Drug screening of α‐amylase inhibitors as candidates for treating diabetes

Abstract In the present study, the identification of potential α‐amylase inhibitors is explored as a potential strategy for treating type‐2 diabetes mellitus. A computationally driven approach using molecular docking was employed to search for new α‐amylase inhibitors. The interactions of potential drugs with the enzyme's active site were investigated and compared with the contacts established by acarbose (a reference drug for α‐amylase inhibition) in the crystallographic structure 1B2Y. For this active site characterization, both molecular docking and molecular dynamics simulations were performed, and the residues involved in the α‐amylase–acarbose complex were considered to analyse the potential drug's interaction with the enzyme. Two potential α‐amylase inhibitors (AN‐153I105594 and AN‐153I104845) have been selected following this computational strategy. Both compounds established a large number of interactions with key binding site α‐amylase amino acids and obtained a comparable docking score concerning the reference drug (acarbose). Aiming to further analyse candidates' properties, their ADME (absorption, distribution, metabolism, excretion) parameters, druglikeness, organ toxicity, toxicological endpoints and median lethal dose (LD50) were estimated. Overall estimations are promising for both candidates, and in silico toxicity predictions suggest that a low toxicity should be expected.


| INTRODUC TI ON
Diabetes mellitus is one of the most common non-communicable metabolic disorders categorized by type-1 or insulin-dependent diabetes, type-2 or non-insulin dependent diabetes mellitus and gestational diabetes mellitus. 1 Among these, type-2 diabetes is a pandemic and has affected almost 200 million people globally, and this figure is likely to rise gradually to 642 million by 2040. It happens when the pancreas stops using insulin effectively, which means the body's cells lose the ability to respond to insulin's efforts to drive glucose into the cells, a condition called insulin resistance leads to a type of metabolic disorder in which the patient has an increased level of sugar (glucose) in blood, which can lead to potential complications that include blindness, heart attacks, strokes, kidney damage, lower limb amputation and nerve damage. The management of blood glucose in patients affected by diabetes mellitus is complex because of personal factors, such as exercise and diet, and the non-specific mode of action associated with oral antihyperglycemic. 2 In the present scenario, enzyme inhibitors can be explored as alternative targets due to their high degree of substrate specificity.
Amylase is a digestive enzyme, mainly produced by the salivary glands and pancreas, serving both endocrine and exocrine functions.
Among the different types of amylase enzymes, one major category of the enzyme is pancreatic α-amylase, which is a calcium metalloenzyme that aids digestion by breaking down polysaccharide molecules into smaller ones, such as glucose and maltose. In addition, the enzyme causes postprandial hyperglycemia and blood glucose levels to rise. By performing the act of inhibiting α-amylase enzyme, it helps in reducing hyperglycemia, obesity and problems such as overweight conditions by reducing postprandial levels of glucose.
Alpha-amylase is the major form of amylase found in humans and other mammals. It is an enzyme that hydrolyses α-bonds of large, αlinked polysaccharides, such as starch and glycogen, yielding shorter chains thereof, dextrins and maltose. 3 As α-amylase is an important target for the treatment of diabetes mellitus and the development of new drugs, scientists are showing great interest in this enzyme. 4 In this regard, 3D computer-aided drug design methods, such as molecular docking and molecular dynamics (MD), take into account ligandreceptor simulation systems. 5 Molecular docking analysis models the interaction between a molecule and a protein at the atomic level.
This information permits us to characterize the behaviour of molecules in the binding sites of target proteins. 6 An essential computational tool for analysing the time evolution of biological structures is MD. The stability and conformation of macromolecules, as well as their host-guest complexes with smaller compounds (i.e. drugs), can be better understood by MD simulations.
Moreover, the role of molecular flexibility, solvent and counter-ions can be assessed, as they are not taken into account in the rigid molecular docking calculations.
In the present study, receptor-based methods are combined in order to design new, potential α-amylase inhibitors for the control and reduction of blood sugar levels and the treatment of type II diabetes.

