Original article
Discovery of new cholesteryl ester transfer protein inhibitors via ligand-based pharmacophore modeling and QSAR analysis followed by synthetic exploration

https://doi.org/10.1016/j.ejmech.2009.12.070Get rights and content

Abstract

Cholesteryl ester transfer protein (CETP) is involved in trafficking lipoprotein particles and neutral lipids between HDL and LDL and therefore is considered a valid target for treating dyslipidemic conditions and complications. Pharmacophore modeling and quantitative structure–activity relationship (QSAR) analysis were combined to explore the structural requirments for potent CETP inhibitors. Two pharmacophores emerged in the optimal QSAR equation (r2 = 0.800, n = 96, F = 72.1, r2LOO = 0.775, r2PRESS against 22 external test inhibitors = 0.707) suggesting the existence of at least two distinct binding modes accessible to ligands within CETP binding pocket. The successful pharmacophores were complemented with strict shape constraints in an attempt to optimize their receiver-operating characteristic (ROC) curve profiles.

The validity of our modeling approach was experimentally established by the identification of several CETP inhibitory leads retrieved via in silico screening of the National Cancer Institute (NCI) list of compounds and an in house built database of drugs and agrochemicals. Two hits illustrated low micromolar IC50 values: NSC 40331 (IC50 = 6.5 μM) and NSC 89508 (IC50 = 1.9 μM). Active hits were then used to guide synthetic exploration of a new series of CETP inhibitors.

Introduction

Atherosclerosis describes the principal progression in arterial dysfunction and remodeling that restricts blood flow to vessels in the peripheral vasculature and is ultimately manifested as coronary artery disease (CAD) [1]. Several epidemiological studies have demonstrated an inverse relationship between serum high-density lipoprotein cholesterol (HDLc) levels and the incidence of ischemic heart disease [2]. HDL mediates the reverse cholesterol transport pathway which removes excess cholesterol from peripheral tissues to the liver for biliary elimination [3].

CETP, a 476-residue glycoprotein, is involved in trafficking lipoprotein particles and neutral lipids, including cholesteryl esters (CE), phospholipids and triglycerides between HDL and low-density lipoproteins (LDL). CETP, as revealed by X-ray crystallography (PDB code: 2OBD, resolution 2.2 Å), has a large highly hydrophobic binding site capable of simultaneously binding up to four lipid molecules [4]. In human plasma, CETP plays a potentially proatherogenic role by moving CE from HDL to very-low-density lipoprotein (VLDL) and low-density lipoprotein (LDL) particles, thereby lowering atheroprotective HDLc and raising proatherogenic VLDLc and LDLc. Apparently, the risk of CAD is proportional to the plasma levels of CETP [5]. In fact, It is quite common within the CAD population to have elevated CETP plasma protein levels that are 2- to 3-fold higher than concentrations typically found in the plasma of normal subjects (1–3 μg/mL) [6].

Evidence exists that the consequences of CETP activity may depend on the metabolic setting, particularly on triglyceride levels. Accordingly, pharmacological CETP inhibition may reduce the risk of CAD in humans, but only in those with high triglyceride levels [5].

The unavailability of satisfactory high resolution crystallographic structures for CETP combined with its prohibitively large binding pocket confined most modeling-related discovery projects to ligand-based approaches particularly quantitative structure–activity relationship analysis (QSAR) [7], [8], [9].

Despite the excellent predictive potential of 3D-QSAR methodologies (e.g., CoMFA and CoMSIA), they generally lack the ability to act as effective search queries to mine virtual three-dimensional (3D) databases for new hits [10], [11].

The continued interest in the development of new CETP inhibitors combined with the lack of adequate CETP crystallographic structures and adequate computer-aided drug discovery efforts in this area, prompted us to explore the possibility of developing ligand-based three-dimensional (3D) pharmacophore(s) integrated within self-consistent QSAR model for CETP inhibitors. The pharmacophore model(s) can be used as 3D search query(ies) to mine 3D libraries for new CETP inhibitors, while the QSAR model helps to predict the biological activities of the captured compounds and therefore prioritize them for in vitro evaluation. We previously reported the use of this innovative approach towards the discovery of new inhibitory leads against glycogen synthase kinase 3β (GSK-3β) [12], dipeptidyl peptidase [13], hormone sensitive lipase (HSL) [14], bacterial MurF [15], protein tyrosine phosphatase 1B (PTP 1B) [16] and influenza neuraminidase [17].

