Computational and pharmacokinetics studies of 1,3-dimethylbenzimidazolinone analogues of new proposed agent against Alzheimer's disease

Abstract

Background

Alzheimer's disease (AD) is a multifactorial disorder that gradually destroys wisdom and memory skills. Currently, this disease can only be treated palliatively. However, the molecular mechanisms underlying this condition remain elusive. Therefore, these treatments are inadequate. Current medications can only increase patient warning signs. Chemical structures were drawn using Chemsketch software. Spartan’14 software was used to optimize the structures using density functional theory (DFT). The PaDEL software was used to generate the descriptors. The genetic function algorithm (GFA) and multi-linear regression (MLR) approaches were used to generate the QSAR model.

Results

In the present study, we performed a computational investigation, molecular docking, and pharmacokinetics analysis of 1,3-dimethylbenzimidazolinone derivatives. The descriptors generated in the model are AATS7i, MATS5p, SpMin7_Bhe, and GATS6c. Compounds 13 and 21 have the best binding scores, 11.2 kcal/mol and 10.8 kcal/mol, respectively, and optimal protein–ligand interactions with AChE. These compounds have brilliant pharmacokinetic and physicochemical properties.

Conclusions

The model was validated and found to have good internal and external assessment parameters: R² = 0.937, $R_{{{\text{adj}}}}^{2}$ = 0.863, $Q_{{{\text{cv}}}}^{2}$ = 0.788, $R_{{{\text{test}}}}^{2}$ = 0.756, LOF = 0.0268, $cR_{{\text{p}}}^{2}$ = 0.677. In summary, these data suggested that compounds 13 and 21 are promising multifunctional agents against AD.

1 Background

Alzheimer's disease (AD) affects modern societies. AD is a product of dementia and a serious health problem worldwide. Today, the number of people living with dementia is estimated at 50.0 million worldwide, and by 2040, it may increase to 130.0 million, which is responsible for approximately 73% of the cases [1]. In the light of these realities, we may expect over 90 million AD patients in the next three to four decades [1]. Deposits [2, 3], oxidative stress [3, 4], and low levels of acetylcholine [4,5,6] have all been proposed as sources of AD, and several drugs have been introduced into the market for the treatment of AD, especially those approved by the US government. They are six in number. Neuronal death cannot be slowed or stopped by any drug. The signs of AD have been improved by the introduction of drugs [7]. A few drugs include donepezil, galantamine, rivastigmine, and memantine [8].

AD remains irredeemable owing to the limited availability and low efficacy of drugs. Many researchers worldwide have intensified their efforts to find effective novel drugs and new biological targets.

To improve the drug improvement process in a more cost-efficient manner and to reduce failures in the final stage, the engagement of CADD techniques became indispensable for the initial stage of the drug discovery process. Valued evidence about the interface pattern between the protein and ligand, as well as the binding affinity, is a parameter for rational drug design, which provides a knowledge-driven method that will provide maximum output. In this study, a structure-based modelling approach was employed, using the 3D structure of the protein target for the screening of potential modulators, followed by biologically tested synthesized compounds and optimization. The best model was selected among the generated models based on statistical validity and was used for structural optimization to enhance potency and identify new chemical compounds in the in silico virtual screening of a large chemical catalogue [9, 10]. The rationale behind the work is Y = A₁x₁ + A₂x₂ + A₃x₃ + B. Y is the activity (pIC₅₀), where “A”s and “x”s are regression coefficients for conforming nonpartisan variables that represent molecular descriptors of molecules; “A”s correspond to “X”s. the last variable, “B”, is the regression constant.

The present study aimed to use QSAR, molecular docking simulation, in silico design, and ADMET properties to identify new drugs that are efficient against AD.

2 Methods

2.1 Data set gathering and activities

The data were obtained from the literature [11] as potential compounds against Alzheimer's disease. Twenty-six derivatives of the compounds were extracted and selected for modelling. A list of the compounds is presented in Table 1. Data are presented as IC₅₀ values (µM). The concentration was converted to the PIC₅₀ by taking the inverse logarithm of each value. The residual value is the difference between the observed and calculated values, as listed in Table 1. The test set is represented by an * sign.

