Molecular determinants of PPARγ partial agonism and related in silico/in vivo studies of natural saponins as potential type 2 diabetes modulators
Abstract
The metabolic syndrome, which includes hypertension, type 2 diabetes (T2D) and obesity, has reached an epidemic-like scale. Saponins and sapogenins are considered as valuable natural products for ameliorating this pathology, possibly through the nuclear receptor PPARγ activation. The aims of this study were: to look for in vivo antidiabetic effects of a purified saponins’ mixture (PSM) from Astragalus corniculatus Bieb; to reveal by in silico methods the molecular determinants of PPARγ partial agonism, and to investigate the potential PPARγ participation in the PSM effects. In the in vivo experiments spontaneously hypertensive rats (SHRs) with induced T2D were treated with PSM or pioglitazone as a referent PPARγ full agonist, and pathology-relevant biochemical markers were analysed. The results provided details on the PSM modulation of the glucose homeostasis and its potential mechanism. The in silico studies focused on analysis of the protein-ligand interactions in crystal structures of human PPARγ-partial agonist complexes, pharmacophore modelling and molecular docking. They outlined key pharmacophoric features, typical for the PPARγ partial agonists, which were used for pharmacophore-based docking of the main PSM sapogenin. The in silico studies, strongly suggest possible involvement of PPARγ-mediated mechanisms in the in vivo antidiabetic and antioxidant effects of PSM from A. corniculatus.
Introduction
Nowadays the cases of hypertension, diabetes and obesity, associated with the metabolic syndrome, have reached an epidemic-like rise and pose a serious global health problem (Simmons et al., 2010; Masuo et al., 2011). The combination of hypertension and diabetes, for example, results in more severe organ pathology than observed in each disease alone. Medicinal plants have historically proven their value as a source of molecules with therapeutic potential, and nowadays still represent an important source for the identification of novel drug leads (Atanasov et al., 2015). In recent decades, herbal remedies and natural products have attracted much attention in the context of prevention or treatment of cardiovascular and metabolic disease. Considering the overall morbidity and mortality related to cardiologic and metabolic disorders, there is a high interest in the discovery of novel compounds as well as novel pharmacological targets that might be effective in the treatment or prevention of metabolic disorders (Waltenberger et al., 2016).The current strategy used for treatment of type 2 diabetes is based on its multifactorial genesis and combines insulin secretagogues and insulin sensitizers to provide a therapeutic effect (Hui et al., 2009). The nuclear peroxisome proliferator-activated receptor γ (PPARγ) is a crucial regulator of glucose and lipid homeostasis and an important pharmacological target for treating metabolic diseases (Berger et al., 2005; Berger and Moller, 2002; Kliewer et al., 1997). The nuclear receptor is particularly involved in the regulation of insulin sensitivity, inflammation, fatty acid storage, and glucose metabolism, and therefore represents an especially interesting pharmacological target which is able to simultaneously modulate several of the underlying pathologies of the metabolic syndrome (Wang et al., 2014a).
PPARγ and its full agonists from the class of thiazolidinediones (e.g. rosiglitazone and troglitazone) had been involved in the therapeutic strategy for T2D treatment (Hauner, 2002), which was not successful due to multiple adverse effects including weight gain, fluid retention, hepatotoxicity, and increased risk of heart failure and myocardial infarction (Villacorta et al., 2009). Therefore, the PPARγ targeting has gradually shifted from design of full agonists, currently associated with adverse effects (Merk and Schubert-Zsilavecz, 2012) toward discovery/development of other PPARγ ligand types: partial agonists (Chigurupati et al., 2015), non-agonists (Choi et al., 2014; Kamenecka et al., 2010), antagonists (Marciano et al., 2015) and multi-targeting cooperative agonists (dual- and pan-PPAR agonists) (Wang et al., 2014b; Fievet et al., 2006; Gonzalez et al., 2007). Furthermore, an increasing research effort has been focused on investigating natural sources of novel PPARγ modulators (Wang et al., 2014a). In particular, attention has been directed towards nutraceuticals originating from plants like saponins. The latter represent a class of chemical compounds found in various plant species, which have been reported to exhibit hypoglycaemic potential in diabetic states (Lee et al., 2000; Liu et al., 2012; Smith et al., 2012; Elekofehinti et al., 2013) and have attracted a lot of interest because of their potent hypolipidemic and insulin- like properties (Bhavsar et al., 2009; Eu et al., 2010; Hu et al., 2012; Lee et al., 2011). There is a large body of information revealing the potential use of the natural saponins in the therapy of type 2 diabetes and hyperlipidemia (Smith et al., 2012; Elekofehinti et al., 2013; Naidu et al., 2015). One of the discussed possible mechanisms of action is related to their activity toward PPARγ, as a crucial regulator of glucose and lipid homeostasis (Kwon et al., 2012; Berger et al., 2005). In a recent study, Montanari et al. (2016) measured the PPARγ binding affinity and transactivation activity of a series of saponins and sapogenins from Medicago species. The triterpenoid sapogenin caulophyllogenin has been identified as a potential PPARγ partial agonist, which has been supported by the first known crystal structure of a sapogenin directly interacting with the nuclear receptor.
