Hyaluronidase Inhibitory Activity of Pentacylic Triterpenoids from Prismatomeris tetrandra (Roxb.) K. Schum: Isolation, Synthesis and QSAR Study

The mammalian hyaluronidase degrades hyaluronic acid by the cleavage of the β-1,4-glycosidic bond furnishing a tetrasaccharide molecule as the main product which is a highly angiogenic and potent inducer of inflammatory cytokines. Ursolic acid 1, isolated from Prismatomeris tetrandra, was identified as having the potential to develop inhibitors of hyaluronidase. A series of ursolic acid analogues were either synthesized via structure modification of ursolic acid 1 or commercially obtained. The evaluation of the inhibitory activity of these compounds on the hyaluronidase enzyme was conducted. Several structural, topological and quantum chemical descriptors for these compounds were calculated using semi empirical quantum chemical methods. A quantitative structure activity relationship study (QSAR) was performed to correlate these descriptors with the hyaluronidase inhibitory activity. The statistical characteristics provided by the best multi linear model (BML) (R2 = 0.9717, R2cv = 0.9506) indicated satisfactory stability and predictive ability of the developed model. The in silico molecular docking study which was used to determine the binding interactions revealed that the ursolic acid analog 22 had a strong affinity towards human hyaluronidase.


Introduction
Hyaluronic acid (HA) is a polymer of varying chain length composed of a repeating dissacharide unit, N-acetylhyaluronic acid, linked via the hexosaminidic bonds in β-(1Ñ4) linkages [1]. It usually consists of 2000-2500 dissacharides to give a molecular mass between 10 6 to 10 7 Da and extended lengths of 2-25 µm [2]. It can be found in an extracellular matrix, especially in soft connective tissues of all vertebrates and in the capsule of some bacteria [2]. HA plays an important role in biological processes such as cellular adhesion, mobility differentiation processes that act as lubricant and shock absorber, regulates water balance and osmotic pressure [3]. It is also the structure-forming molecule in the vitreous humor of the eye, in Wharton's jelly and in joint fluids.

Synthesis
Compounds 4-10 were synthesized from ursolic acid 1. Structural modifications were carried out at positions C-3 and C-28 of ursolic acid 1 (Figure 1). The modifications were carried out either by acetylation, methylation or amino group introduction or by combination of the three methods. Compounds 11-30 were purchased from Chromadex ® which was obtained from natural sources i.e., plants. The structures were confirmed on the basis of their 1 H NMR, 13 C NMR, and ESI-MS spectroscopy data and upon comparison with literature values [25,26].

