In the mammalian central nervous system (CNS), a potential neurotransmitter, glutamate plays very important roles and it is the main excitatory neurotransmitter in the CNS. Glutamate mediates its effects by many families of receptors such as ionotropic glutamate receptors (iGluRs) and metabotrophic glutamate receptors (mGluRs). The iGluRs family contains many types of receptors. Among them, N-methyl-D-aspartate receptors or NMDARs are the receptors that are mainly responsible for learning and memory [15]. Therefore, any disruption in the normal signalling pathway of the NMDARs may lead to the damage of the CNS that may cause the AD to develop.

N-methyl-D-aspartate (NMDA) selectively mediates NMDARs. The NMDARs are encoded by human genes GRIN1, GRIN2A, GRIN2B, GRIN2C and GRIN2D [16]. The NMDARs can be divided into two groups: synaptic and extrasynaptic NMDARs. The activation of synaptic NMDARs leads to synaptic plasticity and cell survival (Figure 1) [17]. However, inappropriate NMDAR signalling leads to injuries in the neuronal system.

In normal condition, upon secretion, glutamate is secreted and binds to NMDAR, thus activates the receptor and mediates the calcium ion transport across the neuron cell. However, during abnormal signalling, inappropriate activation of NMDARs occurs. This causes excessive entry of Na⁺ and Cl⁻ ions into the neuron cells, which is responsible for acute neuronal swelling. Moreover, the excessive entry of Ca²⁺ ions into the post-synaptic neurons causes delayed neuronal degeneration (Figure 2) [18]. Therefore, the entry of excessive levels of ions leads to the toxic condition in the cell and causes neuronal cell death. This leads to the onset of AD. On the other hand, the β amyloid plaques, formed due to AD, selectively activates extrasynaptic NMDARs. The extrasynaptic NMDARs cause the deleterious effects like tau protein phosphorylation and induction of apoptosis by activating caspase-3, which also leads to the onset of AD [19].

Administration of memantine can block the activity of NMDARs by binding with those receptors and thus mediate its therapeutic properties in inhibition of AD [20]. In this experiment, memantine (PubChem CID: 4054) was used to dock against GluN2D (PDB ID: 3OEM), which is a NMDAR or ionotropic glutamate receptor [21].

Glycogen synthase kinase-3β (GSK-3β) is an enzyme kinase that plays important role in the development of AD by phosphorylating the tau protein [22]. Aβ is caused by defective proteolytic processing of amyloid precursor protein (APP). This defection leads to the production and deposition of 42 amino acids long neurotoxic forms of β-amyloid (Aβ) peptides. Three enzymes determine whether the neurotoxic forms of β-amyloid will be formed or not. β-secretase and γ-secretase cleave APP sequentially at the N-terminus and C-terminus, respectively. These cleavages lead to the beta amyloid production and when α-secretase cleaves APP, the possibility of formation of Aβ minimizes [23] [24] [25]. APP is a surface membrane protein that can be processed by two major pathways: non-amyloidogenic pathway and amyloidogenic pathway. In the non-amyloidogenic pathway, the α-secretase and γ-secretase enzymes cleave the transmembrane domain of APP, sequentially. These cleavages give rise to the fragments that are easily degradable [26]. However, in amyloidogenic pathway, the APP is cut by β-secretase and γ-secretase, which form β-amyloid (Aβ) peptides and the Aβ peptides tend to aggregate and form plaques [27]. Microtubule associated protein (MAP) tau is a protein that is found in the neuron cells and their main function is to stabilize the microtubules. They are phosphorylated in lesser extent in the normal adult brain. However, in the AD patients, they are found to be highly phosphorylated. The abnormally phosphorylated tau acquires the shape of paired helical filaments (PHFs) and forms neuro fibrillary tangles (NFTs) with other abnormally phosphorylated tau proteins. These NFTs are insoluble tangles that appear to be accumulated as tangled mass in the brain. NFTs interfere with the normal functions of the neurons by destabilizing the microtubules [28]. One of the proteins responsible for the tau phosphorylation is GSK-3β. The GSK-3 is a serine/threonine kinase enzyme. In the brain, the GSK-3β is responsible for the tau phosphorylation [29]. GSK-3β phosphorylates 36 sites on the tau protein [30]. There is evidence that, Aβ is responsible for the tau phosphorylation [31]. Aβ activates and causes over production of GSK-3β signaling by inhibiting the inhibitory phosphorylation mechanism of this enzyme. Therefore, the formation of Aβ directly causes the over-activation of GSK-3 which in turn hyper-phosphorylate the tau protein and form NFTs. NFTs ultimately result the AD development. Moreover, the formation of NFTs later leads to the apoptosis of the neuron (Figure 3) [32]. A potent inhibitor of GSK-3β is hymenialdisine [10]. Studies have found that another compound named tideglusib can also be used as GSK-3β inhibitor [11]. Moreover, kenpaullone is another compound that has GSK-3β inhibitory activity [12]. Molecular docking has already been performed successfully against the GSK-3β (PDB ID: 1Q5K) using 1,3-disubstituted-1H-pyrazol-5-ols as ligands [33]. In the experiment, docking was performed using hymenialdisine (PubChem CID: 11313622), tideglusib (PubChem CID: 135413546) and kenpaullone (PubChem CID: 3820) as ligands against the GSK-3β (PDB CID: 1Q5K).

