore models (RHHHa, HHHa) comprised hydrophobic centroid and aromatic center These features are consistent with MBX2319 structure, which consisted of aromatic pyridine ring It resulted in an extensive 7Ĩ.-71 stacking
Targeting the hydrophobic trap of AcrB—specifically the aromatic side chain of Phe628—presents a promising strategy for inhibitor design Phe628 is positioned in a narrow, phenylalanine‑lined groove within the substrate‑binding site, forming a key hydrophobic pocket Pharmacophore models that incorporate hydrophobic centroid features can effectively engage this location, indicating that compounds with such characteristics are strong candidates as AcrB inhibitors.
Figure 3.7 Pharmacophore model with the best MBX2319 conformation (having the lowest RMSD) (A) MBX2319 structure, (B) PH411 with MBX2319
Carbonyl cyanide‑m‑chlorophenylhydrazone (CCCP) is a well‑known laboratory efflux‑pump inhibitor (EPI) that functions as an ionophore, collapsing the proton motive force (PMF) by disrupting both its electrical (ΔΨ) and chemical (ΔpH) components This collapse makes bacterial cells metabolically inactive, prompting debate over whether CCCP’s synergistic effects with antibiotics are due to efflux‑pump inhibition or to metabolic shutdown Research shows that CCCP can revive tetracycline activity against *Helicobacter pylori* and *Klebsiella* spp., and it enhances carbapenem efficacy independently of efflux inhibition, supporting the metabolic inactivity hypothesis Nevertheless, CCCP’s toxicity toward mammalian cells limits its use to laboratory studies.
model predict that PApN interacts with residues **Phe35**, **Phe78**, **Phe615**, **Phe628**, **Gly176**, and **Gly11673**, supporting its competitive‑inhibition mechanism Although some evidence suggests that PApN may also increase outer‑membrane permeability, further investigation is needed to confirm this effect.
Unlike compounds that fit pharmacophore models, these agents inhibit via a non‑competitive mechanism and do not compete with substrates CCCP functions as an energy‑dependent efflux pump inhibitor (EPI) that de‑energizes bacterial membranes, while PApN effectively reverses resistance despite its relatively low binding affinity for the AcrB pump compared with substrates such as minocycline, doxorubicin, and novobiocin The modest AcrB affinity of PApN aligns with previous findings that it is rapidly expelled by the pump, and its potent inhibitory effect is attributed to its well‑known outer‑membrane permeabilizing activity, which drives synergistic activity rather than direct substrate competition (Mowla et al., 2019).
In conclusion, compounds comprising at leat ahydrogen bond andtwo hydrophobic centroids are potential inhibitors of AcrB.
Screening on TCM database
Traditional Chinese Medicine (TCM) serves as a comprehensive, searchable database that delivers detailed information on herbal formulae, individual herbs, and their active ingredients, effectively bridging the gap between ancient Chinese healing practices and modern life‑science research By integrating traditional knowledge with contemporary scientific validation, TCM enables researchers, clinicians, and health‑enthusiasts to explore evidence‑based applications of Chinese herbal medicine, fostering innovations in integrative health, drug discovery, and personalized therapeutics This robust platform not only preserves centuries‑old wisdom but also enhances accessibility, ensuring that the therapeutic potential of Chinese herbs is fully recognized and utilized in today’s evidence‑driven medical landscape.
Our database compiles comprehensive information on 41 drugs and diseases extensively researched by modern pharmacology and biomedical science, seamlessly bridging these two knowledge domains by providing detailed drug‑target data and disease‑associated genes and proteins This integrated resource empowers researchers, clinicians, and drug developers to explore therapeutic pathways, identify novel targets, and accelerate the translation of biomedical discoveries into effective treatments.
Traditional Chinese Medicine (TCM) and conventional Western medicine share a core principle: both treat diseases through chemical molecules that interact with dysfunctional proteins While their underlying philosophies differ, screening TCM compounds offers valuable insights for modern drug discovery, helping researchers uncover potential new therapeutics and understand the mechanisms behind drug interactions By integrating TCM‑derived molecules into conventional research pipelines, scientists can accelerate the identification of effective treatments for a wide range of human diseases.