Hyaluronidase Inhibitory Activity
The study was conducted on compounds 1-30. The assay was performed according to the modified Sigma protocol [27]. The IC 50 values for the inhibitory activities of compounds 1-30 are presented in Table 1. The results showed that out of the 30 compounds, only 24 were found to exhibit activity below 2000 µM. Oleanolic acid methyl ester 16 (84.52 µM) and carbonexolone 22 (56.33 µM) showed higher activities compared to ursolic acid 1. showed higher activities compared to apigenin, the positive control. Compounds 25-30 on the other hand were inactive as their inhibitory activities were less than 20% at concentrations of up to 2ˆ10 3 µM (Table 1). All compounds were prepared at an initial concentration of 2000 µM before serial dilution. Thus, if the inhibition towards hyaluronidase at 2000 µM was less than 20%, the compound will be considered as inactive. The highest concentration was prepared less than 2000 µM to avoid the solubility problem.   H  OH  H  H  OH  H  H  NA  30  ------------NA  Apigenin  ------------214.74 a NA-inhibitory activity, 20% at concentration up to 2000 µM; positive control-Apigenin. Values were presented as the mean of three independent experiments performed in triplicate; * Mean for percentage inhibition were significantly different (one-way analysis of variance, p < 0.05); ** Mean for percentage inhibition were significantly different (one-way analysis of variance, p < 0.005).
For the oleanane skeleton, the activity reduced slightly when the methylhydroxyl group was introduced at C-23 (21 vs. 25). However, the activity increased when the methyl group was introduced at C-17 (18 vs. 20). The activity was also increased when the hydroxyl group was introduced at C-16 (15 vs. 20). The C-30 ester derivatives resulted in a great loss of activity (11 vs. 21), while the carboxylation of the same carbon (C-30) increased the activity (21).
Acetylation of 3-OH decreased the inhibitory ability (3). Introduction of a sugar moiety with a glycosidic bond to 3-OH or 28-COOH would either reduce the activity drastically (17) or become inactive, whereas the addition of an oxo group to C-11 either did not improve the activity or reduced it slightly (21 vs. 22 vs. 11). Too many hydroxyl groups as in 28 also resulted in a loss of activity. Significant improvement in the activity was observed when a methylester group was introduced at C-17 (16 vs. 20, 18) or when a carboxypropanoyloxy group, S2, was introduced at C-3 (22).
The inhibitory activity of the ursane skeleton analogues decreased for the 3-oxo, 3-hydroxyimino and 3-acetylate derivatives (4, 5, 6) compared to ursolic acid 1. A similar trend was observed when the 28-OH was substituted with a methyl group (5 vs. 8, 1 vs. 9, 6 vs. 10). This observation suggested that the hydroxyl groups at C-3 and C-28 were essential for the hyaluronidase inhibitory activity. The substitution of a methyl group with a carboxyl group at C-23 decreased the inhibitory activity (14 vs. 26, NA). The introduction of a hydroxymethylene group at C-23 and the addition of a hydroxyl group at C-1 did not affect the activity very much (19 vs. 26, NA), however, the addition of a hydroxyl group at C-19 further decreased the activity (2). The position of the hydroxyl group also affected the activity as it decreased the activity for 2 (C-19) compared to 19 (C-2). The sugar moiety, as usual, would result in a loss of activity (27,29).
It can be concluded that the presence of 3-OH is important for both ursane and oleanane skeletons to inhibit hyaluronidase. Introducing a methyl ester group at C-17 to replace the carboxylic functional group resulted in a different effect on the hyaluronidase inhibition ability of both types of skeletons. Replacement of the 3-OH with hydroxyimino, acetyl and methyl groups lowered the activity. Addition of hydroxyl groups at C-2, C-1 and C-19 or oxo groups at C-11 and C-3, however, would either give no effect or decrease the activity. Introducing a sugar moiety at any position would result in a loss of activity. This could probably be due to the bulkiness of the structure that led to the compounds being unable to reach the active site of the target, which was in the hyaluronidase enzyme. However, the carboxypropionylox group, S2, that replaced the 3-OH, increased the activity. Figure 2 summarizes the structure activity relationship of the PTC compounds .
Even though these simplistic structure activity relations are useful for a single functional group substitutional level, it is not easy to predict the cumulative effect of several substitutions on the hyaluronidase inhibition activity using these relations. Hence, it is necessary to adopt the QSAR approach to analyze and predict the effect of substitution on the activity.

