Exploring the antidiabetic potential of compounds isolated from Anacardium occidentale using computational aproach: ligand-based virtual screening

Abstract

Diabetes mellitus is becoming an important public health challenge worldwide and especially in developing nations. About 8.8 percent of the world adult population has been reported to have diabetes. Glutamine-fructose-6-phosphate amidotransferase 1 (GFAT1) catalyses the first committed step in the pathway for biosynthesis of hexosamines in mammals, and its inhibition has been thought to prevent hyperglycaemia. Dipeptidyl peptidase-4 (DPP-4), on the other hand, degrades hormone glucagon-like peptide-1 (GLP-1), an enzyme that plays a major role in the enhancement of glucose-dependent insulin secretion, making these two proteins candidate targets for diabetes. To find potential inhibitors of DPP-4 and GFAT1 from Anacardium occidentale using a computational approach, glide XP (extra precision) docking, Induced Fit Docking (IFD), Binding free energy of the compounds were determined against prepared crystal structure of DPP-4 and GFAT1 using the Maestro molecular interface of Schrödinger suites. The Lipinski's rule of five (RO5) and ADME properties of the compounds were assessed. Predictive models for both protein targets were built using AutoQSAR. This study identified 8 hit compounds. Most of these compounds passed the RO5 and were within the recommended range for defined ADME parameters. In addition, the predicted pIC50 for the hit compounds were promising. The results obtained from the present study can be used to design an antidiabetic drug.

Supplementary Information

The online version contains supplementary material available at 10.1007/s40203-021-00084-z.

Introduction

Diabetes mellitus, also referred to as diabetes, is a group of metabolic disorders characterized by persistent hyperglycaemia. It is associated with long term complications such as retinopathy, neuropathy and angiopathy (Franks et al. 2010). Diabetes is due to either lack of or resistance to insulin action resulting in type 1 diabetes mellitus (T1DM) and type 2 diabetes mellitus (T2DM) respectively (Shoback and Gardner 2011).

Diabetes is globally the 7th leading cause of death. An estimated 463 million people worldwide had diabetes as of 2019 (8.8 percent of the adult population), with type 2 diabetes responsible for around 90 percent of the cases (Williams et al. 2019). In the same year, about 4.2 million deaths were caused by diabetes (Hu 2011). Diabetes mellitus is becoming an important public healthchallenge (Fakhri et al. 2016; Hassanzadeh et al. 2016; Hu 2011), and is a significant contributor to the large disparities in life expectancy in Nigeria and the world at large (Nguyen et al. 2016).

Presently, pharmacological therapies are focused on increasing the level of insulin in type I diabetes patients, improving insulin sensitivity, slowing carbohydrates delivery and absorption or increasing urinary glucose excretion (Jha and Bhadoriya 2018).

The gut hormone glucagon-like peptide-1 (GLP-1) has an antidiabetic role. It regulates glucose metabolism through several mechanisms, including enhancement of glucose-dependent insulin secretion, slowed gastric emptying, and reduction of postprandial glucagon and of dietary intake or calorie consumption (Ahrén 2005; Deacon et al. 2004; Koliaki and Doupis 2011). However, this important hormone is degraded by the enzyme dipeptidyl peptidase-4 (DPP-4) very fast. This makes DPP-4 an efficient approach for the management of diabetes.

DPP-4 is a type II integral membrane protein also referred to as adenosine deaminase complexing protein 2 or CD26 and consists of 766 amino acids in humans, positioned on the long arm of chromosome 2 (2q24.3) (Abbott et al. 1994; Misumi et al. 1992). While the intracellular N-terminal hydrophilic segment of the type II integral membrane protein consists of 6 amino acids, domain 22 amino acid has a hydrophobic membrane (Hong and Doyle 1990). Another part of the enzyme is the extracellular part containing 738 amino acid sequence. This protein is a protease and it has been observed that its catalytic sites found at the end of C-terminal form a small pocket for catalytic activity (Ludwig et al. 2003; Ogata et al. 1992).