Saponins are among the major bioactive compounds, derived from Astragalus corniculatus Bieb. (Fabaceae) (Krasteva et al., 2006, 2007), which is a perennial, herbaceous flowering plant, native to North Bulgaria, Moldova, Southeast Romania and South Ukraine. Previous research showed a protective effect of purified saponins’ mixture (PSM) against myeloid Graffi tumor (Krasteva et al., 2004; Toshkova et al., 2008). There is no information about the hypoglycaemic activity of A. corniculatus but since its major constituents are flavonoids and saponins such effect can be expected (Ionkova et al., 2014).The increased therapeutic interest to novel PPARγ agonists with high efficacy and improved safety profile is accompanied with multiple in silico studies based on pharmacophore modelling (Al-Najjar et al., 2011; Carrieri et al., 2013; Lu et al., 2006; Guasch et al., 2011, 2012; Sohn et al., 2013; Sharma, 2015). The pharmacophore models have been applied in virtual screening (VS) procedures both in the context of drug discovery (Guasch et al., 2012, 2013, Sohn et al., 2011, 2013; Fakhrudin et al., 2010) and predictive toxicology (Tsakovska et al., 2014; Al Sharif et al., 2016). Most of the PPARγ-related pharmacophore-based database screenings have been particularly focused on discovery of novel drug-like PPARγ agonists to serve as lead molecules (Kirchmair et al., 2007; Markt et al., 2008; Sohn et al., 2011, 2013; Lewis et al., 2015), including those from natural origin (Tanrikulu et al., 2009; Fakhrudin et al., 2010; Goebel et al., 2010; Petersen et al., 2011; Guasch et al., 2012, 2013; Kouskoumvekaki et al., 2013).
In this study we performed in vivo and in silico analyses of saponins and sapogenins as compounds from a chemical class, already exploited in traditional medicines and food additives for ameliorating type 2 diabetes. In the in vivo experiments the antidiabetic and antioxidant potential of PSM, isolated from A. corniculatus was investigated in experimental model of streptozotocine/nicotinamide-induced type 2 diabetes in spontaneously hypertensive rats. The in silico studies focused on analysis of the molecular basis for the gradually changing efficacy of the PPARγ partial agonists and investigation of the two main saponins identified in the PSM in relation of their potential PPARγ partial agonistic mode of action. Our results suggest possible involvement of PPARγ-mediated mechanisms in the in vivo antidiabetic and antioxidant effects of PSM from A. corniculatus.
Overground parts of Astragalus corniculatus Bieb. were collected in July 2016 in Northern Bulgaria. The species was identified by Dr. D. Pavlova from the Department of Botany, Faculty of Biology, Sofia University, where a voucher specimen has been deposited (№ SO-95265). The procedure of obtaining the purified saponin mixture was described earlier (Krasteva et al., 2004). Briefly, the air-dried plant material (400 g) was powdered and exhaustively extracted with 50% EtOH. The extract was filtered, concentrated in vacuo and successively extracted with CHCl3 and EtOAc. The aqueous residue was dissolved in MeOH, filtered and precipitated in acetone to give a crude saponin mixture. The precipitate was separated by column chromatography on a Silica gel eluting with CHCl3–MeOH–H2O (98:72:9) and further purified by gel-filtration on Sephadex LH-20 column (MeOH as eluent) to yield a purified saponin fraction. It was named PSM and contained 70% saponins, based on HPLC analysis (column Luna, RP C18, 5 micron, 250 x 4.6; gradient elution from 10 to 100% MeCN for 30 min, wavelength 204 nm). Two triterpene saponins were isolated and identified as 3β -O-[O-4-oxo-pentopyranosyl-(1→2)-β-D- glucopyranosyl]-21α-hydroxyolean-12-ene-28-oic acid (A) and 21 α-hydroxyolean-12-ene-28-oic acid 3β-4-oxo-pentopyranoside (B) (Figure 1). Complete procedure for structural elucidation of saponins was described earlier (Krasteva et al., 2006). For the animal study, PMS was dissolved ex tempore in double distilled water to obtain a dose of 100 mg/mL (Krasteva et al., 2004). The mixture was highly polar due to the saponin presence, thus freely soluble in water. Most saponins have a good absorption in the GI tract, therefore PSM was given orally with a dosage and a treatment regimen as described in details in previous studies (Krasteva et al., 2004; Vitcheva et al., 2013; Ionkova et al., 2014; Krasteva et al, 2016).