Table 1 Molecular structures of inhibitory compound

2.2 Molecular optimization and generation of molecular properties

All inhibitory compounds were optimized using the 1.1.4 version of Spartan software to attain a steady conformation at minimum energy. Lateral optimization processes were achieved with the help of the mechanical force field and density functional theory (DFT) [12]. V.2.2.0 of PaDEL software was used to generate 1800 descriptors, which simply describe the relationship between the experimental activity and the molecular structure of each molecule.

2.3 Data normalization, pretreatment, and training and test set generation

Each variable influences a good model to generate a good model. Therefore, the descriptor values generated from PaDEL software V2.20 were normalized using Eq. (1) to give the descriptor an equal chance at the inception [13, 14].

$$A = \frac{{X_{i} - X_{\min } }}{{X_{\max } - X_{\min } }}$$

(1)

where X_max and X_min are the maximum and minimum values of each descriptor, respectively. Xi is the value for each descriptor molecule. Afterwards, the normalized data were pretreated with pretreatment software (https://dtclab.webs.com/software-tools) to eliminate redundant molecular properties. The dataset was divided into a ratio of 7:3 that is the training and test sets, using the algorithm of Kennard and Stone, which was incorporated into the DTC lab software. Model and internal validation tests were confirmed using the training set. Meanwhile, the developed model was performed on a test set [15].

2.4 Derivation of the model and models and validation

Multi-variant equations (models) were generated using the multi-linear regression approach (MLR) and genetic function approximation (GFA), and Material Studio software V.8.0 was used to confirm the robustness and predictability of the built model.

2.5 Leverage applicability domain (LAD)

In vividly determining that a compound is an outlier, leverage values and standardized residuals are used together in what is called a Williams graph to describe the LAD of a given model. Influential and outlier molecules present in the dataset were determined using this approach, the leverage (hi) approach is defined in Eq. (2) and was used to define the domain space ± 2.5 for outlier molecules [15].

$$hi = Wi \left( {W^{T} W} \right)^{ - 1} W_{i}^{T}$$

(2)

where W_i represents the matrix of i in the training set. W represents the n × d matrix descriptor for the training set. W^T is the transpose of training set (W) represents the transposed matrix and W_i. The warning leverage k* defined in Eq. (3) is the boundary for establishing the presence of an influential molecule.

$$k^{*} = 3\frac{{\left( {x + 1} \right)}}{A}$$

(3)

where A and x are the total number of training set and descriptors present in the built model.

2.6 Y-randomization validation and Confirmation of the build model

To investigate the robustness of the developed models and determine whether the models were the result of chance correlations [16, 17],

A Y-randomization test was applied to the final model. Here, the activity value is randomly permuted and a new QSAR model will be constructed using the optimum descriptors in the final models as reported in the literature [18]. This new model is expected to report R² and Q² values that are very low for several trials compared to those reported by the original models. For clarity, this procedure was repeated 100 times. The conditions for authenticating both the test and the training set were stated and related to the commonly accepted threshold value shown in Table 6 for any QSAR model [19, 20] to assert the consistency, fitting, stability, strength, and predictability of the developed models.

2.7 Docking studies

2.7.1 Receptor and ligand preparation

The acetylcholinesterase receptor (4MOE) crystal with a low resolution shown in Fig. 1A was downloaded from the PDB (www.rcsb.org). Using the Visualizer software in Discovery Studio, completely imported foreign matter such as cofactors and ligands associated with the enzyme were removed, and the target protein was saved in the PDB format. Next, the target protein saved in PDB format was imported into PyRx software and converted into macromolecules (Fig. 1B). The Spartan software, an optimizing tool, was used to attain the unchanging minimal energy of the benzimidazolinone derivative conformation. The optimized ligands were saved in PDB format, recognized by PyRx software, and converted to micro molecules (pdbqt).

