Exploring the interactive mechanism of acarbose with the amylase SusG in the starch utilization system of the human gut symbiont Bacteroides thetaiotaomicron through molecular modeling

The α‐amylase, SusG, is a principal component of the Bacteroides thetaiotaomicron (Bt) starch utilization system (Sus) used to metabolize complex starch molecules in the human gastrointestinal (GI) tract. We previously reported the non‐microbicidal growth inhibition of Bt by the acarbose‐mediated arrest of the Sus as a potential therapeutic strategy. Herein, we report a computational approach using density functional theory (DFT), molecular docking, and molecular dynamics (MD) simulation to explore the interactive mechanism between acarbose and SusG at the atomic level in an effort to understand how acarbose shuts down the Bt Sus. The docking analysis reveals that acarbose binds orthosterically to SusG with a binding affinity of −8.3 kcal/mol. The MD simulation provides evidence of conformational variability of acarbose at the active site of SusG and also suggests that acarbose interacts with the main catalytic residues via a general acid–base double‐displacement catalytic mechanism. These results suggest that small molecule competitive inhibition against the SusG protein could impact the entire Bt Sus and eliminate or reduce the system's ability to metabolize starch. This computational strategy could serve as a potential avenue for structure‐based drug design to discover other small molecules capable of inhibiting the Sus of Bt with high potency, thus providing a holistic approach for selective modulation of the GI microbiota.


| INTRODUCTION
The resident microbes in the human gastrointestinal (GI) tract profoundly influence host development and health (Flint et al., 2008;Hooper & Gordon, 2001;Marchesi et al., 2016;Vaarala et al., 2008a). Among the resident microbes in the gut ecosystem, Bacteroides is the most wellstudied genus within the dominant Bacteroidetes phylum (Foley et al., 2016). While generally viewed as benign commensal organisms, certain Bacteroides spp. cause serious disease (Wexler, 2007;Zafar & Saier, 2021). B. fragilis (Bf) is the most commonly encountered anaerobic pathogen in the clinic and is a common cause of intra-abdominal abscesses, intra-abdominal sepsis, toxin-mediated diarrhea, and bacteremia (Wexler, 2007;Zafar & Saier, 2021). The organism is also associated with perforated/gangrenous appendicitis, skin/soft tissue infections, septic arthritis, meningitis, endocarditis/pericarditis, and gynecological infections (Wexler, 2007). Enterotoxigenic B. fragilis and other Bacteroides spp. exacerbate several chronic gut-associated diseases, including ulcerative colitis, inflammatory bowel disease, celiac disease, and colorectal cancer (Wexler, 2007;Zafar & Saier, 2021). In addition, several studies have also found an increase in the population of some Bacteroides spp. in patients genetically at-risk for Type I diabetes mellitus (T1D) and patients exhibiting clinical T1D (Abdellatif & Sarvetnick, 2019;Alkanani et al., 2015;Atkinson & Chervonsky, 2012;Brown et al., 2011;Davis-Richardson et al., 2014;Davis-Richardson & Triplett, 2015;de Goffau et al., 2014;Giongo et al., 2011;Han et al., 2018;Knip & Honkanen, 2017;Knip & Siljander, 2016;Leiva-Gea et al., 2018;Mejía-León et al., 2015;Murri et al., 2013;Paun et al., 2017;Vaarala et al., 2008b;Verdu & Danska, 2018;Zheng et al., 2018). The Bacteroides genus encodes a Starch Utilization System (Sus) that is selectively activated to facilitate polysaccharide capture at the cell surface in the presence of starch-based substrates. The Sus of B. thetaiotaomicron (Bt) is the most well-characterized among the species of the Bacteroides genus (Anderson & Salyers, 1989a, 1989b. It consists of eight proteins, SusRABCDEFG, that work in a concerted effort to capture and degrade starch-based polysaccharides ( Figure 1; Cho & Salyers, 2001;Shipman et al., 1999;Tancula et al., 1992). SusR is an inner membrane transcriptional regulator protein which induces up-regulation and transcription of the Sus operon to maximize effective metabolism (Della & Salyers, 1996a). The carbohydratebinding proteins SusD, E, and F and the αamylase SusG comprise the Sus components that