Experiments were performed on 24 male spontaneously hypertensive rats (SHR) with initial body weight 200–250 g, obtained from Charles River Laboratories (Sulzfeld, Germany). The animals were housed in Plexiglas cages (3 per cage) at 20 ±2 °C and under 12/12 h light/dark cycle. Food and water were provided ad libitum. All performed procedures were approved by the Bulgarian Food Safety Agency (BFSA) and the principles stated in the European Convention for the Protection of Vertebrate Animals used for Experimental and other Scientific Purposes (ETS 123) (Council of Europe, 1991) were strictly followed throughout the experiment.All reagents used were of analytical grade. Streptozotocine, as well as other chemicals, beta– nicotinamide adenine dinucleotide 2′-phosphate reduced tetrasodium salt (NADPH), reduced glutathione (GSH), catalase (CAT), superoxide dismutase (SOD), nicotinamide and pioglitazone were purchased from Sigma Chemical Co. (Taufkirchen, Germany). 2,2- Dinitro-5,5 dithiodibenzoic acid (DTNB) was obtained from Merck (Darmstadt, Germany).Prior to induction of diabetes, the rats were fasted for at least 16 h. Type 2 diabetes was induced by i.p. injection of nicotinamide (230 mg/kg bw), dissolved in normal physiological saline and after 15 minutes followed by i.p. injection of streptozotocin (STZ) (40 mg/kg bw), dissolved in
0.1 M citrate buffer, pH 4.5. Pretreatment with nicotinamide ameliorates STZ-induces diabetes eliminating the insulin dependence and thus creating a relevant T2D model (Masiello et al., 1998; Ghasemi et al., 2014; Simeonova et al., 2016). Another group of rats which served as control was injected with citrate buffer alone. Forty-eight hours after STZ and nicotinamide injections, diabetes was confirmed by measuring blood glucose concentrations using an Accu- Chek Glucometer (Roche, Germany) in blood samples taken from tail vein. Rats with blood glucose levels of 12 mmol/L or more were considered to be diabetic and included in the study.
Twenty-four male spontaneously hypertensive rats (SHR) were divided into four groups, each consisting of six animals (n=6). SHR were chosen because in this strain the chemically induced diabetes produces more profound effects than it does in normotensive rats (Fein et al., 1984; Krasteva et al., 2016) and they are considered a suitable model for evaluation and examination of oxidative stress, hypertension and diabetes (Cooper, 1997).Group 1 (SHR C) – Control SHR animals, treated for 21 days with the saline vehicle administered by gavage at 5 mL/kg bw/ day. On day 7 of the experiment the animals received an i.p. injection with citrate buffer.On the 22nd day of the experiment blood for glucose analysis was collected from the tail vain of all animals, then the animals were sacrificed by decapitation and livers were taken for biochemical assays. For all following experiments the excised livers were washed out with cold saline solution (0.9% NaCl), blotted dray, weighed, and homogenized with appropriate buffers.Oxidative damage was determined by measuring the quantity of thiobarbituric acid reactive substances (TBARS), expressed as malondialdehyde (MDA) equivalents as described by Polizio and Pena (2005). GSH was assessed by measuring the non-protein sulfhydryls after precipitation of proteins with trichloracetic acid (TCA), using the method, described by Bump et al. (1983).The antioxidant enzymes activities were measured in the supernatant of 10% homogenates, prepared in 0.05 M phosphate buffer (pH=7.4). The protein content was measured by the method of Lowry et al. (1951). Catalase activity was determined by measuring the decrease in H2O2 absorbance at 240 nm and expressed as µmol/mg protein, as described by Aebi et al. (1974).Superoxide dismutase (SOD) activity was measured by decrease of the rate of epinephrine autoxidation (according to the method of Misra and Fridovich (1972)).Statistical analysis was performed using the statistical software ‘MEDCALC’, v. 12.3 (MedCalc Software bvba, Belgium). Results are expressed as mean ± SEM for six rats in each group. The significance of the data was assessed using the non-parametric Mann–Whitney U test. Values of p ≤ 0.05 were considered statistically significant.
In a search of PPARγ partial agonists comprehensive analysis of the Protein Data Bank has been performed (PDB, www.rcsb.org). The initial set of 152 Protein Data Bank entries for PPARγ of human origin has been restricted to those with a single ligand in the receptor pocket and a partial agonism stated in the corresponding literature source (PDB, www.rcsb.org). Then reported in the literature data on transactivation activity (EC50, µM) and relative maximal activation (Emax, %) of PPARγ partial agonists has been collected (Supplementary Table 1). There is not a generally accepted boundary between full and partial agonists. Some authors set a threshold of 60% (Acton et al., 2005), others 70% (Henke et al., 1998), and thirds define the group with Emax between 50% and 80% as intermediate agonists (Bruning et al., 2007). In this study, we set a 65% threshold below which we considered the ligands as partial agonists and thus restricted the initial pool to 37 PDB entries. The modelling set of PPARγ-ligand complexes was additionally processed in order to consist of uniform experimental data by exclusion of any inter- and intra- laboratory variations in the experimental settings. The ligands’ relative efficacy data were restricted to those measured at the most used in such studies experimental conditions: the HepG2 cell line, the protein construct of chimeric PPARγ with DNA binding domain for the β- galactosidase promoter and the referent PPARγ full agonist rosiglitazone. Although satisfying the criteria for inclusion in the basic set, the only triterpenoid sapogenin in complex with PPARγ
(caulophyllogenin, PDB ID: 5F9B) has been left apart for further optimisation of the pharmacophore-based docking protocol for a virtual screening of natural sapogenins. Thus, the core training set consisted of 9 PPARγ-partial agonist complexes (i.e. basic set). The rest 27 PDB entries, 4 of which presenting two alternative binding modes of the co-crystalized ligand, composed an additional training set of 31 PPARγ-ligand complexes which have been used at the refinement step of the pharmacophore modelling (i.e. refinement set). The common triterpenoid aglycon (sapogenin) of the two major saponins in the studied PSM (Figure 1) was built by modifying the structure of caulophyllogenin as extracted from its complex with PPARγ (PDB ID: 5F9B), followed by atomic partial charges calculation and structural optimization using the MMFF94s force field in MOE (Molecular Operating Environment, CCG Inc., v. 2016.0802). The caulophyllogenin structure used in the re-docking studies was subjected to partial charges calculation only. The correct protonation states of the structures have been verified by predicting their ionization forms at a physiological pH = 7.4 using the ACD/Percepta suite (ACD/Percepta Platform, 2015).