QSAR Model and Its Interpretation
The data set from the hyaluronidase inhibitory activity for compounds 1-24 were converted into log IC 50 in order to improve the normal distribution of the experimental data points [28]. After analyzing the data set using the Heuristic method and Best Multi Linear regression method (BML), it was found that the QSAR model from the BML method was statistically robust compared to the Heuristic method.
The data was divided into the test set (8,11,19,6) and the remaining data into the training set. From the data set which consisted of 20 compounds, eleven models were obtained which consisted of two to twelve descriptors as listed in Table 2. The R 2 values as well as the other statistical values also improved (close to 1). This method managed to avoid over fitting of the regression equations by monitoring the increase of R 2 in the equations with successive number of descriptors involved. The procedure is called the break point technique [28]. The procedure was stopped when the difference between R 2 of the two consequent regression equations was less than or equal to 0.02 [28]. However, the best optimum correlation should have the ratio of the data set compounds to the descriptor at 5:1 [29]. Meaning that, there is a descriptor that represents five data points. Thus, from the data set, the optimum descriptor number was four. The best QSAR model was developed using four descriptors. The p value is less than 0.01 for each descriptor involved in the model generation. These descriptors were selected, as the addition of more descriptors does not lead to any significant improvement in the correlation. A plot of the experimental vs. predicted IC 50 values is depicted in Figure 3 for the 20 PTs (1-5, 7, 9, 10, 13, 24).  The QSAR equation relating the seven descriptors to the hyaluronidase inhibitory activity is given below: The correlation coefficients (R 2 ), cross validated correlation coefficient (R 2 cv ), Fisher criterion values (F), regression coefficient (X) and standard errors for the regression coefficient (∆X) and t-test values corresponding to Equation (1) are given in Table 3. The R 2 value of 0.8579 with the R 2 cv value of 0.7196 showed the good predictive power of the developed model.
The charge distribution-related or electronic descriptor (Q i , E nn ) and quantum chemical descriptor (V h , E c ) were found to influence the activity of PTs in the QSAR equation. The t-test values indicated that the statistical significance of the selected descriptors in the QSAR model, decreased in the order: It is known that the local electron densities or charges determine the mechanism and the rate of most chemical reactions and physico-chemical properties of compounds. The valence electrons in molecules are not fixed to any particular atom but can move around the molecule. The electrons will be more at electronegative atoms compared to electropositive ones, thus resulting in the molecules being partially negative while the others partially positive [30]. In Equation (1), two electronic descriptors were involved. Minimum partial charge for a carbon atom (Zefirov's PC), (Q i ) descriptor or partial charge is important for the ionic interactions between the drugs and its binding site on the receptors. The positive regression in the model in Equation (1) showed that the bigger the partial charge in the molecule, the higher the inhibition towards hyaluronidase.
Total molecular surface area (E nn ) is a combination of the contribution of atomic partial charges to the total molecular solvent-accessible surface area. This descriptor and the minimum partial charge for a carbon atom (Zefirov's PC) (Q i ) suggest the importance of the interaction between the inhibitor molecular surface area with solvent.
Two quantum chemical descriptors were involved in the selected model. The first quantum chemical descriptor was the minimum valency of the H atom (V h ). It describes the atomic valence state for the energies of the given atomic (H) species in the molecule and its fragments [30]. It characterizes the magnitude of the perturbation experienced by an atom in the molecular environment as compared to the isolated atom.
The maximum bond order of a C atom descriptor, E c , is categorized as a valency-related descriptor. This descriptor is related to the strength of the intramolecular bonding interactions and characterizes the stability, conformational flexibility and other valency-related properties of the molecules. This suggests the importance of the CO group for the interaction between the inhibitor and biological receptor.
From the predicted log IC 50 values, it is clear that the QSAR equation generated through the quantum chemical method predicted that the pIC 50 values were very close to the experimental values (Table 4).

Design of a New Potential Pentacylic Triterpene
After repeating for seven times in designing new compounds for this pentacylic teriterpene data set, below is the structure of the new design with the highest activity, which was indicated by the smallest pIC 50 value.
Several structures have been designed continuously until Equation (1) gave the smallest value for log IC 50 . This new compound PTC A consist of a carboxypropionylox group, S2, a hydroxyl group at carbon-12 and a methyl group at carbon-17 which was believed to give rise to the biological activity of PTC A (Figure 4). Equation (1) provided a prediction value of pIC 50 as 1.6183 for this compound, which was the smallest compared to compounds 1-23 and the other designed molecules.

