In silico selection and in vitro evaluation of novel inhibitors of Streptococcus mutans Ag I/II surface protein

Streptococcus mutans is well known for having virulence factors associated with its cariogenic role, such as glucosyltransferases, which have been used as targets for the virtual screening of molecules with inhibitory capacity. The Antigen I/II of S. mutans is involved in the adhesion to the surface of the tooth and the bacterial co-aggregation in the biofilm formation, despite that, this protein has not been used as a target in a virtual strategy search for inhibitors. In this study we identified in silico and evaluated in vitro molecules with adhesion inhibitory potential on S. mutans Ag I/II. A virtual screening of 883,551 molecules was conducted, cytotoxicity analysis on fibroblast cells, S. mutans adhesion studies, scanning electron microscopy analysis for bacterial integrity, and molecular dynamics simulation were also performed. We have found three molecules (ZI-187, ZI-939, ZI-906) that were not cytotoxic and inhibited the adhesion of S. mutans to polystyrene microplates. Molecular dynamic simulation by 300 nanoseconds showed stability of the interaction between ZI-187 and Ag I/II (PDB: 3IPK). This work provides three new molecules that targets Ag I/II and have the capacity to inhibit in vitro the S. mutans adhesion on polystyrene microplates and provides a new computational line for the search and selection of safe inhibitory molecules against different pathogens.


Introduction
In 2016 dental caries was classified as the most prevalent pathology in the world, affecting 2.4 billion people [1,2]. This pathology is one of the oral diseases related to the oral microbiota alteration [3], characterized by perforations or structural damage of the teeth, called carious lesions [4]. There are three well known risk factors for the development of caries: personal factors that are related to socioeconomic status i.e., dental insurance coverage, attitudes and knowledge about oral health and oral hygiene; oral environmental factors such as saliva, fluoride, chewing gum, pH, bacteria, calcium, phosphates and proteins; and factors that directly contribute to the development of caries, like, the tooth, diet (consumption of sugars), bacterial biofilms and time [5].
Oral microorganisms that cannot adhere to a surface are transported by salivary flow out of the mouth and into the digestive tract, but many oral bacteria possess mechanisms of adherence to solid surfaces (co-adhesion), such as coated teeth from salivary films, to squamous surfaces such as epithelial tissue or bacteria that are attached to the surface (co-aggregation) [6]. Ability to form biofilms is one of the main S. mutans characteristics and it has been shown to express virulence factors for biofilm development through two pathways, one dependent and the other independent of sucrose. In the presence of sucrose, cell wall-associated glycosyltransferases (Gtfs) mediate the tight binding of S. mutans to the tooth surface, synthesizing glucans that interact with surface-associated glycan-binding proteins (GBPs) to promote cell-cell aggregation. On the other hand, in the absence of sucrose, S. mutans synthesizes several important adhesins such as antigen I/II (also called SpaP, Pac), which specifically binds to a glycoprotein called SAG (salivary agglutinin) [7] which has been proposed that participate as well in the tooth bacterial adhesion, being fundamental for biofilm formation, due to its ability to promote the aggregation and invasion of the dentin of the collagendependent tooth [8,9].
Currently, to the best of our knowledge, there are not studies of molecules with an antibiofilm activity, which could prevent the formation of the cariogenic biofilm without affecting the normal oral microbiota. However, different strategies have been proposed, for example, molecules derived from natural products, metal ions and oxidizing agents and even antibodies that specifically bind to S. mutans targets (GtfB, GtfC, GtfD, Ag I/II) and inhibit their ability to develop biofilms [10,11]. In recent decades, the virtual search for inhibitors based on structures has taken interest in drug discovery [12][13][14]; for oral microbiology the use of this strategy is relatively new, particularly in the cariogenic context, but several proteins have been proposed that could be used as inhibitory molecules [15]. Gtf-C has been used as a target on the search of molecules with affinity to this protein and for selectively inhibition of S. mutans biofilms formation mainly due to the ability to inhibit the synthesis of expolysaccharides (EPS) in vitro, the biofilm formation and reduce in vivo the caries incidence and severity in a rat model [16,17]. Although the Ag I/II adhesin has been reported to play an important in the early stages of S. mutans biofilm development, participating in adhesion and coaggregation with other bacteria and fungi such as C. albicans, there are no reports of computational studies that use this protein as target, therefore, the aim of this work is to identify in silico molecules with inhibitory effect on S. mutans Ag I/II, , which have no cytotoxic activity on human cells.

