QSAR Analysis of 5-substituted-2-Benzoyl-aminobenzoic acids as PPAR Modulator

A quantitative structure activity relationship (QSAR) study on a series of analogs of 5-aryl thiazolidine-2, 4-diones with activity on PPAR-α and PPAR-γ was made using combination of various thermodynamic, electronic and spatial descriptors. Several statistical regression expressions were obtained using multiple linear regression analysis. The best QSAR model was further validated by leave one out cross validation method. The studied revealed that for dual PPAR-α/γ activity dipole-dipole energy and PMI-Z play significant role and contributed positively for PPAR-γ and PPAR-α activity respectively. Thus, QSAR brings important structural insight to aid the design of dual PPAR-α/γ receptor agonist.


Introduction
Type 2 diabetes is a metabolic disorder that afflicts 120 million worldwide at present and is estimated to rise to over 200 million by the year 2010 1 .In addition to the characteristic combination of insulin resistance and insulin deficiency, the type 2 diabetic often displays cardiovascular risk factors including dyslipidemia (hypertriglyceridemia, low HDL, and small dense LDL), hypertension, and obesity.The recent publication of the United Kingdom Prospective Diabetes Study (UKPDS) 2 has revealed that in Type 2 diabetes, intensive glucoselowering therapy is ineffective at reducing cardiovascular complications, despite decreasing microvascular complications such as retinopathy.
The PPARs (peroxisome proliferator activated receptors) were cloned less than a decade ago and are members of the superfamily of nuclear transcription factors that includes the receptors for steroid, retinoid, and thyroid hormones 3,4 .The PPARs form heterodimers with another nuclear receptor, the 9-cis-retinoic acid receptor (RXR).This heterodimer complex interacts with critical DNA response elements within promoter regions, and when activated by agonist ligand binding, it leads to gene transcription of proteins involved in control of lipid and carbohydrate metabolism.PPAR-γ agonists (e.g., rosiglitazone and troglitazone) have displayed clinical utility for increasing insulin sensitivity and improving glycemic control in Type 2 diabetes.In addition, these compounds have been shown to inhibit atherosclerosis in the mice [5][6][7] .
Studies have shown that thiazolidine-2,4-diones give highly potent in vivo antidiabetic activities 8 .Thiazolidine-2,4-diones (TZDs) are generally selective for PPARγ, although a TZD, KRP-297, with activity at PPAR-α and PPAR-γ was recently disclosed 9 .The structurally related isoxazolidinedione JTT-501 has also been recently reported with PPARγ activity similar to troglitazone accompanied by weak activity at PPAR-α 10 .PPAR-α agonists (e.g., gemfibrozil, fenofibric acid) produce reductions in serum triglycerides and increases in HDL cholesterol, in some cases accompanied by reductions in serum fibrinogen 11,12 .The combined profile of a dual PPAR-α/γ agonist thus appears well suited for treatment of hyperglycemia together with prevention of cardiovascular disease in Type 2 diabetes 13 .Herein we describe the quantitative structure activity relationship (QSAR) study of novel 5-subsituted 2-benzoylaminobenzoic acids as PPARα/γ modulators.This study has resulted in the identification of molecular properties, which significantly correlated with of PPAR-γ and PPAR-α agonist activity.

Experimental
Analogs of 5-substituted 2-benzoylaminobenzoic acids (table 1), as PPAR α/γ modulators were taken from the reported work of Thor et al 14 (excluding compounds with biological activities numerically not well defined).The biological activity data K i values (ligand binding affinity in µM) were converted to negative logarithmic dose in mole (pK i ) for QSAR analysis.The series was subjected to molecular modeling and 3D-QSAR studies using CS Chem-Office Software version 6.0 (Cambridge soft) 15 running on a P-III processor.Structures of all the compounds (table 1) were sketched using builder module of the program.These structure were then subjected to energy minimization using molecular mechanics (MM2) until the root mean square (RMS) gradient value became smaller than 0.1kcal/mol.Å. Minimized molecules were subjected to reoptimization via Austin model-1 (AM1) method until the root mean square (RMS) gradient attained a value smaller than 0.01 kcal/mol.Å using MOPAC.The geometry optimization of the lowest energy structure was carried out using Eigenvector following (EF) routine.The descriptor values for all the molecules were calculated using "compute properties" module of program.
Calculated thermodynamic descriptors included critical temperature (T c ), ideal gas thermal capacity (C p ), critical pressure (P c ), boiling point (BP), Henry's law constant (H), bend energy (E b ), stretch bend energy (SBE), heat of formation (H f ), total energy (TE) and logarithm of partition coefficient (logP).
Multiple linear regression analysis method was used to perform QSAR analysis employing in-house VALSTAT 16 program.The best model was selected on the basis of various statistical parameters such as correlation coefficient (r), standard error of estimation (SE), sequential Fischer test (F).The model was further validated on various statistical parameters like leave one out cross validated square correlation coefficient (Q 2 ) using cross validation method 17 , boot-strapping square correlation coefficient (r 2 bs ), randomize biological activity data test (chance) and test for outliers (Z-score value) which confirm the robustness and applicability of QSAR equation on the structural analogs.