The hexosamine pathway has been identified as one of the biochemical pathways that play an important role in hyperglycaemia and fat-induced insulin resistance (Hawkins et al. 1997; Marshall et al. 1991). In patients with diabetes, an elevated level of blood sugar is mainly metabolized to glucose-6 phosphate, fructose-6 phosphate, and subsequently to the rest of the glycolytic pathway end products. In addition to this, fructose-6-phosphate is also catalysed by the rate-limiting enzyme glutamine:fructose-6 phosphate amidotransferase (GFAT) into glucosamine-6-phosphate through the hexosamine pathway, which produces UDP-N-acetylglucosamine (UDP-GlcNAc), the end product of hexosamine biosynthesis pathway, a substrate for O-linked glycosylation of serine and threonine residues on a variety of cellular proteins, catalysed by O-GlcNActransferase. These series of events lead to induced transcription of Transforming growth factor alpha, Transforming growth factor beta 1 (Kolm-Litty et al. 1998), and Plasminogen activator inhibitor-1 (Du et al. 2000). Since Glutamine-fructose-6-phosphate amidotransferase (GFAT) catalyses the first committed step in the pathway for biosynthesis of hexosamines in mammals, its inhibition has been thought to prevent hyperglycaemia induced elevation in the transcription of TGF-α, TGF-β1 (Kolm-Litty et al. 1998), and PAI-1 (Du et al. 2000), leading to survival from several diabetic complications. Presently, three human isoforms of GFAT have been identified namely GFAT1, GFAT2, and GFAT1L (McKnight et al. 1992; Niimi et al. 2001; Oki et al. 1999). GFAT1 isoform, however, is the ideal target for regulation of obesity and diabetic conditions since it’s predominantly expressed in liver and fat.

Anacardium occidentale is a tropical plant found in different parts of the world principally in Northeastern Brazil (Bitu et al. 2015). The fruit of this plant, typically known as cashew, is used industrially to produce juice. Phytochemical constituents such as flavonoids and carotenoids represent an important content which can be related to significant biological activities which have been used as alternative medicines for primary health care (Kubo et al. 2011; Kulis et al. 2012).While anaesthetic, bactericidal, and insecticidal properties of phytochemicals derived from Anacardium occidentale have been reported by different authors (Castillo-Juárez et al. 2007; Olajide et al. 2004, 2013), its antidiabetic property has also been well established (Jaiswal et al. 2017; Olatunji et al. 2005; Palheta and Ferrira 2018; Ukwenya et al. 2012). Drugs are essential for the prevention and treatment of disease. Human life is constantly threatened by many diseases such as cancer and diabetes. Therefore, ideal drugs are always in great demand (Fuller et al. 2009). The processes involved in drug discovery are strenuous and challenging due to time consumption and the running cost (Mandal et al. 2009). In silico study in medicine is thought to increase the speed of drug discovery while reducing the need for expensive laboratory work and clinical trials. The success rate of potential inhibitors of drug target in in silico experiment has been proven to be promising in in vitro studies (Allegra et al. 2015; Guedes et al. 2014; Röhrig et al. 2010). Presently, Computer aided drug design (CADD) has been explored to develop small molecules as inhibitors/activators of different proteins implicated in the pathogenesis of diabetes mellitus (Elekofehinti et al. 2018; Kikiowo et al. 2020), and so far, close to none has been subjected to clinical trials due to a number of issues, among them is unacceptable efficacy of the molecules against protein target. In this study, phytoconstituents isolated from Anacardium occidentale as identified in different literatures were investigated for their antidiabetic properties against two of the prime targets implicated in diabetes. Some studies have shown that compounds from Anacardium occidentale such as kaempferol and quercetin posses hypoglycaemic properties in vivo and in vitro (Zhu et al. 2007). Hence the main motive for investigating Anacardium occidentale phytoconstituents therapeutic potential against GFAT and DPP-4 using an integrative computational approach.