Target proteins selection
The Ag I/II of S. mutans play an essential role in the etiology and pathogenesis of dental caries. Therefore, the discovery of inhibitors of Ag I/II may facilitate the development of drugs that prevent dental caries. Ag I/II is one of the cell wall-anchored adhesins that mediates attachment of S. mutans to tooth surfaces, recognize salivary glycoproteins, and are also involved in biofilm formation. Finding small-molecule that bind to the Ag I/II may interfere with the function of these adhesine. To address this, we selected a sequence of the S. mutans Ag I/II adhesin and we found a crystal structure from different regions of AgI/II, which corresponds to the A3VP1 region (PDB: 3IPK) [18], C-terminal domain, region V (PDB: 1JMM) [19] and two from the C-terminal domain (PDB: 3QE5) [20] and (PDB: 3OPU) [21] ( Figure S1). According to sequence similarity and coverage results (Table S1), the crystals structures of the proteins 3IPK and 3QE5 were selected, due to their role in S. mutans adhesion to the tooth, through their SAG binding [18] (Figure 1).
Sequence and structural similarity analysis of 3IPK and 3QE5 was performed in order to identify other similar proteins that could be affected by the selected molecules. However, no significant homologies were found, which could be an indication that these molecules would not affect human (Table S2) or bacterial proteins important for the ecological balance of the oral microbiota. On the other hand, we have found sequence homologies with proteins from 23 bacterial species, which are mostly normal inhabitants of the oral cavity, gastrointestinal and genitourinary tracts, but are associated with different pathologies (Table S3, S4). Regarding the structural homologies for 3IPK and 3QE5 proteins, there were no similarities with any human or bacterial proteins, only with S. mutans Ag I/II regions, A3VP1 region (PDB: 3ioxA) [18] and N-terminal and C-terminal interaction complex (PDB: 4tshA) (Table S5) [26].
A fundamental step for the search of molecules based on virtual structures, is the pocket selection, these sites must have typical characteristics such as concave, have a variety of hydrogen bridge donors and acceptors and hydrophobic characteristics [27], otherwise, false negatives may occur when selecting molecules for in vitro assays, since errors may occur when there are no binding sites in the protein or when homology models are used, causing for example, small-volume pockets to be selected that will generate incorrect unions or conformations [28]. For that reason, in this study, two multiservers or specific programs were used for protein ligand binding site prediction, which have different selection algorithms, the MetaPocket uses a consensus method based on the predicted sites of four free access programs LIGSITEcs , PASS, Q-SiteFinder and SURFNET, which are combined to improve the success rate of the prediction and which is based on the geometry and surface of the proteins [29], unlike the COACH that uses the consensus of two methods, one based on the comparison of specific binding substructures (TM-SITE) and the other on the alignment of the sequence profile (S-SITE), for predictions of binding sites based on known proteins [30] ( Figure S2).

Molecules selection
Ten molecules were arbitrarily selected according to their molecular docking score and ten according to the number of pockets interaction. The lowest docking score indicates high affinity between the molecule and the ligand. Therefore, molecules that interacted in several binding sites were selected, because those could have a greater coating of the target protein allowing the inhibition of the two important regions of Ag I/II involved in its adhesin function. Once the selected molecules were reviewed, the molecules ZINC19924906 from the "Small" library was selected for both characteristics, obtaining in this step nineteen molecules, those with the lowest interaction energies, were molecules from the "Natural" library, such as ZINC68568370 with docking score -12.8. Molecules that had a greater number of binding sites, they were from the "Small" library and were coupled to the 12 pockets used for docking, meaning that they have affinity for both proteins 3IPK and 3QE5 ( Table 1).