Method Validation of the Proposed Model
The most important test of the model is its ability to correctly predict the properties of other or new compounds that were not included in the QSAR model. The leave-one-out method technique is based on the difference between the squared cross-validated correlation coefficient (R 2 cv ) and the correlation coefficient (R 2 ) [31,32]. The corresponding R 2 cv for all selected models will be calculated automatically by the validation module, which was implemented in the CODESSA 2.6 package. The value of R 2 cv (0.7196) was found to be close to the value of R 2 (0.8579). The difference was less than 0.3, which suggested good predictive ability of the selected best multi linear model [33].
The external validation is a more reliable way to establish a predictive QSAR model [34]. The correlation coefficient prediction (R 2 pred ), which was based only on the molecules present in the test set, should be more than 0.5. The value of R 2 pred for the training set of this model was 0.5831, which was more than 0.5.

Possible Interactions from an in Silico Molecular Docking Study
Molecular docking study was used to clarify the binding mode and identify the interaction of inhibitors with our targeting protein, human hyaluronidase. The flexible docking result with AutoDock Vina indicated that compound 22 was more active than apigenin with the negative binding affinities range of´8.5 to´7.6, where those of apigenin was in the range of´7.8 to´7.2. The lower (negative) number indicated the stronger binding of the compound with the protein receptor. The selected docked complex was further minimized and visualized for their interactions with CHARMM force field against hyaluronidase in Discovery Studio in Figure 5. In the representation, apigenin and compound 22 in orange, and green, respectively, were superimposed to compare their interactions. A close view of the interactions has been depicted, whereas the orange and green dotted lines represented the pi-pi interactions and hydrogen bonds. The details of the van der Waals (VDW), electrostatic, binding interaction (BI) with amino acids within 4 Å vicinity of the compound and total interaction energy (IE) value are tabulated in Table 5. All the results from binding affinity with AutoDock Vina, BI with amino acid residues within 4 Å, and IE using CHARMM forcefield are in agreement with the experimental results and confirmed the stronger inhibition of compound 22 against hyaluronidase than apigenin. The residues with strong interaction energy below´4 kcal/mol for apigenin are TYR75, TRP321, SER323, TRP324, THR327, where more residue interactions with ASN61, PRO62, TYR75, SER77, GLN78, TYR84, ASP129, TRP321, TRP324 for 22 are found. Apigenin did not bind well with one of the reported binding sites. ASP129, however, the π-π interaction between apigenin with TYR75 and no hydrogen bonding were observed.

Chemicals and Instruments
All chemicals were obtained from commercial sources (Aldrich, Merk, and Sigma, Darmstadt, Germany) and used without further purification. Solvents were used without further purification or drying, unless stated otherwise. Reactions and isolations were monitored using thin layer chromatography (TLC) (aluminum supported silica gel 60 F254 plates were used for TLC). TLC spots were visualized under ultra-violet light (254 and 365 nm). The plates were then sprayed with 10% sulphuric acid followed by heating using a hot plate to detect the presence of phenolics and terpenes, which were indicated by the presence of colourful spots. Several packing materials were used for column chromatography i.e., MCI gel CHP 20P, Sephadex LH-20, Chromatorex ODS, silica gel 60 (70-230 Mesh ASTM or equivalent to silica gel of size 0.063-0.200 mm). The infrared spectra were recorded on a Perkin Elmer Spectrum 100 Fourier Transform Infrared (FT-IR) spectrometer (Perkin Elmer, Waltham, MA, USA) equipped with a mid-infrared deuterated triglycine sulphate (DTGS) detector. NMR analyses were carried out on a Bruker DRX 300 NMR spectrometer (300 MHz for 1 H NMR and 75 MHz for 13 C NMR, Bruker corporation, Billerica, MA, USA) system with deuterated pyridine (C 5 D 5 N). The mass spectra were obtained using an LTQ Orbitrap Mass spectrometer (Thermo Fisher Scientific, Bremen, Germany) equipped with an electrospray ionization probe by employing either a negative or positive ion mode, whichever could afford the best limits of detection for the compounds.

