Structure-based virtual screening of mangiferin derivatives with antidiabetic action: a molecular docking and dynamics study and MPO-based drug-likeness approach

Abstract

Extracts from Mangifera indica leaves and its main component, mangiferin, have proven antidiabetic activity. In this study, mangiferin and its natural derivatives Homomangiferin (HMF), Isomangiferin (IMF), Neomangiferin (NMF), Glucomangiferin (GMF), Mangiferin 6’-gallate (MFG), and Norathyriol (NRT) were compared regarding their action on Diabetes mellitus (DM), employing docking and molecular dynamics (MD) simulations to analyze interactions with the aldose reductase enzyme, the precursor to the conversion of glucose into sorbitol. Notably, HMF showed significant affinity to residues in the active site of the enzyme, including Trp 79, His 110, Trp 111, Phe 122, and Phe 300, with an energy of − 7.2 kcal/mol, observed in the molecular docking simulations. MD reinforced the formation of stable complexes for HMF and MFG with the aldose reductase, with interaction potential energies (IPE) in the order of − 300.812 ± 52 kJ/mol and − 304.812 ± 52 kJ/mol, respectively. The drug-likeness assessment, by multiparameter optimization (MPO), highlighted that HMF and IMF have similarities with polyphenols and glycosidic flavonoids recently patented as antidiabetics, revealing that high polarity (TPSA > 180 Ų) is a favorable property for subcutaneous administration, especially because of the gradual passive cell permeability values in biological tissues, with P_app values estimated at < 10 × 10⁻⁶ cm/s. These compounds are metabolically stable against metabolic enzymes, resulting in a low toxic incidence by metabolic activation, corroborating with a lethal dose (LD₅₀) greater than 2000 mg/kg. In this way, HMF showed a systematic alignment between predicted pharmacokinetics and pharmacodynamics, characterizing it as the most favorable substance for inhibiting aldose reductase.

Supplementary Information

The online version contains supplementary material available at 10.1007/s13205-024-03978-9.

Introduction

Diabetes mellitus (DM) is a chronic disease that poses a significant global health concern due to its high prevalence. It results in hyperglycemia, which is characterized by high blood glucose levels. In 2019, diabetes affected 460 million people worldwide, making it the eighth leading cause of death and disability. By 2021, the number of people affected by diabetes had risen to 529 million. It is predicted by researchers that the number of registered cases could reach approximately 629 million by 2045 and increase by 59.7% by 2050, potentially affecting up to 1.31 billion people (Ong et al. 2023).

According to the International Diabetes Federation (IDF) report of 2021, diabetes was more prevalent in men than women (6.5% vs 5.8%), with a ratio of 1.14 between the sexes. Approximately 485 million people worldwide, nearly half of adults aged 20–79 with diabetes, were unaware of their diagnosis. Type II diabetes affects around 508 million people, accounting for 96.0% of all cases of the disease (International Diabetes Federation 2017; Chaudhury et al. 2017; Ogurtsova et al. 2022; Ong et al. 2023).

Unawareness of DM can result in various issues, worsening the disease and causing heart, nerve, kidney, and eye problems. Additionally, it can lead to emotional stress and anxiety for both the person with diabetes and their family. Late diagnosis of diabetes can even result in mortality (Castro et al. 2021). DM poses a significant burden on healthcare systems, with global expenditures surpassing US$966 billion in 2021 and potentially exceeding US$1,054 billion in 2045 (Ong et al. 2023).

Aldose reductase, encoded by the AKR1B1 gene, is an enzyme that reduces glucose to sorbitol. Sorbitol dehydrogenase then transforms it into fructose using NAD + . This reaction plays a fundamental role in the build-up of sorbitol in organic tissue cells, contributing to hyperglycemia. Inhibiting aldose reductase activity with bioactive substances is a potential treatment for diabetes (Taslimi et al. 2018; Demir et al. 2018).

Under conditions of high blood sugar, glucose molecules enter the polyol pathway and use NADPH as a cofactor. This results in a significant decrease in NADPH levels. This condition is concentrated in the pancreas of diabetic patients, leading to increased intracellular oxidative stress and irreversible pancreatic damage. Aldose reductase inhibition plays a crucial role in reducing NADPH consumption, which in turn reduces the pathways for the formation of GSH-based metabolites that use NADPH as a cofactor. This inhibition also reduces the concentration of H₂O₂ free radicals, protecting cells from oxidative stress (Thakur et al. 2021). Therefore, it constitutes a fundamental pathway in DM therapy.

Mangifera indica leaves and bark have been traditionally used in many cultures to treat diabetes through infusions and decoctions (Ediriweera et al. 2017; Senthilkumar et al. 2020). Studies conducted in vitro and in vivo confirm their effectiveness in reducing blood sugar levels (Saleem et al. 2019; Villas Boas et al. 2020; Mistry et al. 2023), making natural products a useful therapeutic strategy for treating DM. Both the leaves and mangiferin, its primary biomolecule, possess antidiabetic properties. These properties are observed through the inhibition of α-amylase and α-glucosidase, as demonstrated in studies by Kulkarni and Rathod (2018) and Ngo et al. (2019). The bark and leaves of Mango (Mangifera indica L.) contain chemical constituents that exhibit various therapeutic activities, including antioxidant (Samadarsi and Dutta 2020), cardioprotective (Jiang et al. 2020), and antidiabetic (Gu et al. 2019) properties.

Among the substances present in M. indica, phenolic compounds stand out, mainly xanthones, with the C-glycosylxanthone known as mangiferin (MF) as the majority, in addition to its derivatives n-homomangiferin (HMF), isomangiferin (IMF), neomangiferin (NMF), glucomangiferin (GMF), mangiferin 6'-gallate (MFG) and a glycosylated derivative of noratiriol (NRT) that coexist in smaller quantities (Imran et al. 2017; Wu et al. 2010). These xanthones have an aromatic chromone chemical structure linked to catechol and a glycoside substructure. They interact with various pharmacological targets (Gu et al. 2019; Dutta et al. 2023). Recent patent registrations indicate that NRT analogs retain their pharmacokinetic characteristics, such as metabolic stability and synthetic accessibility. Furthermore, glycosylated phenolic compounds, such as MF, have shown significant potential in inhibiting α-glucosidase and aldose reductase in the hyperglycemia evolution cycle across various rodent species (Quadri et al. 2019).

Bioactive compounds possess a wealth of biological activities. However, an increasing number of tests are carried out with molecules that have inappropriate chemical properties. This requires new strategic approaches to optimize the selection process of new medicines and promote the reduction of unfavorable tests on laboratory animals (Cerqueira et al. 2015). Structure-based virtual screening is a bioinformatics approach that estimates the biochemical effect of bioactive molecules in natural products. Park et al. (2008) and Zhao et al. (2023) have used it to direct molecular docking and dynamics simulation studies, in vitro assays with cell lines expressing selected enzymes, and in vivo efficacy.

The aim of this study is to evaluate the inhibitory potential of natural compounds derived from mangiferin on the enzyme aldose reductase, which is responsible for converting glucose into sorbitol. The study uses molecular docking techniques and molecular dynamics (MD) simulations to analyze the interactions between the compounds and the enzyme's active site. The study also examines how structural modifications affect the pharmacokinetics of these compounds and, additionally, seeks to compare the potential efficacy of various natural derivatives of mangiferin in the treatment of DM.

Materials and methods

Structure-based virtual screening

The two-dimensional representation of the chemical structures of NRT (PubChem CID 5281656), HMF (PubChem CID 5491388), MF (PubChem CID 5281647), IMF (PubChem CID 5318597), NMF (PubChem CID 6918448), GMF (PubChem CID 101090205), and MFG (PubChem CID 131752602) shown in Fig. 1 was obtained from the PubChem repository (https://pubchem.ncbi.nlm.nih.gov/), where the linear notations of SMILES were submitted to the SwissTargetPrediction server—from Swiss Institute of Bioinformatics (SIB) (http://www.swissadme.ch/)—configured to perform a 3D/2D structural similarity test with more than 300,000 compounds deposited in the ChEMBL database with characterized activity in the organism Rattus norvegicus, to estimate the therapeutic pharmacodynamic pathway, expressed in classes of biological targets and specific biological receptors (Daina et al. 2019).

Fig. 1.