Figure 3.8 Virtual screening result of TCM
According to the previous of Pharmacophore model results, 3 selected models would be used to screen The process and summarized results ofTCM database was displayed in Figure 3.7
The TCM database contains 57,423 compounds, and after rigorous screening, 18,435 natural compounds satisfied Lipinski’s five‑rule criteria From this subset, 11,320 compounds matched the PH4 5 pharmacophore model, while 11,825 matched the PH4 9 model, and 10,016 progressed to the subsequent virtual‑screening stage The top‑ranking molecules with the lowest root‑mean‑square deviation (RMSD) are listed in Table 3.3, and the optimal pharmacophore model featuring the best Nagarine conformations is illustrated in Figure 3.8.
Table 3.3 Virtual screening on TCM
PH4_5 model with the best
PH49 model with the bestNagarine conformation.
PH4_11 model with the bestNagarine conformation.
Figure 3.9 Pharmacophore models with the best Nagarine conformations
DrugBank is a curated pharmaceutical knowledge base that powers precision medicine, electronic health records, and drug discovery worldwide Leveraging the robust DrugBank Platform, it offers a comprehensive suite of products designed for researchers, clinicians, and industry professionals With a global presence across multiple sectors, DrugBank supports advanced therapeutic development and personalized healthcare solutions, making it an essential resource for precision medicine initiatives and electronic health record integration.
Figure 3.10 Virtual screening result of Drugbank
Drugbank organized all the important information about drugs into a single unified resource Screening on Drugbank opens opportunities to understand unapproved use ofanapproved drug.
According to the previous of Pharmacophore model results, 3 selected models would be used to screen The process and summarized results of Drugbank was displayed in Figure 3.9
Out of the 8,823 compounds listed in DrugBank, 56 candidates met the stringent validation criteria (RMSD ≤ 0.5) and matched three pharmacophore models, qualifying them for subsequent virtual screening These selected compounds displayed the lowest RMSD values, as detailed in Table 3.4, and were evaluated using the most predictive pharmacophore model derived from Dexamethasone isonicotinate conformations This focused screening workflow highlights the power of RMSD‑based filtering and pharmacophore alignment in identifying promising drug‑like molecules from large chemical databases.
Table 3.4 Virtual screening on Drugbank
PH4_5 model with the best
PH4_9 model with the best
PH4_11 model with the best
Figure 3.11 Pharmacophore models with the best Dexamethasone isonicotinate conformations 3.6 Screening on Chrontolaena odorata Asteraceae, Solatium torvum Solanaceae, Vernonia amygdalina Asteraceae and Glinus oppossitiflius Molluginaceae
14 compounds extracted from Chromolaena odorata Asteraceae, Solatium torvum Solanaceae, Vernonia amygdalina Asteraceae and Glinus oppossitiflius
Molluginaceae were determined by 3 models, 10 compounds matchedthree models: PH4 5, PH4 9 and PH4_10.Their structures were exhibited in the Figures 3.3
They are Saponins extracted from Solatium torvum Solanaceae namely S-l ( neochlorogenin 6-O-a-L-rhamnopyranosyl-(1—>3)-p-D-quinovopyranoside), S-2 ((25S)-6a-hydroxy-5a-spirostan-3-on 6-O-a-L-rhamnopyranosyl-( 1 —>3)-0-D- quinovopyranoside), S-3 (Solanolactoside A), Saponin extracted from Glinus
47 oppossitiflius Molluginaceae namely G-8( 3-O-p-D- xylopyranosyl- spergulageninA) and G-9(Spergulin A).
Flavonoids extracted from Glinus oppossitiflius Molluginaceae namely G-10 (vitexin), Spegulacin, speculin B Flavonoids extracted from Chromolaenaodorata
Asteraceae namely Kaempferol-3-O-rutinoside and Ọuercetin-3-O-rutinoside. Flavonoids extracted from Vernonia amygdalina Asteraceae namely Apigenin-2 and Apigenin.