Fig. 1

A Protein target (4MOE) and B prepared receptor

2.7.2 Docking of receptor and ligand

PyRx virtual screening software was used to perform ligand–receptor interactions and molecular docking of benzimidazolinone derivatives. AutoDock Vina and AutoDock 4.2 in PyRx software were used as the docking software. Discovery Studio Visualizer software, V 2016, was used to create and investigate the docked results [21].

2.8 Drug-likeness and prediction of ADMET

Hypothetically derived statistical models are generated by in silico approaches for the determination of absorption distribution, metabolism excretion, and toxicity (ADMET) parameters, which are produced by relating the structural features of compounds. They have been measured in a given assay to their biological response, and are now commonly used because of their low resource requirements [21]. Therefore, two out of twenty-six compounds that showed reasonable binding modes with AChE based on similar binding patterns to donepezil and advanced negative binding affinity, were subjected to physicochemical studies. Studies were carried out using the Swiss online ADMET web tool [22] to evaluate the model for physicochemical properties, pharmacokinetics, drug-likeness, and medicinal chemistry friendliness, including in-house proficient methods such as the BOILED-egg, ILOGP, and Bioavailability Radar of the selected compounds. A diagram of WLOGP versus TPSA was plotted to determine the blood–brain barrier (BBB) and human gastrointestinal absorption ability of the compounds [22].

3 Results

3.1 QSAR analysis

The best model for predicting the derivatives of 1, 3-dimethylbenzimidazolinone against AD was successfully achieved by adopting the techniques of the computational method. A dataset of 26 molecules was split into 20 training sets and 6 test sets using the method of Ambure and his research group [23]. QSAR models were derived using the MLR technique from 20 training set compounds, which also served as a data set for internal validation tests and the validation of the model was conducted on the test set.

3.2 Constructed model

$$\begin{aligned} {\text{PIC}}_{50} & = 11.585405732 \times {\text{AATS7i}} + 0.212053417 \times {\text{MATS}}5p + 0.230548471 \\ & \quad \times {\text{SpMin7}}\_{\text{Bhe}} {-}0.753332018 \times {\text{GATS6c}} + 2.057162 \\ \end{aligned}$$

The experimental activities reported in the literature and the theoretical activities calculated for all the anti-Alzheimer compounds are presented in Table 1. The residual value, which is the difference between the experimental and theoretical activities, was observed to be significantly low. A low residual value indicates predictability of the model. The best descriptors that efficiently described the anti-Alzheimer compounds in relation to their experimental activities selected by the GFA approach are reported in Table 2.

Table 2 Descriptors used in the model

3.3 Mechanistic information of descriptors in the model built

Table 2 provides a comprehensive description of the molecular descriptors in the constructed model.

3.4 Derivation of the model and models and validation

Table 3 shows the strength and direction of influence of each descriptor in the constructed model, assessed by determining the standard regression coefficient (Xj) and mean effect (AE) [24]. The molecular properties highlighted in Table 4 have correlation coefficients of 0.6. Table 5 presents the validity results of the internal assessment to ensure that the model is reliable. The results confirmed the stability and robustness of the model as valid because the calculated parameters were all in full agreement with the general validation criteria (Table 5).

Table 3 Statistical consideration to validate the descriptors

Table 4 Validation of the descriptors using Pearson's correlation matrix

Table 5 Validation parameters to confirm the built model

Table 6 shows Y-Randomization parameters while Table 7 presents the threshold and coefficient of Y-randomization and Table 8 shows validation results for external assessment.

Table 6 Y-randomization parameters test

Table 7 Random models parameters

Table 8 External validation

Table 9 Pharmacokinetics of compounds 13 and 21

Figures 2 and 3 show graph of calculated activity versus observed activity of training set and graph of calculated activity versus observed activity of test set, respectively.

Fig. 2

Graph of calculated activity versus observed activity of training set

Fig. 3

Graph of calculated activity versus observed activity of test set

The William’s graph shows the domain of applicability space (AD) shown in Fig. 4. The leverage value of compounds 10 and 15 was observed to be higher than the h* = 0.75 (i.e. warning leverage).