Two-dimensional representation of the chemical structures of Norathyriol (NRT), Homomangiferin (HMF), Mangiferin (MF), Isomangiferin (IMF), Neomangiferin (NMF), Glucomangiferin (GMF), and Mangiferin 6’-gallate (MFG)

The support vector machine model trains on known compounds with known protein interactions from the ChEMBL database and the input compound. Structural similarity is estimated using the Manhattan distance (d) (Eq. 1).

$$d=\sum _{s=1}^N\left|x_s,-,y_s\right|$$ 1

This equation compares each pair of vectors x and y in a two-dimensional space (bond between each pair of atoms) between two different molecules, considering the 20 most stable conformations (s) of each molecule (Gfeller et al. 2014). The SwissTargetPrediction server database, provided by SIB, contains a primary dataset from ChEMBL. This dataset considers ligands with up to 80 heavy atoms that have the potential to form a ligand–protein complex with measured activity values, such as inhibition constant (K_i) and inhibitory concentration (IC₅₀), of less than 10 μM in all assays. The database contains a robust base of more than 300,000 characterized compounds for three commonly tested organisms: Homo sapiens, Mus musculus, and Rattus norvegicus (Daina et al. 2019; Gfeller et al. 2014). For this study, we chose to focus on Rattus norvegicus due to its genetic similarity to Homo sapiens and the reliability of the databases (Kleandrova et al. 2015; Muñoz-Antoli et al. 2014).

Ligand preparation for docking simulations

To carry out the molecular docking simulations, the compounds underwent electronic and structural optimization using the force field parameters of the classical mechanics Merk Molecular Force Field (MMFF94). This structural optimization made the compounds more stable for the start of the simulations (Halgren 1996). The optimization was performed using the Avogadro^® software (Hanwell et al. 2012), configured for cycles of 1000 interactions of the steepest descent algorithm. Pre-hydrogenated structures and intramolecular hydrogen interactions were considered.

The MMFF94 force field is commonly used in molecular modeling studies to achieve the lowest potential energy state of chemical compounds, making it an essential tool for planning drug candidates. This force field has been associated with structural modifications of compounds (Halgren 1996; Jakhar et al. 2020).

Molecular docking simulation and data output

The Human aldose reductase protein and complex with LIT ({[5-(5-nitro-2-furyl)-1,3,4-oxadiazol-2-yl]thio}acetic acid) this is an organic compound that belongs to the nitrofuran class, available in the Protein Data Back virtual repository—PDB, under code PDB ID 2ikh, structure with resolution: 1.55 Å, determined through X-ray diffraction, being classified as oxidoreductase for Homo sapiens, According to research done by Steuber et al. (2007), LIT is an inhibitor of aldose reductase. LIT interacts with amino acids that are crucial for inhibition and is considered a model for predicting new drug candidates due to its interaction with the protein pocket (Steuber et al. 2007). For the proper molecular docking simulation, the AutoDockVina code (Trott and Olson 2009), was used, where the calculations were configured under the Lamarkian Genetic algorithm (LGA) (Marinho et al. 2020); The three-dimensional coordinates of the analyzed protein present values referring to center 15.605 x, 0.463 y, 21.764 z; the size parameter, equal to 126 Å (x), 104 Å (y) and 122 Å (z), with the total spacing equal to 0.431 Å.

To validate the molecular fit, molecular redocking of the co-crystallized inhibitor LIT was carried out. For each compound analyzed, 50 simulations were carried out, providing 20 positions for forming the ligand–protein complex for each simulation, with the provision of affinity energy values equal to or less than (− 6.0 kcal/mol) (Shityakov and Foerster 2014; de Oliveira et al. 2021) and Root Mean Square Deviation (RMDS) equal to or less than (2.0 Å) (Yusuf et al. 2008).

The results of molecular docking simulations indicate that the strength of interaction between the ligand and the protein is determined by the distance between them. Strong interactions are defined as distances less than 2.5 Å, moderate interactions occur at distances between 2.5 Å and 3.1 Å, while distances greater than 3.1 Å are considered weak interactions (da Rocha et al. 2023).

Molecular dynamics

For the MD studies presented in this work, the results were obtained using the GROMACS 2020.4 software (Berendsen et al. 1995), configured with the CHARMM 36 force field for standardization purposes (Hughes et al. 2008). The complexes resulting from the molecular docking calculations were used as the starting point for the MD simulations. The best poses obtained from molecular docking were selected as the starting point for the MD simulations. All ligands used in the MD simulations were parameterized from the SwissParam online server (https://www.swissparam.ch/). The protein system in complex with each ligand was inserted into CHARMGUI to add water molecules from the TIP3P model (Boonstra et al. 2016), and then neutralized by inserting Na⁺ (27) and Cl⁻ (25) ions. With the neutralization and solvation of the system, geometric optimization must be carried out using Steepets Descent algorithms and a series of gradients. To carry out MD simulations, it is necessary to use some algorithms that aim to minimize the energy in the system under study (Coretti et al. 2018), following a tolerance of the number of steps 5000 and the energy equivalent to 10 kJ mol⁻¹ nm⁻¹.

To balance MD, it was simulated in 1 ns and then divided into two parts. The first part was subjected to a temperature of 310 K, using the V-rescale method to keep the temperature adjusted and controlled during the simulation (Nagasundaram et al. 2017). The second part suffered direct interference from pressure, parameterized to keep the system stable with a pressure value equal to 1.0 bar throughout the simulation (Martoňák et al. 2003). The realization of MD situations was standardized to order a time scale of up to 200 ns by the Leap Frog integrator (Van Gunsteren and Berendsen 1988; Hu et al. 2023). Some studies analyzed indicate that the proper use of the integrator and the Parrinello–Rahman method demonstrates greater stability of the system throughout the entire simulation (Martoňák et al. 2003; Biçak et al. 2019).

RMSD calculations are performed, where z, y, and z coordinates are used (Brüschweiler 2002), demonstrated in (Eq. 2), which outlines the interactions between molecules through their coordinates, squared aiming at the positive value and then the sum is carried out, followed by division by the number s of atoms (N), obtaining the average and subsequently using the root to remove the effect of the power of 2, of all RMSD values.

$$\text{RMSD}=\sqrt{\frac{1}{N}{\sum }_{i=1}^N{\left(x_{i1},-,x_{i2}\right)}^2+{\left(y_{i1},-,y_{i2}\right)}^2+{\left(z_{i1},-,z_{i2}\right)}^2}$$ 2

To calculate Root Mean Square Fluctuation (RMSF), the entire movement of a certain group of atoms or just one specific atom is analyzed throughout the entire MD simulation, using as a comparison the initial state of the 200 ns (Case et al. 2005), (Eq. 3). From this point onwards, all the complexes formed by the aldose reductase enzymes with mangiferin and its derivatives were analyzed throughout 200 ns in the simulation, and the rigidity and flexibility of the enzyme were analyzed. All conformational variations due to fluctuation were identified.

$${\text{RMSF}}_{\text{atom}}=\sqrt{\frac{1}{t}{\sum }_{i=1}^t{\left(x_i^{\mathit{ref}},-,x_i\right)}^2+{\left(y_i^{\mathit{ref}},-,y_i\right)}^2+{\left(z_i^{\mathit{ref}},-,z_i\right)}^2}$$ 3

MPO-based drug-likeness

The chemical structure of the compounds was represented in two dimensions (see Fig. 1) using the academic license software MarvinSketch^® version 23.11.0 by Chemaxon^© (https://chemaxon.com/marvin). This software was used to calculate the physicochemical properties associated with the pharmacokinetic profile of the substances. Equation 4 shows the use of the drug-likeness scoring criterion with substances registered in patents in recent years of the Medicinal Chemistry Evolution 2018 (MCE-18) descriptor.

$$\text{MCE}18=\left(\text{AR},+,\text{NAR},+,\text{Chiral},+,\text{Spiro},+,\frac{Fsp3+\text{Cyc}-\text{Acyc}}{1+Fsp3}\right)Q^1$$ 4

The fraction of sp3 hybridized atoms is distributed between cyclic structures, including aromatic rings (AR) and non-aromatic rings (NAR), and acyclic structures, including chiral and spiro centers. Each attribute is scored from 0 to 1 according to the desirability function. The final score expresses the systematic balance between structural complexity and satisfaction of the trends observed in new drug candidates in recent years. Compounds with an MCE-18 threshold between 45 and 78 are more efficient as enzyme inhibitors, ligands for G protein-coupled receptors (GPCRs), and modulators of ion channels. These compounds are formed by Fsp3-rich substances that are larger and more polar than commercialized drugs and have low affinity for kinases, indicating high effectiveness as a drug and low toxic response. It is important to note that Ivanenkov et al. (2019) have not tested these compounds for toxicity and efficacy in humans.