The assignment of the correct ionisation states and the positioning of hydrogen atoms in the X- ray PPARγ protein structures, were performed with the MOE tool “Protonate 3D”. The generalized Born/volume integral electrostatics model was used for the titration free energies’ optimization regarding all titratable group and physiologically relevant conditions were set (temperature: 310 K; pH = 7.4; ion concentration: 0.152 mol/L). The addition of hydrogen atoms by this application is related to the determination of: the rotamers of –SH, –OH, –CH3 and –NH3 groups; the ionisation states of acids, bases and metals; the tautomers of imidazoles and carboxylic acids; the protonation state of metal ligand atoms; and the element identities in His and the terminal amides (Asn, Gln).
The analysis of the ligand-receptor interactions in the PPARγ complexes was performed using the MOE tool “Ligand Interactions” to identify potential hydrogen bonds (HB), salt bridges, hydrophobic interactions, cation-π, sulphur-lone pair, halogen bonds and solvent exposure between the ligand and the receptor-interacting entities. The latter include HB residues, close, but non-bonded residues (approaching the ligand, but not having any strong interactions, i.e., HBs), solvent molecules and ions. The tool relies on probability criteria derived from a large training set, particularly for the identification of the protein-ligand HB interactions (Labute, 2001). In this study the default HB scores (expressed as percentages) and directionality notation were applied. The initial pharmacophore hypotheses (pharmacophore queries) were derived on the basis of 9 selected PDB complexes of PPARγ with its partial agonists which were superposed on a template structure (PDB ID: 2I4P) by the C-alpha atoms. The “Pharmacophore Query Editor” tool in MOE was used to generate query features based on three categories of automatically detected ligand annotations: atom, projected and centroid, the latter including bioisosteres. The atom annotations are located directly on an atom of a molecule and typically indicate a function related to protein-ligand binding (these include the H-bond donor (Don), the H-bond acceptor (Acc), cation (Cat), anion (Ani), metal ligator (ML) and hydrophobic atom (HydA).
The projected annotations are located along implicit lone pair or implicit hydrogen directions and are used to annotate the location of possible partners for hydrogen bond or metal ligation or possible R-group atom locations. These include: projected donor (Don2), projected acceptor (Acc2), projected metal ligator (ML2) and ring projection (PiN). The centroid annotations are located at the geometric centre of a subset of the atoms of a molecule: aromatic (Aro), pi-ring (PiR) and hydrophobic (Hyd). By defining a non-zero radius encoding the permissible variation in the pharmacophore query’s geometry these annotation points were converted into query features.The evaluation and refinement of the features involved additional set of 31 PDB complexes of PPARγ partial agonists.The protein structure of the complex with the sapogenin caulophyllogenin (PDB ID: 5F9B, Resolution = 2.25 Å) has been used for docking. The docking site has been defined by ligands’ atoms. The placement method has been set to “Pharmacophore” and the London dG scoring function has been applied to score the generated poses of the docked ligands (Molecular Operating Environment, CCG Inc., v. 2016.0802). The common aglycon of the two PSM saponins was used for the docking studies, since a potential metabolic transformation of these compounds is suggested in vivo. Such evidence is available from the literature for the biotransformations of the steroidal saponin anemoside B4 in rats (Wan et al., 2017).
Results and discussion
The effect of PSM, isolated from Astragalus corniculatus on blood glucose in diabetic SHRs is presented in Table 1. Compared to non-diabetic SHR control rats, streptozotocine/nicotinamide administration increased blood glucose nearly twice (by 160%). In the diabetic rats treated with PSM the blood glucose levels were significantly lower (by 41%, p<0.05) when compared to diabetic rats. What has to be noted is that the antidiabetic effect of PSM was comparable with the effect of pioglitazone. The latter decreased the blood glucose by 40% (p<0.05) compared to the diabetic rats. Our results are in good support of the results reported by Bhavsar et al. (2009). They proved that saponins and a methanol extract isolated from Helicteres isora are effective against hyperlipidemia, hyperglycemia and hyperinsulinemia in the obese and diabetic C57BL/KsJ-db/db mice. Their effects are discerned by reducing the expression of glucose-6- phosphatase (G6Pase) in the liver and fatty acid binding protein 4 (FABP4) in adipose tissue while increasing the expression of adipsin, PPARγ and Glut4 (glucose transporter type 4) in the adipose tissue. Platyconic acid which was one of the six saponins isolated from Radix Platycodi and studied for antidiabetic properties, effectively increased insulin-stimulated glucose uptake in 3T3-L1 adipocytes in vivo (streptosotozine-induced diabetic mice fed a high fat diet) and in vitro (3T3-L1 adipocytes), possibly in part by working as a PPARγ activator (Kwon et al., 2012). Furthermore, an improved hormonal regulation of the insulin sensitivity has been reported as possible mode of action of the PPARγ agonist telmisartan. The latter has been associated with the nuclear-receptor mediated increase in the expression of adiponectin and its receptors (Shen et al., 2017).Type 2 diabetes is characterized with increased oxidative stress, and decreased enzymatic and non-enzymatic antioxidant defense (Giacco and Brownlee, 2010; Tiwari et al., 2013; Haluzík and Nedvídková, 2000).
