Investigation of the anti-TB potential of selected propolis constituents using a molecular docking approach

Human tuberculosis (TB), caused by Mycobacterium tuberculosis, is the leading bacterial killer disease worldwide and new anti-TB drugs are urgently needed. Natural remedies have long played an important role in medicine and continue to provide some inspiring templates for drug design. Propolis, a substance naturally-produced by bees upon collection of plant resins, is used in folk medicine for its beneficial anti-TB activity. In this study, we used a molecular docking approach to investigate the interactions between selected propolis constituents and four ‘druggable’ proteins involved in vital physiological functions in M. tuberculosis, namely MtPanK, MtDprE1, MtPknB and MtKasA. The docking score for ligands towards each protein was calculated to estimate the binding free energy, with the best docking score (lowest energy value) indicating the highest predicted ligand/protein affinity. Specific interactions were also explored to understand the nature of intermolecular bonds between the most active ligands and the protein binding site residues. The lignan (+)-sesamin displayed the best docking score towards MtDprE1 (−10.7 kcal/mol) while the prenylated flavonoid isonymphaeol D docked strongly with MtKasA (−9.7 kcal/mol). Both compounds showed docking scores superior to the control inhibitors and represent potentially interesting scaffolds for further in vitro biological evaluation and anti-TB drug design.

by honeybees mainly upon collection of plant secretions, such as resins and sticky exudates on leaf buds and plant wounds. The word propolis is derived from Greek, in which pro means "at the entrance to" and polis means "community" or "city". Bees use propolis as a construction and repair material to seal gaps, smooth out internal walls in their hives and as an antiseptic coating to generally protect from external contamination. Propolis has a highly variable chemical composition depending on the geographical location from where it is collected. For instance, propolis from temperate regions of the world is rich in phenolic compounds derived from poplar tree exudates whereas bees in tropical countries have different plant sources at their disposal resulting in propolis types rich in other phytochemicals such as prenylated flavonoids and benzophenones, lignans, terpenoids and phenolic lipids [9][10][11][12][13] . Propolis has a long history of use as a folk remedy to treat a variety of ailments 14 . Numerous scientific studies have been carried out to investigate its medicinal properties, including anti-inflammatory 15 , immunostimulant 16 , anti-oxidant 17 , antitumour 18 , neuroprotective 19 and antimicrobial activity 12,20,21 . Interestingly, propolis has been used as an ingredient in traditional cures for tuberculosis [22][23][24][25] . Previous in vitro studies have demonstrated that extracts of propolis could inhibit the growth of M. tuberculosis as well as synergise the effect of established antitubercular drugs such as isoniazid, rifampicin and streptomycin 26,27 . It has also been observed that propolis inhibited the development of TB by lowering necrosis formation in granulomas of M. tuberculosis-infected animals 28 .
Several enzymes involved in vital physiological functions in M. tuberculosis have been identified as novel attractive molecular targets for anti-TB drug development [29][30][31][32] . Here, we used a guided docking approach with AutoDock Vina to predict the interactions between selected propolis constituents and four of these essential mycobacterial enzymes, namely pantothenate kinase (MtPanK, type 1) 33 , decaprenylphosphoryl-β-D-ribose 2′-epimerase 1 (MtDprE1) 34 , protein kinase B (MtPknB) 35 and β-ketoacyl acyl carrier protein synthase I (MtKasA) 36 . Molecular docking is a popular tool used in the virtual screening of small molecules (ligands) against proteins (targets) and several studies have successfully used AutoDock Vina to investigate the interactions of natural products against specific protein targets, including mycobacterial enzymes [37][38][39][40][41] . The docking of propolis constituents towards MtPanK, MtDprE1, MtPknB and MtKasA, however, has never been reported.