Methodology

Preparation of ligand and protein target structure

The three-dimensional crystal structure of the GFAT1 (PDB ID: 2V4M) and DPP-4 (PDB ID: 3D4L) were downloaded from Protein Data Bank (Berman et al. 2000), and further prepared by protein preparation wizards which is available in Glide(2018c) as previously described by Iwaloye et al. (2020a). The protein preparation wizards panel have two components namely, pre-process and refinement tab. In the pre-process tab, create zero bonds to metals and create disulfide bonds options were selected, bond orders were assigned, waters were deleted from 5.0 Å of het groups, het states was set at pH 7.0 ± 2.0, and hydrogen was added. In the next step, water molecules were removed and hydrogen atoms were added to the structure. Minimization was performed until the average root mean square deviation of the non-hydrogen atoms reached 0.3 A˚. Consequently, a total of 71 phytoconstituents of Anacardium occidentale identified in literatures (Salehi et al. 2019; Bhakyaraj Singaravadive, 2012) were uploaded on the workspace of Maestro (2018a) and prepared for docking using Ligprep panel (2017). Low-energy 3D structures with correct chiralities were generated. The possible ionization states for each ligand structure were generated at a physiological pH of 7.2 ± 0.2. While the ligands were minimized using an OPLS3 force field (Harder et al. 2016), other options were set to their default.

Molecular docking

Glide tool in maestro v11.8 (2018a) was employed to perform the molecular docking of the prepared ligands into the active site of the crystal protein structures. Glide utilizes a hierarchical series of filters to search for possible locations of the ligand in the active-site region of the receptor, while the shape and properties of the receptor are represented on a grid by several sets of fields that progressively provide more accurate scoring of the ligand poses. An extensive conformational search is performed using a heuristic screen that rapidly eliminates unsuitable conformations (e.g. long-range internal H-bonds or sterical clashes). The entire amount of poses generated is then hierarchically classified, refined and further minimized into the active site grid before being finally scored using the proprietary GlideScore function, defined as in Eq. (1).

$$\begin{array}{cc}\hfill \mathrm{G}\mathrm{Score}& =0.065\times \mathrm{vdW}+0.130\times \mathrm{Coul}+\mathrm{Lipo}\hfill \\ \hfill & +\mathrm{Hbond}+\mathrm{Metal}+\mathrm{BurryP}+\mathrm{RotB}+\mathrm{Site}\hfill \end{array}$$ 1

The ligand docking panel used the grid file generated by picking the co-crystallized ligand at the active site of the protein which defines the area around the active site in terms of coordinates x, y and z. Extra precision (XP) was assigned as the scoring function. The XP scoring functions integrate the basic GlideScore functions with additional terms: to model solvation, an explicit water term was used, parameterized on several known protein–ligand complexes; similarly, additional parameterized rewards were calculated for buried hydrophobic pockets occupied by hydrophobic ligand groups.

Validation of molecular docking results

The docking protocol was further validated by docking native ligand (co-crystal ligand) with the prepared crystal structure of both GFAT1 and DPP-4 to validate the docking efficiency by determining the root mean square deviation (RMSD) were determined. An RMSD value of 1.62 Å and 0.935 Å for GFAT1 and DPP-4 (Fig. 1a, b), respectively showed the docking procedure is reproducible (Iwaloye et al. 2020b).

Fig. 1.

Docking pose of co-crystallized ligands with its native form showed a RSMD value of 0.935 Å for DPP-4, and b RMSD value of 1.62 Å for GFAT1

Induced fit docking (flexible simulation)

Induced Fit Docking (IFD) is an in silico based method that uses Glide and the Refinement module in Prime to accurately predict ligand binding modes and concomitant structural changes in the receptor (Sherman et al. 2006). The conventional molecular docking procedure allows the binding of the flexible ligands into the catalytic site of a rigid receptor while in reality, a receptor freely rotates on binding to a ligand. IFD protocol has made the docking process more realistic than the conventional docking.

Free energy calculation using prime

Prime Molecular Mechanics/Generalized Born Surface Area (MM-GBSA), also known as a post-scoring approach, is a key and appealing computational method used to analyse the accuracy of docking processes. The docked poses with the lowest Glide score (best docking score) were minimized using MM-GBSA approach as implemented in the Prime module of the Schrödinger suite with default settings (Jacobson et al. 2002, 2004). The final output obtained from the above process is ΔG_bind, a mixture of different kinds of energies in which polar and nonpolar energies contribute separately. ΔG_bind was computed as:

$$\mathrm{\Delta }{\text{G}}_{\text{bind}}=\mathrm{\Delta }{\text{E}}_{\text{MM}}+\mathrm{\Delta }{\text{G}}_{\text{Solv}}+\mathrm{\Delta }{\text{G}}_{\text{SA}}$$

where ΔE_MM is the difference in energy between the complex structure and the sum of the energies of the ligand and unliganded protein using the OPLS3 force field. ΔG_solv isthe difference in the GBSA solvation energy of the complex and the sum of the solvation energies for the ligand and unliganded protein. Lastly, ΔG_SA is the difference in the surfacearea energy for the complex and the sum of the surface area energies for the ligand and uncomplexed protein.