reside on the outer membrane (Arnal et al., 2018). Degraded starch (released oligosaccharides) are imported into the periplasmic space by SusC where they are further cleaved by an intermembrane αamylase, SusB, and a neopullulanase, SusA (Della & Salyers, 1996b). Among the enzymes described in the Sus of Bt, SusG is the main outer membrane-bound αamylase that cleaves glycosidic linkages in starch-based polysaccharides at the cell surface (Altschul et al., 1997;Martens et al., 2008). Thus, we hypothesized that small molecule competitive inhibition against this enzyme might impede the entire Bt Sus and eliminate the ability of Bacteroides species to metabolize starch. Our previous study described a non-microbicidal strategy for inhibiting the growth of members of the Bacteroides genus (Santilli et al., 2018). We reported that acarbose ( Figure 2b) can selectively inhibit the growth of Bt and Bf strains by preventing their ability to metabolize potato starch and pullulan (Santilli et al., 2018). This study offered a potential therapeutic strategy for modulating the gut microbiota by means of targeted growth inhibition in a potent yet non-lethal manner. Subsequently, we also demonstrated similarly potent growth inhibition against B. dorei clinical isolates from T1D patients (Santilli et al., 2019).
Here, we report a computational strategy using density functional theory (DFT), molecular docking, and molecular dynamics (MD) simulation to explore the interactive mechanism between acarbose and SusG at the atomic level in order to gain insight into how acarbose shuts down the Bt Sus. We hypothesized that SusG inhibition represents a major pathway for Sus inhibition by acarbose based on the fact that SusG is the membrane-bound αamylase that cleaves glycosidic linkages in starch-based polysaccharides before the metabolites enter the cell for further degradation. Indeed, a Bt mutant bearing a deletion of the SusG enzyme cannot metabolize starch (Shipman et al., 1999). Further, SusG has been crystallized with acarbose cleavage products (i.e., maltose and acarviosin) bound within its active site ( Figure S10; Koropatkin & Smith, 2010). Therefore, the present study had two main objectives. First, we aimed to investigate whether acarbose binds orthorsterically to SusG using a molecular docking model, and secondly, we explored the dynamic behavior of SusG in the presence of acarbose to gain insight into the binding mechanism of acarbose on SusG using molecular dynamics calculations. This computational strategy focusing on understanding the interactions between acarbose and SusG is not only important from the standpoint of uncovering the mode of action for acarbose's potent and selective inhibitory activity against Bacteroides spp., but it could also inform future structure-based drug design efforts to discover other small molecule scaffolds that are capable of inhibiting the Sus with high potency.

| METHODS
Full details for all experimental methods and additional supporting data (i.e. Figures S1-S10; Tables S1 and S2) are provided in the Supporting Information.

| Small molecule model and geometry optimization
SusG belongs to a family of GH13 amylases, composed of A, B, and C domains and a carbohydrate-binding module family 58 (CBM58) that share structural features with other amylases (Birt et al., 2013;Hii et al., 2012). The crystal structure of the catalytically inactive SusG D498N mutant (PDB 3K8L; Koropatkin & Smith, 2010) shows maltoheptaose bound within its active site (Figure 3a). To ensure the reliability and efficiency of the docking model, maltoheptaose was used as a reference compound and was taken through all the computational processes. Full (unconstrained) geometry optimization of the chemical structures of maltoheptaose and acarbose (Figure 2a,b) was carried out in the gas phase (no implicit solvent) using the Gaussian 09 suite with DFT calculation, employing the B3LYP/6-311G (d,p) level of theory (Bergner et al., 1993). B3LYP/6-311G (d,p) is a commonly used functional level of theory for geometry optimization and transition state study based on its good accuracy and moderate computational time for carbohydrate systems (Huang et al., 2016;Kim et al., 2011;Parthasarathi et al., 2011;Wang et al., 2017).