| Computational strategy to identify novel α-amylase inhibitors
In order to identify novel α-amylase inhibitors for the treatment of type-2 diabetes mellitus, we depicted a structure-based drug design strategy to select potential candidates to inhibit α-amylase. For this purpose, pancreatic α-amylase co-crystallized with the reference antidiabetic drug acarbose was retrieved from the Protein Data Bank (PDB: 1B2Y). 7 The steps followed were: 1. Virtual high-throughput screening: a molecular docking screening was performed in order to identify potential α-amylase inhibitors. Three different commercial databases were considered: Academic and Natural Compounds collection (from Specs) 8 and Natprod collection (from Discovery Systems, Inc.). 9 From over 380,000 molecules, 16 potential α-amylase inhibitors were selected.
2. Validation of the binding site: the magnitude and nature of interaction with the reference drug was investigated. From blind and site-specific docking, as well as MD simulations, enzyme residues interacting with the acarbose could be tracked, and used as references in the next step. The drug stability in the binding pocked could also be verified. Acarbose was selected as a reference antidiabetic drug in order to check if there is a similarity between the experimental location of α-amylase and the calculated location of the candidate inhibitors.
3. Candidate drugs interaction with α-amylase: a blind and site-specific docking analysis was performed in order to check whether these 16 potential α-amylase inhibitors will interact preferentially with the reported binding site with therapeutic activity.
For the tasks above, molecular docking and MD calculations were performed as described below.

| Molecular docking
Molecular docking studies play a major role in determining whether millions of synthesized compounds are effective drug substances.
The calculation of the interaction of a chemical compound on a protein structure allows the prediction of different binding modes.
Since the docking process brings powerful insights for explaining the enzyme-ligand relationship, it plays an important role in enzyme inhibition studies. 6 For the present study, in order to predict the potential activity of novel α-amylase inhibitors, the structure of human pancreatic alpha-amylase in complex with the carbohydrate inhibitor acarbose (PDB: 1B2Y) 7 was retrieved from the RCSB PDB. 10 This model provides a good balance between 3D protein structure resolution (3.2 Å) and accurate information (R-value free: 0.217) about the reference drug (acarbose) binding site.
Docking calculations were performed using AutoDockVina 11 and Schrödinger 12 software suite molecular modelling packages (version 2021-3), using default parameters unless otherwise reported.
Two types of docking calculations were performed: site-specific docking and blind docking. Site-specific docking focuses on a reduced region of the protein that is selected as the preferred binding active site, so that all Monte Carlo moves are centred in this region.
Here, the binding site was defined by the crystallographic position of the co-crystallized antidiabetic drug (acarbose). Blind docking considers the whole protein as potential sites for binding, and allows obtaining, as a result, a shortlist (called 'modes') of the potential active sites within the protein for the ligand. If some of these modes correspond to the active site of known drugs with demonstrated therapeutic action, it is an additional indication that the molecule being tested could be a good candidate with the same potential therapeutic action.
Therefore, although knowing the catalytic site of an enzyme is crucial, it is still important to perform a parallel blind docking or unguided analysis to study other potential interactions of the compound with other binding sites of the target protein. This information provides insight into the specificity of the drug with respect to the pharmacological target. By employing this strategy, we can evaluate if the most favourable binding site of potential drug candidates is within the protein's active site. Blind docking is, therefore, an excellent benchmark to test the accuracy of our computational methods. Since the docking site is known, blind docking should render the active site as one of the most energetically favourable for binding, and this condition has been satisfied by acarbose.
Moreover, a vital component of the molecular docking pipeline is the scoring function, as it determines the fitness of sampled poses (modes). 13 The scoring function is used to evaluate the binding affinity of the ligand to the protein, and it is based on various factors such as van der Waals interactions, hydrogen bonding, electrostatic interactions and solvation effects. 14 Albeit the score value obtained from molecular docking is not always accurate, it is still a useful tool for predicting and ranking the binding affinity of different ligands to a macromolecule, being widely used in drug discovery studies.
In addition to all these important and useful features of molecular docking studies, they should be supported by MD. The molecule, which is buried inside the protein structure, is also interacting with the solvent. The thermal effects on the kinetic energy of the molecule and on the protein conformation could lead to an activated jump of the molecule between different interaction sites that can be studied using MD calculations.