We employed the HYPOGEN module from the CATALYST software package to construct numerous plausible binding hypotheses for CETP inhibitors [18]. Subsequently, genetic function algorithm (GFA) and multiple linear regression (MLR) analysis were employed to search for an optimal QSAR that combines high-quality binding pharmacophores with other molecular descriptors and capable of explaining bioactivity variation across a collection of diverse CETP inhibitors. The optimal pharmacophores were further validated by evaluating their ability to successfully classify a list of compounds as actives or inactives by assessing their receiver-operating characteristic (ROC) curves. Subsequently, the optimal pharmacophores were complemented with tight shape constraints to enhance their ROC profiles. Thereafter, the resulting shape-complemented pharmacophores were used as 3D search queries to screen several available virtual molecular databases for new CETP inhibitors. Active hits were employed as guides to synthesize new series of active CETP inhibitors.

CATALYST models drug–receptor interactions using information derived from the ligand structures [18], [19], [20], [21], [22], [23], [24], [25], [26]. HYPOGEN identifies a 3D array of a maximum of five chemical features common to active training ligands that provides relative alignment for each input molecule consistent with binding to a proposed common receptor site. The conformational flexibility of training ligands is modeled by creating multiple conformers that cover a specified energy range for each input molecule [16], [21], [22], [23], [27], [28], [29], [30], [31].

The SHAPE module in CATALYST is a shape-based similarity searching method. The Van der Waals surface of a molecule (in a certain conformation) is calculated and represented as a set of points of uniform average density on a grid. The surface points enclose a volume on the grid. The geometric center of the set of points is computed along with the three principal component vectors passing through the center. The maximum extents along each principal axis and the total volume are calculated. These provide shape indices that can be compared with the query and used in an initial screening step to eliminate poor matches from further consideration [32]. CATALYST pharmacophores, with or without shape constraints, have been used as 3D queries for database searching and in 3D-QSAR studies [21], [23], [27], [32].

Section snippets

Pharmacophore modeling

The literature was extensively surveyed to collect diverse CETP inhibitors. A dataset of 118 N,N-disubstituted-3-amino-2-propanol derivatives (1–118, Table A under Supplementary material) was used for pharmacophore modeling and subsequent QSAR analysis [33], [34], [35], [36]. The conformational space of each inhibitor was sampled utilizing the poling algorithm implemented within CATALYST [21], [22], [23], [37].

The pharmacophoric space of CETP inhibitors was explored employing three structurally

Conclusions

This work includes elaborate pharmacophore exploration of CETP inhibitors utilizing CATALYST-HYPOGEN. QSAR analysis was employed to select the best combination of molecular descriptors and pharmacophore models capable of explaining bioactivity variation across an informative list of training compounds. The successful pharmacophores were complemented with strict shape constraints to optimize their receiver-operating characteristic (ROC) curve profiles. The best binding hypotheses were used as 3D

Software and hardware

The following software packages were utilized in the present research.

Pharmacophore modeling and QSAR analysis were performed using CATALYST (HYPOGEN module) and CERIUS2 software suites from Accelrys Inc. (San Diego, California, www.accelrys.com) installed on a Silicon Graphics Octane2

Acknowledgments

This project was partially sponsored by the Faculty of Graduate Studies (Ph.D. Thesis of Reema Abu Khalaf). The authors wish to thank the Deanship of Scientific Research and Hamdi-Mango Center for Scientific Research at the University of Jordan for their generous funds. The authors are also indebted to national cancer institute for freely providing hit molecules for evaluation.

References (56)

  • R.P. Sheridan et al.

    Why do we need so many chemical similarity search methods?

    Drug Discov. Today

    (2002)
  • S.W. Homans

    Water, water everywhere — except where it matters? Drug Discov

    Today

    (2007)
  • C.A. Lipinski et al.

    Experimental and computational approaches to estimate solubility and permeability in drug discovery and development settings

    Adv. Drug Del. Rev.

    (2001)
  • G.K. Hansson

    Inflammation, atherosclerosis, and coronary artery disease

    N. Engl. J. Med.

    (2005)
  • A.R. Tall et al.

    Is it time to modify the reverse cholesterol transport model?

    J. Clin. Invest

    (2001)
  • X. Qiu et al.

    Crystal structure of cholesteryl ester transfer protein reveals a long tunnel and four bound lipid molecules

    Nat. Struct. Mol. Biol.

    (2007)
  • S.M. Boekholdt et al.

    Plasma levels of cholesteryl ester transfer protein and the risk of future coronary artery disease in apparently healthy men and women

    Circulation

    (2004)
  • R. McPherson et al.

    Plasma concentration of cholesteryl ester transfer protein in hyperlipoproteinemia

    Arterioscler. Thromb

    (1991)
  • P. Hanumantharao et al.

    QSAR analysis of analogs of bis[2-(acylamino) Ph] disulfides, 2-(acylamino) benzenethiols and S-[2-(acylamino) Ph] alkanethioates as antihyperlipidemic agents

    Ind. J. Chem., Sect. B, Org. Chem. Includ. Med. Chem.