The geometry optimization was done to ensure the proper arrangement of all bond lengths, bond angles, and torsional angles in space since the drawn chemical structures are not necessarily energetically favorable. After the energy refinement of maltoheptaose and acarbose chemical structures, vibrational frequency calculations performed on the optimized conformers ( Figure 2c) indicated that the stationary points corresponded to minima on the potential energy surface. The molecular electrostatic potential (MEP) calculated from the above DFT level of theory for the optimized acarbose structure ( Figure 2d) shows an appreciable surface charge distribution characterized by a positive electrostatic potential (blue surface) and a negative electrostatic potential (red surface). Interestingly, the active site of SusG exhibits a similar pattern of electrostatic potential ( Figure S2). This suggests that acarbose may have the desired electrostatic complementarity to the active site of SusG. However, it is important also to note that the ligand must have the desired shape complementarity in addition to the electrostatic properties to bind effectively at the receptor's active site (Kangas & Tidor, 2001). Interestingly, some evidence does suggest that acarbose may have a desirable shape complementarity to the active site of SusG since the optimized acarbose conformer is comparable to the X-ray structures of the conformers of acarbose cleavage products (i.e., maltose and acarviosin) bound within the active site of SusG ( Figure S1).

| Molecular docking
To gain insight into how acarbose interacts with the SusG protein, we employed molecular docking calculations using Autodock Vina as an efficient approach to probe these interactions at the atomic level. To ensure that our molecular docking technique is reliable and efficient in predicting the binding properties between acarbose and the SusG protein, we first docked the optimized maltoheptaose pose with mutant SusG (PDB 3K8L) with the specific docking parameters and scoring functions described in the experimental section. The mutant SusG was used in this study since it was cocrystallized with maltoheptaose, where the Asp-498 which plays a critical role in enzyme catalysis was mutated to Asn-498. Thus, this study provides an unbiased comparison with acarbose modeled with the same mutant SusG.
The docking model predicts strong binding of maltoheptaose to the active site of SusG with a free energy of binding of −8.7 kcal/mol. This binding pose reappeared four times out of the nine poses calculated from the scoring function in the docking model (Table S2), demonstrating the convergence of the docking result. Again, to ensure the reproducibility of the docking results, we changed the exhaustiveness parameter (default 8) to 32, 100, and a maximum value of 1000. Interestingly, there were no significant changes in the conformations and binding affinities predicted under these conditions from the scoring function in the docking model (Table S2). Notably, the conformation with the best scoring result compared favorably with the experimental crystal pose of maltoheptaose bound to the catalytically inactive SusG D498N mutant (Figure 3b,c). To quantify and validate the reliability of the docking model, the docked maltoheptaose pose with the best score was rigidly superimposed with the experimental crystal pose (Figure 3d), demonstrating a root mean square deviation (RSMD) of 1.57 Å. These results suggest that the docking model is reliable for predicting bound conformers within this system. Additionally, the number of Degrees of Freedom (DOF) for acarbose and maltoheptaose calculated from the aforementioned DFT level of theory are 255 and 444, respectively. Since the DOF associated with acarbose is significantly smaller than maltoheptaose, we surmised that the docking model would also be amenable to acarbose since it has fewer accessible conformations.
Thus, the optimized acarbose pose was docked at the binding site of SusG (PDB 3K8L) using the same docking parameters and scoring functions described in the experimental section. The docking analysis ( Figure 3e) shows that acarbose binds strongly to the active site of SusG with a predicted binding energy of −8.3 kcal/mol (Table S3). This free energy of binding was comparable to that calculated for the SusG-maltoheptaose complex (−8.7 kcal/ mol), indicating both ligands to be tight binders of SusG. This is not surprising since both maltoheptaose and acarbose metabolites have been complexed experimentally with SusG (Koropatkin & Smith, 2010).
