ABSTRACT: A series of novel phosphoguanidine and phosphopyrazine derivatives were synthesized and characterized by IR, 1H, 13C and 31P NMR spectroscopy, one of which ((C6H10)2P(O)NH(C4H3N2)) was crystallized and investigated by X-Ray crystallography. The activity and the mixed–type mechanism of aforementioned compounds were evaluated using the Ellman and Lineweaver–Burk methods on acetylcholinesterase (AChE) enzyme. Quantitative structure-activity relationship (QSAR) models including stepwise multiple linear regression (SW–MLR) and artificial neural networks (ANN) together with Docking were applied to study the factors affecting the inhibition of acetylcholinesterase activity. Docking Analysis showed an efficient AChE inhibition capacity due to multiple binding patterns in peripheral anionic site (PAS) and anionic-subsite (AS) of AChE through several favorable interactions including hydrophobic, hydrogen bonding and the π-π stacking. QSAR results demonstrated that the electronic chemical potential (µ) has major effect on inhibitory behavior of compounds. DFT calculations to prove Docking and QSAR results which confirm clear trends of electronic properties.
KEYWORDS: Acetylcholinesterase inhibitors; Phosphoguanidine; Phosphopyrazine; Docking; QSAR; DFT calculations.
* Corresponding author’s E-mail address: gholi_kh@modares.ac.ir
■ INTRODUCTION
Acephate is applied mono-phosphoramidate insecticide groups, which has been extensionally utilized against agricultural pests.1 The mechanism of this compound is that it affects the acetylcholinesterase enzyme (AChE) of the target. Despite a large number of its benefits, the usage of this product is accompanied with serious repercussions, ranging from mutagenicity through cytotoxicity to cancer growth, on non-target species including humans and animals.2-4 Consequently, researchers seek out to find innovative strategies to diminish the major drawbacks on human and animal organisms.
A large number of natural and synthetic bioactive compounds possess guanidines and pyrazine functionality within their structural motifs.5,6 These physiologically active compounds with wide-spread activity exhibition, including anticancer agents7,8, anti-inflammatory9, antibacterial10,15, histamine H2 receptor antagonists11, glutamate release inhibitors12,13, antimicrobial14, fungicidal15, anti-malarial16 and herbicidal activities.17 Phosphorylation of various drugs increases their biological activity.18 Phosphorus substituted guanidine compounds can represent agricultural and medicinal applications or can be used as a tool in organic synthesis. 19
Recently, we have reported that the phosphoryl groups of the acephate analogus phosphoramidite tend to have a better interaction with active sites of the enzyme than carbonyl groups (Figure 1). The polarization and the acidity of NH are also among the main factors in inhibiting of an AChE enzyme.20 Twenty-Six novel compounds were synthesized and characterized by 1H, 13C, 31P NMR and IR spectroscopy as well as 9 other compounds were selected form literature to study the factors which affect the inhibition of acetylcholinesterase activity. The general formula of all compounds are (R)(R’)P(X)–Y, that X = O or S, Y = HN–C(=NH)N(CH3)CH2CO2H, C4H3N2–NH, (CH3)2C4HN2–NH, Cl2C4HN2–NH, (CH3)ClC4HN2–NH or C7H5N2–NH, C4H3N2–NH, R = (C6H5,OC6H5 or OCH3,OCH2CH3) and R’ = (R, Y). The solid-state structure of (C6H10) 2P (O) NH (C4H3N2) (compound 35) was determined by X-ray crystallography. The inhibition properties of the acephate derivatives were evaluated in vitro on a model of human AChE (hAChE) by means of the modified Ellman's method. Compound 4 is the most prominent one among the other ligands and Lineweaver– Burk plot of this compound reveals a mixed type and reversible mechanism. Molecular docking and QSAR analysis (SW–MLR and SW–MLR–ANN) were used to find the most efficient parameters and enhance understanding of the mechanism of AChE inhibition.
■ MATERIALS AND METHODS
Synthesis. The synthesis pathway of 26 novels phosphoguanidine and phosphopyrazine compounds is represented in Scheme 1. All the synthesized compounds (1–35) were characterized by 1H, 13C, 31P NMR and IR spectroscopy, (see supplementary information).
Enzymatic Experiments. Method of Ellman used to perform human acetylcholinesterase activity measurements. 21 The effects of different concentrations of the inhibitor were also surveyed. The method of measuring AChE activity is described in supplementary information. Furthermore, the IC50s (median inhibitory concentration) was determined using the GraphPad Prism 7 software. (Table 7). Enzymatic activities of AChE were determined at different ATCh (Acetylthiocholine) concentrations, and fixed concentrations of DTNB (5, 5-dithiobis (2-nitrobenzoic acid) under the experimental conditions. The Km (Michaelis constant) and Vmax (maximum reaction rate) values in the presence and absence of inhibitors were estimated from Lineweaver–Burk plots. Inhibitory constant Ki and KI were calculated too (Figure 2). A mixed inhibitor displays an affinity for both the free enzyme (Ki) and the enzyme-substrate complex (KI). Thus, KI > Ki, the inhibitor preferentially binds to the free enzyme and its mechanism was termed mixed competitive–noncompetitive inhibition. More detailed discussion has been excellently described elsewhere. 22
Crystal structure determination. X-ray data of compound 35 were collected at 120 K on a Bruker SMART 1000 CCD area detector with graphite monochromated Mo-K radiation (k =0.71073 Å). The structure was refined with SHELXL-97 by full-matrix least squares on F2. 23 The positions of hydrogen atoms were obtained from the difference Fourier map. Routine Lorentz and polarization corrections were applied and an absorption correction was performed using the SADABS program for these structures. 24 Crystal data and experimental detail analysis of X-ray are given in Table 1. CCDC 1582927 contains the supplementary crystallographic data for the compound.