    (2005)
  • M.A. Kelkar et al.

    3D QSAR studies of inhibitors of cholesterol ester transfer protein (CETP) by CoMFA, CoMSIA and GFA methodologies

    Med. Chem. Res.

    (2004)
  • M. Akamatsu

    Current state and perspectives of 3D-QSAR

    Curr. Top. Med. Chem.

    (2002)
  • M.O. Taha et al.

    Pharmacophore modeling, quantitative structure–activity relationship analysis, and in-silico screening reveal potent glycogen synthase kinase-3beta inhibitory activities for cimetidine, hydroxychloroquine, and gemifloxacin

    J. Med. Chem.

    (2008)
  • I.M. Al-masri et al.

    Discovery of DPP IV inhibitors by pharmacophore modeling and QSAR analysis followed by in silico screening

    Chem. Med. Chem.

    (2008)
  • M.O. Taha et al.

    Combining ligand-based pharmacophore modeling, QSAR analysis and in-silico screening for the discovery of new potent hormone sensitive lipase inhibitors

    J. Med. Chem.

    (2008)
  • A.M. Abu Hammad et al.

    Pharmacophore modeling, quantitative structure–activity relationship analysis, and shape-complemented in silico screening allow access to novel influenza neuraminidase inhibitors

    J. Chem. Inf. Model.

    (2009)
  • Catalyst User Guide

    (2005)
  • P.W. Sprague et al.
  • D. Barnum et al.

    Identification of Common Functional configurations among molecules

    Chem. Inf. Comput. Sci.

    (1996)
  • Cited by (56)

    • Pharmacophore modeling of JAK1: A target infested with activity-cliffs

      2020, Journal of Molecular Graphics and Modelling
      Citation Excerpt :

      GFA implements evolutionary operations of crossover and mutation to select combinations of pharmacophores/descriptors and deliver them separately to multiple linear regression (MLR) or k-nearest neighbour (kNN) analyses to evaluate their abilities (i.e., descriptor combinations) to explain bioactivity variation among collected JAK1 inhibitors. Details of the implemented GFA are shown in Supporting Information section SM7 [32–55,66]. MLR assumes linear, parametric relationship connecting certain group of molecular descriptors with the biological activities of the corresponding training list of compounds [39,57].

    • Ligand-based computer aided drug design reveals new tropomycin receptor kinase a (TrkA) inhibitors

      2018, Journal of Molecular Graphics and Modelling
      Citation Excerpt :

      The total pharmacophoric, 2D, and 3D properties are presumed as potential factors influencing the activity of the hypothesized optimum TrkA inhibitor, including several powers and other functions of the raw inputs. Accordingly, the GFA-MLR approximations were applied to find the subset of terms that correlates best with the activity since each column in the QSAR spreadsheet is considered as a new descriptor to compete in the context of QSAR analysis [22,24,28–33]. The protocol “Genetic Function Approximation Model” implemented in DS 4.5 was used to search for the finest descriptive GFA-QSAR regression equation.

    • Discovery of new selective cytotoxic agents against Bcl-2 expressing cancer cells using ligand-based modeling

      2016, Chemico-Biological Interactions
      Citation Excerpt :

      The training compounds in each subset were selected with bioactivities extending over 3 logarithmic cycles to further assure wide structural diversity within each training subset. Moreover, training subsets were selected in such a way that differences in Bcl-2 inhibitory activities among their member compounds are attributable primarily to the presence/absence of binding features and not due to steric shielding or electron withdrawing/donating functional groups [5,6,8,26,57,59,60]. HYPOGEN was instructed to explore pharmacophore models of 4 and 5 binding features and ignore models of less than 4 features (see Sections SM.1.1, SM.1.2, SM.1.3, SM.1.4 and Table S3 under Supplementary Materials).

    • Ligand-based modeling followed by in vitro bioassay yielded new potent glucokinase activators

      2015, Journal of Molecular Graphics and Modelling
      Citation Excerpt :

      This problem combined with the fact that cluster representatives (i.e., best pharmacophores produced during HYPOGEN exploration) had similar statistical properties (Table 4) impelled us to use Genetic algorithm/multiple linear regression-based QSAR (GFA-MLR-QSAR) modeling to select the most optimal combination of pharmacophore models and two-dimensional descriptors that can explain variation in GK bioactivation among the gathered activators (1–30, Table 1) [13,14,16,35]. The fit values obtained by fitting the representative pharmacophores (Table 4) against the entire list of collected GK activators (Table 1) were enrolled as descriptors in GFA-MLR-QSAR (see Section 2.1.6) [24c]. We randomly selected 6 molecules (labeled in Table 1) and used them as external testing molecules for validating the QSAR models (determination of r2PRESS, see Section 2.1.6).

    View all citing articles on Scopus
    View full text