Plant Material
P. tetrandra was collected from Setiu, Terengganu on the 29 January 2006 and was deposited in the herbarium of the Forest Research Institute Malaysia (FRIM) with a herbarium specimen number of FRI 50080. The sample was identified by an FRIM botanist. The plant material was dried in an oven at 40˝C, divided into different parts and ground.
The cut and oven-dried roots (2.2 kg) were ground and extracted with 5 L of MeOH by soaking three times at room temperature. The concentrated extract (100 g) was suspended in H 2 O and partitioned with EtOAc (5 L, 3ˆ). The EtOAc layer after drying under reduced pressure (15.6 g) was chromatographed over silica gel and eluted using petroleum ether with increasing amounts of acetone to give eight fractions. Fraction 3 was further fractionated and purified by a silica gel column eluting with a hexane-CHCl 3 eluent system to afford 3β-acetylolean-12-en-28-oic acid 3 (20 mg, 100:0 v/v).

Synthesis of Ursolic Acid Analogues
Ursolic acid 1 was selected as the lead compound and seven derivatives 4-10 were prepared from the semisynthesis of it. The modification methods were carried out as described by Ma et al. [16], with a slight modification in the solvents and reagents used. The reaction scheme used for the synthesis of the ursolic acid 1 derivatives is shown in Figure 1.

3-Oxo-urs-12-en-28-oic Acid (4)
Pyridinium chlochromate (PCC) (161.3 mg, 0.75 mmol) was added to a solution of ursolic acid 1 (112.3 mg, 0.25 mmol) in a acetone-dicholoromethane (5:5, 10 mL) mixture. After stirring at room temperature until the reaction was almost complete (monitored with TLC), the mixture was concentrated and partitioned with H 2 O and CH 2 Cl 2 . The CH 2 Cl 2 layer was concentrated and purified by silica gel column chromatography eluting with hexane: acetone (95:5 v/v) to give 4 Yield 79.34%, white amorphous powder. 1 (6) Ursolic acid 1 (120.0 mg, 0.26 mmol) was treated overnight with acetic anhydride (534.0 mg, 5.32 mmol) and pyridine at room temperature and worked up with 10% HCl, NaHCO 3 , followed by separation using a separating funnel to get the CH 2 Cl 2 layer. MgSO 4 was added to absorb water from it. The solution was filtered and rinsed using CH 2 Cl 2 . It was purified over a silica gel column eluted using chloroform: petroleum ether (90:10 v/v) to give 6. Yield 50%, white amorphous powder. 1 (9) To a stirred solution of ursolic acid 1 (10.0 mg, 0.02 mmol) in approximately 10 mL of toluene: MeOH (3:2), a solution of TMSCHN 2 (trimethylsilane diazomethane) in hexane was added drop wise until the yellow color persisted. The mixture was stirred at room temperature and concentrated. It was purified over a silica gel column with hexane: CHCl 3 (70:30 v/v) as the eluent to give 9. Yield 95%, white amorphous powder. 1 (7) Methyl ursolate 9 was treated with PCC in the same manner as was carried out for compound 4 to obtain compound 7. The resultant was purified over a silica gel column eluted with chloroform: petroleum ether (90:10 v/v). Yield 90%, colorless crystal. 1  3.4.6. 3-Hydroxyimino-urs-12-en-28-oic Acid Methyl Ester (8) A solution of 7 (80.0 mg, 0.17 mmol) and hydroxylamine hydrochloride (108.0 mg, 1.56 mmol) in pyridine was heated for 2 h at 50˝C. It was then cooled to room temperature and diluted with CH 2 Cl 2 followed by washing with 10% HCl (3ˆ). It was then dried over anhydrous Na 2 SO 4 and concentrated under reduced pressure. It was purified over a silica gel column using CHCl 3 : hexane (40:60 v/v) as the eluent to give 8 Yield 95%, white amorphous powder. 1 (10) Compound 9 was treated with TMSCHN 2 using the similar method, which was used for the preparation of 9 to give 10. Yield 95%, colorless crystal. 1

Hyaluronidase Inhibitory Assay
The assay was performed following the modified Sigma protocol [27].