Results
The propolis constituents investigated in this study represent some structurally diverse compounds that we grouped into four main categories, namely flavonoids, terpenoids, simple phenolics and miscellaneous substances including a pterocarpan, a phenylethanoid derivative, five stilbenes and four lignans. Known molecules, that had been reported previously in the literature as inhibitors of the target enzymes and for which the nature and role of the binding site residues were known from their available complexes with the proteins, were used as controls. In order to validate the docking conditions prior to virtually screening the propolis constituents, each control inhibitor was retrieved from its co-crystallised complex and re-docked using the AutoDock Vina software against the relevant target. Then, a docking score for each propolis compound was calculated to estimate its binding free energy towards MtPanK, MtDprE1, MtPknB and MtKasA (Table S1). The docking score values obtained for compounds within each phytochemical class were compared to the scores of the control inhibitors for each target in order to select molecules with the lowest energy values that ranked higher than the chosen control inhibitors against the target proteins. We observed that none of the propolis constituents exhibited scores that ranked better than the controls neither against MtPanK nor MtPknB. Instead, only docking to MtKasA and MtDprE1 gave useful scores. Thus, the prenylated flavanones isonymphaeol D, isonymphaeol C and isonymphaeol B showed strong docking scores towards MtKasA (−9.7, −9.6 and −9.5 kcal/mol, respectively) superior to the control inhibitor thiolactomycin (−7.9 kcal/mol). The Ki of isonymphaeol D for MtKasA was estimated at 0.07 μM (control was 1.62 μM). Isonymphaeol D also showed a strong predicted binding towards MtDprE1 (−10.1 kcal/mol) compared with the control inhibitor 0T4 (−9.2 kcal/mol). Among the terpenoids, we observed that the oleanane-type triterpene β-amyrin acetate showed some affinity for MtDprE1 (−9.9 kcal/mol) and ranked better than 0T4 (−9.2 kcal/mol). In the simple phenolics group, (+)-chicoric acid exhibited a strong binding score (−9.5 kcal/ mol) for MtKasA compared with thiolactomycin (−7.9 kcal/mol). The stilbene 5-((E)-3,5-dihydroxystyryl)-3-((E)-3,7-dimethylocta-2,6-dien-1-yl) benzene-1,2-diol showed a docking score against MtKasA (-9.4 kcal/mol) that also ranked better than thiolactomycin while the lignan (+)-sesamin docked strongly to MtDprE1 with a score (−10.7 kcal/mol) and predicted Ki (0.01 μM) better than 0T4 (−9.2 kcal/mol and Ki of 0.18 μM) ( Table 1).
Specific interactions were further explored to understand the nature of the intermolecular bonds formed between selected compounds and the binding site residues for the four studied enzymes (Table S2). The binding poses obtained for the best binding ligands isonymphaeol D and (+)-sesamin were visually inspected and are depicted in Figs 1 and 2, respectively. We observed that isonymphaeol D showed some key molecular interactions with the key residues Pro280, Phe402 and His311 of MtKasA ( Fig. 1) and (+)-sesamin interacted with the key residues Cys387, Ser59 and Gly117 of MtDprE1 (Fig. 2). The best score towards MtPanK was observed for the flavonoid pinobanksin-3-(E)-caffeate (−10.0 kcal/mol) but protein-ligand interactions were not investigated for this compound as this ranked lower than the score obtained for the control inhibitor ZVT (−10.9 kcal/mol). The highest affinity towards MtPknB was observed for the flavonoids pachypodol and pinobanksin-3-(E)-caffeate (−9.1 kcal/mol) but again this ranked lower than the score obtained for the control inhibitor mitoxantrone (−10.8 kcal/mol) and was not investigated any further.