ADME predictions

Absorption, Distribution, Metabolism, and Excretion analysis constitutes the pharmacokinetics and physicochemical properties of a drug molecule (Nisha et al. 2016). This study used Qikprop(2018b) to predict significant descriptors of drug-likeness such as Lipinski rule of 5 (RO5), binding to human serum albumin (QPlogKhsa), IC₅₀ value for the blockage of HERG K + channels (QPlogHERG) and Human Oral Absorption.

Automated QSAR

AutoQSAR is a machine-learning algorithm provided by Schrödinger suite that builds and applies QSAR models through automation (Dixon et al. 2016). In order to build a predictive model, AutoQSAR takes the 1D, 2D and 3D structural data of a molecule along with a property (e.g. IC₅₀) to be modeled, as an input. It will then compute the fingerprints and descriptors using machine-learning statistical methods for creating a predictive QSAR model. The experimental dataset containing 39 DPP-4 inhibitors with their respective IC₅₀ was retrieved from CHEMBL database online server (www.ebi.ac.uk/chembl/), by blasting the FASTA sequence of the DPP-4. In the same vein, 19 known inhibitors of GFAT1 with their respective IC₅₀ were retrieved from different literature reviews (Bobzin et al. 2000; Bolin et al. 2006; Floquet et al. 2007; Le Camus et al. 1998a, b; Qian et al. 2011). The compounds were uploaded into the workspace of Maestro and prepared by Ligprep. QSAR models were built using the structure of inhibitors with their respective IC₅₀.

Results and discussion

Molecular docking analysis

The primary aim of docking study is to obtain accurate predictions of the ligand conformation and orientation within a targeted binding site (Guedes et al. 2014). Prepared compounds were refined using a two-step hierarchical process, which glide SP and XP are docking. The use of these hierarchical filters and associated parameters was earlier discussed in different studies (James and Ramanathan 2018; Rohini and Shanthi 2018; iwaloye et al. 2020c). The co-crystallized ligand of each protein target was used for the basis of comparison. Figure 2 showed the chemical structure of hit compounds against GFAT1 and DPP-4. These compounds were proved to be more active than other phyto-compounds derived from Anacardium occidentale. Docking score and Induced fit score of the hits compounds against GFAT1and DPP-4 are shown in Table S2 and S3. Kaemferol 3-O-Beta-d-Xyloside and myricetin recorded the highest binding affinity with GFAT1 with docking score of − 7.521 kcal/mol and − 7.440 kcal/mol, respectively. The results of glide XP docking score of hit compound with DPP-4, however, showed that delphinidin and myricetin recorded the highest affinity (− 9.332 kcal/mol and − 8.906 kcal/mol). Consequently, we further made use of Schrodinger’s innovative method known as induced fit docking for modeling conformational changes induced by ligand binding (Tables 1, 2, 3).

Fig. 2.

Chemical structure of hit compounds

Table 1.

Induced fit docking and interacting residues of hit molecules and co-crystallized ligand GFAT