Molecular docking. The crystal structures of the Protein-ligand complexes were selected from the PDB (1B41)25 and prepared by performing the Preparation Wizard in Schrödinger's Maestro suite.26 Final structure minimization was performed with default settings using the OPLS-3 force field.27-29 All studied ligands were sketched and cleaned using the builder tools option implemented in the Maestro. The ligands were optimized with the Gaussian 03 (B3LYP/6–311+G**) and the LigPrep utility (pH 7.4). The resultant structures of protein and ligand molecules were used for docking in Glide. 30,31 Docking to the macromolecule was carried out using the extra precision (XP) mode34. The XP Glide scoring function, which is a modified version of the Chemscore function, was used to assess the binding of small molecules to the protein. 32 The XP scoring function contains four main components that account for i) the Coulomb energy of the interacting atoms (Ecoul), ii) the van der Waals energy of atoms (EvdW), iii) a collection of terms that favor binding interactions (Ebind), and iv) a collection of terms that hinder binding interactions (Epenalty). 33,34
XP Glide Score = Ecoul + EvdW + Ebind + Epenalty (1)
Ebind = Ehyd_enclosure + Ehb_nn_motif + Ehb_cc_motif + EPI + Ehb_pair + Ephobic_pair (2)
Epenalty = Edesolv + Eligand_strain (3)
In equations (2) and (3), Ehyd_enclosure is the hydrophobic enclosure reward, Ehb_nn_motif is the term for neutral-neutral hydrogen bonds in a hydrophobically enclosed environment, Ehb_cc_motif is the term for special charged–charged hydrogen bonds, EPI is the π-stacking/π-cation reward, Ehb_pair is the Chemscore-like pair hydrogen-bond term, Ephobic_pair is the pair lipophilic term, Edesolv is the water desolvation energy term and Eligand_strain is the strain energy term.34 By applying equation below
ΔGbinding = −RT ln Kbinding (4)
free energy of binding, ΔGbinding, was calculated regarding Kbinding. In this equation, R is the gas constant (1.987 cal K−1 mol−1) and T is the absolute temperature, assumed room temperature (298.15 K).34
Descriptors Calculation for QSAR. Constructing numerical descriptors of a set of molecules to build QSAR models is essential. A descriptor can represent a quantitative property that depends on the structure of the molecule.35 The uncertainty of experimental measurements has no effect on theoretical descriptors, so one can calculate them for not synthesized compounds. A large number of molecular descriptors were calculated via HyperChem, Dragon, and Gaussian 03.36
Some chemical parameters including hydrophobic coefficient (Log P), and molecular polarizability (MP) were calculated using HyperChem software. Different functional groups, topological, geometrical and constitutional descriptors for each molecule were calculated by Dragon software. Gaussian 03 was also employed for calculation of different quantum chemical descriptors including, energies of highest occupied molecular orbital (HOMO), lowest unoccupied molecular orbital (LUMO), the energy difference between LUMO and HOMO (ΔEL−H), dipole moment (DM), charges on atoms (Qi) and total energy (E), electrophilicity (ω), molecular volume (Mv), polarizability (PL, the charge difference between the atoms in functional groups), electronic chemical potential (μ), chemical hardness (η), Chemical softness (S), and chemical power (Cπ) (see Table 4,5).37,38
Chemometric Methods. The correlation between biological activity and structural properties were obtained by using multiple linear regression and artificial neural network, as linear and non-linear regression methods, respectively. A stepwise multiple linear regression (SW–MLR) model was built by using the calculated molecular descriptors. This method has been used for variable selection and model development in biological systems.39 RMSELOO has been applied as fitness function for developing the SW–MLR model. It is common that many MLR models will be resulted using the stepwise multiple regression procedures; so the best one should be selected. It is general to consider four statistical parameters for this purpose. These parameters are the correlation coefficient (R), root mean square error (RMSE) for the training and validation procedures, F statistic, and number of descriptors. A valid MLR model is one that has low RMSE and number of descriptors, high R2 and F values. 39 Consequently, the model should have a high predictive ability. Finaly, the best descriptors selected using SW–MLR have been used as input variables for development of nonlinear ANN model, in this work.
ANN is artificial systems simulating the function of the human brain. Three components create a neural network: the processing elements or nodes, the topology of the connections between the nodes, and the learning rule by which new information is encoded in the network. The ANN was first developed for nonlinear modeling and it cannot be used for the variable selection. However, this method can be used to measure a nonlinear dependency between the independent and dependent variables. Although there are a number of different ANN models, one of the most commonly used types of ANN in QSAR is the three-layered feed-forward network. For generating a network as a regression model some parameters like the numbers of nodes in the hidden layer, the type of input variables, type of transfer functions and the optimum number of training iteration should be determined. The theory of SW–MLR–ANN method has been adequately described elsewhere.40 In this work, SW–MLR method was used for the variable selection and the selected descriptors were used as input for development of (SW–MLR–ANN) model. The network with the less RMSELOO contains the best set of descriptors and optimum number of nodes in hidden layer for the modeling. In order to evaluate the generated models, leave-one-out cross-validation (LOO–CV) and Leave-multiple-out cross-validation (LMO–CV) approaches were used. In LOO–CV, one compound is left in each step as prediction set and the model is developed using the remaining molecules as the training set. Also, a group of six compounds randomly selected to perform authentication method of leave-6-out (L6O) cross-validation. Then the pIC50 of this group was predicted by the models developed by using the remaining observations as the training set. 39