Then, Pfizer's biopharmaceutical classification system was applied to relate physicochemical attributes with pharmacokinetic descriptors based on multiparameter optimization (MPO), as shown in Eq. 5:

$$\text{MPO}={\sum }_{i=1}^N{w}_k{T}_k\left(x_k^0\right)$$ 5

The desirability functions (T(x)) are determined by a weighting factor (w) which ranges from 0 to 1 for each physicochemical property (k). If the property is within a unilateral limit (x_k ≤ x_a) or bilateral limit (x_b < x_k ≤ x_a), then w = 1. If the property is outside a unilateral limit (x_b < x_k) or bilateral limit (x_a < x_k < x_b), then w → 0. The limits include: intrinsic lipophilicity (logP ≤ 3), buffer lipophilicity (logD ≤ 2), molecular weight (MW ≤ 360 g/mol), Topological Polar Surface Area (40 < TPSA ≤ 90 Ų), H bond donors (HBD ≤ 1) and pKa of most basic center ≤ 8.0 (N = 6 properties), resulting in a score that varies from 0 (poor drug-likeness) to 6 (good drug-likeness), where MPO values > 4 are associated with drugs with greater pharmacokinetic viability of oral absorption, metabolic stability, cell viability and safety to the central nervous system (CNS) (Johnson et al. 2009; Wager et al. 2010). The MPO scores were related to pharmacokinetic descriptors predicted by the consensus absorption, distribution, metabolism, excretion and toxicity (ADMET) test between the ADMET lab 2.0 (https://admetmesh.scbdd.com/), ADMET boost (https://ai-druglab.smu.edu/admet), PreADMET (https://preadmet.qsarhub.com/adme/), and SwissADME (http://www.swissadme.ch/), which include apparent permeability (P_app) by the Madin-Darby Canine Kidney cells (MDCK) model, skin permeability (logK_p), P-glycoprotein substrate (P-gp), human intestinal absorption (HIA), plasma protein binding (PPB), intrinsic clearance rate (Clint, u) and human hepatotoxic response (H-HT).

Predicted LD₅₀ in administration routes

Toxicity prediction was made using quantitative structure–activity relationship (QSAR) models based on quantitative neighborhood of atoms (QNA) descriptors from the online server GUSAR Online—Way2Drug (https://www.way2drug.com/gusar/index.html). The QNA descriptors can establish correlations between the connectivity matrix (C), ionization potential (IP), and electronic affinity (EA) for each atom of every molecular fragment in the Way2Drug database. The server analyzes the organization of each pair of P and Q atoms in the two-dimensional space (read by SMILES) to obtain a biological activity spectrum based on topological descriptors, such as unsaturation (the number of bonds between each two atoms), H bonds, and molecular volume (the sum of the atomic volumes). The QSAR properties that predict lethal dose (LD₅₀) thresholds are a combination of the three physicochemical attributes and three descriptors: C, IP, and EA (Lagunin et al. 2011).

The dataset includes approximately 10,000 compounds from the training set with favorable therapeutic activity for more than 3000 characterized mechanisms of action, 57 toxic and adverse effects (hepatotoxicity and mutagenicity, for example), 199 inhibitory compounds and substrates of cytochrome P450 isoforms (CYP450)—associated with first-pass metabolism—and more than 40 drug transporter proteins, to estimate the LD₅₀ for different routes of drug administration, including oral, intravenous (IV), and subcutaneous (SC) (Druzhilovskiy et al. 2017).

The Pearson coefficient is utilized to assess the similarity and dissimilarity between molecular fragments in a server database and an input compound provided in SMILES notation. An ideal applicability domain (AD) has an average value of 0.7 or higher. The server then employs QSAR and QNA descriptors to calculate LD₅₀ values in log10 (mmol/kg). The values are subsequently converted into relative LD₅₀ values in mg/kg. LogLD₅₀ values above 0.5 are considered to be within the applicability domain. Model validation is performed using leave-twenty-percent-out cross-validation, repeated until a 20% data discard threshold (R²_L20%Out) is reached. The training and test sets are divided into 80% and 20% proportions, respectively. Lagunin et al. (2011) conducted an analysis that included physicochemical attributes, QNA descriptors, QSAR descriptors, and LD₅₀ values. Results were compared to predictors of P-gp inhibitor and substrate and CYP450 isoforms from the ADMET predictive test.

Results and discussion

Structure-based virtual screening

Structure-based virtual screening is a technique used in computer-aided drug design (CADD) to identify potential biological targets for small molecules. This screening method is based on the chemical structure of the molecules. The term CADD refers to computational techniques that use molecular modeling to discover new drugs (Sant'Anna 2002). This method uses a three-dimensional comparison to compare the input compound with compounds previously recorded in datasets. The compounds are fitted by a support vector machine model. The training sets and the input compound are combined in a vector analysis that produces similarity checks. These similarity checks lead to the selection of the target protein. This method can predict potential therapeutic targets for further virtual investigations, such as molecular docking and MD (Daina and Zoete 2019).

The reference organism chosen was Rattus norvegicus because of its physiological and genetic similarity to humans. It is also a widely used animal model in biological trials, providing a vast database. This similarity is crucial in understanding biological processes, particularly those involving glycosylated derivatives. These derivatives play a fundamental role in signaling and regulating proteins, especially at an intracellular level. Kleandrova et al. (2015) and Muñoz-Antoli et al. (2014) have both highlighted this. Researchers have used a technique to discover new candidate compounds for antidiabetic drugs by inhibiting aldose reductase. This has resulted in the selection of substances with more favorable properties. When combined with other predictive techniques, such as druglikeness functions and ADMET estimation, it is possible to design a substance with the lowest possible toxicological risk for future experimental tests (da Rocha et al. 2022).

After conducting similarity test analyses (see Fig. 2a) using the structure-based virtual screening model, it was observed that the HMF and IMF ligands exhibit similarity with at least 10 compounds with a 3D structure (green bar) that are known to be active against the enzyme aldose reductase in Rattus norvegicus (by homology). On the other hand, NRT and NMF exhibit the greatest 2D similarities (red bar). The analysis supports the prediction of the target classes. It was observed that glycosylated derivatives of NRT have G protein-coupled receptors (GPCRs) and ion channels in at least 50% of their biological interactions (Fig. 2b).

Fig. 2.

Virtual screening of A target classes and B 3D/2D similarity prediction against aldose reductase

The gene encoding aldose reductase in Rattus norvegicus is homologous to human aldose reductase. Barski et al. (2008) showed that human AKR1B1 has genetic similarities with rodent type 1B3 isoforms, while human AKR1B10 has low genetic homology with the 1B7 and 1B8 genes. Table S1 highlights the comparable structures of several glycosidic flavonoids, polyphenols, and even MF, with 3D similarity scores ranging from 0.75 to 0.91. Hyperoside (CHEMBL251254) is a glycoside located on the chromone ring of a flavonoid and is in the preclinical stage within the ChEMBL database. The MF (CHEMBL464825), a natural glycoside in the preclinical phase documented in the ChEMBL database, has also demonstrated antidiabetic effects in Rattus norvegicus (Muruganandan et al. 2005).

Results docking simulations

Redocking between aldose reductase and LIT inhibitor

Regarding molecular docking simulations, the complex formation between the target protein and the LIT compound showed RMSD and affinity energy values equal to 1.822 Å and − 7.9 kcal/mol, respectively. The values presented by the theoretical inhibitor followed the parameters established for selecting favorable complexes in molecular docking simulations.

The complex resulting from the redocking process established an energetic interaction with a value of − 7.9 kcal/mol. Figure 3A shows the interactions between the LIT inhibitor and the active site of aldose reductase. Hydrophobic interactions were observed with the aromatic amino acid residues Trp 79, Phe 115, and Phe 122. These interactions were weak in intensity, as the donor–acceptor distances were greater than 3.1 Å. However, the amino acid Trp 111 showed a strong donor–acceptor interaction, with a distance of 2.23 Å, as observed by Imberty et al. (1991) (Table 1). Figure 3C displays the heat map for the ligand–protein interactions exerted by the LIT inhibitor. The colors range from the most intense blue, representing 2.20 (Å), to white, representing 5.00 (Å), as the interaction distance increases, as reported by Steuber et al. (2007). It is important to note that the compound exhibited π-stacking interactions with the heteroaromatic substructure of Trp 111 residue. This interaction was particularly strong due to the oxygenated pentacyclic ring linked to the NO₂ group, as shown in Fig. 3A.

Fig. 3.

A interaction of the LIT inhibitor with the residues present in the active site; B Docking positions of the LIT inhibitor at the Aldose Reductase Human; C distance between the ligand-receptor represented by a spectrum with the residues interacting with the residues of Human Aldose Reductase

Table 1.