S/no.	Pubchem ID	Docking score	Induced fit docking	No of H− bond	Interacting residues	Predicted pIC50	ΔGBind
1	kaemferol 3-O-Beta-d-Xyloside	− 7.521	− 748.09	4	ASP446, GLN440, THR394, GLU579	6.107	− 40.58
2	Myricetin	− 7.440	− 748.41	3	ASP446, THR394, SER395	6.209	− 36.99
3	Quercetin-3-O-beta-d-arabinofuranoside	− 7.140	− 752.41	7	THR444, ASP446, SER441, CYS392, VAL490, SER492	6.044	− 36.56
4	Delphinidin	− 6.907	− 747.80	4	THR394, GLU579, GLU487	6.203	− 26.06
5	Gallic acid	− 6.880	− 746.13	5	GLN440, SER395, SER441, VAL490	6.236	− 31.51
6	Quercetin-3-O-d-galactopyranoside	− 6.628	− 749.46	7	ASP446, SER441, GLN440, SER395, THR394, GLU579	6.044	− 34.67
7	( +)-Catechin	− 6.117	− 747.80	3	SER395, GLU487, VAL490	6.025	− 42.60
8	Protocatechuic acid	− 6.020	− 743.55	5	THR444, SER441, GLN440, SER439, THR394	6.371	− 24.69
9	Epigallocatechin	− 5.982	− 749.75	8	THR394, SER395, GLN440, SER441, GLU579, SER492	5.997	− 45.73
10	Naringenin	− 4.841	− 750.82	4	THR394, CYS392, SER492, GLU487	6.269	− 33.29
11	(−) epicatechin	− 4.788	− 748.11	6	CYS392, SER395, GLN440, GLU487, GLY489	6.025	− 45.72
12	Co-crystallized ligand	− 8.698	− 749.73	8	GLN440, SER395, VAL490, SER492, GLU579	− 5.031	− 41.80

The boldened amino acid residues represent pi-pi iteraction

Table 2.

Induced fit docking and interacting residues of hit molecules and co- crystallized ligand with DPP-4

S/no.	Pubchem ID	Docking score (3D4L)	Induced fit docking	No of H-bond	Interacting residues	Predicted pIC50	ΔGBind
1	kaemferol 3-O-Beta-d-Xyloside	− 8.081	− 1335.01	4	PRO550, GLY549, TYR666, SER209	5.076	− 40.16
2	Myricetin	− 8.906	− 1336.63	6	GLU206, VAL207, TYR547, TYR666, PRO550, PHE357	4.628	− 38.58
3	Quercetin-3-O-beta-d-arabinofuranoside	− 8.857	− 1336.66	6	GLU206, GLU205, PHE357, TYR585, TYR547, TYR662	5.028	− 31.75
4	Delphinidin	− 9.332	− 1335.39	6	GLU205, GLU206, ASN710, ARG125, TYR666, TYR547, TYR666	4.546	− 57.95
5	Gallic acid	− 4.891	− 1322.10	5	TYR670, TYR666, TYR547	4.830	19.05
6	Quercetin-3-O-d-galactopyranoside	− 8.416	− 1335.78	6	GLU206, ARG125, TYR547, PRO550, TYR662, PHE357	5.156	− 42.64
7	( +)-Catechin	− 6.448	− 1333.57	4	GLU205, GLU206, ARG122, TYR666	4.883	− 39.42
8	Protocatechuic acid	− 4.386	− 1321.16	3	PRO550, TYR666, TYR547, TYR547	4.877	19.44
9	Epigallocatechin	− 7.681	− 1335.56	6	GLU206, TYR670, PRO550, SER630, TYR662	4.779	− 16.01
10	Naringenin	− 5.723	− 1334.81	4	GLU206, TYR670, GLN553, TYR585	4.989	− 30.88
11	(−) epicatechin	− 7.204	− 1336.04	3	GLU205, GLU206, HIS740, TYR666	4.883	− 21.22
12	Co-crystallized ligand	− 7.670	− 1335.63	6	GLU206, VAL207, ARG125, TYR662, TYR662, TYR666	− 4.102	− 40.46

Table 3.

Prediction of ADME properties by QIkprop

S/no.	Pubchem ID	RuleOfFive	HOA	QPlogKhsa	QPlogHERG	QPPMDCK	PSA
1	kaemferol 3-O-Beta-d-Xyloside	1	2	− 0.606	− 5.427	4.151	177.65
2	Myricetin	1	2	− 0.488	− 5.008	2.149	164.858
3	Quercetin-3-O-beta-d-arabinofuranoside	2	1	− 0.746	− 5.265	1.171	199.057
4	Delphinidin
5	Gallic acid	0	2	− 0.985	− 1.413	4.294	115.079
6	Quercetin-3-O-d-galactopyranoside	2	1	− 0.895	− 5.375	0.953	217.198
7	( +)-Catechin	0	2	− 0.415	− 4.774	19.439	116.961
8	Protocatechuic acid	0	2	− 0.903	− 1.493	12.821	93.683
9	Epigallocatechin	1	2	− 0.552	− 4.543	7.055	137.168
10	Naringenin	0	3	− 0.031	− 4.975	54.4	100.802
11	(−) epicatechin	0	3	− 0.035	− 4.963	55.288	100.26