Construction of a QSAR Model
All the molecular structures were built by the Chem3D Ultra package, and the structures were optimized using the MM2 force field. The lowest energy conformations obtained by molecular mechanics calculations were optimized by the quantum chemical semi empirical RM1 (Recife Model 1) method [35]. The RM1 method was selected for our calculations because the average errors in the prediction of enthalpies of formation, dipole moments, ionization potentials, and inter atomic distances, using the RM1 methods were found to be less than the average errors given by AM1, PM3 and PM5 methods. The MOPAC program [36] was used to do semi empirical molecular orbital calculations, by passing the RM1 parameters via the keyword EXTERNAL in MOPAC along the keyword AM1. The optimized structures were found to be in good agreement with the available crystal structure of 10 reported earlier [37].
An input file, which contained the data obtained from self-consistent field (SCF), thermodynamics, force and molecular structure calculations for each structure together with the activity value (log IC 50 ) was prepared. All the data files were loaded into the CODESSA 2.6 for further calculation of topological, conventional, geometrical, electrostatic, quantum chemical and thermodynamic descriptors. More than 450 descriptors could be calculated from this program. The descriptors were further analyzed for linear dependence.
The good statistical methods that could select appropriate descriptors and the best quality correlation are essential in developing the QSAR/QSPR models. In this study, two methods were used to obtain the QSAR equation i.e., Heuristic and the Best Multi Linear regression method (BML). The Heuristic model could work fast and could be applied on a no limit data set. It could either give good correlation from the data or several best regression models. The algorithm of this method follows several steps, summarizing as it eliminates the descriptors with bad or missing values followed by the highly intercorrelated descriptors. The best multi-parameter regression models, which were developed from the remaining descriptors, will come with optimum values of statistical criteria consisting of regression correlation coefficient (R 2 ), the cross-validation (R 2 cv ), and the F-value. Compared to the Heuristic method, the BML method was more thorough and thus takes a longer time to complete. This method also limits the number of experimental values (150 structures) and descriptors (300 descriptors). The best two, three and etc. parameter regression models were based on the highest R 2 value. The models will be constituted from the selected non-collinear descriptors. After the initial analysis, the equation from BML was selected as the best equation based on the statistical parameters such as correlation coefficient (R), standard deviation (s), and F values. The BML method builds a single correlation using all selected descriptors in order to find the best regression model.
In developing a good QSAR model, it is important to decide when to stop adding descriptors. The technique is the so-called "breaking point" which helps to control the model expansion and thus, in turn, improves the statistical quality of the model [28]. Initially, any addition of any independent variables in the QSAR model can lead to improvement in the R 2 value in the consequent regression.
However, if the addition of the new descriptor does not significantly improve the R 2 value, then the added descriptor does not contribute any new information to the model. If the increase in R 2 value is less than 0.02, then the QSAR model described by such regression equation is considered as the best model.
The obtained model was validated to test the internal stability and predictive ability by using the internal test procedure. It employed the leave-one-out method (LOD). In the calculation for the cross validation regression coefficient (R 2 cv ), each molecule in the data set was eliminated once. The activity of the eliminated molecule was predicted by using the model developed by the remaining molecules. The cross validation regression coefficient (R 2 cv ) can be calculated by the following formula [38]; where Y exp and Y pred were activities of molecule in the test data and Y is the average activity of all the molecules in the data set. The LOD method technique is based on the difference between the squared cross-validated correlation coefficient (R 2 cv ) and correlation coefficient (R 2 ). The small difference suggested a good predictive ability of the QSAR model. When the data set is divided into the training and test sets, a model is generated from the training set compounds. The model should be validated through the external validation using the parameters like R 2 pred . It could be defined as The value of R 2 pred should be more than 0.5.