Details of molecular docking simulations between mangiferin and its derivatives for the Human aldose reductase receptor, which include RMSD values and details on the types of ligand-receptor interactions

Compound	RMSD (Å)	Residue	Distance (Å)	Type
LIT*	1.822	Trp 79	3.35	Hydrophobic
		Phe 115	3.67	Hydrophobic
		Phe 122	3.44	Hydrophobic
		Trp 111	2.23	H-Bond
		Leu 300	3.20	H-Bond
		Trp 111	3.60	π-Stacking
		Trp 111	3.67	π-Stacking
		His 110	4.46	Salt Bridges
NRT	0.960	Val 27	3.66	Hydrophobic
		Phe 122	3.52	Hydrophobic
		Leu 300	3.85	Hydrophobic
		Trp 20	2.96	H-Bond
		His 110	2.61	H-Bond
		Trp 111	2.18	H-Bond
		Trp 79	4.85	π-Stacking
HMF	1.163	Tyr 48	3.33	Hydrophobic
		Trp 111	3.75	Hydrophobic
		Leu 300	3.19	Hydrophobic
		Val 47	3.01	H-Bond
		Gln 49	2.91	H-Bond
		His 110	1.83	H-Bond
		Trp 20	4.58	π-Stacking
		Trp 20	4.37	π-Stacking
IMF	1.843	Val 47	3.82	Hydrophobic
		Trp 219	3.61	Hydrophobic
		Leu 300	3..25	Hydrophobic
		Trp 20	3.16	H-Bond
		Trp 20	4.59	π-Stacking
MF	1.623	Trp 219	3.63	Hydrophobic
		Asp 224	2.78	H-Bond
		Trp 295	2.47	H-Bond
		Ala 299	2.73	H-Bond
		Ala 299	2.28	H-Bond
		Leu 301	1.85	H-Bond
		Ser 302	3.40	H-Bond
GMF	1.915	Ala 299	3.74	Hydrophobic
		Lys 194	2.42	H-Bond
		Arg 296	2.88	H-Bond
		Arg 296	2.18	H-Bond
		Glu 314	2.48	H-Bond
		Arg 296	3.46	Salt Bridges
NMF	1.114	Ala 299	3.42	Hydrophobic
		Leu 301	3.65	Hydrophobic
		Arg 217	2.21	H-Bond
		Arg 217	2.77	H-Bond
		Trp 219	2.47	H-Bond
		Trp 219	2.17	H-Bond
		Asn 294	2.27	H-Bond
		Trp 295	2.05	H-Bond
		Arg 296	1.94	H-Bond
		Arg 296	2.48	H-Bond
		Arg 296	2.15	H-Bond
MFG	1.359	Arg 296	3.74	Hydrophobic
		Arg 296	3.88	Hydrophobic
		Ala 299	3.50	Hydrophobic
		Leu 300	3.90	Hydrophobic
		Trp 219	3.64	H-Bond
		Trp 219	3.29	H-Bond
		Trp 295	2.89	H-Bond
		Arg 296	2.74	H-Bond
		Arg 296	1.98	H-Bond
		Ala 299	2.19	H-Bond
		Tyr 309	2.19	H-Bond
		Phe 311	2.30	H-Bond
		Trp 219	4.74	π-Stacking
		Trp 219	3.77	π-Stacking
		Trp 219	4.68	π-Stacking

*Control ligand used in molecular docking simulations

Molecular docking of Norathyriol derivatives against aldose reductase

For the ligands NRT, HMF, MF, and IMF, their characteristics are associated with variations relating to the “R” groups, which are divided into R1, R2, and R3, highlighted in Fig. 1, and for the NRT ligand the three groups are substituted. are divided as: (R1-H, R2-H, and R3-H), as for the HMN compound, the difference is associated only with R1 and R2, where, (R1-CH₃, R2-Gly, and R3-H), for the MF ligand, the pharmacophore presents the following substitutions in its groups (R1-H, R2-Gly, and R3-H), and finally, the IMF ligand showed its variations for (R1-H, R2-H, and R3-Gly) explained in Fig. 1, still referring to the compounds, the NMF ligand is highlighted with the presence of two groups (Gly), present in the replacement of R2, and a replacement of the hydroxyl group, in relation to the GMF ligand it is highlighted that in the In the R2 group, two groups (Gly) are replaced, associated with each other, but for MFG, the substitution referring to R2 contains the presence of a group (Gly), associated with the methyl gallate structure. These characteristics may be directly associated with the possibilities of favorable interactions with aldose reductase, corroborating the interactions analyzed in molecular docking.

At the end of the simulations with the NRT derivatives, it was possible to observe that all ligands exhibited an RMSD within the stipulated statistical range, with values below 2.0 Å (Yusuf et al. 2008). The ligands presented the following RMSD values: NRT—0.960 Å, HMF—1.163 Å, MF—1.623 Å, IMF—1.843 Å, NMF—1.114 Å, GMF—1.915 Å and MFG—1.359 Å, as indicated in Table 1. Here, it is worth highlighting that the ligands NRT, HMF and IMF showed interactions in common with the inhibitor LIT, which include residues of Trp 79, His 110, Trp 111, Phe 122, and Phe 300, indicating that these ligands interact with the active site aldose reductase (Fig. 4A). The heat map in Fig. 4B shows all the reference residues in front of the NRT, HMF and IMF compounds. The heat map illustrates variations in distances, with the most intense blue representing a distance of 1.80 Å, indicating a strong interaction. As the distance increases, the intensity decreases, with distance values of 5.00 Å, shown in white, indicating weak interactions, while distances of up to 10.00 Å are considered extremely weak interactions, shown in red. The heat map shows that the NRT and HMF ligands have comparable interaction strengths with residues Trp 20, His 110, and Trp 111.

Fig. 4.

A interactions carried out by the derivative ligands NRT, HML, and IML against the enzyme; B Heatmap of the distances of interactions between residues and ligands; C Affinity energy values of ligands NRT, HML, and IML

Figure 4C presents a graphic representation of the affinity energy values, in kcal/mol, for the ligand–protein complex and the glycosylated HMF derivatives. The ligand–protein complex exhibited the best energy value of − 8.2 kcal/mol, while the glycosylated HMF derivatives showed a value of − 7 kcal/mol. The interaction energy between the ligand and the protein was found to be − 6.7 kcal/mol, while the energy required to form the complex was 2 kcal/mol. It is important to note that these values are objective and based solely on the presented data. When discussing ideal energy trends, the comparison parameter should be set at values equal to or lower than − 6.0 kcal/mol. Complexes formed with these values demonstrate a greater affinity between the ligand and the protein (da Fonseca et al. 2023; Shityakov and Foerster 2014).

Among the other ligands analyzed, which were complexed in a region distinct from the protein's active site, interactions with 13 residues present in another cavity of the protein were observed. However, these interactions tended to be with the active site residues, specifically Ala 299 and Tyr 309 (Fig. 5A), through H bond interactions. The heatmap in Fig. 5C indicates the distances of the interactions with the compounds MF, GMF, NMF and MFG for the two highlighted residues. The heatmap's distance representation follows the same scale as that used in Fig. 4B. The intensity ranges from deep blue (1.80 Å), indicating strong binding, to white (5.00 Å), representing weak interaction. Weak interactions are shown in red with a distance of 10.00 Å, with only two residues present in the protein's active site. The data presented in Fig. 5D show that MFG has the highest binding energy value of − 8.7 kcal/mol, followed by GMF with a value of − 8.3 kcal/mol, and NMF with a value of − 8.2 kcal/mol. MF has an affinity energy value of − 7.0 kcal/mol.

Fig. 5.

A Interactions between the amino acids of the protein and the ligands MF, GMF, NMF, and MFG; B region of interest between the ligands and the protein cavity; C heat map of interactions with residues of interest that interact with the cited ligands; D binding energy values in kcal/mol

Molecular dynamics

Interaction potential energy

Interaction Potential Energy (IPE) is a parameter that measures the binding energy of ligand–receptor complexes. It quantifies the order of stability in the system based on the affinity of the molecules involved. The energies studied include Leonard–Jones and Coulomb. Leonard–Jones demonstrates all interactions between molecules and atoms through the Van der Waals interaction. Coulomb relates the use of charges through electrostatic interactions. When these charges are added together, the total interaction energy originating from the ligand–receptor system is found (Ribeiro et al. 2016).

For all the molecular dynamics simulations, the energy values are duly presented in Table 2, where the crystallized inhibitor LIT has an energy value equal to − 168.760 ± 44 kJ/mol, while the compounds analyzed in the research can be highlighted with the following values, where the ligand (NRT) has an energy value equal to − 118. 726 ± 39 kJ/mol, (HMF) equal to − 300.812 ± 52 kJ/mol, (IMF) with an energy of − 216.483 ± 33 kJ/mol, (MF) equal to − 212.292 ± 34 kJ/mol, (GMF) with an energy of − 139. 069 ± 39 kJ/mol, (NMF) with a result of − 242.940 ± 45 kJ/mol and finally the compound (MFG) with an energy value equal to − 304.812 ± 52 kJ/mol.

Table 2.

Values of interaction power energies (IPE) of systems evaluated in molecular dynamics

Compound	Energy interaction power (kJ mol−1)	Standard deviation
LIT	− 168.760	± 44
NRT	− 118.726	± 39
HMF	− 300.812	± 52
IMF	− 216.483	± 33
MF	− 212.292	± 34
GMF	− 139.069	± 39
NMF	− 242.940	± 45
MFG	− 304.812	± 52

Source. Prepared by the author

The results show that the complex between aldose reductase and the MFG analog has an energy of − 304.812 ± 52 kJ/mol. This complex has the best energy yield and is therefore the most stable. However, the compound HMF forms a complex with aldose reductase in the receptor's active site, resulting in the second-best energy order with an IPE value of − 300.812 ± 52 kJ/mol (Table 2). On the other hand, the interaction between the receptor and the NRT ligand resulted in the least stable system (− 118.726 ± 39 kJ/mol).