As can be seen, there were 7 Saponins and 3 Flavonoids matched Pharmacophore models Pharmacophore models with the best S-2 conformations was shown in
Table 3.5 Pharmacophore models with the best S-2 conformation were shown in
In conclusion, extracted compounds matched the pharmacophore models had hydrophobic carbon chains and hydroxyl forming hydrogen bond They are Saponin and Flavonoid derives, which are potential inhibitors ofAcrB.
Table 3.5 Virtual screening on extracted compounds
((25S)-6a-hydroxy-5a-spirostan-3-on 6-O-a-L rhamnopyranosyl-( 1 —>3)-P'D- quinovopyranosid)
PH4_5 model with the best
PH4_9 model with the best S-2 conformation
PH4_11 model with the best S-2 conformation. RMSD=0.31
Figure 3.12 Pharmacophore models with the best S-2 conformations
In this study, a set of 119 compounds extracted from 12 scientific articles was used to create eleven novel pharmacophore models for Escherichia coli AcrB inhibitors—an effort not previously undertaken The models were generated with the Pharmacophore Elucidator tool in MOE 2008.10 and feature aromatic center (Aro)/π‑ring center (πR), hydrophobic centroid (H), and hydrogen‑bond acceptor projection (Acc2) The resulting pharmacophore patterns include RHHH, RHHHa, RHHa, HHHa, and HHaa, with the top three models—two HHHa variants and one HHaa variant—demonstrating the highest predictive performance for AcrB inhibition.
Predictive pharmacophore modeling of the AcrB efflux pump revealed that doxorubicin, lanatoside C, and MBX2319 are potential inhibitors, highlighting the models’ strong predictive ability The analysis shows that compounds possessing two hydrophobic centroids combined with a hydrogen‑bond acceptor moiety exhibit high affinity for the AcrB protein, indicating that such structural features are key determinants for effective AcrB efflux pump inhibition.
Using top pharmacophore models, a virtual screening of 57,423 compounds—from the Traditional Chinese Medicine (TCM) database, 14 plant‑derived extracts, and 8,823 DrugBank medicines—identified 442 TCM compounds, 56 DrugBank molecules, and 10 plant extracts that met the pharmacophore criteria These hits all contain hydroxyl groups capable of forming hydrogen bonds with key AcrB residues and hydrophobic centroids, perfectly matching the defined pharmacophore features and highlighting their potential as AcrB inhibitors in drug discovery.
All screening natural compounds , specifically compounds can extract from plant materials present in Viet Nam, extracted compounds from Solatium torvum
Solanaceae, Chromolaena odorata Asteraceae and Glinus oppossitiflius
The Department of Pharmacognosy is investigating bioactive compounds from the Molluginaceae family alongside approved medicines using advanced molecular docking techniques on the AcrAB‑TolC crystal structure to evaluate ligand‑protein binding affinity Promising candidates identified from these in silico studies will subsequently undergo in vitro biological assays against *E. coli* strains that over‑express the AcrAB‑TolC efflux pump, providing critical insight into their potential as effective antimicrobial agents.
Aparna, V., Dineshkumar, K., Mohanalakshmi, N., Velmurugan, D., & Hopper, w
(2014) Identification of natural compound inhibitors for multidrug efflux pumps of Escherichia coli and Pseudomonas aeruginosausing in silico high- throughput virtual screeningand in vitro validation PloS one, 9(7), e 101840
Blair, J M., Webber, M A.,Baylay, A J., Ogbolu, D o., & Piddock, L J (2015)
Molecular mechanisms of antibiotic resistance Nature reviews microbiology,
Blanco Torres, p (2019) Inducible and acquired antibiotic resistance in
Dror, O., Schneidman-Duhovny, D., Inbar, Y., Nussinov, R., & Wolfson, H J.