The T2D-related changes of the oxidative stress marker MDA, the levels of the cell protector GSH and the activity of the antioxidant enzymes SOD and CAT, as well as the effects of PSM and pioglitazone on these parameters are shown in Table 2. T2D oxidative stress induction was discerned by statistically significant (p<0.05) increase in MDA production by 27%, decreased GSH levels by 20% and decreased activity of SOD and CAT by 12% and by 18%, respectively. These results are in good agreement with the effects of diabetes on MDA quantities in plasma, serum, and many other tissues in patients reported in the literature (Bandeira et al., 2012).In our study PSM and pioglitazone administration in SHR with induced T2D, having pronounced and comparable hypoglycaemic effects, resulted in statistically significant and comparable reduction in MDA formation by 15% and 11%, respectively, as well as they restored the GSH levels and the antioxidant enzymes’ activity. When compared to SHR diabetic animals, PSM treatment increased the GSH levels by 24% (p<0.05) and the activities of SOD and CAT - by 9% (p<0.05) and 24% (p<0.05), respectively.The oxidative stress-relieving effect of PSM could be explained by the free radical scavenging, and SOD and CAT activity regulation by saponins, as previously shown for similar compounds isolated from different plant sources (Alli Smith and Adanlawo, 2014; Akinpelu et al., 2014). The antioxidant potential of PSM has been proved in enzyme- and non-enzyme-induced lipid peroxidation in liver microsomes, isolated from spontaneously hypertensive and normotensive rats (Simeonova et al., 2010). On the other side, although the antioxidant effects of pioglitazone have been observed in diabetes, it did not exert any direct antioxidant action in vitro and in non- diabetic animals (Kakadiya et al., 2010; Yin et al., 2013; Surapaneni and Jainu, 2014; Chandrasekar et al., 2015a, 2015b; Abd El-Twab et al., 2016; Ogunlana et al., 2017).
It has been shown that the indirect antioxidant action of pioglitazone involves a PPARγ-mediated mechanism (Sun et al., 2016). The protective role of the PPARγ activation against oxidative stress has been explained by a redox homeostasis modulation through the mitochondrial uncoupling protein 2 (UCP2) (Chan et al., 2010; Cabrera et al., 2012; Wang et al., 2014b), known targets of PPARγ (Villarroya et al., 2007). Increased UCP2 expression has been shown to decrease the superoxide anion levels (Cabrera et al., 2012). Another possible mechanism of the PPARγ-dependent antioxidant response, has been reported to be mediated through its targets retinol-binding protein 7 and adiponectin (Hu et al., 2017), the latter being shown to suppress ROS induced by high glucose (Ouedraogo et al., 2006).
The similar patterns of the hypoglycaemic and oxidative stress-relieving effects of the PSM and pioglitazone suggest that they could share common mechanisms of antidiabetic and antioxidant action involving PPARγ signalling. Therefore, the in vivo results for the PSM-mediated effects prompted the in silico studies of the two major saponins in the mixture in order to investigate their possible relevance to PPARγ modulation.The number of the crystallographic complexes of PPARγ with its partial agonists, with known relative efficacy data are constantly increasing, in the PDB. This offers new opportunities for interpretation of the gradually changing effects of partial agonists’ subclasses exhibiting different binding modes by a comparative analysis of their protein-ligand interactions.
The analysis of the collected data clearly outlined two subclasses of partial agonists that can be discriminated relying on the correspondence of the Emax-based classification of the basic set (“strong”, 33% – 50% and “weak”, 10% – 27%) to the clustering according protein-ligand interactions’ patterns (Table 3) and the occupation of different subregions of the receptor’s pocket (Figure 2)agonists’ pattern reported in our earlier study (Tsakovska et al., 2014). Figure 2A illustrates that the strong partial agonists occupy either pocket Arm I (between the activation helix H12 and helix H3, dark blue sticks) or Arms II and III (between the β-sheet, helices H3 and H8, black sticks, see Figure 2A and B for the assignments of the structural elements and arms) thus differing from the full agonists which simultaneously participate in interactions in Arms I and II (Figure 2A, C). This could explain the lower capacity of the strong partial agonists to stabilise H12, compared to the full agonists, and their submaximal efficacy (Emax ≤ 65%). The weak partial agonists occupy the region between the β-sheet, H3 and the Ω-loop, partially overlapping the binding site of the strong partial and the full agonists. Their positioning in the receptor pocket is mediated by interactions with Arg288 and Ser342 (Table 3). It is suggested that this binding mode stabilises H3 and favours protein contacts involving H3 and the H11-H12 loop, thus indirectly stabilizing H12 (Hughes et al., 2014).