RMSD scale and analysis of the most stable complexes

The MD simulations were performed in triplicate for each complex, with each run represented by a different color: black for the first run, red for the second, and green for the third. This pattern was maintained for all complexes evaluated in the study, as shown in Fig. 6.

Fig. 6.

A RMSD results of the LIT compound, theoretical protein inhibitor, B RMSD of the NRT compound, C RMSD values with the HMF ligand, D RMSD in triplicates of the system formed with the IMF compound, E RMSD with the MF ligand, F RMSD result in complex with the NMF ligand, G RMSD of the GMF ligand, and H RMSD values with the MFG compound

The ligands were selected for molecular dynamics simulations, using the compound LIT, an aldose reductase inhibitor from Fig. 6A, as a comparative parameter. In the inhibitor triplicates, it was observed that the system demonstrated an initial instability presented in the second simulation due to a large variation in its RMSD. However, at 100 ns, the three simulations presented similar conformations until the end of the dynamics.

The first simulation highlighted in black, the second (red), and the third (green), demonstrated that the stabilization of the system occurs in approximately 100 ns, with an RMSD variation of 2.1 Å, maintaining the conformational consistency of the three independent simulations, demonstrating equivalent and stable results until the end of the MD simulations, corroborating the prediction potential regarding the method used, despite the second simulation having presented a large initial variation, however, in 100 ns the three simulations demonstrated the existences of similar thermodynamic, entropic and conformational similarities until the end of the simulation, thus validating the presence of a stable system.

From the MD simulations, it is observed that at the end of 200 ns, the first and second simulations showed similar conformational variations and around 2.0 Å, while the third simulation (green) revealed a slightly larger variation, reaching 2.5 Å. However, upon comparing the conformational and structural variations (Fig. 6B), it becomes apparent that the system formed by the NRT ligand, despite having a low RMSD variation, exhibited a significant conformational discrepancy toward the end of the simulation. This indicates that the analyzed system did not yield favorable results for stabilization.

Referring to the system formed with the NRT ligand (Fig. 6B), the RMSD values indicate conformational stability throughout the entire simulation. The first and second simulations were similar, with an RMSD of around 2.0 Å, demonstrating low conformational variation. However, the third MD simulation showed a small variation starting at approximately 60 ns, with an RMSD of around 2.5 Å, which persisted for the remainder of the simulation. Although the simulations show minor discrepancies, it is important to note that the conformational divergence at the end of the simulation indicates that the complex formed by the NRT ligand lacks structural stability during the MD triplicates.

For the system formed with the HMF ligand, as shown in Fig. 6C, three simulations were carried out. Small conformational variations occurred, ranging from 1.75 Å to approximately 2.0 Å. The three runs showed a high degree of similarity at the end of the simulation, indicating low conformational variation. This demonstrates the stability of the system.

Figure 6C shows that the three independent MD runs for the system formed with the HMF ligand were similar to each other. Stabilization was observed within the first 60 ns and continued until the end of the simulation (200 ns), with conformational variations equal to 1. During the simulation, the system formed by the presence of HMF demonstrated low variation and high stability, with all three runs presenting values of approximately 2.0 Å at the end of the 30 ns period. It is important to note that the conformational variation between the three runs was also low.

With the MD results, the relationship between the protein and IMF is observed in Fig. 6D, by triplicate simulations. The RMSD variations were analyzed, where the three runs presented similarity up to 80 ns, after that, each run demonstrated a specific variation, which indicates a low reliability of prediction, in relation to the system formed by the presence of the IMF ligand.

Still dealing with Fig. 6D, it is highlighted that the peaks related to the RMSD conformational variations are scored in three runs, where an initial similarity is observed up to 80 ns, but from this point onwards it is noted that each run showed a different behavior. distinguishing, where at approximately 175 ns, it is noted that for the first run (Black), the peak reached 2.5 Å, for the second (Red) run the value of its conformational variation reached only 2.0 Å, and dealing with the third simulation, it was observed the smallest variation being expressed as a value equal to 1.5 Å. Thus, the system formed by the presence of the IMF ligand presents a large variation when comparing the three runs; therefore, it does not demonstrate stability of the analyzed system.

The RMSD values referring to the MD simulations for the MF compound are observed in Fig. 6E, where the results of each run performed are evaluated, observing that at the beginning of the simulation there was already a very high variation reaching approximately 3.1 Å for the third run, the first run and the second showed approximately 1.5 Å, in 30 ns, therefore, demonstrating a high conformational variation expressed by the RMSD values, thus, demonstrating a low stability of the system with the presence of the MF ligand.

As shown in Fig. 6E, the system with the MF ligand exhibited a difference in conformational variation between the first two runs, highlighted at 25 ns. The first run showed a variation of 1.2 Å, while the second showed a value close to 1.9 Å. The third run had a higher value of 3 Å. At the end of the simulations, the first run presented a value of 2.3 Å, the second 2.8 Å, and the third 3.5 Å. These variations indicate that the system with the presence of the MF ligand exhibited significant oscillation. Therefore, the analyzed system does not demonstrate stabilization potential according to the MD predictions.

For the system formed between the protein and the NMF ligand, Fig. 6F, the three simulations show increasing and cohesive values among themselves, but, when approaching 100 ns, it can be observed that the three runs performed for the complex demonstrate behaviors distinct, with high conformational variations being expressed in the analyzed system, thus highlighting that the analyzed system does not have stability when analyzing its RMSD values.

Still present in Fig. 6F, the behavior of the three runs carried out by the NMF ligand is demonstrated, initially highlighting a uniformity between all simulations from 20 to 100 ns; and from this point onwards, each run takes on a different conformation, where the first run (Black) presents a variation equal to 2.5 Å, the second (Red) presents a value of approximately 1.5 Å, and the third run (Green), presents a variation equal to 1.75 Å. But at the end of the MD simulations, it is observed that the first run demonstrated a variation equal to 1.75 Å, the second showed an approximate value of 1.3 Å, and the third run at the end of the simulation showed a value equal to 2.25 Å, the data of RMSD, point out that the analyzed system presents a potential conformational variation compared to the analysis in triplicates, corroborating the potential instability of the analyzed system.

The MD simulations showed that the system formed with the GMF compound, present in Fig. 6G, presented two distinct behaviors, where the first run (Black) and the second (Red) were quite similar throughout the entire MD, but, when evaluating the third run, it is noticeable that in approximately 65 ns, there is a high variation in its conformation, thus completely differing from the behavior observed in the first two runs, therefore, the RMSD data indicate that the system formed with the presence of the GMF ligand, demonstrate instability when assessed by MD.

Still inserted in Fig. 6.G, the similarities that exist in the three runs are highlighted, from the beginning of the simulation until approximately 60 ns, but, from 65 ns onwards, it is observed that the third run demonstrates a large conformational variation, reaching a peak of 3.3 Å approximately in 85 ns, while the first and second runs show RMSD values equal to 1.9 Å and 1.3 Å, respectively. At the end of the simulations, it is observed that the first run presented a value equal to 1.7 Å, the second with a value of 2.0 Å, while the third one presents the greatest variation, equal to 3.7 Å. Therefore, the simulations in triplicates demonstrated that the system formed by GMF ligand presents high variations in its simulations, mainly referring to the third run, and thus, demonstrating that the system appears to be unstable, through MD simulations.

Regarding the RMSD analysis for the system formed with the MFG compound, Fig. 6H is shown. The similarities in the triplicates and their low variations stand out even considering each individual simulation. Therefore, the system presents good prediction reliability with little variation. The graph shows the three runs where the first and second in colors (black and red) respectively, demonstrated great similarity in their simulations, showing similarity between 25 ns and up to 200 ns with variations of up to 2.2 Å, while the third simulation (green) showed a small increase at 30 ns, stabilizing at 50 ns and remaining at 2.1 Å that in a time of 125 ns the three simulations reach a similar conformational state that lasts until the end of the MD (200 ns) being a of the systems that stood out from the RMSD analyses.

In short, the RMSD analyses of all systems showed more clearly which systems do or do not have reliability and similarity when compared individually using the triplicate simulation method. Therefore, the systems formed with the HMF and MFG ligands can be highlighted, as they demonstrate similarity in individual calculations in triplicates as well as the LIT cocrystallized ligand system.

Analysis RMSF

In this way, studies focused on flexibility aim to obtain an understanding regarding the interference caused by the presence of the ligand inserted in the system, which can be analyzed in Fig. 7; the fluctuation studies mainly evaluate how the ligand interferes in a certain region of the protein where, when inserted, when the ligand interacts directly with the residue present in the protein, a structural deformation occurred in the interaction region, highlighting greater flexibility in the region, due to the reduction of intermolecular forces, and thus, promoting greater flexibility of the protein being reflected in the RMSD variations (Mossé et al. 2008).