| Molecular dynamics simulations
According to the literature, there is good agreement between computational and experimental measurements of macromolecular dynamics. 15 Since receptor dynamics and flexibility play a key role in their interaction with candidate drugs, MD simulations can be a powerful tool for drug discovery studies. To complement the docking study and further evaluate the changes in enzyme and drug conformations, MD simulations were performed with pmemd.cuda software from the AMBER16 simulation package. 16 The α-amylase was described by the ff99SB-ILDN force field, 17 since it presents good agreement with experiments when in conjunction with TIP4P-Ew 18 water model. 19,20 General Amber Force Field 21 was considered to model drug molecules, as it is compatible with Amber force fields such as f99. 21 An all atom modelling of the system was considered, where hydrogen atoms for the macromolecule and drug were added with pdb4amber program, from Ambertools package.
Enzyme-ligand complexes were solvated with TIP4P-Ew water molecules, considering a solvation shell of at least 10.0 Å with respect to any solute atom. Na + and Cl − ions were added to the simulation box to neutralize the system and to simulate a saline solution (0.15 M), as previously reported in the literature. 22 AM1-BCC partial charges 23,24 were calculated for the enzyme-acarbose complexes.
First, energy minimization of the simulation boxes was performed, followed by temperature and pressure equilibration of the systems. The temperature was gradually increased to 298.15 K during an MD simulation of 100 ps at the NVT ensemble. Then, the thermalized system was simulated in the NPT ensemble for 250 ps and at 1 bar. Temperature and pressure control was ensured by Langevin thermostat 25 and Monte Carlo barostat 26 for orthogonal boxes, respectively. Next, the production run was performed, with trajectories being computed over 135 ns. Every MD simulation considered a time step of 2.0 fs, which was possible due to constraints applied to bonds containing hydrogen atoms, by the SHAKE algorithm. 27 To avoid finite-size effects, periodic boundary conditions were considered. Finally, a cutoff of 9.0 Å was set for the nonbonded interactions, while electrostatic interactions were treated with particle mesh Ewald 28 method.
The hydrogen bonds (HB) established during the simulations, mass-weighted root mean square deviation (RMSD), mass-weighted radius of gyration (R g ), gyration tensor and 3D histogram of drug conformations were computed from trajectories by cpptraj 29 program, available in the AmberTools. Additional analyses are included in Section S2 of Supplementary Information (SI).

| Computational toxicology study
In current drug development studies, the potential of a novel chemical is commonly evaluated initially through virtual tools. 30 The aim of this analysis is to predict organ toxicity, toxicological endpoints, median lethal dose (LD 50 ), druglikeness and ADME (absorption, distribution, metabolism, excretion) parameters of the selected molecules. The enzyme-drug complexes were screened using in-silico Pre-ADMET software 31 and SwissADME 32 to estimate their overall ADME properties and toxicity hazards. Furthermore, ProTox-II software 33 was used to estimate the organ toxicities and toxicological end points of molecules and their LD 50 (median LD 50 , the amount which causes the death of 50% of a group of test animals) values.

| RE SULTS AND D ISCUSS I ON
We outline here the strategy to identify novel α-amylase inhibitors with antidiabetic activity by employing two different approaches: molecular docking studies and MD. Initially, virtual high-throughput screening was conducted to generate a potential list of α-amylase inhibitors. Subsequently, the binding site for α-amylase antidiabetic activity was validated. Once the active binding site has been validated, potential antidiabetic compounds were docked (blind and site-specific docking) with α-amylase so that the amino acid interactions and docking score (DS) values were determined. Only molecules with both low DS and the same amino acid interaction pattern with α-amylase as acarbose will be selected as molecules with potential therapeutic antidiabetic activity.
Then, MD simulations were performed in order to identify and assess the stability of the ligandα-amylase interaction determined by molecular docking.

| Virtual screening of potential α-amylase inhibitors
A site-specific molecular docking study was performed for the reference α-amylase inhibitor (acarbose) as well as over 380,000 compounds from three different commercial databases. The Schrödinger software, 12 employing the OPLS3e force field, 34

