Molecular docking studies of gyrase inhibitors: weighing earlier screening bedrock

Abstract

For any antimicrobial assay, a standard drug is used to compare the bactericidal efficiency of the bioactive compound under screening. The standard drugs have different targets that may be intracellular or membrane located. The location of the target is believed to be determining the bioactivity of the drug depending on the drug's access to its target. Therefore, different drugs must have a different magnitude in exhibiting the biological effect. However, in most of the published literature about the screening of bioactive compounds on antimicrobial activity, generally, the standard drug is randomly chosen while comparing against the bioactive compound of interest. Further, the antimicrobial activity is inferred by comparing the randomly chosen standard drugs without knowing the physicochemical parameters of the standard drug and the test molecule. It is just like an unfair comparison of the impact of a bullet with the impact of an explosive in a combat scene. Computer-based strategies for structure-based drug discovery presents a valuable alternative to the costly and time-consuming process of random screening. The docking studies provide better insights into the binding mechanism of substrate and inhibitor at the molecular level. The evaluation of such a comparison of bioactive compounds against randomly selected standard drugs through a customized virtual screening pipeline showed 57% false positives, 18% true positive, 17% true negative, 8% false-negative results. This study directs for mandatory cheminformatics-based assessment of the bioactive compounds before choosing the standard drug to compare with.

Supplementary Information

The online version contains supplementary material available at 10.1007/s40203-020-00064-9.

Introduction

The increase of many deadly diseases like infections by multidrug-resistant bacteria implies re-inventing the wheel on drug discovery (Bagchi et al. 2013). Infectious diseases are still one of the most important causes of mortality in humans especially in children (Isitua et al. 2016). They are the second most deadly diseases in the world, only behind cancer. Worldwide, 1.7 million people succumb to bacterial infections every year.

For the past 2 decades, high throughput screening has served as the most common approach to identify antibacterial compounds for further development, whether for use alone or in combination with other compounds. Bioactive compounds have been used from plants, animals, and microbes from time immemorial as an important weapon against bacterial infections. There exist, different classes of secondary metabolites that are bactericidal in different mechanisms. The bioprospecting procedure is done in two ways, classical pharmacology and reverse pharmacology (Kraus 2008). Antibacterial bioassays such as disk-diffusion, well diffusion, and broth or agar dilution, and other more recent methods that are not widely used to date, such as methods using flow cytometry and bioluminescence are being employed for determining the efficacy of moleule under test (Rashid et al. 2001).

Quite often in the published literature (Supplementary table), it has been observed that the bioactive compounds are screened against standard drug which may or may not relate chemically with the molecule under testing. This kind of unfair comparison results in underreporting or over the interpretation of the efficacy of the bioactive compound under test. To address this problem, one can employ a wide range of screening strategies, firstly, to understand the structure and class of the bioactive compound and secondly to choose an appropriate standard molecule that belongs to the same class of the bioactive compound under study. One such method makes use of computational tools including molecular docking and virtual screening (Wu et al. 2003).

Virtual screening (VS) has emerged as an important tool in identifying bioactive compounds through computational means, by employing knowledge about the protein target or known bioactive ligands (Hara et al. 1997). Virtual screening has appeared as an adaptive response to the massive throughput synthesis and screening paradigm as a necessity to counter the sharp increase in the cost of drug discovery along with the high rate of failure in the clinical trials and it has forced the computational chemistry community to develop tools that screen against any given target and/or property of millions or perhaps billions of molecules in a short period (Hara et al. 1997). Recently, many reports have shown that the field of computer-aided virtual screening can successfully assist in discovering new potential drug candidates with the motive of increasing the chance of discovery and decrease in time, labor, and money (Duch et al. 2007). Thus, VS approaches have gained immense popularity and have become an integral part of industrial and academic research, directing drug design and discovery. Although the VS is employed for the identification of bioactive compounds, the same pipeline can be used to re-evaluate the wet lab screening procedure which is mentioned in the previous paragraph.

The present study uses a computational pipeline to evaluate the sanity of comparison of the bioactive molecules tested with their corresponding standards with special reference to DNA gyrase inhibitors and weighs the choice of standards for antibacterial activity screening.