Fig. 7.

A fluctuation of amino acid residues with the ligands LIT (black), NRT (red), HML (green), and IMF (blue); B fluctuation of the other ligands MF (black), GMF (red), NMF (green), and MFG (blue)

As observed in the RMSF graph present in Fig. 7A, all the flexibility variations exerted by the inhibitor LIT, and by the ligands NRT, HML and IMF are presented, where all ligands are inserted in the region of the inhibitor observed in Fig. 4A, the simulations of RMSF pointed out that the highest levels of flexibility were exerted by the ligands LIT and the compound NRT, followed by (HMF) having the smallest conformational variation of RMSD compared to the compounds studied and finally the ligand IMF, having a fluctuation similar to HMF. Referring to the graph in Fig. 7B, all fluctuations of the ligands MF (black), GMF (red), NMF (green) and MFG (blue), which are inserted in the same region of the protein, were analyzed individually, however, different to the inhibitor, as can be seen in Fig. 5A. Among the simulations, the four compounds showed similarity in their fluctuations, presenting some divergences between the systems analyzed, corroborating their high RMSD values present in previous discussions, highlighting the stability of the systems formed by the inhibitor LIT, HMF, and MFG.

The graph in Fig. 7A shows the values for the highest intensities of amino acid flexibility with respect to each compound analyzed individually. For the compounds (LIT, NRT, HMF, and IMF), it can be seen that the inhibitor LIT and the compound NRT are similar in that they show a high intensity with variations in three sets of residues, the first being residues 118–133 (Pro 118, Gly 119, Lys 120, Glu 121, Phe 122, Phe 123, Pro 124, Leu 125, Asp 126, Glu 127, Ser 128, Gly 129, Asn 130, Val 131, Val 132, Pro 133), the second 292–308 (Tyr 292, Asn 293, Arg 294, Asn 295, Trp 296, Arg 297, Val 298, Cys 299, Ala 300, Leu 301, Leu 302, Ser 303, Cys 304, Thr 305, Ser 306, His 307, Lys 308) and the third 310–316 (Tyr 310, Pro 311, Phe 312, His 313, Glu 314, Glu 315, Phe 316), reaching scales close to or greater than 4.0 Å, demonstrating the high conformational variation exerted by the ligands associated with these residues.

In relation to the compounds HMF (green) and IMF (blue), shown in Fig. 7A, higher values were observed in the set of residues 213–225 (Leu 213, Gly 214, Ser 215, Pro 216, Asp 217, Arg 218, Pro 219, Trp 220, Ala 221, Lys 222, Pro 223, Glu 224, Asp 225), a greater flexibility can be observed which may be associated with the coupling of the HMF and IMF ligands, resulting in greater flexibility in the system. Compared to the inhibitor LIT and the compound NRT, the ligands (HMF and IMF) show greater variations in RMSF, present in the set of residues mentioned above, with variations of 4.2 Å and 2.8 Å, respectively. However, residues Phe 122, Val 298 and Cys 299, which are amino acids present in the active site of aldose reductase and suffered great flexibility due to interaction with the inhibitor LIT and NRT, stand out in relation to all the flexibility values.

When relating to the RMSD values present in Fig. 6, it is highlighted that the indices of greatest flexibility refer to the LIT inhibitor and the NRT compound, with the highest indices of fluctuation variations, corroborated with what was observed in the RMSD variations, in slightly higher overall when compared to the HMF and IMF ligands, which had smaller RMSD variations and consequently a reduction in their fluctuations, compared to the protein amino acids. Thus, demonstrating a greater favorability regarding the HMF ligand due to its more cohesive results where it demonstrates a greater potential for stability and a good index of flexibility with the amino acids present in the active site.

Referring to the graph in Fig. 7B, the ligands MF (black), GMF (red), NMF (green) and MFG (blue) were analyzed, with their fluctuation values, highlighting the formation of the complex with the presence of the compound MF, highlighting variations with values around 2.0 Å, highlighted by two sets of residues where the first is composed of residues 17- 27 (Gly 17, Leu 18, Gly 19, Thr 20, Trp 21, Lys 22, Ser 23, Pro 24, Pro 25, Gly 26, Gln 27) and the second 42–50 (His 42, Ile 43, Asp 44, Cys 45, Ala 46, His 47, Val 48, Tyr 49, Gln 50), the analysis of the fluctuations with the presence of MF compounds, pointed out an increased variation in their fluctuation present in the set of residues 300–307 (Ala 300, Leu 301, Leu 302, Ser 303, Cys 304, Thr 305, Ser 306, His 307), reaching a variation close to 6.9 Å, these high variations are associated with considerable conformational changes, corroborating the variations presented in the RMSD graphs and possibly affecting the stability of the evaluated system.

When treating the set of residues 213–222 (Leu 213, Gly 214, Ser 215, Pro 216, Asp 217, Arg 218, Pro 219, Trp 220, Ala 221, Lys 222), inserted in Fig. 7B, large variations in complexes formed with the presence of individual ligands (MF, GMF and NMF), demonstrating more flexible regions, reaching variation values close to 3.0 Å, and for the MFG compound, there was a smaller conformational variation in the analyzed region, this behavior was also observed in the regions composed of the two groups of residues, the first being composed of 17–27 (Gly 17, Leu 18, Gly 19, Thr 20, Trp 21, Lys 22, Ser 23, Pro 24, Pro 25, Gly 26, Gln 27) and the second set 217–233 (Asp 217, Arg 218, Pro 219, Trp 220, Ala 221, Lys 222, Pro 223, Glu 224, Asp 225, Pro 226, Ser 227, Leu 228, Leu 229, Glu 230, Asp 231, Pro 232, Arg 233).

It should be noted that the smaller variations in conformational fluctuation highlighted in Fig. 7B, preferably for MFG, corroborate its respective RMSD variation, demonstrating low oscillation as seen previously, in direct comparison with the other ligands, which presented higher fluctuation values, for specific residues, greater conformational variations are noted in RMSD studies, highlighting the possibility of potential instability associated with systems with the individual presence of the compounds (MF, GMF and NMF).

Despite being in the same system, but with different ligands, it is clear that each ligand presents its individual interference in the studied system, thus dismantling its potential for stability, structural modifications, followed by its flexibility and rigidity.

Analyses regarding H bond occupancy

The results of the MD simulations indicate that all hydrogen bonds were present in the ligand-receptor complexes. The values reported represent the percentage of simulation time during which the bonds were observed, with a range from 0 to 100%. For the purpose of analysis, only values equal to or greater than 5.0% of the simulation time were selected. The interaction between the ligand and the analyzed protein through H-bonds may have an impact on both the stability and structural changes of the system. The initial interaction site of the ligand was determined through molecular docking calculations. Therefore, the stability of the system can be assessed by analyzing the frequency of H bond interactions (Kumar et al. 2020; Morris et al. 2020; Zikri et al. 2020).

For analytical purposes, we have calculated all H bond values exerted by the theoretical inhibitor LIT with the target protein. The relationship between the atoms involved in the formation of 63 hydrogen bonds was analyzed. However, only 10 of these bonds (Ala 299_Acc, Thr 113_Don, Thr 304_Don, Thr 304_Don, Trp 111_Don, Leu 300_Don, Thr 113_Acc, Thr 113_Don, and Thr 309_Don) were found to significantly contribute to the stability of the system, as illustrated in Fig. 8. The protein's inhibitor is believed to form hydrogen bonds with several residues in its active site, such as Ala 299, Tyr 309, and Trp 111. It is worth noting that the bond with Trp 111 is considered the most significant, as it occupies the site for 200 ns and contributes to the system's stability. These findings suggest that the inhibitor could be a promising candidate for further research. The RMSD and RMSF results suggest that the system formed with the LIT inhibitor has a high potential for stability, which can be used as a comparison criterion for the other ligands evaluated in the research.

Fig. 8.

H-Bond occupancy values of the LIT inhibitor, in relation to the protein residues: Don = donor; ACC = acceptor

The occupancy values of the studied ligands were analyzed. Ligands NRT, HMF, MF, and IMF formed several hydrogen bonds with respective results of 171, 25, 258, and 211. However, only a few values were observed within the established parameters. The NRT binder had an occupancy value of 5.19% due to its interaction with the Leu 5 residue. The HMF compound showed a higher occupancy with a result of 14%. Through interaction with the Trp 111 residue, which is present in the protein's active site, MF exhibited a significant occupancy of 6. This result demonstrates the confidence in the accuracy of the findings. The hydrogen bond formed with the amino acid Ala 299 in the protein's active site accounts for a significant portion of the observed effect, specifically 79%. Additionally, two relevant values were observed for the IMF ligand regarding the occupation of the hydrogen bond with the protein residue Val 47 (10.39% and 7.59%), which is not present in the analyzed enzyme's active site. The presented results are of great significance, particularly highlighting the HMF ligand, which occupies a greater space through hydrogen bonding. This suggests a structural advantage aimed at enhancing the stability of the system, which is supported by the RMSD and RMSF values.