Additionally, we predicted the K i of acarbose as 0.84 μM using the expression that relates the inhibition constant (K i ) to the predicted Gibbs free energy of binding in a rigid-receptor docking: (Hopkins et al., 2014;Iman et al., 2015;Onawole et al., 2017) where R is the universal gas constant (1.985 × 10 −3 kcal/ mol·K) and T is the temperature (298.15 K). Since acarbose acts as a competitive inhibitor (DiNicolantonio et al., 2015;Rosak & Mertes, 2012;Wehmeier & Piepersberg, 2004) the half maximal inhibitory concentration (IC 50 ) could be predicted from the calculated K i using the expression, K i = IC 50 /2 (Cer et al., 2009;Haupt et al., 2015). This predicts the IC 50 of acarbose from the docking model to be 1.68 μM. Interestingly, this result is consistent with the IC 50 value of 1.73 ± 0.4 μM determined for acarbose from the experimental bioassay study when acarbose was treated against Bf cultured in the presence of potato starch as the sole carbon source (Santilli et al., 2019). Even though the predicted IC 50 of acarbose from the docking model is consistent with the experimental IC 50 value, it should be noted that the AutoDock Vina carries a standard error of about 2.85 kcal/ mol based on the calibration of the scoring function used in predicting the free energy of binding with 1300 receptorligand complexes from PDBbind (Trott & Olson, 2010). Thus, the predicted IC 50 value of acarbose from the docking model taking into consideration the intrinsic error may lie in the range of 0.13-20 μM.
In addition, the free energy of binding of a ligand in a rigid-receptor docking predicted from in silico calculations only indicates whether a ligand binds well at the active site of a receptor. This means that the docking results should be used cautiously when predicting whether a well-scored ligand would act as a strong antagonist or agonist with respect to SusG. Therefore, the docking result should not be over-interpreted until further validations of bioassays have been performed (Chen, 2015).
Analysis of the docking pose (Figures 3e,f) revealed that acarbose bound to the active site of SusG in a conformation similar to the bioactive conformer of maltoheptaose bound to the active site of SusG from the X-ray crystal pose (Figure 3b). Extensive protein-glycan hydrogen bond interactions were observed between the polar groups at the active site of SusG and the maltose and acarviosin units of the bound conformer of acarbose. This was expected due to the remarkable electrostatic complementarity exhibited by acarbose and the active site of SusG ( Figure S2). At the nonreducing end (Glc2) of the maltose unit, there were hydrogen bond interactions between O-2 with ASN-498 (3.3 Å), O-3 with GLU-431 (2.8 Å), ARG-386 (2.9 Å), and HIS-497 (3.4 Å), and O-6 with HIS-154 (3.1 Å) (ΔG∕RT) and ASP-388 (2.9 Å) where distances are measured from the heavy atom to hydrogen. The O-6 at the reducing end (Glc1) also engages in hydrogen bonds with GLU-431 (3.0 Å) and HIS-392 (3.3 Å). The acarviosin unit of acarbose stacks against the hydrophobic face of HIS-112, the phenolic side chain of TYR-114, and is positioned by a network of hydrogen bonds.
Intriguingly, the bound conformer of acarbose in the active site of SusG is bound in a higher energy conformation where the nonreducing end (Glc2) of the maltose unit has undergone a ring flip to facilitate a number of hydrogen bonding interactions with HIS-154, ASP-388, ARG-386, GLU-431, HIS-497, and ASN-498 (Figure 3e,f). To confirm the energy state of the bound conformer, we performed a frequency calculation on the bound and the optimized conformers of acarbose ( Figure S3 and Table S1) using the B3LYP/6-311G (d,p) level of theory. The results revealed that the energy difference between these two conformers is approximately 2.13 kcal/mol. However, we speculate that the extensive hydrogen bonding formed between the bound conformer of acarbose and the residues at the active site of SusG may serve as a stabilizing effect to compensate for the internal energy penalty as a result of the ring flip in the bound conformation of acarbose.