Molecule -protein interaction and solubility analysis
The presence of one or two hydrogen bridges was confirmed (Table S7) using the interaction complexes in the pockets established by the two predictors between the selected molecules and the 3IPK and 3QE5 proteins. This finding could indicate that the interaction between each molecules ZI-187, ZI-939 and ZI-906 with both proteins, would have very stable couplings, resulting in a possible inhibition of the S. mutans adhesion, since it has been shown that hydrogen bridges regulate and facilitate molecular interactions [31][32][33][34][35]. In addition, water solubility analysis of the molecules was conducted using a consensus from 3 methods resulting in a low aqueous solubility of the molecules (Table S8).

Cytotoxicity and antimicrobial assays
It was found that molecules at concentrations of 100 µM have no effect on periodontal ligament fibroblast cells growth and the cells treated with molecules ZI-187 (P= 0.7372), ZI-939 (P= 0.8) and ZI-906 (P= 0.7964) ( Figure S5). However, cells treated with each molecules showed changes in size and granularity. Therefore, it is important to analyze other human cell lines and add complementary analysis such as incorporation of DIOC6 for the mitochondrial membrane potential measurement [36] or apoptosis tests as Annexin V [37]. In addition, for antimicrobial assays molecules at concentrations of 1000 -100 -10 µM co-cultured experiments with S. mutans LT-11 or C. albicans -NCPF 3179, did not affect their growth (P= <0.001) ( Figure S6 and S7, respectively).

Adhesion assays
It was established that treating S. mutans LT11 cells for three hours with all the molecules selected (ZI-187, ZI-939, ZI-906) inhibited the surface adhesion to a polystyrene microwell plate.  Additionally, an agglutination phenomenon was observed when molecules with the bacteria was mixed, only at concentrations where the inhibition was affected (Figure 4). ZI-187 was the only molecule that maintained adhesion inhibition above 90% at a concentration of 100 µM, hence this was selected for scanning electron microscopy. Reduction of adherent bacteria was evident, and we did not observe morphological changes in S. mutans following treatment with ZI-187, bacteria had intact cell structure and round shapes with smooth edges ( Figure 5).   There are multiple chemical strategies that could limit the development of dental biofilm, however, most of them can have side effects on teeth, soft tissues or killing oral microbiota, which show the need for specific therapies for cariogenic bacteria. Several studies have focused on blocking two important mechanisms for S. mutans biofilm development, such as avoiding sucrose-dependent or sucrose-independent adhesion and interference of cellular signaling "Quorum sensing" [38]. This study was carried out avoiding the sucrose-independent adhesion way, which has been aimed mainly at blocking sortase A, a transpeptidase involved at the anchoring of cell surface proteins, including Ag I/II, through the LPXTG motif. It has been found that several molecules can reduce biofilm formation, through the inhibition of the sortase A [39,40], such is the case of the natural phenol curcumin, (Curcuma longa) with which it has been reported inhibition S. mutans sortase A activity with IC50: 10 µM and a MIC of 175 µM [41]. However, despite the multiple benefits of this molecule, some toxic effects have also been evidenced related to the high doses as result from its use as a supplement in the diet [40,41]. Another natural product is named Morin, it has an inhibitory effect against S. mutans SrtA with IC50: 27.2 µM [42,43]. Morin other natural product, has an inhibitory effect against S. mutans SrtA with IC50: 27.2 µM [44], similar to the IC50 obtained with our molecules ZI-187 and ZI-939 (IC50: 27,6 y 28,3 µM, respectively). However, the antimicrobial activity of Morin has also been reported against S. mutans [45], differently from the molecules found in this study that did not present cytotoxic or antimicrobial activities.