^aNumber of violations of Lipinski’s rule of five (Range: maximum is 4)

^bHumanOralAbsorption: Predicted qualitative human oral absorption: 1, 2, or 3 for low, medium, or high

^cQPlogKhsa: Prediction of binding to human serum albumin. Range from –1.5 to – 1.5

^dQPlogHERG: Predicted IC50 value for blockage of HERG K + channels. concern below –5

^eQPPMDCK: Predicted apparent MDCK cell permeability in nm/sec. MDCK cells are considered to be a

good mimic for the blood–brain barrier. QikProp predictions are for non-active transport. < 25 poor, > 500 great

^fPSA Van der Waals surface area of polar nitrogen and oxygen atoms. Range from 7.0 – 200.0

This method offered more accuracy in the binding of a ligand with its protein target because it enables the receptor to change its binding pocket so that it conforms more closely to the ligand shape and binding mode (Venkatachalam et al. 2003). The results of the IFD score revealed that quercetin-3-O-beta-d-arabinofuranoside and naringenin bind better with GFAT1 than other hit compounds. Additionally, myricetin and quercetin-3-O-beta-d-arabinofuranoside showed a more favourable binding affinity with DPP-4 than other compounds, recording an IFD score of − 1336.66 kcal/mol and − 1336.63 kcal/mol respectively. The results of the molecular docking analysis clearly showed the inhibitory potential of the hit compounds against GFAT1 and DPP-4.

Pose orientation of hit compounds in the binding pocket of GFAT1 and DPP-4

It is important to investigate the interaction of proposed inhibitors with crucial amino acids residue located at the active site of the proteins, as this will provide evidence of their inhibitory properties. The Ligand Interaction panel of Maestro was used to visualize the 2D interaction of the compounds with the protein target. GFAT1 active sites are known to be conserved from bacteria to humans. The human GFAT1 isomerase domain consists of two sugar isomerase sub-domain: both are composed of five stranded parallel beta-sheets surrounded by two or three alpha-helices (Ruegenberg et al. 2020). An experiment by Ruegenberget al. demonstrated that residues of both isomerase domains contribute to the binding to inhibitors (2020). Here, the hits showed similarity in binding by interacting with hydrophobic, negatively charged and polar amino acid residues and 2D ligand interaction diagram of hit compounds are shown in Table S1 and Figs. 3, 4, 5.

Fig. 3.

The 2D ligand interaction diagram of kaemferol 3-O-Beta-d-Xyloside with GFAT1

Fig. 4.

The 2D ligand interaction diagram of Epigallocatechin with GFAT1

Fig. 5.

The 2D ligand interaction diagram of Naringenin with GFAT1

Two binding sites are found in the human DPP-4 namely; S1 site and S2 site. S1 site is a long hydrophobic region that allows uncharged amino acids to fit in receptor and it has a catalytic triad Ser-630, Asn-710 and His-740. In contrast, the long lipophilic region called S2 site involves key interaction with Glu-205, Glu-206 dyad and Arg-125 (Chakraborty et al. 2014; Kuhn et al. 2007; Watanabe et al. 2015). The substrate-specific S1 site is made up of highly hydrophobic residues such as Tyr-547, Tyr- 631, Val-656, Trp-659, Tyr-662, Tyr-666, and Val-711 (Chakraborty et al. 2014). In our present study, three compounds (Quercetin-3-O-d-galactopyranoside, Quercetin-3-O-beta-d-arabinofuranoside and Quercetin-3-O-d-galactopyranoside) formed pi-pi stacking with the amino acid residue Phe-357, while ( +)-Catechin, kaemferol 3-O-Beta-d-Xyloside and Delphinidin also showed pi–pi interaction with Tyr-666 (Figs. 6, 7, 8).

Fig. 6.

The 2D ligand interaction diagram of kaemferol 3-O-Beta-d-Xyloside with DPP-4

Fig. 7.

The 2D ligand interaction diagram of Quercetin-3-O-d-galactopyranoside with DPP-4

Fig. 8.