The conformation of the active site of SusG is similar to many enzymes of the family of GH13 αamylases. The amino acid residues Asp-388, Glu-431, and Asn-498 at the active site of SusG are conserved among all GH13 αamylase homologs (Braun et al., 1995;Davies et al., 1999;Gloster et al., 2008;Henrissat, 1991;Henrissat et al., 1995;Henrissat & Bairoch, 1993;Rydberg et al., 1999Rydberg et al., , 2002. These residues play critical roles in the general acidbase double-displacement catalytic mechanism involving the formation and hydrolysis of the covalent glycosylenzyme intermediate ( Figure S4; Li et al., 2005;Nahoum et al., 2000;Qian et al., 2001). Asp-388 acts as a catalytic nucleophile that attacks at C1 of the sugar anomeric carbon to form the β-glycosyl-enzyme intermediate followed by subsequent hydrolysis of the intermediate by a water molecule activated by Glu-431. The Asn-498, which mutated from Asp-498, is believed to regulate the pKa of the general acid-acid catalytic process or stabilize the oxocarbenium ion in the transition state of this catalytic mechanism ( Figure S4; Koropatkin & Smith, 2010). From the X-ray crystal pose of SusG (Figure 3b), Asp-388 is positioned 3 Å from the anomeric carbon of Glc4, thus confirming its role as the catalytic nucleophile that forms the β-glycosyl-enzyme intermediate. Glu-431 also interacts with O-4 of Glc5 and likely acts as a proton donor to the leaving αglucan chain ( Figure S4; Koropatkin & Smith, 2010). Interestingly from our docking model (Figure 3e), the two anomeric centers in acarbose are positioned 8 and 9 Å away from the reach of the catalytic nucleophilic residue Asp-388, as a result of the ring flip in the nonreducing end (Glc2) of the maltose unit. This suggests that the binding conformation of acarbose at the active site of SusG may prevent the hydrolysis of acarbose by the enzyme. This dynamic property of acarbose at the active site of SusG supports its role as a competitive inhibitor. Also, it provides structural features essential for the design of tight SusG binders.

| Molecular dynamics simulation
To further validate the binding conformation of acarbose and the strong binding interactions of the SusG-acarbose complex from the molecular docking, we performed a molecular dynamics (MD) simulation of the mutant SusG-acarbose complex for 100 ns following the procedure described in the experimental section. The MD simulation was also carried out for the mutant SusGmaltoheptaose complex as a reference. The MD simulations not only help to validate the binding interactions from the docking study but also provide insight into the binding mechanism, structural dynamics, and the stability of the SusG-acarbose complex under physiological conditions. This would enhance our understanding of the structural and dynamic properties of the SusG catalytic site for effective drug design.

| Conformation and simulation systems stability
To check the stability of the SusG-acarbose complex, we first analyzed the RMSDs of the Cα atoms of the amino acid residues in the protein for both the SusG-acarbose and SusG-maltoheptaose complexes. From Figure 4a both complexes appear to attain equilibrium after 30 ns and fluctuated around 3.5 ± 0.35 Å thereafter. It is interesting to note that even though both complexes equilibrated after 30 ns, fluctuations were significantly visible until 70 ns. This suggests rapid conformational transitions in both SusG-acarbose and SusG-maltoheptaose complexes during the initial 70 ns of the simulation.
To address the issue of the structural compactness and rigidity of the SusG-acarbose and SusG-maltoheptaose complexes, the solvent-accessible surface area (SASA) and the radius of gyration were monitored in both complexes throughout the MD simulation. The SASA indicates the amount of the protein accessible to both the bulk solvent and the ligand. A significantly increasing SASA would indicate an expansion of the protein volume due to the solvation of the hydrophobic core as the protein unfolds, and that was not observed here. From Figure 4b, both complexes attained equilibrium after 30 ns and fluctuated around 26,250 Å 2 characterized by a lower SASA toward the end of the simulation. This suggests a decrease in the flexibility of the protein after both systems stabilize, thus indicating higher compactness and rigidity of the proteinligand complexes.