Molecular dynamics simulations (MD)
Using molecular dynamics simulation, we have found that the complex 3IPK/ZI-187 attained a high stability after 300ns, during this time the 3IPK protein did not have strong changes when it was coupled with ZI-187 ( Figure 6A), which could indicate that the interacting molecule moves on the pocket, but not drastically, this means that it does not destabilize the complex; in agreement with the RMSD result, no significant fluctuation of amino acid residues was observed, however, between residues 550-600 and 800-850, some differences were showed between the APO protein for 3IPK (black) and the complex (red) (Figure 6B), which could be due to the specific 3IPK residues that interact with ZI-187, since different types of interactions of ZI-187 with residues Leu 553 -Asp 554 -Thr 586 -Val 587 and Lys 811 -Lys 812 -Asn 814 -Ile 815 -Trp 816 were identified ( Figure S8). In addition, 1 and 2 H-bonds were observed between the complex during the simulation time ( Figure  6C) and a constant protein structure compactness in the complex ( Figure 6D), these two parameters would support the stability of the interaction between ZI-187 and 3IPK, giving a possible explanation for the molecule's ability to inhibit the adhesion of S. mutans. Finally, the present work allowed to establish a virtual search strategy and selection pipeline for adhesion inhibitory molecules of a cariogenic bacteria but that could be applicable to any pathogen. This study, to the best of our knowledge, is the first report that uses the S. mutans Ag I/II as a target, since the previous studies were mainly performed on Gtfs or sortase A proteins. Three molecules were selected ZI-187, ZI-939 and ZI-906, without showing any cytotoxic effect on periodontal ligament fibroblasts or antimicrobial activity on S. mutans or C. albicans. However, it is suggested to perform assays in other different human cell lines, as well as on other microorganisms of oral cavity importance to evaluate in a much wider range other possible effects. It was also found, as expected, that the molecules selected had a significant effect in terms of reduction of the S. mutans surface adhesion as a single microorganism, but it is very important to carry out complementary studies on multispecies biofilms models, also to identify through transcriptomic analysis if there are variations on the expression of adhesion genes dependent and independent of sucrose when treating bacteria with the selected molecules, as well as genes that participate in cell signaling during the biofilm development process. Similarly, an in vivo cariogenic model should be established in order to insight the anti-cariogenic capacity of these molecules.

Target proteins selection
3D protein structures for the virtual search were selected using the Ag I/II of S. mutans sequence AFR75221.1 (NCBI https://www.ncbi.nlm.nih.gov/) and the 3D SWISS-MODEL software (http://swissmodel.expasy.org/ ). 3D structures of two Ag I/II protein fragments 3IPK (PDB-ID) [18] and 3QE5 [20], were used for a sequence and protein structure similarity analysis using a BLAST-P (Protein Basic Local Alignment Search Tool -NCBI) and the FAT-CAT server (Flexible structure Alignment by Chaining Aligned fragment pairs allowing Twists) [46] to search for similar (rigid) protein structures, using similarities only with a P-value <0.05. Subsequently, for 3IPK and 3QE5 proteins an analysis of binding sites "Pockets" using meta-servers MetaPocket 2.0 [29] and COACH [30] was carried out. 3D structures of the proteins were obtained in PDB format and edited in AutoDockTools 4.0 (http://mgltools.scripps.edu) [47]. Molecular docking of molecules with AgI/II protein fragments, was performed at the Texas Advanced Computing Center (TACC: Texas Advanced Computing Center; Austin, TX) using 3 libraries "ZINC (Lrg)" of ~ 642,759, Library "ZINC (Sm)" of ~ 46,702 molecules, and "ZINC Natural Cmpds (Large)" of 194,090 from the ZINC15 database [48], a total of ~ 883,551 molecules.

Molecules selection:
Two methodologies were used for molecules selection, one according to the molecular docking score and other according to the number of pockets in which molecules interacted, classification of molecules that interacted in most pockets was carried out using a script executed in the R-studio package (Version 1.0.156); ten molecules were arbitrarily selected from each methodology. An in silico analysis was performed using the QuikProp application (Version 3.2) from Schrödinger software [49], according to [50] with some modifications, in order to analyze pharmacokinetic profiles such as absorption, distribution, metabolism and excretion (ADME). Molecules that had more than two violations of the Lipinski's Rules and that did not comply with more than two of the standard physical-chemical parameters established by 95% of the known drugs, according to Schrödinger's QuikProp program repositories, were excluded.
For computational toxicity prediction of the molecules, the Protox-II program was used [51], for acute toxicity, hepatotoxicity, cytotoxicity, carcinogenicity, mutagenicity and immunotoxicity. In addition, the molecules lethal dose 50 (LD50) (mg/kg) was calculated and classified according to the Globally Harmonized System of Classification and Labeling of Chemicals (GHS). For this analysis, chlorhexidine was used as the reference drug, one of the most commonly prescribed antiseptic agents in dentistry; the experimental LD50 values of oral administration in mice of this drug were taken from the Pfizer chlorhexidine technical data sheet [52]. Molecules above class 3 and without any probability of toxicity were selected.