The initial pharmacophore models for the two subclasses of partial agonists were outlined using the basic set. After evaluation of the features, using the refinement set, their radius, coordinates or type were edited. The final arrangements of the strong- and weak-partial agonists’ pharmacophore features are presented in Figure 3 and a detailed description of the features is given in Supplementary Table S2.
contacts with several amino acid residues in this region (Ser289, His323, Tyr 327, and His449) are reflected by a single pharmacophore feature, i.e. F3 (Figure 3B). However, such restricted spatial arrangement of the possible HB interactions is not observed for the smaller and more mobile weak partial agonists.
In particular, the F2 and F3 HB acceptor features of the “weak” pharmacophore model (Figure 3D) describe HB acceptor contacts with more than one amino acid residue, either direct (Arg288 or Ser342), or through a bridging water molecule (Leu288 or Glu343). On the other side, Arg288 due to its high conformational mobility and Ser342, due to the peculiar spatial arrangement of the sidechain and backbone groups justified the generation of more than one pharmacophoric HB acceptor feature (F1 and F3 for Arg288, and F2 and F3 for Ser342). As already mentioned, the different strong partial agonists may occupy particular subareas of the region characteristic for the full agonists (Figure 2A), which is described by the two alternative and partially overlapping arrangements of the “strong” pharmacophore’s features F1-F4 and F3-F6 (Figure 3B). The “weak” pharmacophore’s features F6 and F7 reflect the possible alternative localization of hydrophobic/aromatic ligand substructures observed in different complexes (Figure 3D).Considering these specific interactions, the pharmacophore models were designed to be redundant so that the pharmacophore features to cover all of them. However, the possible partial agonists are not expected to match simultaneously all the features but at least 3 of them.During the refinement, ligands from 8 complexes of the strong and 2 complexes of the weak partial agonists’ refinement sets were considered as outliers, because their Emax-based classification did not correlate to the subclass-characteristic ligand’s binding mode, i.e. the number of pharmacophore features matched in the pharmacophore of the other Emax-based subclass (cross-matches) was higher than those matched in the own Emax-based subclass(Supplementary Table 3). Analysis of these outliers allowed for further interpretation of the experimental data (biological and crystallographic). In the case of the PDB ID 4CI5, two different binding modes of a single ligand are observed.
Relying on the activity data, we classified the ligand as a weak partial agonist (Emax = 21%), which was further confirmed by the localization of the ligand in the PPARγ pocket (chain B). However, according to our concept for a subclass-specific binding mode of the partial agonists, the same ligand in chain A can be associated with the strong partial agonists’ subclass. This PDB complex supports the possibility that switching between alternative binding modes could result in a submaximal effect at a molecular level. Whether the experimentally measured efficacy is related to the weak partial agonist binding mode only or it is a result of a gradual transition between the two observed poses passing through an array of multiple intermediate binding modes remains unclear. The strong partial agonist-like binding mode refers to one acceptor and one aromatic features of the “strong” pharmacophore and an aromatic feature of the “weak” pharmacophore (Supplementary Figure 3). It should be noted that the two aromatic features are partially overlapping. The alternative binding mode matches equal number of features – 3 (1 acceptor and 2 aromatic) from the two pharmacophores. The same issue, related to the alternative binding modes, is valid for the only covalently bound (to Cys285) agonist within the modelling set, (PDB ID 3X1I). The ligand from this complex has been reported to possess Emax = 55% and thus was initially included in the strong partial agonists’ refinement set. Nevertheless, the ligand with a weak partial agonist’s binding mode covers 4 features of the corresponding pharmacophore while the other pose matched only 2 features per each pharmacophore model. Alternative binding modes were observed also in the two chains of PDB entry 2YFE, both ligands occupying the sub-region typical for the weak partial agonists and matching 3 or 5 features of the “weak” pharmacophore.