The pathway of β-secretase enzyme involves the abnormal proteolytic processing of APP. The cleavage of the APP protein by β-secretase leads to the formation of β-amyloid (Aβ) plaques [34].

The formation of Aβ is decided by the activities of three enzymes: α-, β- and γ-secretases. The APP protein can be cleaved by two major pathways: non-amyloidogenic pathway and amyloidogenic pathway. In non-amyloidogenic pathway, the transmembrane portion of APP protein is cleaved sequentially by α- and γ-secretases. These cleavages don’t lead to the formation of Aβ. Since α-secretase cleave within the Aβ region, the Aβ formation never occurs [26]. However, in the amyloidogenic pathway, the abnormal cleavage of APP is carried out sequentially by β- and γ-secretases and the β-secretase cuts the APP protein at a site 99 amino acids away from the C-terminus, leaving the C-terminal portion of the protein in the membrane, called C99. This newly generated C99 fragment contains the first amino acid of the Aβ plaque, at the newly generated N-terminus. Then γ-secretase cuts the C99 between 38th and 43th amino acids and liberates the Aβ peptides, which later aggregate together with other Aβ peptides and form plaques. This Aβ plaque formation is one of the main reasons behind the AD onset (Figure 3) [35].

One of the current approaches to treat AD is the use of β-secretase inhibitors that can inhibit the activity of β-secretase [36]. In Silico, studies have already been conducted against β-secretase (PDB ID: 2OHM) using 1,3-disubstituted-1H-pyrazol-5-ols as the inhibitors [33]. Another compound, named dihydrospiro[dibenzo[a,d][7]annulene-5,4’-imidazol], has been patented as an potent inhibitor of β-secretase [13]. In our study, docking study was performed with dihydrospiro[dibenzo[a,d][7]annulene-5,4’-imidazol] (PubChem CID: 24983268) against β-secretase (PDB ID: 2OHM).

Abnormal phosphorylation of tau protein is one of the main reasons of AD development [37]. Many enzymes are responsible for such type of phosphorylation, for example, glycogen synthase kinase 3 (GSK-3), cyclin-dependent kinase 5 (CDK-5), cAMP-dependent protein kinase A etc. The DYRK1A (dual specificity tyrosine (Y)-phosphorylation-regulated kinase 1A) is a recently discovered enzyme that is also responsible for the abnormal phosphorylation of tau protein. This enzyme is expressed from the DYRK1A gene of 21^st chromosome. This enzyme exhibits dual specificity. First, the enzyme autophosphorylates itself on the tyrosine 321 residue for activation and then phosphorylation of the target protein occurs [38].

The abnormally phosphorylated tau acquires the shape of paired helical filaments (PHFs) and forms NFTs that are insoluble and appear to be accumulated as tangled mass in the brain (Figure 3) [28].

Current treatment focusing on DYRK1A enzyme uses various inhibitors that can bind to the DYRK1A enzyme and inhibit its activity. Some of the inhibitors are: harmine, harmol, 1-methyl-4-phenylpyridinium etc. [14]. Docking was performed using the inhibitors: harmine (PubChem CID: 5280953), harmol (PubChem CID: 68094), 1-methyl-4-phenylpyridinium (PubChem CID: 39484) against the DYRK1A enzyme (PDB ID: 2VX3).