Discussion
In this study, we investigated the anti-TB potential of a range of propolis compounds using a guided molecular docking approach with a view to characterise their affinity towards the four mycobacterial enzymes, MtPanK, MtDprE1, MtPknB and MtKasA. The rationale for the selection of these particular proteins was that these are key enzymes required for M. tuberculosis to grow and survive within the eukaryotic host. They are also involved in a variety of essential mycobacterial pathways such as cell wall biogenesis, cofactor biosynthesis and signal transduction. They are absent in mammalian cells, which makes them highly selective and attractive 'druggable' targets for mycobacterial diseases, and they represent some newly-validated emerging targets against which no marketed drug is currently available [33][34][35][36] . Pantothenate kinase type I from M. tuberculosis (MtPanK) is an enzyme that catalyses the first step in the biosynthesis of the cofactor Coenzyme A (CoA) by converting pantothenate (vitamin B 5 ) to 4′-phosphopantothenate 33 . Serine/threonine protein kinases, such as protein kinase B (MtPknB), which is implicated in the regulation of mycobacterial cell morphology, play an important role in signal transduction pathways and allow M. tuberculosis to grow and survive successfully within the host 35,42 . As the mycobacterial cell wall is a complex structure comprising layers of peptidoglycan, arabinogalactan, lipoarabinomannan and some mycolic acids, two key protein targets in the M. tuberculosis cell wall biosynthesis, β-ketoacyl acyl carrier protein synthase I (MtKasA) and decaprenylphosphoryl-β-D-ribose 2′-epimerase 1 (MtDprE1), were also included in this study 34,36 . The presence of mycolic acids is a unique feature of the mycobacterial cell wall. These very-long chain fatty acids, which have been linked with the ability of mycobacteria to survive in the host and to resist many antibiotics, are produced through the activity of a range of fatty acid synthases (FAS). MtKasA is one of the enzymes of the mycobacterial type II FAS pathway, which is only found in bacteria 36 . A closer look at the interactions between isonymphaeol D and MtKasA reveals that the prenylated tail of this flavonoid binds to the hydrophobic pocket of MtKasA that contains Pro280 and Phe402. Furthermore, a strong hydrogen bond (contact distance 2.77 Å) was observed between the C-4′ phenolic oxygen of isonymphaeol D and a nitrogen of the His311 residue at the active site, in close similarity to what has been previously described as the mode of binding of the TLM control 36 . The mycobacterial cell wall enzyme decaprenylphosphoryl-β-D-ribose 2′-epimerase 1 (MtDprE1) participates in the biosynthesis of two fundamental mycobacterial cell wall components, namely arabinogalactan and lipoarabinomannan 43 . The Cys387 in the active site of MtDprE1 has been identified as a critical residue for the binding, through a covalent bond, of the control inhibitor 0T4 (also called CT325) 34 . In the case of (+)-sesamin, the interactions observed were not via covalent bonds but involved a π-sulfur interaction with Cys387, and strong hydrogen bonds between oxygens of the methylenedioxy and the tetrahydrofuran moieties and Ser59 and Gly117 (contact distances 3.08 and 2.94 Å, respectively).
The purpose of molecular docking is to use scoring algorithms to estimate the likelihood that a given compound will bind to a protein target. We have identified the lignan (+)-sesamin and the prenylated flavonoid isonymphaeol D from propolis as being the best predicted binding ligands for MtDprE1 and MtKasA, respectively. Interestingly, isonymphaeol D displayed a strong predicted binding towards both enzymes, which suggests that it is a particularly promising agent as it has been demonstrated that the odds of successfully discovering active compounds using structure-based virtual screening methodologies are greater when a single compound can target multiple proteins 63 . Both (+)-sesamin and isonymphaeol D showed docking scores ranking higher than those obtained for the known control inhibitors of the target proteins and had predicted activities at the target sites lower than 0.1 μM. There was, however, a lack of correlation between the strong predicted affinity of (+)-sesamin for MtDprE1 (score of −10.7 kcal/mol and Ki of 0.01 μM) and its observed (moderate) activity It has been previously reported that a direct correspondence between in silico molecular docking results and in vitro biological parameters cannot always be established. This can be due to the fact that some compounds are not able to go through the complex mycobacterial cell wall, or the characteristics of the binding site where inhibition takes place is different in vivo 64 . Isonymphaeol D and (+)-sesamin may not be used as such clinically. However, as most bioactive natural products, they represent some potentially interesting "hits" 65 that can be further structurally optimised for the design of new anti-TB drugs and they warrant further in vitro biological evaluation.

Methods
Ligand selection. The  Ligand and protein preparation. Each ligand structure was drawn using ChemOffice v.15.1 and geometry optimised using MM2 energy minimisation 80 . All rotatable bonds present on the ligands were treated as

Identification of binding site residues.
Previous studies were used to identify the nature and the role of the binding site residues for MtPanK type 1 33 , MtDprE1 34 , MtPknB 35 and MtKasA 36 . Specific amino acids involved in ligand/protein interactions were also confirmed following the analyses of the PDB crystal structures available for each target protein in complex with either natural substrates or control inhibitors (Table S3).   81 . To validate the accuracy of the docking and to allow a comparison between docking scores, all co-crystallised inhibitory ligands were re-docked into the corresponding protein structures. Different orientations of the ligands were searched and ranked based on their energy scores. Our docking protocol was able to produce a similar docking pose for each control ligand with respect to its biological conformation in the co-crystallised protein-ligand complex. We further visually inspected all binding poses for a given ligand and only poses with the lowest value of RMSD (simply root-mean-square deviation) (threshold < 1.00 Å) were considered to gain a higher accuracy of docking. The Lamarckian Genetic Algorithm was used during the docking process to explore the best conformational space for each ligand with a population size of 150 individuals. The maximum numbers of generation and evaluation were set at 27,000 and 2,500,000, respectively. All other parameters were set as default. As the active binding sites and some control inhibitors for our four selected mycobacterial enzymes have been well-characterised in previous studies [33][34][35][36] , we decided to use a guided docking approach to increase docking efficiency 37 by sampling each ligand conformation (including re-docking of the control inhibitors) in each protein binding site and then ranking these conformations using a scoring function to predict the best protein-ligand binding affinities (calculated as the predicted binding free energies ΔG bind in kcal/mol) ( Table S1). The lowest binding free energy (i.e. best score of the docking pose with the least root mean square deviation) indicated the highest predicted ligand/protein affinity. The Auto Dock Vina docking scores of these selected propolis constituents which ranked higher than a control inhibitor were further used to calculate the predicted inhibition constants (Ki values) of selected compounds against a given target (Table 1) 85 . Specific intermolecular interactions with the targets (  Table 2. Grid box parameters selected for target enzymes, based on binding site residues a . a Spacing and exhaustiveness values were set up at 1 Å and 9, respectively in all cases.