Among the outliers remained also the PDB 4F9M ligand which has been reported to possess relative efficacy of 64 or 38% in two separate studies. This ligand, however, occupied the weak partial agonists sub-pocket matching 5 out of 7 points of the corresponding pharmacophore and only 3 of the “strong” pharmacophore. The difference of 26% in the Emax values of this ligand, tested in different experimental settings, justifies the initially postulated criteria for selection of the basic set in the current study (Section 2.2.1.).The remaining outliers in the strong agonists’ subset include the following complexes with Emax> 30% and “weak” pharmacophore feature matches reported in the brackets: 2Q6S (60%, 5), 3LMP (47%, 4), 3H0A (34%, 3), and 3B1M (33%, 5). The ligand from the complex 2YFJ (Emax = 18%) is located at the interface of the two binding modes, matching 1 “weak” and 4 “strong” pharmacophore features (Supplementary Table 3).Figure 4 summarises the numbers of simultaneously matched / cross-matched features of both pharmacophores for the two partial agonists’ subclasses excluding the complexes of the binding mode outliers D. Exclusion of the outliers allowed for a clearer estimation of the features’ matching frequency (Figure 5, C and D). For the weak partial agonists’ pharmacophore, the most frequently matched features were F2 (Acc), F4 and F5 (Hyd|Aro) while for the “strong” pharmacophore these were F3 (Acc), F2 and F4 (Hyd|Aro). The high rate of cross-matches even after removal of the outliers could be explained by the partial overlap of the “weak” and “strong” pharmacophore features as follows: F7 (Don&Acc; “strong”) overlaps F2 and F3 (Acc; “weak”); features F4 (Hyd|Aro) from the two pharmacophores share common subarea; and the Hyd|Aro features F5 and F7 from the “weak” pharmacophore cover entirely the corresponding F6 from the “strong” pharmacophore.
However, the tendency for prevalence of the cross-matches with the other pharmacophore is typical for the strong partial agonists’ pharmacophore while for the “weak” pharmacophore features higher rate of class-specific matches vs. cross-matches with the other pharmacophore is recorded. This observation is not surprising considering that the strong partial agonist “pass through” the active site preferred by the weak partial agonists before reaching their own active site.Attempts have been made to differentiate between full and partial agonists in previous studies. Ligands’ similarity studies have suggested possible clustering of PPARγ partial agonists (Seto et al., 2010; Guasch et al., 2012). Additionally, Vidović et al. (2011) identified a partial agonist-like ligand cluster within a binding mode similarity dendrogram based on analysis of co-crystallised PPARγ modulators. Recently Lewis et al. (2015) selected criteria for filtering the full agonism activity type. Moreover, Guasch et al. (2012b), developed separate pharmacophore models for full and partial agonists of PPARγ to apply them in VS of natural ligands with partial agonism.
The in silico studies, published in the literature, with focus exclusively on partial agonist-based pharmacophore modelling using ligand-, structure- and shape-based approaches are summarised in Supplementary Table 4. These 19 pharmacophore models were compared with the two models reported in the current study and the rate of matching with the individual pharmacophore features was analysed (Figure 6, Supplementary Table 5). The pharmacophore models proposed by us are composed by features which in а different arrangements match some of the pharmacophore points previously identified by other authors. Overall, 6 out of 7 features in the strong partial agonists’ pharmacophore (Figure 6A) and 5 out of 7 features in the weak partial agonists’ pharmacophore (Figure 6B) were confirmed by previous studies. Analysis of the features of the strong / weak partial agonists’ pharmacophores underlined F6 / F4 and F7 / F2 as the most frequently reported by other authors Hyd|Aro and H-bond-related features, respectively (Figure 6). The pharmacophore points F1 (Hyd|Aro) from the “strong” pharmacophore, and F1 (Acc) and F6 (Hyd|Aro) from the “weak” pharmacophore have not been reported for partial agonists by other authors partial agonists and antagonists. In their study the importance of the ligand’s localization within the binding cavity of PPARγ for determination of the activity class has been emphasised.
The current study stresses on a more detailed subdivision of the partial agonists into strong and weak ones and involves pharmacophore generation and refinement based on a large training set. Analysis of positions and interactions of the strong partial agonists questions the general understanding that the location of the partial agonists in the pocket is associated only with ligand’s positioning close to the H3 and the β-sheet while the features closer to H12 are typical only for full agonists (Lewis et al., 2015). Our analysis, illustrates that partial agonists can also be located between H12 and H3, indicating a more complex behaviour of the partial agonists in the binding site. Thus, HB acceptors’ clustering in the vicinity of H12 and the stabilisation of H3 give rise to alternative mechanisms for PPARγ partial activation. These are addressed by the pharmacophore models suggested by us through involvement of the most typical features of the two subclasses of partial agonists. Furthermore, as observed in the analysis of the outliers, the molecular determinants of partial agonism, rather than being strictly associated with a single sub- region of the pocket, could be related to an ensemble of binding modes involving subpopulations of the interactions exhibited by all activity types.
In summary, we outlined two subclasses of PPARγ partial agonists using a clearly defined criterion for partial agonism to differentiate between strong and weak partial agonists. Such approach has not been reported so far. Moreover, both groups of partial agonists occupy preferentially different regions of the receptor binding site with some common partially overlapping pharmacophore points.In order to gain insights on the putative molecular mechanism underlying the hypoglycaemic action of the studied purified saponins’ mixture we performed in silico analyses with the aglycon of the two main saponins identified. First, the structure of caulophyllogenin (PDB ID: 5F9B) has been re-docked in the receptor binding site of the corresponding protein structure in order to optimise the docking protocol. Multiple simulations have been performed by applying various “weak” pharmacophore models differing in the number and type of the pharmacophore features considered (all features, exclusion of features 1, 6, and 7 one at a time or simultaneously). Each simulation was performed twice and produced identical results. The dockings generated ligand poses similar to the X-ray pose. The frequency of reproducing the experimentally observed binding mode and the scoring ranges varied among the different docking protocols as shown in Supplementary Figure 1. The 7-point pharmacophore-based docking (F1-F7) did not reproduce the initial pose.