Upon analysis of the NMF compound presented in Fig. 9A, it was determined that 172 hydrogen bonds had been formed. Despite the limited number of interactions, the residues Trp 219_Dor, Arg 296_Acc, His 312_Dor, Lys 221_Dor, Trp 219_Acc, Leu 301_Dor, Ala 299_Acc, and Thr 304_Dor have been established with a time greater than 5%. These findings demonstrate the significance of these interactions in the context of the study. The system's stability potential was confirmed by all interactions, with Thr 296 (88.96%) being the most frequent hydrogen bond, followed by Ala 299 (7.04%). However, only one interaction occurred with a residue present in the protein's active site, specifically with Ala 299.

Fig. 9.

H-Bond occupancy values of NMF (A), GMF (B), and MFG (C) ligands; in front of protein residues. Don donor, ACC acceptor

The interaction between the GMF ligand and the protein Human Aldose Reductase resulted in the formation of 211 hydrogen bonds. However, only six interactions were identified as the most relevant to the stability of the analyzed system, with occupancy values equal to or greater than 5% of the MD simulation. These interactions were produced by chemical bonds with the residues Arg 296. It is worth noting that these findings demonstrate the crucial role of Arg 296 in the stability of the system. Although the residues inserted into the active site of the enzyme were not highlighted among all the observed hydrogen bonds, the occupancy values indicate that the ligand is a weak drug proponent with little conformational variation. Notably, the residues with the highest occupancy values were Ser 217_Acc (9.49%), Glu 314_Acc (22.03%), Asn 294_Acc (6.59%), Phe 311_Acc (10.84%), and Leu 301_Dor (8.74%).

When evaluating the interaction with the MFG ligand in Fig. 9C, 113 hydrogen bonds were observed, due to the relationship between the ligand and protein residues, but 16 occupancy values occurred within the parameter established through the interaction with the residues Trp 219_Dor (40.17%), Asp 224_Acc (6.34%), Arg 296_Dor (7.39%), Arg 296_Acc (9.39%), Trp 295_Acc (36.31%), Tyr 309_Dor (5.24%), Trp 219_Acc (8.39%), Ala 299_Acc (7.84%), Leu 301_Dor (11.99%), Thr 304_Dor (6.64%), Trp 295_Dor (6.04%), Cys 298_Acc (8.89%), Pro 225_Dor (11.34%), Asp 224_Dor (6.34%), Tyr 189_Acc (9.64%), Asp 308_Acc (12.79%). Given the occupancy values, the system tends to demonstrate greater stability, given that the complex formed with the MFG ligand exerted a greater number of interactions with significant values, presenting values greater than 5%. Among all the interactions selected, it is important to highlight the two highest values related to the occupation of the hydrogen bond, through the amino acids Trp 219_Dor and Trp 295_Acc, as they remain stable for a more extended period of time in MD. The system showed that hydrogen interactions occurred with three amino acids in the active site (Cys 298, Ala 299 and Tyr 309), demonstrating an affinity with the catalytic region of the enzyme.

Simulation studies and their predictive power can present results that can estimate, through a virtual screen, potential drug proponents, evaluating specific behaviors that are associated with each ligand (Dhiman and Purohit 2023). In the molecular docking simulations, the favorable behaviors related to each ligand are highlighted, through the interactions carried out with the protein aldose reductase, being directly compared with the LIT inhibitor, corroborating with the data from the MD molecular docking evaluates whether the ligand remains in the region estimated by docking, adding variables, such as RMSD, RMSF and H bond values, facilitating and refining the comparison with the inhibitor, over 200 ns, and thus confirming the inhibition potential of the protein, mainly associated with the ligands HMF and MFG, being the main candidates against aldose reductase.

MPO-based drug-likeness

Fsp3 and medicinal chemistry spectrum

The patented compounds have unique physicochemical characteristics that are intriguing and can help us better understand their properties and applications in medicinal chemistry. Recent patent registrations have shown a physicochemical singularity in a series of compounds. These compounds have a notable increase in the proportion of carbon atoms with sp3 hybridization (Fsp3) compared to their portion of aromatic or unsaturated compounds. This suggests a greater prevalence of single bonds and tetrahedral arrangement, as well as an increase in the number of aliphatic rings present in the molecular structures of these compounds (Ritchie et al. 2011; Ivanenkov et al. 2019). These compounds tend to increase in molecular mass when structural complexity allows for synthetic accessibility. This results in improved solubility and distribution of small molecules in blood plasma, as it reduces their interaction with lipoproteins (Wei et al. 2020).

Figure 10 shows a 3D graph indicating that NMF and GMF compounds have 2 NAR within a MW range of > 500 g/mol. This strongly influences Fsp3, which is 0.48 for the aromatic substructure of NRT (Fsp3 = 0.0) and in the formation of the 10 chiral centers susceptible to torsion. The MFG compound has a substituted gallic acid substructure that results in 4 AR, distributed in a MW of 574.45 g/mol, reducing Fsp3 to 0.23 (Tables 3, 4). The compounds showed an MCE-18 score greater than 100, which suggests that their structural complexity makes it difficult to investigate their drug-like properties further.

Fig. 10.

Spectrum of the relationship between Fsp3 and structural complexity in estimating the gau of similarity of NRT glycosylated analogs to patent-pending substances according to MCE-18 trends

Table 3.

Distribution of Fsp3 between aromatic (nAR) and non-aromatic (nNAR) substructures, chiral and spirocyclic centers in the MCE-18 score and the estimation of oral bioavailability (F%)

Compound	Fsp3	Chiral	Spiro	nAR	nNAR	MCE-18 score	F%
NRT	0.00	0	0	3	0	18.00	41.51
HMF	0.35	5	0	3	1	91.00	33.67
MF	0.32	5	0	3	1	91.80	32.08
IMF	0.32	5	0	3	1	91.80	33.66
NMF	0.48	10	0	3	2	123.92	31.89
GMF	0.48	10	0	3	2	123.92	32.55
MFG	0.23	5	0	4	1	115.94	35.35

Table 4.

Physicochemical properties calculated and applied to the Pfizer, Inc. multiparameter optimization (MPO) system

Compound	logP	logD	MW (g/mol)	TPSA (Å2)	HBD (OH)	pKa	MPO score
NRT	2.40	0.31	260.20	107.22	4	− 1.69	4.43
HMF	− 0.22	− 1.91	436.37	186.37	7	− 1.69	3.45
MF	− 0.36	− 2.92	422.34	197.37	8	− 2.98	3.55
IMF	− 0.36	− 2.80	422.34	197.37	8	− 2.98	3.55
NMF	− 2.63	− 5.03	584.48	276.52	11	− 3.65	3.00
GMF	− 2.13	− 4.69	584.48	276.52	11	− 3.65	3.00
MFG	1.22	− 1.42	574.45	264.13	10	− 3.70	3.00

However, the monoglycosylated analogs with a molecular weight of less than 500 g/mol, namely HMF, MF, and IMF analogs, showed a consistent correlation between unsaturation and structural complexity. The total of 1 NAR and 3 AR resulted in only 5 chiral centers due to Fsp3 in decreasing order. The MCE-18 score of 91.0 indicates that these compounds follow the trends of new pharmaceutical products registered in patents in recent years. This score is a result of the combination of 0.32 of the glycoside and 0.35 of the structural increase of the methoxyl radical (–OCH₃) present in HMF.

The compound's aqueous solubility can be increased by increasing the molecular sp3 portion of the unsaturated fraction. This, in turn, can reduce the oral bioavailability capacity (F%) (Yang et al. 2012), as shown in Fig. 10 in the heatmap. The glycosylated analogs' estimated F% is highest for NRT (41.51%), with F% values ranging from 31–35% (Tables 3, 4).

Predicted pharmacokinetics profile

The new chemical–medicinal singularity is characterized by an increase in intrinsic polarity in new compounds used for producing medicines. This is due to the presence of more polar functional groups (HBD or HBA groups) or a greater number of heteroatomic bonds (Ivanenkov et al. 2019). According to Wager et al. (2016), compounds that have low lipophilicity (logP < 3) and high topological polarity (TPSA > 75) are associated with high cell viability and greater safety to the CNS (Hughes et al. 2008). Additionally, reducing the number of HBD groups can enable passive permeability (P_app MDCK > 10 × 10⁻⁶ cm/s) and decrease the hepatic clearance rate (CL_int,u < 100 mL/min/kg), which can help estimate compounds with optimized oral bioavailability. It was observed that polar compounds, which are not commonly commercialized drugs, have a reduced effect against kinases. This reduces the toxic response of new compounds registered in patents and increases the efficiency of compounds as modulators of GPCRs, ion channels, and the inhibition of enzymes (Ivanenkov et al. 2019). Therefore, this descriptor is essential in selecting safer substances (Hughes et al. 2008).