Results and Discussion
When data set was subjected to multiple linear regression analysis, in order to ascertain QSAR between binding affinity at PPAR-γ receptor as dependent variables and physiochemical descriptor as independent variable, several multivariant equations were obtained.The statistically significant equation with coefficient of correlation (r) =0.899 was considered as model for PPAR-γ binding affinity (table 2 2), the value of Q 2 ≥0.3 in cross validation method corresponds to a confidence limit greater than 95%, which minimized the risk of finding significant explanatory equation for the biological activity just by mere opportunity.The value of cross-validated squared correlation co-efficient (Q 2 =0.632), predictive residual sum of square (S PRESS =0.308) and standard error of predictivity (S DEP =0.267) suggested good predictive ability of the biological activity with low S DEP .The r 2 bs =0.807 is at par with the conventional squared correlation coefficient (r 2 ), indicating that no single compound much more/less contributed to the model.Randomize biological activity data test (Chance < 0.001) revealed that the result was not based on chance correlation (table 3).The model was further tested for outlier by Z-score method no compound was found to be outlier suggested that the model is able to explain the structural diversify analogs, which is helpful in designing of more potent compounds using physiochemical parameters.
The study revealed that for PPAR-γ binding affinity, dipole-dipole energy of the compound contributed positively which is responsible for the interaction of molecular polar portion with receptor.Substitution of group that is favorable for DDE may be enhances binding affinity of molecule with receptor.While vander Waals -1,4-energy contributed negatively to the activity, explains the depth of the attraction potential and how easy it is to push atoms together to the pocket of PPAR-γ receptor at helix-3.Dipole moment at Xcomponent contributed negatively to the PPAR-γ receptor binding affinity.
The correlation was also established between PPAR-γ transactivation potency (EC 50 ) of fifteen compounds of the series with physiochemical properties to explore the parameters, which play significant role in the agonist activity.Various correlated equations were obtained which account for more than 75% variance in activity, the equation 2 was considered as model for transactivation potency.
For PPAR-α only eight compounds having numerically well defined binding affinity are subjected to regression analysis in order to explore the physiochemical properties responsible for the activity.Eqn. 2 was considered as statistical significant model for PPAR-α, which account for more than 73.8% of the variance in the binding affinity (table 4 & figure 5).pKi = 3.573e-005( ± 8.682e-006)* PMI-Z + 4.281 (eqn.3)n=8, r=0.859, r 2 =0.738,SE =0.098, F=16.935 For PPAR-α binding affinity model having correlation coefficient value (r ≥ 0.859) and significantly low standard error of estimation (SE = 0.098).The data showed better statistical significance >98% with F (1,6) = 16.935against the tabulated value for sequential Fischer test at 98% (F 1,6 α 0.02 = 13.7).The model was further subjected for leave one out cross validation method, the value of (Q 2 =0.243), (S PRESS =0.166) and (S DEP =0.144) suggested moderate predictive ability of the biological activity (table 4 & figure 6).Randomize biological activity data test is less than 0.001.The model also shows that no compound is outlier.Equation 3revealed that for PPAR-α binding affinity, principal moment of inertia of Z component contributed positively suggested that substitution of bulkier group is favorable for the activity.

Observed pK i
Predicted pK i

Figure 1 .Figure 2 .Figure 3 .Figure 4 .Figure 5 .Figure 6 .
Figure 1.Plot between observed pK i and calculated pK i using eqn 1 for PPAR-γ and figure1), The model showed overall internal statistical significance level better than 99.9% as it exceeded the tabulated F

Table 3 .
QSAR statistics of significant equations

Table 4 .
PPARNegative logarithm of observed activity in mole taken from reference 14. ‡ Calculated activity data.Residual of observed and calculated activity data.† † Z-score data for screening of outliers.↑ Leave one out predicted activity data.‡ ‡ Residual of observed and leave one out predicted activity data.