From Figure 4c, SusG-acarbose and SusGmaltoheptaose complexes showed consistent radius of gyration values after both systems equilibrated at 30 ns during the simulation. These results indicate tight protein-ligand packing and suggests that both complexes are stable after each ligand binds at the active site of the protein. Again, we assessed the flexibility of the amino acid residues at the receptor's active site that are involved in interactions with the ligand by analyzing the root mean square fluctuation (RMSF). Lower RMSF values of key residues suggest stronger binding interactions in the protein-ligand complex (Krieger & Vriend, 2015). From Figure 4d, both complexes showed comparable per-residue RMSF throughout the simulation. Minimal fluctuations were observed in key residue regions ranging from 111 to 154 and 380 to 554, which are involved in the protein-ligand binding interactions. This result reflects strong binding interactions and stability of both complexes.
where K B is the Boltzmann constant and T is the temperature (300 K). The P(N 1 , N 2 ) term represents the probability distribution of the molecular system along the reaction coordinates (N 1 , N 2 ). The reaction coordinates can be selected from a number of possibilities, including contact distances between two atoms, the RMSD, the radius of gyration, the dihedral angle, principal component (PC) analysis, etc. The FEL provides holistic insight into the conformational variability of the protein along with the Gibbs free energy during the MD simulation (Bhardwaj & Purohit, 2022). We visualized the conformational variability of each proteinligand complex in each system by plotting the Gibbs FEL against the RMSD and radius of gyration (Rg) as the two reaction coordinates ( Figure 5). The red color of the basins represents conformational states with higher energy, and the deeper blue color of the basins represents conformational states having lower energy. Both SusG-acarbose and SusGmaltoheptaose complexes showed similarly shallow basins of less stable conformational states accompanied by broad deep basins of comparatively more stable conformational states. This is not surprising since acarbose and maltoheptaose assume similar binding interactions with residues at the active site of SusG from the docking model and MD simulation (Figures 3b,c,e and 7a). Interestingly, the deep basins are centered around 0.35 and 3.2 nm of the RMSD and the radius of gyration respectively, which correlate with the equilibrated states for both SusG-acarbose and SusG-maltoheptaose complexes after 30 ns during the MD simulation (Figure 4a,c). These results provide support for the stability of the SusG-acarbose and SusG-maltoheptaose complexes during the MD simulation, thus, further reinforcing the conformational stability of the system.
In addition, the structural stability of the protein in both complexes was evaluated by performing a Define Secondary Structure of Proteins (DSS) analysis through the MD simulation. The DSS provides detailed information about the stability of the secondary structures (i.e., αhelices, βsheets, coil, turn, 3 10 -Helix, and Pi-Helix) during the simulation. From the DSS analysis ( Figures S5  and S6), αhelices and βsheets were dominant and consistent throughout the simulation for both SusG-acarbose and SusG-maltoheptaose complexes. This is not surprising since αhelices and βsheets are the most stable secondary structures of proteins. On the contrary, coil and turns occurred infrequently along with traces of 3 10 -Helix, and Pi-Helix, which are comparatively unstable and thus do not frequently occur in proteins. This result suggests that the conformational stability of the SusG protein is maintained after each ligand binding in both complexes.

| Analysis of structural and conformational variability from MD simulation
To gain insight into the binding mode and conformational variability of the SusG-acarbose complex, we analyzed the trajectories at different time intervals throughout the MD simulations. These results revealed that acarbose remained in the binding pocket and interconverted between the gas-phase optimized and docked conformations throughout the simulation (Figure 6 and Figure S7). Interestingly, acarbose remained in the optimized conformation between 10 to 30 ns and oscillated between the gas-phase optimized and the docked conformers between 30 and 60 ns, after which it remained consistently in the docked conformation until the end of the simulation at 100 ns. This is not surprising since the gas-phase optimized and docked conformers of acarbose are structurally close ( Figure S3). We surmise that the conformational switches between these two conformers observed between 30 and 60 ns of the MD simulation may proceed through a lower energy barrier. Maltoheptaose, on the contrary, remained in the bioactive conformation with positional fluctuations in atoms throughout the simulation ( Figure S8).