| Αlpha-amylase binding site validation: Blind docking versus site-specific docking
The structure of the α-amylase co-crystallized with the reference antidiabetic drug acarbose (PDB: 1B2Y) contains specific amino acids that interact preferentially with acarbose and are related to its pharmacologic activity. A heavy benchmark for the software should be to test if a blind docking calculation of acarbose in αamylase is able to find the specific site of interaction as found in the crystallized structure. For this task, we have used AutodockVina.
Water and other additional molecules present in the crystallographic structure were removed from α-amylase for the calculations. The enzyme's polar hydrogens were explicitly considered, and Kollman partial charges 35  The results are shown in Figure 1. In the calculations, only the nine most stable locations will be selected, called from 'mode 1' to 'mode 9', according to the notation employed by AutoDock-Vina software, with 'mode 1' being the most stable and the others corresponding to increasingly weaker compound-protein interactions.
'Mode 5', among the most stable, is reasonably close to the experimental location of acarbose.
Considering the large number of possible binding sites throughout the entire protein, 'mode 5', among the most stable, is still reasonably close to the experimental location of acarbose. It is important to note that the use of more accurate methods to a high-throughput screening has a prohibitive computational cost. For instance, a direct application of quantum mechanics to protein systems is unfeasible due to the large molecular size of proteins, 38 and as a consequence, there is no general quantum mechanics-based method for simulating protein dynamics. On the other hand, even if considering a well parameterized force field, the utilization of MD simulations for a high-throughput screening is extremely expensive computationally, specially from the need of enhanced sampling methods for a proper simulation of the enzyme conformational space. Therefore, we have to come down to the limited accuracy of molecular docking simulations using force fields, 39 in order to explore a wide variety of new candidate drugs.
An analysis of specific interactions between α-amylase and acarbose in mode 5 (blind docking) is shown in Figure 2A Table S2), and they are included as a reference in Table 2. A comparison with other compounds will allow an assessment of their potential therapeutic activity. TA B L E 1 High-throughput site-specific docking score (DS) values (Schrödinger software) of potential α-amylase inhibitors across three databases of compounds. For reference, the value of acarbose is also included. The partial overlap of the mode 5 configuration (from blind docking) in the actual binding site of the enzyme is confirmed by comparing the interacting residues depicted in Figures 2 and 3.
Among the 12 residues interacting with the site-specific docking geometry ( Figure 2B; Table 2), six were detected in the mode Pi-Pi:

Movies of the MD simulations can be accessed by the QR codes
in Figure S2 and they illustrate that, structurally, the three acarbose locations simulated do not lead to significant changes in the enzyme, as shown in Figure S4A  Acarbose geometry from 1B2Y crystallographic structure is composed by a disubstituted amine, containing a disaccharide and a trisaccharide group ( Figure S1A). Although all acarbose chains are initially in contact with the enzyme surface in the initial configuration of the 'mode 1' simulation ( Figure S3A), the disaccharide moiety presented considerable mobility, being completely solvated after the first 16 ns of simulation ( Figure S3B), corresponding to the first significant change observed in the drug RMSD plot (yellow curve of Figure S4B). Nonetheless, the solvation of this disaccharide group is probably not the preferred state, as in the remaining time simulated, such a group re-established interactions with the enzyme surface, but with conformations different from the original ( Figure S3C).
Considering the co-crystallized structure as a reference, by relaxing the acarbose structure in the α-amylase binding pocket during the site-specific docking, the number of HB increases from 6 to 11 ( Figure 4A,B, respectively). When the co-crystallized structure is simulated through MD ( Figure 4C), a higher number of different HBs is observed (involving 19 residues), as a logical consequence of adding flexibility to the system. Even so, the main interactions (with colour tending to magenta in Figure 4C) correspond to the residues already accounted for in the site-specific docking geometry (such as ASP-300, GLU-233, THR-163, and GLN-63). Thus, MD findings support the docking results that will be considered in the following section as the reference interactions.
Since acarbose therapeutic effect seems to suggest a relatively strong binding to α-amylase, a stable active site can be expected from the MD simulations. This is in agreement with the binding free energy in solution obtained from the MD trajectories (Table S1, Section In summary, although it is considerably important that various modes from blind docking analysis of a drug are found near the enzyme active site, these MD results reinforce the importance of using site-specific docking as a criterion for the energy and the interactions between enzyme residues and candidate drugs.