Materials and methods

Data collection

All Ph.D. theses containing antimicrobial activity studies of the bioactive compounds available at the library of Kuvempu University (Jnana Sahyadri campus) were referred for the structure of the bioactive compounds. Apart from this to increase the size of the compound library, earlier literature was also used. The scientific web was searched via J-Gate which is considered as an electronic gateway to global e-journal search. In the present work, 145 ligands were screened, among them, 139 molecules were selected from the literature survey of J-GATE research articles, and 7 molecules selected from the survey of Kuvempu University life science Ph.D. theses as a reference. The selection of the data set was based on the following parameters:

1. The compound should be elucidated structurally by either NMR or X-ray diffraction

2. The pure compound should be used for antibacterial screening against a known standard

3. The results of the screening should be present in the public domain as a paper or thesis

Preparation of ligand datasets

The ligand datasets were drawn in Marvin sketch and smiles were generated. These smiles were submitted to the Molinspiration server for the validation of the Rule of Five predictions. Molinspiration supports the internet chemistry community by offering free on-line services for calculation of important molecular properties as well as prediction of bioactivity score for the most important drug targets (GPCR ligands, kinase inhibitors, ion channel modulators, nuclear receptors). The ligands were drawn using Marvin Sketch, and the 3D coordinates were generated. The pharmacophore parameters and ADME-Tox properties of ligands were examined with OSIRIS DATAWARRIOR software and at the SWISSADME server. The results were tabulated and PAST software was used for the pharmacophore-based clustering of the ligands based on the Jaccard Similarity index method. The nexus file generated during the clustering was used to generate the cladogram using GeneiousPro 4.5 software.

The molecules were energy minimized using Avogadro with the UFF force field resulting in the geometrical optimization and energy minimization of the molecule. The molecules were saved in the “filename.pdb” which is the input format for Autodock software.

Identification and target preparation

The target protein for this study was chosen based on two criteria.

• Essentiality of the protein for the survival of the organism

• Frequency of the standard drug molecule used in antibacterial screening reports.

Super target is a database that helps to identify the druggable protein of considerable sequence conservation so that the target protein occurs in a wide range of bacteria. Search for the antimicrobial target in the Super target database leads to the selection of DNA gyrase as a potential target for further studies because it is involved in DNA supercoiling as an indispensable part of the replisome. The structure of the gyrase enzyme differs to a larger extent between the bacteria and humans, hence avoiding the scope for cross inhibition and side effects. Most of the molecules taken from the published reports are compared with ciprofloxacin as the standard drug which is a well-known inhibitor of DNA gyrase. The structure file PDB ID:2XCT was retrieved from the protein data bank (www.rcsb.org/pdb) in the “filename.pdb” format. The structure file was prepared for docking according to previously reported protocol (Trott and Olson 2010).

Virtual screening

Screening for the energy minimized ligands was done by docking all the molecules against DNA gyrase PDB ID: 2XCT. In the present work, 145 ligands are screened, among them, 139 molecules were selected from the literature survey of J-Gate research articles, and 7 molecules are selected from the survey of Ph.D. theses submitted to the School of Biosciences, Kuvempu University.

AutoDock Vina is an efficient and open-source program for protein–ligand docking and claims to improve the average accuracy of the binding mode predictions compared to the earlier version AutoDock 4.2. It is improved in terms of speed and accuracy of docking with a new scoring function, efficiently optimized for multithreading (Mahesh and Satish 2008). Therefore, the AutoDock Vina was used for virtual screening. The grid box was set with a grid volume of 8, 14, and 14 for X-axis, Y-axis, and Z-axis respectively with the grid center values of 3.194, 43.143, 69.977 respectively for X, Y, and Z centers. The results were tabulated based on the binding energy and physical contacts of the sorted ligands. The interactions were visualized in 2D using LigPlot + (V 1.4.5). The ligand–protein interactions were further visualized in 3D with PyMol.

All the ligands were sorted according to docking results and the ligands are spread into four quadrants—true positive, true negative, false positive, and false-negative.

Results and discussion

Bioprospecting of flora and fauna has served as a source for the identification of numerous therapeutic agents. Both ethnopharmacology based as well as random screening have been proven to be trusted tools for the discovery of new bioactive compounds especially the novel antibiotics. (Kuzu et al. 2014; Lee et al. 2012; Masadeh et al. 2016). In the light of alarming side effects of synthetic drugs, there is a demand, now more than ever for the search of new bioactive compounds from natural sources which make higher plants an ideal source for antimicrobial compounds (Duch et al. 2007; Siddiqui et al. 2019).