In the Fig. 11, the radar plot shows that the natural product NRT is in a physicochemical space occupied by compounds commonly absorbed in the gastrointestinal tract, as predicted by Pfizer's biopharmaceutical classification system (Wager et al. 2016). This alignment is based on the logP and TPSA attributes. When evaluating glycosylated analogs, it is estimated that they have a low HIA and low risk of distribution to the CNS. This is mainly due to the high polarity resulting from the OH groups of the substructures of HMF, MF, and IMF monoglycosides (TPSA between 180 and 200 Ų), NMF and GMF diglucosides (TPSA 276.52 Ų) and the glycoside derived from gallic acid MFG (TPSA = 264.13 Ų). The results confirm the predicted pharmacokinetic descriptors. The ligands HMF and IMF, which were the most effective in the molecular docking simulations, showed a P_app MDCK predicted to be in the order of 2.4 × 10^–7 cm/s and 5.5 × 10⁻⁸ cm/s, respectively. This suggests a low passive permeability (Table 5), which limits the gastrointestinal absorption of these compounds. However, all analogs have a P_app range below the ideal for good oral bioavailability (< 10 × 10⁻⁶ cm/s).

Fig. 11.

Alignment between logP and TPSA for drug-space estimation related to human intestinal absorption (HIA) and CNS distribution of NRT and its glycoside analogs

Table 5.

Pharmacokinetic descriptors predicted by consensus prediction of ADMET between ADMET lab 2.0, ADMET boost, Pre ADMET and Swiss ADME platforms

Compound	Papp MDCK (cm/s)	logKp (cm/s)	Pgp%	HIA%	PPB%	CLint,u (mL/min/kg)	H-HT
NRT	3.9 × 10−6	− 6.40	47.76	73.38	91.39	33.80	0.07
HMF	2.4 × 10−7	− 8.99	42.66	69.94	81.80	37.38	0.16
MF	3.7 × 10−7	− 9.14	42.85	70.23	85.70	36.30	0.13
IMF	5.5 × 10−8	− 9.14	43.50	69.33	84.44	36.71	0.20
NMF	1.7 × 10−7	− 11.41	42.25	59.15	72.89	40.13	0.13
GMF	8.1 × 10−9	− 11.65	41.65	58.88	81.19	39.86	0.13
MFG	4.8 × 10−9	− 9.63	47.88	62.05	86.62	48.55	0.08

*The best binders from molecular docking tests are highlighted in bold

The estimated CL_int,u values are around 37 mL/min/kg, which directly reflects the high aqueous solubility of the HMF and IMF compounds. This is because these substances have a lower predicted hepatic clearance than the more polar derivatives NMF, GMF and MFG (Table 5). The HMF and IMF analogs are metabolically more stable than the other MF derivatives, resulting in a low risk of hepatotoxicity (probability < 0.2). However, these compounds have a higher clearance range than most oral drugs available commercially. Stepan et al. (2013) reported a CL_int,u of less than 8.0 mL/min/kg, which makes oral administration impractical.

On the other hand, low lipophilicity strongly influences the distribution of these compounds between blood plasma and biological tissues (Pires et al. 2018). Due to their physicochemical properties not meeting the standards for orally absorbed drugs, such as small molecular mass compounds (Johnson et al. 2009), it is recommended to administer these compounds via the SC route (Kinnunen and Mrsny 2014; Ortiz-Zamora et al. 2022). Drugs that are not very soluble in water may have difficulty being absorbed when taken orally. Therefore, local application may be a better strategy for drugs that need to act on adipose tissue (Pires et al. 2018; Van De Waterbeemd and Gifford 2003).

The pharmacokinetic prediction showed that Fsp3 and the water-soluble nature of glycosylated derivatives significantly increase the free molecular fraction of these substances in blood plasma when the degree of PPB is less than 90% (Table 5). The compounds HMF and IMF are absorbed through the skin at a gradual rate due to their logKp values of around − 9.0 cm/s, while the more polar analogs NMF and GMF (logKp < − 10.0 cm/s) are not. This enables the administration of these compounds through the skin as an alternative route.

Predicted LD₅₀ in administration routes

The QSAR model used to predict LD₅₀ used here considers QNA descriptors for molecular fragments that are CYP450 inhibitors and substrates that result in mutagenic and carcinogenic damage in the case of assessing acute oral toxicity and molecular fragments capable of binding to transport proteins, such as plasma proteins and P-gp (Lagunin et al. 2011). The consensus ADMET test showed that glycosylated analogs tend to undergo passive efflux by P-gp if administered orally, with a probability of around 43%, mainly due to the intrinsic polarity of the OH groups (Didziapetris et al. 2003), which may reduce the oral bioavailability of these compounds. However, they have shown resistance to phase I metabolism by the main isoforms of CYP450 (2C9, 2D6 and 3A4), indicating a toxic response through metabolic activation is unlikely (Table 6). However, these substances are poorly absorbed in the intestine, highlighting the need to select a secondary administration route.

Table 6.

Estimated metabolism and transporter properties in probability scores and LD₅₀ values in mg/kg predicted by QNA descriptors

Compd	Pgp%	CYP450 isoforms inhibition			CYP450 isoforms substrate			Predicted LD50 in mg/kg
		2C9	2D6	3A4	2C9	2D6	3A4	Oral	IV	SC
HMF	42.66	0.01	0.02	0.02	0.35	0.17	0.01	2079	1360	2066
MF	42.85	0.02	0.02	0.02	0.22	0.15	0.01	1175	1380	2611
IMF	43.50	0.03	0.01	0.03	0.21	0.15	0.01	4492	2051	2440
NMF	42.25	0.00	0.00	0.00	0.08	0.13	0.00	3199	2484	5191
GMF	41.65	0.00	0.00	0.00	0.08	0.12	0.00	2910	2648	3575
MFG	47.88	0.15	0.00	0.02	0.10	0.14	0.00	2050	1241	988.2

IV intravenous, SC subcutaneous

With the results, it is possible to note that MF showed a predicted LD₅₀ value of 1175 mg/kg, similar to the value present in the literature that did not show observed hematological changes or any severe toxic response in rats by oral administration (250–1000 mg/kg) (Mei et al. 2023). With this parameterization, it is observed that all glycosylated analogs showed LD₅₀ values > 2000 mg/kg for oral administration (Fig. 12), indicating that the daily oral dose administered has a low susceptibility of causing an acute toxic response due to metabolic activation (Gonella Diaza et al. 2015).

Fig. 12.

Prediction of LD₅₀ (in mg/kg) for glycosylated NRT analogs for different administration routes: oral, intravenous (IV), and subcutaneous (SC)

Although they are pharmacologically safe when administered orally, they are pharmacokinetically unviable and poorly absorbed in the gastrointestinal tract. In this way, the SC route becomes the more viable form of secondary administration than the IV route, with values of order > 2000 mg/kg (Fig. 12), except the gallic acid derivative MFG, with a value of LD₅₀ IV 1241 mg/kg > LD₅₀ SC 988.2 mg/kg (Table 6). This is essential data for selecting an alternative route of new antidiabetic drugs since these substances can inhibit aldose reductase in the adipose tissue that expresses the protein via subcutaneous administration, as with other medications such as insulin (Thiagarajan et al. 2022).

Conclusion

Among all the compounds evaluated via molecular docking, it was possible to observe that the ligands NRT, HMF and IMF interact with a more significant portion of the residues in the active site of aldose reductase. However, MD simulations indicate that the MFG compound showed better stability when complexed with the receptor. However, it interacts in a catalytic site distinct from the LIT inhibitor, raising the hypothesis of an allosteric modulating effect. On the other hand, the HMF analog obtained the second-best energy yield in MD simulations, acting as a competitive inhibitor concerning LIT, exhibiting the most effective model in the selective inhibition of aldose reductase. In line with the results of predictive pharmacokinetics and toxicology, it was possible to observe that the HMF analog is an ideal candidate for inhibition of aldose reductase by subcutaneous administration due to the strong influence of the balance between high polarity and ideal distribution of Fsp3, which reduces the degree of binding with metabolic proteins and lipoproteins.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

Conceptualization, FFSL, VMO, CHAR, and MMM; supervision, ESM and SMM; writing original draft preparation, FFSL, VMO, MNR, CHAR, and FNML; reviewing and editing of the manuscript, MMM, ESM, and SMM; all authors read and agreed to the final version of the manuscript.

Funding

This study was financed by National Council for Scientific and Technological Development (CNPq) and Cearense Foundation for Supporting Scientific and Technological Development (Funcap). Márcia Machado Marinho acknowledges financial support from the PDCTR (CNPq/Funcap) (Grant#: DCT-0182–00048.02.00/21 e 04879791/2022).

Data availability

All the data generated or analyzed during this study are included in this published article.

Declarations

Conflict of interest

The authors declare that they have no conflict of interest.

Ethical approval

Not applicable.

Consent to participate

Not applicable.

Consent for publication

Not applicable.