Fig. 4

Standardized residual activity versus observed activity

3.5 Molecular docking studies

The results of the binding energy of the selected benzimidazolinone derivatives (compounds 13 and 21) and the reference drug to vital target proteins (AChE), implicated in the pathogenesis of AD from the molecular docking study are, respectively, shown in Figs. 5, 6, and 7.

Fig. 5

A and B show the 3D and 2D docking interactions between ligand 13 and receptor, respectively

Fig. 6

A and B showed the 3D and 2D docking interactions between ligand 21 and receptor, respectively

Fig. 7

Show the 3D and 2D interactions between donepezil and acetylcholinesterase

3.6 Drug change assessment of selected compounds

We performed in silico ADMET studies on the two molecules reported in Tables 9 and 10 to supplement the results of 3D-QSAR and docking studies. Table 11 displays the physicochemical properties of compounds 13 and 21.

Table 10 Drug-likeness and medicinal chemistry of compounds 13 and 21

Table 11 Physicochemical properties of compounds 13 and 21

The predicted bioactivity scores of the selected compounds obtained Molinspiration software v2018.03 Chemoinformatics tools are given in Table 12.

Table 12 Bioactivity score of the compounds

Figure 8 exhibits the oral bioavailability graph of the two compounds on base of the six features discussed in physicochemical properties while Fig. 9 shows points located in the BOILED-Egg’s, yolk of molecules 13 and 21.

Fig. 8

Radar plots of six drug-likeness parameters used to calculate the oral bioavailability of the studied molecules

Fig. 9

BOILED-egg plot

4 Discussion

4.1 QSAR analysis

The considered model was carefully chosen and reported because it is statistically fit with the following assessment parameters as compared to other constructed model: R² of 0.937, $R_{{{\text{adj}}}}^{2}$ of 0.863, $Q_{{{\text{cv}}}}^{2}$ of 0.788, $R_{{{\text{test}}}}^{{2}}$ of 0.756, LOF of 0.0268, and $cR_{{\text{p}}}^{2}$ of 0.677. The selected model was found to have passed the minimum recommended values for validation of good QSAR models as reported in [24]. The details of the descriptors used in the model are listed in Table 2.

4.2 Mechanistic information of descriptors in the model built

Table 2 provides a comprehensive description of the molecular descriptors of the constructed models. Furthermore, the model showed a positive contribution from the descriptors AATS7i, MATS5p, and SpMin7_Bhe, but a negative contribution from the descriptor GATS6c. This means that an increase in the magnitude of AATS7i, MATS5p, and SpMin7_Bhe descriptors will positively influence the prediction of PIC₅₀ with the negative influence of the GATS6c descriptor. However, the AATS7i descriptor has the highest contribution, and is the most significant descriptor to be considered in the design of new hypothetical compounds. In addition, the signs of the regression coefficients for each descriptor indicated the direction of influence of the descriptors in the models, such that a positive regression coefficient associated with a descriptor will augment the activity profile of a compound, while the negative coefficient will diminish the activity of the compound.

4.3 Derivation of the model and models and validation

The influence of each descriptor in the constructed model was assessed by determining standard regression coefficients Xj and AE [24]. Table 3 shows the strength and direction, as well as size and symbols for the Xj and AE values with which each descriptor influences the activity model. The connection between the descriptors and the activity of each compound was determined using ANOVA. The probability value for each descriptor at the 95% confidence level was found to be p $< \frac{1}{20}$ 1, as shown in Table 3. Several statistical investigations were conducted on the calculated molecular properties in order to assess their validity.

The VIF was evaluated to define the extent of correlation between each descriptor. Generally, a VIF equal to 1 ≥ 5 signifies the non-existence of inter-correlations present in each of descriptor. However, VIF ≥ 10 implies that the developed model is unsteady [25]. The VIF for each descriptor in the built model, which was found to be less than 5, as reported in Table 3, affirms that the descriptors were meaningfully orthogonal to each order because there was no inter-correlation between them. Therefore, the alternative hypothesis is accepted. This implies a direct connection between the biological activity of each compound and the descriptors that influence the built model.