In the post-genomic era, with the advent of molecular docking and simulation methods, virtual screening has been used as a tool to predict the probable mechanism of ligand-receptor interaction. The toxicity and efficacy and mode of action can also be predicted nowadays using appropriate servers or software tools. Hence computer-aided drug discovery is an integral part of modern drug development programs and screening programs as well. The issue of unfair comparison between the bioactive compounds and standards earlier discussed can be attempted by using computational methods to develop a strategy for choosing the right standard for the molecule under study.

Computer-based methods are becoming increasingly important and complementary to wet laboratory experiments in studying the structure and function of biomolecules (Spížek et al. 2010). Molecular docking is a frequently used tool in structure-based rational drug design. Although early efforts were hindered by limited possibilities in computational resources, due to recent advances in high-performance computing, virtual screening methods became more and more efficient. These methods have contributed to the development of several drugs and drug candidates that advanced to clinical trials. Examples include lead compounds to prevent myocardial infarction, to treat HIV infection, Alzheimer’s disease, rheumatoid arthritis, and many other diseases (Clark 2008; Jorgensen 2009). Further, docking is often used in conjunction with scoring functions to predict binding affinities of ligands in virtual screening experiments (Lage et al. 2018) and in studying structure–activity relationships to prioritize the synthesis of new compounds (Wheeler et al. 2012).

The bioactive compounds isolated from natural sources are often screened for their antibacterial activity. While screening, the bioactive compounds are generally compared with standard drug, and based on the comparison of inhibition zone formed by standard drug and the bioactive compounds, the potency of the bioactive compound is inferred. Since, nothing is known about the physicochemical properties of the bioactive compound, comparing with randomly chosen standard drug results in an unfair comparison of its potency. Here we present a case study of Ciprofloxacin as a standard drug and how computational resources can be deployed for evaluating the unfair comparison and interpretation involving comparison of bioactive compounds with Ciprofloxacin.

Ciprofloxacin is one of the new generation fluorinated quinolones structurally related to nalidixic acid. It is a broad-spectrum antibacterial drug to which most Gram-negative and Gram-positive bacteria are susceptible or moderately susceptible because it attacks DNA Gyrase, an enzyme crucial for DNA replication, being an integral part of replisome complex (Hughes et al. 2011; Mahesh and Satish 2008). Bacteria seldom develop resistance to ciprofloxacin as established both in vitro and clinically (Ghazal et al. 1992). Several clinical trials have shown that both oral and intravenous ciprofloxacin administration has confirmed the potential of the molecule for its use in a broad range of infections (Campoli-Richards et al. 1988; Ghazal et al. 1992).

The survey of theses and research articles in the public domain culminated in listing 145 ligands. Out of 145 ligands, antibacterial activity was done without a standard drug for 40 ligands (Supplementary datasheet). All the 145 ligands were included in the virtual screening pipeline. Details of the source of ligands have been tabulated (supplementary datasheet). It is interesting to note that, several compounds recurrently reported, including 2-hydroxytrideca-36-dienyl-pentanoate, catechin, chlorogenic acid, kaempferol, naringenin, quercetin, rutin, sitosterol, and β-sitosterol-3-O-β-d-glucopyranoside are from different plants. In each of the studies, these compounds have been compared to different standard compounds (studies cited in Supplementary data). This clearly shows that the same compounds, though from different plants, were compared with different standards either due to inadequate literature review or lack of skill of cheminformatics analysis to decide the choice of standards. The molecular docking study led to several different kinds of binding behavior of ligands and they are classified into four classes or quadrants.

1. True positive Molecules having good gyrase inhibition property and tested against ciprofloxacin in wet-lab screening reports. Few ligands and their statistics are given in Table 1.

   Mearnsetin is an O-methylated flavanol, whereas ellagic acid, syringic acid, protocatechuic acid are phenolic acids. Protocatechuic acid, caffeic acid, and p-coumaric acid have been reported to possess anti-microbial effects when compared with Ciprofloxacin in earlier studies (Mahesh and Satish 2008; Siddiqui et al. 2019 Mar 1; Yang et al. 2014). Mearnsetin has been shown to exhibit antimicrobial effect in comparison with levofloxacin (Biswas et al. 2012). The same ligands have performed well pharmacodynamically in the current docking study (Figs. 1 and 2). Hence, they are categorized as true positive molecules because the standard drug used both in vivo and in silco comparison has the same target i.e. DNA Gyrase (Ulloora et al. 2013).

2. False-positive Molecules having good gyrase inhibition property and tried and reported against other standards rather than ciprofloxacin. Few ligands and their statistics are given in Table 2.