(2009) Novel approach for efficient pharmacophore-basedvirtual screening: methodand applications.Journal of chemicalinformation andmodeling,
Edward, w Y., Aires, J R., & Nikaido, H (2003) AcrB multidrug efflux pumpof
Escherichiacoli: composite substrate-binding cavityof exceptional flexibility generates its extremely wide substrate specificity Journalof bacteriology, /55(19), 5657-5664.
Fernandez, L., & Hancock, R E (2012) Adaptive and mutational resistance: role of porins andefflux pumpsin drugresistance Clinical microbiologyreviews,
Kineses, A., Varga, B.,Csonka, A., Sancha, s., Mulhovo, s., Madureira, A M.,
Spengler, G (2018) Bioactivecompounds from the African medicinal plant Cleistochlamys kirkii as resistancemodifiers in bacteria Phytotherapy
Larners, R p., Cavallari, J F., & Burrows, L L (2013) The efflux inhibitor phenylalanine-arginine beta-naphthylamide (PApN) permeabilizesthe outer membrane of gram-negative bacteria PloSone, 5(3), e60666.
Lionta, E., Spyrou, G., KVassilatis, D., & Coumia, z (2014) Structure-based virtual screening for drug discovery: principles, applications and recent advances Currenttopics in medicinalchemistry, 14(\6), 1923-1938.
Mowla, R., Wang, Y., Ma, s., & Venter, H (2018) Kinetic analysis of the inhibitionof the drug efflux protein AcrB using surface plasmon resonance
Biochimica et Biophysica Acta (BBA)-Biomembranes, 1860(4), 878-886.
Nikaido, H., & Pages, J.-M (2012) Broad-specificity efflux pumps and their role in multidrug resistance of Gram-negative bacteria FEMS microbiology reviews, 36(2), 340-363.
Pages, J.-M., & Amaral, L (2009) Mechanisms ofdrugefflux and strategies to combatthem: challenging the efflux pump of Gram-negative bacteria
Biochimica et Biophysica Acta (BBA)-Proteins andProteomics, 1794(5),
Pao, s s., Paulsen, I T., & Saier, M H (1998) Major facilitator superfamily
Rao,M., Padyana, s., Dipin, K., Kumar, s., Nayak, B., & Varela, M (2018).
Antimicrobial compounds ofplant origin as efflux pump inhibitors: new avenues for controlling multidrug resistant pathogens J Antimicrob Agents,
MolecularOperating Environment (MOE) 2008.10 Chemical Computing Group. Retrieved from https://www.chemcomp.com/Products.htm
Seukep, A J., Kuete, V., Nahar, L., Sarker, s D., & Guo, M (2019) Plant-derived secondary metabolites as the main source of efflux pumpinhibitors and methods for identification JournalofPharmaceutical Analysis.
Sharma, A., Gupta, V K., & Pathania, R (2019) Efflux pump inhibitors for bacterial pathogens: From bench to bedside The IndianJournal of Medical Research, 149(2), 129.
Shriram, V., Khare, T., Bhagwat, R., Shukla, R., & Kumar,V (2018) Inhibiting bacterial drug efflux pumps via phyto-therapeutics to combat threatening antimicrobial resistance Frontiers inMicrobiology, 9, 2990.
Su, C.-C., Nikaido, H., & Edward, w Y (2007) Ligand-transporter interaction in the AcrB multidrug efflux pump determined by fluorescence polarization assay FEBSletters, 581(25),4972-4976.
Yang, S.-Y (2010) Pharmacophore modeling and applications in drugdiscovery: challenges and recent advances Drugdiscoverytoday, /5(11-12), 444-450. Zgurskaya, H I., Lopez, c.A., & Gnanakaran, s (2015) Permeability barrier of
Gram-negative cell envelopes and approaches to bypass it.ACS infectious diseases, 7(11), 512-522.
Kobylka, J.,Kuth, M s., Muller, R T., Geertsma, E R., & Pos, K M (2020)
AcrB: a mean, keen, drug effluxmachine Annals of theNew York Academy of Sciences, 1459(1), 38-68.
Appendix 1 Testing set AP-2 Appendix 2 Training set AP-27