Data Set
A total of 24 anti-inflammatory compounds were used to develop the QSAR model. Out of these compounds, compounds 1, 2 and 3 were isolated from P. tetrandra, compounds 4-10 were from the semi synthesis of ursolic acid 1, while compounds 11-30 were purchased from Chromadex. All compounds were evaluated for their inflammatory activity of the hyaluronidase enzyme, and the data was calculated as IC 50 values. The data was converted into log IC 50 values which were used instead of IC 50 to improve the normal distribution of the experimental data points [28].

Descriptors
Four descriptors were found to influence the activity of PTs towards the hyaluronidase inhibitory activity. Two electronic descriptors were involved in Equation (1). The first electronic descriptor was the min partial charge for a carbon atom (Zefirov's PC) (Q i ). It could be defined as [39]: whereby X i is the atomic electronegativity given by; whereby X 0 i is the electronegativity of the isolated atoms and n is the number of atoms in the first coordination sphere of a given atom "i".
The other electronic descriptor was the molecular surface area, E nn .
Two chemical descriptors were involved in the selected model. Max bond order of a C atom, E c is defined as [30]: (6) whereby the first summation is performed over all occupied molecular orbitals (ni denotes the occupation number of the ith MO), and the two other summations over µ and υ, the atomic orbitals belonging to the C atoms in the molecule. MO coefficients are denoted as c iu and c jv . The min valency of an H atom, the final quantum chemical descriptor is the max electron-electron repulsion for a C-H bond (E ee ). It could be defined as [30]: E ee pABq " ÿ µ,vPA ÿ λ,σPB P µv P λσ xµv |λσ y where P λσ and P µv are the densities of the matrix element over atomic basis tµvλσu and xµv |λσ y is the electron repulsion integrals on atomic basis tµvλσu.

Molecular Docking Study
Molecular docking was used to predict and clarify the interaction of the complex between the most active ursolic acid analogue, compound 22 in Table 1, and hyaluronidase in comparison to the positive control apigenin. Apigenin was taken from the RCSB protein data bank (PDB ID = 4HKK, [40,41] where compound 22 was further modeled from apigenin and optimized with density functional theory under b3lyp/6-311g(d,p) basis set using Gaussian09 [42]. The high resolution of the crystal structure of human hyaluronidase-1, a hyaluronan hydrolyzing enzyme involved in tumor growth and angiogenesis was obtained from the RCSB protein data bank (PDB ID = 2PE4) [41,43]. The waters and ligands were removed from the original crystal structure. Then, the initial structure was modified according to the CHARMM force field with partial charge Momany-Rone [44] and minimization of the structures was performed with RMS gradient tolerance of 0.1000 kcal/(molˆAngstrom) satisfied. Flexible docking of compound 22 and apigenin into the targets was performed using AUTODOCK VINA [45] to Asp 129 (one of the binding sites) of human hyalurodinase. A 30ˆ30ˆ30 Å point grid was used. The low free energy complex structures were further minimized and analyzed. Detailed interaction energy was investigated by calculating binding energies using the protocol from Discovery Studio (Accelry Inc., San Diego, CA, USA, 2.5.5). This enabled us to estimate the residue interaction energy between the hyaluronidase and compounds.

Conclusions
A series of structurally related PTs were developed from the semi synthesis of ursolic acid 1. Together with the isolated and commercial sources of the derivatives, a total of 30 compounds/derivatives were evaluated for their inhibition towards hyaluronidase. However, only the data from 20 compounds were used to develop the QSAR model using the quantum chemical approach together with statistical analysis.
Equation (1) provided a measure of the influence of the changes in the size, shape, intermolecular hydrogen bonding, and the binding of the H, C-O and C-H atoms of the pentacylic triterpenes investigated here on the inhibition towards hyaluronidase. This QSAR equation can be used to predict the hyaluronidase inhibitory activity of new PT compounds, thus providing an efficient approach to design and development of new bioactive PT compounds.