   Quercetin 3-O-rhamnoside and other ligands were compared with ampicillin, streptomycin, and gentamycin which targets ribosome (Hara et al. 1997). However, from Table 2 it is clear that they are not ribosome binders but are gyrase inhibitors as evident from the pharmacodynamics performance is on par with the ligands in the true positive quadrant (Figs. 3 and 4). Therefore, these ligands have been unfairly compared with ribosome binders than being compared with DNA gyrase inhibitors for the efficacy study. These ligands are positive for gyrase inhibition but are falsely compared with ribosome binders which classify them as false positive ligands.

3. False-negative Molecules that are not gyrase inhibitors but have been reported against ciprofloxacin.

   Friedelin and other compounds do not show any physical contacts in the form of hydrogen bonds, therefore, these ligands though are not DNA gyrase ligands, have been tried and tested against Ciprofloxacin for which they are negative. Friedelin, Oleanonic acid, and Betulinic acid are pentacyclic triterpenoids that have been investigated for their pharmacological activity and none of the pentacyclic triterpenoids have shown to be DNA gyrase inhibitors (Figs. 5 and 6) (Chung et al. 2014; Lee et al. 2012). Dehydroabietylamine has been shown to inhibit pyruvate dehydrogenase kinase (Isitua et al. 2016). This indicates that Table 3 represents an unfair comparison where ligands that do not bind to DNA gyrase are being compared with Ciprofloxacin which is a DNA gyrase inhibitor.

4. True Negative Molecules that are neither gyrase inhibitors nor tried against ciprofloxacin.

Table 1.

Molecules under the true positive quadrant

Ligand name	Standard drug	Target of the standard	Binding energy	#H-bond	Interacting residues
Mearnsetin	Ciprofloxacin	DNA gyrase	− 4.7	5	Ser1085, Asp512, Gly459, Asp437, Arg1122
3,3′-di-O-methylellagic acid			− 5.2	4	Gly1082, Ser1085, Arg1122, Asp437
3,3′-di-O-methylellagic acid 4′-O-β-d-glucopyranoside			− 4.8	4	Ser1085, Gly1082, Arg1122, Asp437
Caffeic acid			− 4	4	Asp437(2), Gly459, Asp512
p-Coumaric acid			− 4	4	Gly459, Asp437, Gly1082, Arg1122
3,3′-di-Omethyl ellagic acid 4′-O-β-d-xylopyranoside			− 5.4	3	Asp437, Lys460, Arg458
Syringic acid			− 3.7	3	Arg1122, Asp437(2)
Protocatechuic acid			− 3.6		Arg1122, Gly459(2)

Fig. 1.

2D and 3D protein and ligand interaction of Mearnsetin

Fig. 2.

2D and 3D protein and ligand interaction of 3,3′-di-O-methylellagic acid

Table 2.

Molecules under the false-positive quadrant

Ligand name	Standard drug	Target of the standard	Binding energy	#H-bond	Interacting residues
Quercetin 3-O-rhamnoside	Streptomycin	30s Ribosomal subunit	− 5.8	7	Asp437, Gly459, Gly1082, Arg1122, Ser1084
Isorhamnetin 3-O-glucoside			− 5	7	Asp437, Ser1084, Arg1122, Gly1082, Asp512
Luteolin			− 4.8	7	Ser1085, Gly1082, Asp437, Gly459, Asp512
Luteolin 7-O-glucoside			− 5.4	6	Lys460, Gly1082, Ser1085, Ser1084
Naringenin			− 4.8	5	Gly1082, Ser1085, Asp512, Asp437
Loganic acid	Gentamycin		− 4.8	6	Ser1085, Gly459, Asp437, Asp512
Gentiopicroside			− 5.3	5	Ser1085, Gly1082, Asp437, Gly459
Swertiamarin			− 5	5	Asp437, Gly459, Ser1085, Gly1082, Ser1084
Amoroswertin			− 5	5	Ser1084, Gly1082, Asp512, Asp437, Gly459
Kaempferol			− 4.7	5	Asp512, Gly1082, Ser1085, Asp437

Fig. 3.

2D and 3D protein and ligand interaction of Quercetin 3-O-rhamnoside

Fig. 4.

2D and 3D protein and ligand interaction of Isorhamnetin 3-O-glucoside

Fig. 5.

2D and 3D protein and ligand interaction of Friedelin

Fig. 6.