Moving forward, we analyzed the binding pose of the gas-phase optimized conformer of acarbose at the active site of SusG from the MD simulation (Figure 7a). The protein-glycan interactions observed in this binding pose were similar to the binding patterns from the docked model of acarbose and SusG (Figure 3e). Remarkably, however, the catalytic nucleophilic residue Asp-388 is positioned 2.8 Å from the anomeric carbon of Glc2 of the maltose unit. This suggests that Asp-388 may attack C1 of Glc2 to form acarviosine-glucosyl enzyme intermediate (Figures 7a  and 8a; Koropatkin & Smith, 2010;Li et al., 2005;Qian et al., 2001). Interestingly, Glu-431 interacts with O-4 of Glc1, consequently protonating the glycosidic oxygen of the cleaved αglucose residue via general acid-base catalysis. Also, Asn-498 is positioned 3.0 Å from the anomeric carbon of Glc2, and it is in close proximity to both Asp-388 and Glu-431, thus, supporting its role as a pKa regulator of the general acid/base catalysis (Figures 6a and 7a). Our data provide support for the previously reported work where Asp-388 was covalently linked to the acarviosineglucose unit with maltose residue occupying the active site of SusG after complexation of acarbose with SusG ( Figure S9; Koropatkin & Smith, 2010).
However, given that maltose was found at the active site rather than glucose after the complexation of SusG with acarbose is interesting ( Figure S9). As proposed by the authors, either SusG cleaves maltose rather than glucose from the reducing end of acarbose, or the observed maltose unit was a result of the cleavage of a pentasaccharide generated from an infrequent transglycosylation of acarbose at the active site of SusG (Koropatkin & Smith, 2010). Based on the result from our MD simulation (Figure 7a), we speculate that acarbose may undergo a transglycosylation reaction (Figure 8a,b) to generate the pseudo-pentasaccharide. Subsequently, SusG cleaves the maltose unit from the reducing end of this pseudo-pentasaccharide to generate the acarviosine-glucose and maltose units observed at the active site of SusG (Figure 8c and Figure S9). Our hypothesis is based on the fact that many αamylase homologs of the family GH13 perform transglycosylation of acarbose followed by acid-base hydrolysis to produce various metabolites at the active site (Braun et al., 1995;Gloster et al., 2008;Henrissat et al., 1995). To test our hypothesis, we ran an MD simulation of the pseudo-pentasaccharide bound to SusG. Indeed, analysis of the binding pose of the SusGpentasaccharide complex (Figure 7b) clearly shows that the catalytic nucleophilic residue Asp-388 is positioned 2.9 Å from the anomeric carbon of Glc3 from the reducing end of the pseudo-pentasaccharide. This suggests that Asp-388 attacks C1 of Glc3 to form the acarviosine-glucosyl enzyme intermediate. Again, Glu-431 is positioned 3.1 Å from C1 of Glc3 and interacts with the glycosidic oxygen O-4 of Glc2, likely supplying hydrogen to the cleaved maltose residue. This result is interesting and suggests that the gasphase optimized conformation of acarbose may facilitate transglycosylation by SusG leading to the trapped acarviosin-glucose intermediate observed at the active site ( Figure S9) which may act as an irreversible inhibitor to the enzyme. However, we speculate that this may not be a major pathway through which acarbose inhibits SusG since the transglycosylation may be an infrequent event because the products of such reaction were not detected experimentally (Koropatkin & Smith, 2010). In addition, the gas-phase optimized conformer of acarbose was not observed during the last two-thirds of the MD simulation F I G U R E 8 Proposed general acid-base double-displacement catalytic mechanism of SusG and acarbose at the active site. (a) Proposed mechanism for the formation of acarviosine-glucosyl enzyme intermediate from acarbose at the active site of SusG. (b) A possible mechanism for transglycosylation of acarbose to form pseudo-pentasaccharide species. (c) Proposed mechanism for the formation of acarviosine-glucosyl enzyme intermediate and maltose species from pseudo-pentasaccharide at the active site of SusG.
after the system has stabilized (Figure 4a and Figure S7), which further supports the fact that the gas-phase optimized conformer of acarbose infrequently occurs during the sampling of the conformational space in the MD simulation. Nonetheless, analysis from the binding pose of the gas-phase optimized conformation of acarbose (Figures 7a,b and 8) provides a mechanistic explanation for the acarviosine-glucose intermediate trapped at the active site of SusG ( Figure S9; Koropatkin & Smith, 2010). We believe that the docked conformation of acarbose (Figure 3e), which also occurred as the dominant conformation from the MD simulation ( Figure S7), represents the main mechanism by which acarbose inhibits SusG. This conformation prevents hydrolysis by the main catalytic residues and fully occupies the active site, thus preventing the natural substrate from accessing the main catalytic residues.