Due to the advancements of various computer softwares, it is now possible to simulate the biological environment with the aid of various softwares without even using the original biological environment. Molecular docking is a technique that places a possible ligand molecule in the binding site of a suspected target protein. Molecular docking acts on algorithms that determine the potential interactions between macromolecules like protein-protein interactions, protein-drug interactions etc. These algorithms examine the orientational and conformational degrees of freedom of ligand molecules within the binding pocket of the target molecules and generate scores to select the best possible pose of the ligands for ranking them in correct order [39] [40].

ADME/T test determines the ADME/T properties of means a drug or candidate molecule. The ADME/T test determines how a candidate drug molecule may be absorbed, distributed, metabolized and excreted as well as its toxicological properties. ADME/T-test is one of the prerequisites for a potential candidate molecule to become a successful drug [41] [42].

In this experiment, eight drug molecules: memantine, hymenialdisine, tideglusib, kenpaullone, dihydrospiro[dibenzo[a,d][7]annulene-5,4’-imidazol], harmine, harmol and 1-methyl-4-phenylpyridinium, which have already been patented or developed in trials, have been used to dock against four proteins: NMDAR, glycogen synthase kinase-3 (GSK-3), β-secretase and dual specificity tyrosine (Y)-phosphorylation-regulated kinase 1A (DYRK1A), respectively, to study their potential interaction in a search for the best possible drug compound.

Three dimensional structures of NMDAR (PDB ID:3OEM), GSK-3β (PDB ID:1Q5K), β-secretase (PDB ID:2OHM) and DYRK1A (PDB ID:2VX3) were downloaded (sequentially) in PDB format from the Protein Data Bank (http://www.rcsb.org/) online server (Figure 4). The proteins were then prepared and refined using the Protein Preparation Wizard in Maestro Schrödinger Suite 2015-1. Bond orders were assigned and hydrogens were added to heavy atoms. Selenomethionines were converted to methionines as well as all the waters were deleted. Finally, the structure was optimized and then minimized using force field OPLS_2005. Minimization was done setting the maximum heavy atom RMSD (root-mean-square-deviation) to 30 Å and any remaining water less than 3 H bonds to non water was again deleted during the minimization step.

The 2D conformations of memantine (PubChem CID: 4054), hymenialdisine (PubChem ID: 11313622), tideglusib (PubChem CID: 135413546), kenpaullone (PubChem CID: 3820), dihydrospiro[dibenzo[a,d][7]annulene-5,4’-imidazol] (PubChem CID: 24983268), harmine (PubChem CID: 5280953), harmol (PubChem CID: 68094) and 1-methyl-4-phenylpyridinium (PubChem CID: 39484) were downloaded (sequentially) from PubChem (http://www.pubchem.ncbi.nlm.nih.gov/) (Figure 5). These structures were then prepared using the LigPrep function of Maestro Schrödinger Suite 2015-1. Minimized 3D structures of ligands were generated using Epik2.2 and within pH 7.0 ± 2.0. Minimization was again carried out using OPLS_2005 force field which generated 32 possible stereoisomers.

Grid usually confines the active site to shortened specific area of the receptor protein for the ligand to dock specifically. In Glide, a grid was generated using default van der Waals radius scaling factor 1.0 and charge cutoff 0.25 which was then subjected to OPLS_2005 force field. A cubic box was generated around the active site (reference ligand active site). Then the grid box volume was adjusted to 15 × 15 × 15 for docking test.

SP adaptable glide docking was carried out using Glide in Maestro Schrödinger Suite 2015-1. The Van der Waals radius scaling factor and charge cutoff were set to 0.80 and 0.15 respectively for all the ligand molecules. Final score was assigned according to the pose of docked ligand within the active site of the receptor. The ligand with the lowest glide docking score was considered as the best ligand. The docking results are listed in Table 1. After successful docking, the 2D representations of the best pose interactions between the ligands and their respective receptors were generated using Maestro-Schrödinger Suite 2015-1 (Figure 6). The 3D representations of the best pose interactions between the ligands and their respective receptors were obtained using Discovery Studio Visualizer (Figure 7).

The molecular structures of every ligands were analyzed using SWISSADME server (http://www.swissadme.ch/) to confirm whether they obey Lipinki’s rule of five or not, along with some other properties. Various physicochemical properties of ligand molecules were calculated using OSIRIS Property Explorer (https://www.organic-chemistry.org/prog/peo/). The drug likeness properties of the selected ligand molecules were analyzed using SWISSADME server (http://www.swissadme.ch/) as well as the OSIRIS Property Explorer (https://www.organic-chemistry.org/prog/peo/). The results of drug likeness property analysis are summarized in Table 2 [45]. The ADME/T for each of the ligand molecules was carried out using an online-based server ADMET-SAR (http://lmmd.ecust.edu.cn/admetsar1/predict/) to predict their various pharmacokinetic and pharmacodynamic properties including blood brain barrier permeability, human abdominal adsorption, Caco-2 permeability, Cytochrome P (CYP) inhibitory capability, carcinogenicity, mutagenicity etc. The result of ADME/T for all the ligand molecules is depicted in Table 3.