| Molecular docking analysis of potential antidiabetic compounds with α-amylase: Blind and site-specific docking approach
In this section, we will expand upon the molecular docking analysis of the potential α-amylase inhibitors described in sections above.
Using both blind (best mode #1) and site-specific docking (using the experimental location 1B2Y), the main interactions of acarbose with the amino acid residues of α-amylase have been determined.
The calculations have been performed in AutoDockVina software, considering the AD4 scoring function, which uses potentials derived from early versions of AMBER force field. 40 Consequently, the following results were described by models compatible with those used in the MD simulations. Amino acid interactions and DS with αamylase of the 16 potential antidiabetic compounds are reported in Table S3, which also contains the reference values for acarbose that were calculated previously.
If we analyse both DS and the relevant amino acid interactions (best-performing candidates reported in Table 2, and a larger list of candidates in Table S3), two compounds stand out from the rest: AN-153I105594 and AN-153I104845. AN-153I105594 presents a large number of interactions with the relevant amino acids of α-amylase amino acids and a favourable DS (Table 2), both in site-specific (−10.8 kcal/mol; LEU165, HIS305, TRP59, TYR151, TYR62) and blind   docking (−10.3 kcal/mol; LEU165, HIS305, TRP59, TYR151). This is shown in Figure 5A. Seven of the nine modes whose geometry is analysed are very close to the targeted specific site (the active site of the protein). AN-153I104845 also presents a large number of interactions with the relevant amino acids of α-amylase amino acids and a favourable DS (Table 2), both in site-specific (−8.1 kcal/ mol; GLY306, HIS305, GLN63, TRP59, TYR62, GLU233) and blind docking (−7.4 kcal/mol; THR163, ASP300, HIS305, LEU165). This is shown in Figure 5B. Eight of the nine modes whose geometry is analysed are very close to the targeted specific site. This makes them good candidate drugs, since both molecules are likely to bind to the protein in the active site and interact with the same amino acid residues as those of acarbose.
Hederacoside C also presented a significantly stabilizing DS in tion. In spite of this significant shortcoming, Hederacoside C could also be considered as a candidate drug for α-amylase inhibition.

| Computational toxicology study
In silico preliminary studies were performed with three enzymedrug complexes (acarbose, AN-153I105594 and AN-153I104845) to estimate their toxicity, as well as the ADME properties. The computed druglikeness and ADME parameters are summarized in Table 3, which should be evaluated using Lipinski's rule of five.
According to the Lipinski's rule of five, the molecular weight (MW) must be, ideally, less than 500, the Log p value must be less than five, and there must be fewer than 5 and 10 HB donors and accep-  Table 3 indicate that these compounds are noncarcinogenic in nature, and have no effect on mutagenicity and cytotoxicity. Note: Toxicity class goes from 1 (toxic) to 6 (non-toxic). Caco-2 permeability is the permeability of compounds through a human intestinal epithelial cell barrier.
Abbreviations: ADME, absorption, distribution, metabolism, and excretion parameters; BBB, blood-brain barrier; GI, gastrointestinal absorption; HBA, number of hydrogen bonds acceptors; HBD, number of hydrogen bond donors; HIA, percentage of human intestinal absorption; LD 50 , median lethal dose; Log P, logarithm of the partition coefficient; MW, molecular weight; TPSA, total polar surface area.
TA B L E 3 Calculated ADME/Tox and druglikeness properties of AN-153I105594, AN-153I104845 and acarbose.
Some results of hepatotoxicity and immunotoxicity give activity, but the overall computationally predicted toxicological profile of these compounds (in particular, toxicity class and LD50) suggests AN-153I105594 and AN-153I104845 may not be toxic. Despite toxicity promising results, more sophisticated assays should be performed, and are beyond the scope of this work.

| CON CLUS IONS
In order to explore the potential interactions between amino acids and α-amylase, a combination of structure-based drug design computational techniques, molecular docking and MD methodologies were utilized. By analysing the amino acid interactions observed between acarbose and the target enzyme (PDB: 1B2Y), key interactions relevant to developing antidiabetic pharmacological activity, similar to the reference drug, were identified.
To identify suitable candidates with potential antidiabetic activity, a high-throughput docking screening involving over 380,000 compounds was performed. Based on DS and considering the cocrystallized binding site of acarbose reported in the literature, 16 molecules were selected as potential α-amylase inhibitors.
To investigate if some of these molecules could interact favourably with different binding sites within α-amylase, a blind docking approach was employed. Among the 16 previously identified compounds, two (AN-153I105594 and AN-153I104845) were predicted to exhibit the most favourable binding at the active site.
Furthermore, the interactions between these two compounds and the amino acids of α-amylase were found to be reasonably similar to those observed with acarbose. Therefore, these compounds can be considered potential α-amylase inhibitors, similar to the reference drug, acarbose.
The selected candidates were also evaluated by their toxicity and ADME parameters employing in silico methods. Both candidate drugs met Lipinski's criteria, as they possess physicochemical properties expected for a medicament. Likewise, other desired properties, such as a high intestinal adsorption and a low toxicity, were predicted to AN-153I105594 and AN-153I104845 compounds, highlighting them as suitable candidates for further theoretical and experimental studies.
This work is sought to present a valuable computational strategy for the selection of novel α-amylase inhibitors, as in silico studies are powerful tools to give useful insights and guide subsequent experimental work, such as in vitro α-amylase inhibition essays. writing -original draft (lead); writing -review and editing (lead).