2D and 3D protein and ligand interaction of Oleanonic acid

Table 3.

Molecules under false negative quadrant

Ligand name	Standard drug	Target of the standard	Binding energy	#H-bond	Interacting residues
Friedelin	Ciprofloxacin	DNA gyrase	− 5.6	No	No
Oleanonic acid			− 5.5	No	No
Betulinic acid			− 4.7	No	No
Casuarinin			− 5.7	No	No
Dehydroabietylamine			− 4.3	No	No
2-Methoxy 6-decyl benzoquinone			− 3.6	No	No

The true negative group of ligands (Table 4) represents the ligands which are neither DNA gyrase inhibitors nor compared against ciprofloxacin as evident by their binding statistics (Figs. 7 and 8). There is a complete absence of physical contact between the enzyme and the ligand under testing.

Table 4.

Molecules under the true negative quadrant

Ligand name	Standard drug	Target of the standard	Binding energy	#H-bond	Interacting residues
(2,3,10,11)-Dimethyl ene dioxy tetrahydro proto berberine	Netilmicin	30s ribosomal subunit	− 5.4	No	No
Oxysanguinarine			− 5.3	No	No
11-Hydroxy-10-methoxy-(2,3)-methylenedioxytetrahydroprotoberberine			− 5.2	No	No
Oxynitidine			− 5	No	No
Rhoifoline B			− 4.9	No	No
8-Methoxynorchelerythrine			− 4.7	No	No
8-Methoxynitidine			− 4.7	No	No
8,9,10,12-Tetramethoxynorchelerythrine			− 4.6	No	No
5-Methoxydictamnine			− 4.5	No	No
Allanxanthone A	Gentamicin		− 4.8	No	No
Isovitexin			− 5	No	No
Alpha-Amyrin	DMSO		− 5.8	No	No

Fig. 7.

2D and 3D protein and ligand interaction of (2,3,10,11)-dimethyl-ene-dioxy-tetrahydro protoberberine

Fig. 8.

2D and 3D protein and ligand interaction of Oxysanguinarine

The pharmacophore of all the ligands which belongs to all the quadrants is given in Table 5 in electronic supplementary material. A cladogram was developed based on the pharmacophore of the ligands using the Jaccard Similarity Index (Fig. 9). Jaccard similarity methods organize the entities into a clade considering the similarity of the pharmacophore descriptors or attributes about different ligands under comparison.

Fig. 9.

Pharmacophore based clustering of ligands. TP ligands of true positive quadrants, FP ligands of false positive quadrant, FN ligands of false negative quadrant, TN ligands of true negative quadrant

The interesting observation of TP and FP ligands appearing in the same group signifies that the ligands of TP and FP quadrant possess similar pharmacophore descriptors. It is interesting to observe that the molecules of the true negative quadrant are aligned into a single clade except for TN8 and TN9. This suggests that the pharmacophore-based clustering of ligands using Jaccard’s Similarity index can be a good tool for the clustering of ligands. Since the clustering is based on the similarity between the data entities, the cladogram indicates two important facts, firstly, the clustering of true negative molecules into a single clade suggests that the method of clustering is correct and secondly, the intermixing of false-positive and true positive ligands suggests that they are similar concerning pharmacodynamics and cheminformatics parameters. The cheminformatic parameters found to influence the pharmacodynamics behavior of the ligands very much support the argument of the present study. Further, it also suggests that the bioactive compound and the standard drug used for comparison can be chosen by thorough cheminformatic analysis.

The pie chart (Fig. 10) drawn by taking the number of ligands into account shows only 18% of the total number of ligands to be true positives which is a cause for great concern. A large chunk of the surveyed ligands belongs to false negative (57%) which indicates that the antimicrobial screenings are being done without considering the structure–activity relationship (SAR) of the ligands. Therefore, this study recommends doing a proper SAR analysis, and cheminformatics assessments of the isolated bioactive compounds which may help in the identification of appropriate standards to compare with.

Fig. 10.

Pie chart showing the percentage of ligands in all four quadrants

Conclusion

The present study expresses serious concern about the procedure of antimicrobial activity being done. It uses virtual screening coupled with cheminformatic analysis to prove the mistakes in the selection of the standard drug for the comparison of antimicrobial activity. This study also gives the pipeline for the cheminformatics and virtual screening steps to be followed for deciding the standard to compare with given bioactive compounds.

Supplementary Information

Below is the link to the electronic supplementary material.