All the selected ligand molecules were docked successfully against NMDAR, GSK-3β, β-secretase and DYRK1A.

Memantine generated docking score (binding energy) of −4.075 Kcal/mol and glide energy of −9.918 Kcal/mol, when docked against NMDAR. Memantine formed 1 hydrogen bond with proline 170 residue of NMDAR and the distance was 2.03 Å (Table 1 and Figure 8).

Hymenialdisine generated docking score of −8.079 Kcal/mol and glide energy of −45.218 Kcal/mol, when docked against GSK-3β. Hymenialdisine formed 6 hydrogen bonds with lysine 85, valine 135 (formed 2 hydrogen bonds), tyrosine 134 (2 bonds) and aspartic acid residues of GSK-3β and the distances were 2.73, 2.86, 1.88, 2.86, 2.49 and 2.40 Å, respectively (Table 1 and Figure 8).

Tideglusib generated docking score of −6.445 Kcal/mol and glide energy of −36.290 Kcal/mol, when docked against GSK-3β and generated 2 hydrogen bonds with the target protein GSK-3β (Table 1 and Figure 8).

Kenpaullone generated docking score of −7.545 Kcal/mol and glide energy of −35.502 Kcal/mol, when docked against GSK-3β. Kenpaullone formed 1 hydrogen bond with valine 135 of GSK-3β and the distance was 2.03 Å (Table 1 and Figure 8).

Dihydrospiro[dibenzo[a,d][7]annulene-5,4’-imidazol] generated docking score of −4.742 Kcal/mol and glide energy of −36.295 Kcal/mol, when docked against β-secretase. Dihydrospiro[dibenzo[a,d][7]annulene-5,4’-imidazol] formed 3 hydrogen bonds with aspartic acid 32, threonine 73 and aspartic acid 228 residues of β-secretase and the distances were 1.70, 2.50, and 2.79 Å, respectively (Table 1 and Figure 8).

Harmine generated docking score of −7.570 Kcal/mol and glide energy of −30.172 Kcal/mol, when docked against DYRK1A. Harmine formed 2 hydrogen bonds with leucine 241 residue of DYRK1A and the distances were 2.02 and 2.41 Å (Table 1 and Figure 8).

Harmol generated docking score of −6.583 Kcal/mol and glide energy of −31.214 Kcal/mol, when docked against DYRK1A. However, harmol did not generate any hydrogen bond with the target protein DYRK1A (Table 1 and Figure 8).

1-methyl-4-phenylpyridinium generated docking score of −5.214 Kcal/mol and glide energy of −16.037 Kcal/mol, when docked against DYRK1A and generated 3 hydrogen bonds with the target protein DYRK1A, unlike harmol (Table 1 and Figure 8).

Lipinski’s rule of five demonstrates that the acceptable ranges of the best drug molecule for all the five parameters are: molecular weight: ≤500, number of hydrogen bond donors: ≤5, number of hydrogen bond acceptors: ≤10, lipophilicity (expressed as LogP): ≤5, molar refractivity from 40 to 130 [46]. All the ligand molecules followed the Lipinski’s rule of five without any violation. The results are listed in Table 2.

None of the molecules violated the Veber and Egan rules. Only hymenialdisine violated the Ghose filter factor. However, memantine, harmol and 1-methyl-4-phenylpyridinium violated the Muegge rule of druglikeness properties. All the molecules showed similar bioavaibility score of 0.55.

Tideglusib showed the lowest LogS value of −5.09, whereas, 1-methyl-4-phenylpyridinium had the highest LogS value of 0.47. Kenpaullone and dihydrospiro[dibenzo[a,d][7]annulene-5,4’-imidazol] showed similar LogS values, −4.47. Hymenialdisine and harmol showed slightly similar LogS values of −1.94 and −2.18, respectively. Memantine and harmine had LogS values of −3.02 and −4.05, respectively.