The 2D ligand interaction diagram of Quercetin-3-O-beta-d-arabinofuranoside with DPP-4

These interactions were demonstrated by a similar study (Jha and Bhadoriya 2018). We further observed hydrogen bond interaction of two positively charged Glu-205 and Glu-206 with the hit, which has been reported to be involved in active site construction (Aertgeerts et al. 2004). Additional interactions of the proposed inhibitors with Arg-125, Tyr-662 and Tyr-547have been previously established as key residues responsible for DPP-4 inhibition (Pantaleão et al. 2015).

Binding free energy calculation

The post docking program called MMGBSA which calculates ligand binding energies and ligand strain energies for a set of ligands and a single receptor by employing the combination of molecular mechanics calculations and salvation models was used to validate the binding affinity of respective compounds with the target protein. Since Genheden and Ryde argued that output obtain through MMGBSA for binding energies calculations were found to be highly efficient (Genheden and Ryde 2015), several studies have demonstrated MMGBSA post docking program as the most reliable for rating the affinity of a ligand on binding to its protein target (Maffucci et al. 2018; Sun et al. 2014). Table S2 and S3 showed the evaluation of the free energy of binding for each hit-protein complex. All the compounds except protocatechuic acid have a favourable binding with the proteins. Epigallocatechin and (−) epicatechin, however, recorded the most favourable ΔG_Bind with GFAT1 (− 45.73 kcal/mol and − 45.72 kcal/mol), while quercetin-3-O-d-galactopyranoside in complex with DPP-4 was more favourable than others (− 42.64 kcal/mol). In terms of binding free energy, the major energy contributors were identified as van der Waals (∆Gvdw), Coulomb interaction (∆GColulomb), Hydrogen bond (ΔGHbond) and lipophilic energy (∆GsolLipo) that enhance the binding affinity of the compounds towards the binding pocket of the protein (Table S1 and S2).

Prediction of ADME by Qikprop

Besides the evaluation of binding affinity and binding free energy, other properties including drug-like properties such as the number of hydrogen donor, hydrogen bond acceptor and molecular weight were also determined. These parameters are important molecular properties as formulated by Lipinski et al. (1997). The formulated rule known as Lipinski’s rule of five (RO5) stated that for a compound to be considered as druglike, the drug must not violate more than one of the following attribute: molecular weight < 500 Da, octanol–water partition coefficient < 5, hydrogen bond donor ≤ 5 and hydrogen bond acceptor ≤ 10 (Table S3) (Lipinski et al. 1997). As Gombar et al. reckoned, a prospective drug that obeys the RO5 in a clinical trial has lower attrition rates and has increased chance of passing all phases of the clinical trial. Here it’s interesting to note that all the hit compounds with the exception of quercetin-3-O-beta-d-arabinofuranoside and quercetin-3-O-d-galactopyranoside are considered as druglike. Furthermore, other parameters were further scrutinized and predicted by Qikprop and the results are shown in Fig. 3. Among them are Human Oral Absorption, Prediction of binding to human serum albumin, Predicted apparent MDCK cell permeability in nm/sec and Van der Waals surface area of polar nitrogen and oxygen atoms. Interestingly, all these parameters fell within the recommended range. The compounds’ pharmacokinetics properties were aslso predicted via SWISSADME server (Table 4); (−) epicatechin, Naringenin, Protocatechuic acid, ( +)-Catechin, Gallic acid and Delphinidin had high GI absorption which simplifies that that can be easily absorpbed by gastro intestinal tract. Five most important Iso-enzyme of cytochrom p450 (CYP450) which are responsible for the metabolism of drugs/metabolites were also taken into consideration. They are CYP1A2, CYP2C19, CYP2C9, CYP2D6 and CYP3A4. All the lead compounds with the exception of Myricetin and Naringenin showed moderation against these iso-enzymes (Table 5).

Table 4.