It should be noted that in silico analysis has played a significant role in structure-based drug discovery and also, in some cases, provides an in-depth understanding of the dynamics of complex bio-macromolecular systems. Nonetheless, they are not without limitations, and this study is no exception. Despite the success achieved through Autodock Vina in predicting binding poses of receptorligand complexes from molecular docking, it should be noted that the calculations are based on rigid receptor docking and, as a result, will correspond to a single receptor conformation out of the many different conformations accessible to the receptor upon ligand binding. Thus, the most stable and bioactive conformation of ligand at the receptor's active site may not always be sampled through this docking approach, which is a major challenge. The docking process was done in the absence of water molecules, which sometimes act as bridging molecules between the ligand and receptor; thus, this model is limited when describing how small molecules (ligands) interact with proteins in biological systems. In addition, like any other in silico calculation, the scoring functions from the docking model have some level of approximations, and some intermolecular interaction terms, such as solvation effects and entropic changes, may not be predicted perfectly accurately (Sethi et al., 2020). This may lead to some intrinsic and expected error in the free energy of binding predicted for receptor-ligand complexes. To improve and validate the result from the docking model, we employed molecular dynamic simulations that rely on robust scoring functions for analysis. However, it should be noted that MD simulations also have inherent limitations since the classical forcefields rely on some approximations, necessary for computational viability, relative to a fully ab initio quantum description. Despite the inherent limitations associated with these in silico analyses, they remain efficient and effective in studying complex dynamic biomacromolecular systems as utilized in the present study.

| CONCLUSIONS
Taken together, our data and observations from the docking model and MD simulation provide strong evidence that acarbose binds orthosterically to SusG, and sheds light on the inhibition mechanism of acarbose with SusG. Our results show that acarbose adopts a higher energy binding conformation where the nonreducing end (Glc2) of the maltose unit has undergone a ring flip to prevent hydrolysis by the main catalytic residues Asp-388, Glu-431, and Asn-498 at the active site. We believe that this conformational ensemble of acarbose represents the main mechanism by which acarbose inhibits SusG since it fully occupies the active site and prevents the natural substrate from accessing the main catalytic residues.
Interestingly, the MD simulation revealed that the higher energy bound conformer of acarbose may transition to a lower energy bound conformer that interacts with the main catalytic residues via a general acid-base double-displacement catalytic mechanism leading to the formation of the acarviosine-glucose enzyme intermediate at the active site. Although previous work has revealed the formation of the acarviosine-glucose enzyme intermediate of SusG with acarbose and has attributed this occurrence to an infrequent transglycosylation of SusG (Koropatkin & Smith, 2010), the present study provides an in-depth understanding of how this interactive phenomenon may occur between acarbose and SusG.
In summary, our results emphasize the full understanding of the dynamic property of acarbose at the active site of SusG and also provide structural features essential for the design of effective SusG inhibitors. Our data suggest that a small molecule competitive inhibition against the SusG protein might impede the entire Bt Sus and eliminate the ability of Bacteroides spp. to metabolize starch. Nonetheless, there is a possibility that acarbose could also interact with other enzymes in the Sus suite of Bt to unleash its inhibitory effect. Currently, we are using this computational strategy to probe the interactions between acarbose and other proteins in the Sus operon. Finally, we are also using this computational strategy in combination with experimental bioassays to uncover other small molecules capable of inhibiting the Bt starch utilization system.

ACKNOWLEDGMENTS
We gratefully acknowledge the Clemson University Creative Inquiry Program and the Clemson University Departments of Chemistry and Biological Sciences for financial support. We are grateful to the Clemson University Palmetto Cluster for computational resources.

DATA AVAILABILITY STATEMENT
Data available upon request.