The molecular properties are highlighted in Table 4 with correlation coefficients of < ± 0.6 correlation coefficient between them, which indicates that all properties were annulled of multicollinearity.

Table 5 shows the validity results of the internal assessment to guarantee that the model is reliable. The results confirmed the stability and robustness of the model as valid because the calculated parameters were all in full agreement with general validation criteria (Table 5).

Figures 2 and 3 show the graphs of calculated activity versus observed activity of a training set and a graph of calculated activity versus observed activity of a test set, respectively. It can be observed that the values of the test sets are in close agreement with the training set values.

William’s graph shows the LAD, as shown in Fig. 4. The leverage values of compounds 10 and 15 were observed to be higher than h* = 0.75 (is, warning leverage). Thus, it can be inferred that compounds 10 and 15 are influential molecules. Moreover, it was also observed that all the compounds were within the defined space of 2.5, which indicates that no compound is said to be an outlier.

4.4 Molecular docking studies

The results of the binding energy of the selected benzimidazolinone derivatives (compounds 13 and 21) and the reference drug to a vital AChE implicated in the pathogenesis of AD from the molecular docking study are, respectively, shown in Figs. 5, 6, and 7, with the values for compounds 13 and 21 having a higher binding affinity for AChE than the approved drug. In order to study the binding mode and selectivity of most active compounds 34 and 38 with AChE (PDB code: 4MOE), Compound 13 was docked into the active site of the AChE domain protein with a splendid binding score of − 11.2 kcal/mol involving the following interactions: conventional hydrogen bonding with Val281, Phe75, Met164, and carbon hydrogen bond–bonding, Asp76, Pro56, and Leu547Ala320, Ile77, and π-Alkyl interactions Trp168, Cys583, Pro594, His548 interactions are shown with their 3D interactions and surface interactions in Fig. 5. Compound 21 was docked into the active site of the AChE domain protein with an excellent binding score of 10.9 kcal/mol involving the following interactions: conventional hydrogen bonding with Asp211, Pro248, Lys411, and carbon hydrogen bonds Arg245, Pro431, Asp430, and π-σ bonding Arg245 and π-Alkyl interactions Arg245, Pro427, interactions shown with its 3D interactions and surface interactions are shown in Fig. 6. The reference compound docked into the active site of the AChE domain protein with an excellent binding score of 9.4 kcal/mol, involving the following interactions: Van der Waals bond Ser216, Carbon Hydrogen bond Thr208, Arg245, Glm207, Pi Cation interaction Lys411,—interaction Thr244, Amide-bonding Glu215 and π-Alkyl interactions are shown with their 3D interactions and surface interactions in Fig. 7.

4.5 Drug change assessment of selected compounds

To supplement the findings of the 3D-QSAR and docking studies, we conducted in silico ADMET analyses of the two molecules listed in Tables 9 and 10. The ability to reach targets in bioactive form was assessed using the http://swissadme.ch and http://biosig.unimelb.edu.au/pkcsm/web platforms. The technologies of these webs employ a reasonable degree of certainty, as false-positive results are common in biochemical assays for small molecules [26].

Table 9 indicates that the absorption characteristics of compounds 13 and 21 are capable of oral availability owing to optimal cell permeability. Human intestinal absorption and skin permeability were > 0.8, > 80% and < −2.5, respectively, are shown in Table 9.

The volume of distribution of the studied compounds was > 0.45 which indicates that the drug is distributed in the plasma and describes the extent of drug distribution. Together with the unbound fraction, which labels the portion of free drug in plasma that may extravasate, these are two of the most significant pharmacokinetic medication parameters. These two parameters have an adequate plasma distribution profile, with a fraction of the unbound drug between 0 ≥ 0.157. These values show that the molecules can circulate well and present an important unbound fraction in the plasma, thus becoming available to interact with the pharmacological target. The two compounds can penetrate the central nervous system owing to the volume of distribution and fraction unbound as indicated in Table 9.