Molecule -protein interaction and solubility analysis
ZINC19835187 (Database code Zinc15), ZINC19924906 and ZINC19924939, (ZI-187, ZI-906 and ZI-939, respectively) were selected and the H-bonding interactions between the molecules and the two protein fragments 3ipk -3qe5 were identified [53] using Biovia Discovery Studio program [54]. The swissADME web server (http://www.swissadme.ch/) was used to predict water solubility characteristics of the molecules , using a consensus of three methods Log S (ESOL), Log S (Ali) and Log S (SILICOS-IT) [55].

Adhesion assay:
S. mutans-Lt11 (UB579 WT) was cultured in BHI broth (BD®) overnight at 37 °C shaking at 250 RPMI. The culture medium was discarded and bacteria were washed with phosphate buffer solution (PBS 1X) by centrifugation at 3000 RPM for 10 minutes. Subsequently, bacterial suspension in PBS 1X, OD600nm= 1 was measured using an Epoch™ Microplate spectrophotometer (BioTek®) and 180 µl were inoculated into a 96-well microplate (NEST®, Ref: 701001) with 20 µl of each molecule at 200, 100, 50, 25, 12.5 µM and incubated for three hours at 37 °C shaking at 250 RPM [58]. S. mutans SpaPstrain (mutant for AgI/II also called SpaP) and S. mutans-Lt11 treated with DMSO 10% were used as control. Plate wells were washed with water and adherent cells were stained by adding 200 µL 0.05% crystal violet for 15 min, washed and measured by absorbance at 600 nm after addition of 30% glacial acetic acid.

Data analysis
All experiments were performed in triplicate and reproduced three separate times. Cell viability percentages were reported as negative % PI ± SEM and analyzed by one-way ANOVA, followed by a Dunnett's multiple comparison test. In S. mutans and C. albicans growth and adhesion inhibition analysis, the OD data were normalized to percentages and analyzed using D'Agostino & Pearson omnibus and Shapiro-Wilk normality test; subsequently the nonparametric Kruskal-Wallis test with Dunnett's multiple comparisons against the controls S. mutans and C. albicans treated with DMSO 10%. Finally, the half maximal inhibitory concentration (IC50) was calculated. GraphPad Prism version 6.0 (GraphPad Software, La Jolla California USA, www.graphpad.com) was used and values of P <0.05 were considered statistically significant.

Scanning Electron Microscopy (SEM):
Surface adhesion assays were performed on Thermo Scientific Nunc Lab-Tek and Lab Tek II Chamber Slides, using S. mutans -Lt11 untreated and treated with molecule ZI-187 (100 µM). After incubation for three hours at 37 °C shaking at 250 RPM, each sample was washed three times with PBS 1X , then the samples were fixed with 2,5% glutaraldehyde (0.1M PBS) for 24 hours at 4 °C. Finally, the slides were washed three times with distilled water and dehydrated by immersion in solutions of ascending concentrations of ethanol 70, 90 and 100% (10 minutes each) and dried overnight in a laminar flow cabinet. The samples were covered with gold and visualized using a FEI QUANTA-200TM scanning electron microscope with a variable range acceleration voltage of 1-30 KV.

Molecular dynamics simulations (MD)
MD simulations were carried out using the GROMACS 4.5.5 package [59]. Molecule ZI-187 was docked to the 3IPK protein pocket with the highest binding affinity (P1 by COACH predictor). The ZI-187/3IPK complex and the 3IPK protein in its APO state (reference state), were used as initial coordinates for MD simulations. Finally, both systems were subjected to a 300ns production stage, using a 2fs time step. The equilibrations and productions were carried out using a temperature of 310K (36.85 °C) and a 1 bar pressure. Descriptors such as the RMSD (Root of the mean square deviation), the RMSF (Root of the mean square fluctuation) and hydrogen bonds present in the protein-ligand complex were followed with the tools contained g_rms, g_rmsf and g_hbond, respectively.