All 30 poses of the ligand generated from each docking simulation have been further subjected to analysis of the binding mode and the characteristic receptor-ligand interactions have been recorded. Although the RMSD of 2Å is a generally accepted cut-off value for selecting the best docking poses (Ballante and Marshall, 2016), an array of important protein- ligand interactions was observed for poses slightly beyond this threshold. The contacts which these poses exhibited were in agreement with the “weak” subclass’ HB pattern outlined in the analysis of the PPARγ-ligand complexes (Section 2.2.1.). Moreover, а satisfactory reproduction of the binding mode was maintained. Therefore, we considered all poses falling within a threshold extended up to 2.5Å. The rate of the poses successfully reproducing the X-ray pose varied within the different docking protocols. Their corresponding docking scores covered the whole range of the top-30 poses. Analysis of sub-populations within the top-15, allowed for selection of a virtual screening protocol resulting in the highest number of successful poses within a narrowed range of the best docking scores (Supplementary Figure 1). Based on the validation over the X-ray structure of caulophyllogenin, the docking protocols involving pharmacophore models F2-F5, F2-F6 and F2-F7 were selected for further in silico analyses of the main sapogenin found in the experimentally tested PSM. The docking simulations resulted in arrays of top-30 poses with scores ranging between -11.6 and -8.5 (Supplementary Figure 2) and a high frequency of reproduction of caulophyllogenin’s binding mode as follows: 24 (F2- F5), 23 (F2-F6) and 28 (F2-F7) poses. The PSM sapogenin’s poses resulting from the different docking simulations were observed to participate in protein-ligand interactions with Tyr327, Gly284, Met364 and Ser342, the latter being typical for the weak partial agonists’ subclass.
Putting together the in vivo studies of the purified saponins’ mixture and the in silico simulations of the PSM sapogenin’s interactions, our results direct to the potential modulation of PPARγ by partial agonistic activity as one of the possible mechanisms related to the hypoglycaemic effect of the studied saponins. These results are in accordance with the already observed hypoglycaemic effects of other representatives of this chemical class (Lee et al., 2000; Liu et al., 2012; Smith et al., 2012; Elekofehinti et al., 2013) and with the PPARγ-mediated mechanisms of the pioglitazone’s antidiabetic and antioxidant action demonstrated also in our study. Our finding for the possible PPARγ weak partial agonistic mode of action of the investigated PSM sapogenins is in accordance with the study of Petersen et al. (2011) who have applied a pharmacophore-based virtual screening in the PPARγ pocket and have discovered a very similar triterpenoid sapogenin (oleanoic acid) which PPARγ partial agonistic activity has experimentally been confirmed (~20% Emax). Furthermore, the structural similarity between the studied sapogenin and caulophyllogenin (PDB ID 5F9B) with proven relative efficacy of 9.4% (Montanari et al., 2016) and the high rate of reproduction of the caulophyllogenin’s binding mode by the PSM sapogenin in the docking simulations, support the relevance of the studied PSM to PPARγ. However, the action of the compounds as PPARγ agonists remains to be further proven experimentally, for example by the use of receptor binding and cell-based transactivation studies. Not only could such experiments serve as a bridging evidence between our in silico predictions and the observed in vivo effects but also a quantitative comparison/ranking of the studied sapogenins and their derivatives will be possible based on these assays.
Conclusions
In line with the global tendency for natural product-based drug discovery and development we combined in vivo and in silico methods for: (i) investigation of the potential PPARγ-mediated mechanism of antidiabetic effect of saponins from A. corniculatus and (ii) identification of the underlying molecular determinants of PPARγ partial agonism. An emphasized antidiabetic effect comparable to that of the referent PPARγ agonist pioglitazone was observed upon treatment of spontaneously hypertensive rats with a purified saponins mixture from A. corniculatus.Significant oxidative stress-relieving effects, with a pattern similar to that of the pioglitazone, were also reported. Two pharmacophore models of PPARγ partial agonists located in discrete sub-pockets of the receptor were developed in the context of a relative efficacy-based discrimination between strong and weak partial agonists. The results from a pharmacophore- based docking of the major PSM sapogenins, using the weak partial agonists’ pharmacophore, suggested a potential PPARγ-mediated mechanism as a molecular basis of the observed antidiabetic and oxidative stress-relieving effects of the PSM. Nevertheless, the action of the PSM compounds as direct PPARγ agonists still remains to be directly proven experimentally (e.g. by using a PPARγ antagonist). This warrants further in vivo / in vitro studies involving single saponins, sapogenins and their in silico simulations (e.g. molecular dynamics) in the receptors’ pocket. Such studies will allow for additional characterisation of the potential pharmacological effects of the studied saponins and further interpretation of the data in the context of different disease models, relevant to the metabolic syndrome. The developed docking protocol can be applied for a pharmacophore-based virtual screening of other natural saponins and has the potential to direct the scientific effort toward in vitro testing and structure optimisation of natural triterpenoids as potential Sapogenins Glycosides T2D modulators.