As shown in Table 9, cytochrome P450 (CYP) molecules are indispensable information sources, this superfamily of isoenzymes is key players [27]. The synergy between CYP and P-gp can process small molecules to improve the protection of tissues and organisms [28]. Estimation of therapeutic molecules that are major isoform substrates (CYP2D6, CYP3A4, and CYP1A2) [29, 30]. Therefore, compounds 13 and 21 were brilliant CYP substrates.

Finally, from Table 9, the expected values of the total clearance, which measures the efficiency of the body in eliminating a drug, indicate that the two compounds have a noble renal elimination and are not substrates of the renal organic cation transporter 2 (OCT2). In conclusion, the compounds passed the AMES and Minnow toxicity tests, and did not present any particular toxicity problems. The overall analysis of Table 9 highlights that compounds 13 and 21 could be outstanding candidates as drugs, or could lead to further studies and manipulations.

Table 10 displays other drug-likeness rules, such as Ghose, Veber, Egan, and Muegge violations, and all are satisfied by the two molecules that will provide the lead like rule with high affinity in high-throughput screens that allow for the discovery and exploitation of additional interactions in the lead-optimization phase [31,32,33,34]. Furthermore, the PAINS model, which was designed to exclude small molecules that are likely to produce false positives in biological assays, paid no attention to compounds 13 and 21.

Table 11 lists the physicochemical properties of compounds 13 and 21. The compounds passed Lipinski’s rule of five [35]. This further demonstrates the druggability of these compounds.

The predicted bioactivity scores of the selected compounds obtained using the Molinspiration software v2018.03 Chemoinformatics tools are given in Table 12. For the bioactivity score, the G protein coupled receptor (GPCR) ligand was active for both compounds, with value of 0.32 and 0.17. In [36], modulators of ion channels permit charged particles across cell membranes, and they are an important receptor in the healing tract. These two compounds were found to be active. All the compounds were moderately active as kinase inhibitors. In a previous study [37]. Nuclear receptors play a combinatorial role in inflammation and immunity. The two compounds were active, but compound 13 was more active in terms of protease inhibitor bioactivity scores. The two compounds were moderately active. The values for enzyme inhibition showed that all compounds were highly active. The activity score profile of the compounds showed that they were biologically active and have physiological influence.

Figure 8 illustrates the oral bioavailability graph of the two compounds on the basis of the six features discussed in physicochemical properties. The compounds have shown results within these limits, and these two compounds have good physio-chemical profile, a necessary parameter in drugs or clinical trials. The six physicochemical properties are lipophilicity, size, polarity, solubility, flexibility, and saturation. Descriptors were used on each axis to define physicochemical range [38, 39]. The pink region was the drug-like consideration of the molecule in the radar graph. The compounds obey Lipinski’s rule of five. Therefore, the two compounds displayed values within the interval known for medicine.

Figure 9 shows points located in the BOILED-Egg’s yolk that signified the molecules predicted to passively permeate through the blood–brain barrier, whereas the egg white was relative to the molecules predicted to be passively absorbed by the gastrointestinal tract. Blue dots indicate the molecules which emanate from the central nervous system with the aid of P-glycoprotein. Overall, the plot showed that compounds 13 and 21 have excellent bioavailability.

5 Conclusions

A well-validated and robust QSAR model was constructed for compounds 13 and 21 with anti-Alzheimer activity against the AChE protein receptor. The computational studies, QSAR, docking, and ADMET studies proved that these compounds have high binding affinity towards the targeted AChE protein receptor and revealed the binding energies of these compounds, which showed that there is plausibility of inhibiting the AChE protein. Overall, the present study acts as evidence to prove that these compounds from 1.3-dimethylbenzimidazolinone derivatives have the capacity to inhibit the AChE protein receptor, which also paves the way for the two compounds to be screened for AChE protein. In other words, this study with the QSAR, docking, and drug-like properties in primarily pharmacokinetic studies will be positioned as a foundation for carrying out in vitro and in vivo studies in the future.