Prediction of compounds’ pharmacokinetics by SWISSADME server

S/no.	Entry name	MW	GI absorption	Rotatable bonds	CYP1A2 inhibitor	CYP2C19 inhibitor	CYP2C9 inhibitor	CYP2D6 inhibitor	CYP3A4 inhibitor
1	Kaemferol 3-O-Beta-d-Xyloside	418.35	Low	3	No	No	No	No	No
2	Myricetin	418.24	Low	1	Yes	No	No	No	Yes
3	Quercetin-3-O-beta-d-arabinofuranoside	366.36	low	4	No	No	No	No	No
4	Delphinidin	303.24	High	1	Yes	No	No	No	No
5	Gallic acid	170.2	High	1	No	No	No	No	Yes
6	Quercetin-3-O-d-galactopyranoside	463.38	Low	4	No	No	No	No	No
7	( +)-Catechin	290.27	High	1	No	No	No	No	No
8	Protocatechuic acid	306.27	High	1	No	No	No	No	No
9	Epigallocatechin	458.37	low		No	No	No	No	No
10	Naringenin	272.25	High	1	Yes	No	No	No	Yes
11	(−) epicatechin	290.27	High	1	No	No	No	No	No

Table 5.

Best model generated for GFAT1 and DPP-4

Best model generated for GFAT1
S/n	Model code	R2	RMSE	Q2
1	Kpls_desc_47	0.8208	0.1887	0.8658

Best model generated for DPP-4
S/n	Model code	R2	RMSE	Q2
1	Kpls_linear_15	0.7771	0.5799	0.7303

QSAR analysis

The purpose of the AutoQSAR is to automatically produce robust QSAR models with minimal user input or understanding. This program splits the curated dataset randomly into the desired percentage of the training set and test set. The models are built on each training set from all possible combination of machine learning method, and sets of independent variables that are supported by each machine learning methods, and it could produce up to 400 models where each mode assigns a quality model score to each model based on its performance on the training and test set (An et al. 2013).Models are scored based on the equation below.

Score M = accuracy test [1.0 – (accuracy train – accuracy test)] (Li et al. 2015).

Where: Accuracy is defined between intervals 0–1.

A perfect prediction is represented by 1 and incorrect prediction as 0.00. The equation accepts models that exhibit high accuracy with respect to the test and train and penalize inaccurate models (Sastry et al. 2010).

As previously described, the autoQSAR program was used to build a different model for GFAT1 and DPP-4. The observed activities and predicted activities of the training set and test set in the negative logarithm of inhibitor concentration(pIC₅₀) for both GFAT1 and DPP-4 was represented in Tables S4 and S5. The top model represented in Table S4 was selected, and itwas employed in this study. For GFAT1, the top model Kpls_desc_47 recorded R² of 0.8208, RMSE of 0.1887 and Q² of 0.8658. Consequently, the top model for DPP-4 (Kpls_linear_15) recorded R² of 0.7771, RMSE of 0.5799 and Q² of 0.7303. The scatter plot depicting predicted pIC₅₀ versus experimental pIC₅₀ for the best-generated model is shown in Figs. 9 and 10. It is worth noting that the hit compounds are observed to have promising predicted pIC₅₀. kaemferol 3-O-Beta-d-Xyloside, Quercetin-3-O-beta-d-arabinofuranoside and Quercetin-3-O-d-galactopyranoside had predicted pIC50 of 5.076, 5.028 and 5.156 for DPP-4, and predicted pIC50 of 6.107, 6.044 and 6.044 for GFAT1.

Fig. 9.

Scatter plot analysis of best model predicted from AutoQSAR for known inhibitors of DPP-4

Fig. 10.

Scatter plot analysis of best model predicted from AutoQSAR for known inhibitors of GFAT1

Conclusion

Here we identified hit compounds from Anacardium occidentale as potential inhibitors of two prime therapeutic targets (GFAT1 and DPP-4) in the management of diabetes. Although molecular docking, binding free energy, QSAR and ADME properties revealed gallic acid, naringenin and kaemferol 3-O-Beta-d-Xyloside as promising compounds, these results are not sufficient to affirm their therapeutic potential against GFAT1 and DPP-4, further experiments are needed to attest to the inhibitory potentials of these compounds.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

VOU: conceptualization, experiments design, data analysis, data presentation, writing and reviewing. SAA: data analysis, data presentation, writing and reviewing. OOE: Experiments design, data analysis, data presentation, writing and reviewing.

Declarations

Conflict of interest

Authors declare no conflict of interest.

Ethics approval

Not Applicable