Memantine, kenpaullone and harmol had 1 hydrogen bond acceptor each. Both tideglusib and harmine had 2 hydrogen bond acceptors. Moreover, both hymenialdisine and dihydrospiro[dibenzo[a,d][7]annulene-5,4’-imidazol] had 3 hydrogen bond acceptors. However, 1-methyl-4-phenylpyridinium didn’t show any hydrogen bond acceptor. Memantine, dihydrospiro[dibenzo[a,d][7]annulene-5,4’-imidazol] and harmine had 1 hydrogen bond donor each, whereas kenpaullone and harmol had 2 and hymenialdisine had 4 hydrogen bond donors. However, tideglusib and 1-methyl-4-phenylpyridinium didn’t have any hydrogen bond donor.

Dihydrospiro[dibenzo[a,d][7]annulene-5,4’-imidazol] showed the highest molar refractivity of 106.82 and 1-methyl-4-phenylpyridinium had the lowest score of 55.47, although memantine had also very close score to 1-methyl-4-phenylpyridinium (55.68). Harmine and harmol had quite similar scores (65.06 and 61.39, respectively). Hymenialdisine, tideglusib and kenpaullone had scores of 82.32, 97.43 and 86.73, respectively.

Hymenialdisine possessed the largest topological polar surface area (TPSA) (112.37 Ų) and 1-methyl-4-phenylpyridinium had the lowest area of 3.88 Ų. Tideglusib and dihydrospiro[dibenzo[a,d][7]annulene-5,4’-imidazol] had almost similar scores of 72.24 Ų and 67.92 Ų, respectively. Memantine had quite low score of 26.02. The results showed by the rest of the molecules were: 44.89 (kenpaullone), 37.91 (harmine) and 48.65 (harmol). Tideglusib exhibited the highest druglikeness score of 2.98 and the lowest solubility score (−7.1), however, its drug score was very low (0.15) and it showed very high tumorigenic and mutagenic activity. Tideglusib had no effect on reproductive system and irritation. Harmol should be the best molecule in this regard since it showed good druglikeness score of 2.63 and very good drug score of 0.91, second highest solubility score (−2.42) and it did not exhibit any of the deleterious effects. Kenpaullone showed druglikeness score of 1.08, solubility score of −5.14 and drug score of 0.33, however, its reproductive effectivity was quite high, although it didn’t have any irritant, tumorigenic and mutagenic properties. Dihydrospiro[dibenzo[a,d][7]annulene-5,4’-imidazol] also showed relatively good druglikeness score (1.34) and drug score (0.63) as well as moderate solubility (−4.32) and it had no reproductive effectiveness or irritant, tumorigenic and mutagenic properties. Other molecules, memantine, hymenialdisine, harmine had druglikeness scores of −0.8, 0.39, 0.35, respectively and drug scores of 0.6, 0.73 and 0.56, respectively. Memantine and hymenialdisine showed no harmful effect, however, harmine was quite mutagenic. However, the druglikeness score, drug score, solubility score, reproductive effectiveness, irritation properties, tumorigenic and mutagenic properties of 1-methyl-4-phenylpyridinium are not determined yet.

The results of ADME/T test of selected ligand molecules are listed in Table 3. All the ligand molecules showed the ability to cross the Blood-Brain Barrier (BBB) and gave positive results in human intestinal absorption (HIA). Only memantine and 1-methyl-4-phenylpyridinium showed Caco-2 permeability. All the molecules were proved to be P-glycoprotein Inhibitors.

All the molecules were non-substrate of CYP450 2C9 and CYP450 2D6. However, hymenialdisine, kenpaullone and dihydrospiro[dibenzo[a,d][7]annulene-5,4’-imidazol] were substrates of CYP450 3A4, whereas other molecules were non-substrates. Memantine, tideglusib and 1-methyl-4-phenylpyridinium were non-inhibitors of CYP450 1A2, however, the other ligand molecules were inhibitors. Only tideglusib showed inhibition of CYP450 2C9. Kenpaullone, harmine and harmol were the inhibitors of CYP450 2D6 and only tideglusib was the CYP450 2C19 inhibitor. Kenpaullone and harmine were the inhibitors of CYP450 3A4. Tideglusibe, kenpaullone, harmine and 1-methyl-4-phenylpyridinium showed high CYP inhibitory promiscuity. Others showed low CYP inhibitory promiscuity.

Only harmine and harmol showed AMES-toxicity. Though all the ligand molecules were non-carcinogenic, all of them were not readily biodegradable and all of them showed level-III oral acute toxicity.