Structure‐Based Pharmacophore Screening for Natural‐Product‐Derived PPARγ Agonists

Peroxisome proliferator-activated receptors (PPARs) are transcription factors that play a critical role in lipid signalling and immunomodulation and functionally interact with other nuclear receptors, like PXR and NF-kB, in the regulation of lipid metabolism. Therefore, agonists of PPARa and PPARg serve as therapeutic agents for the treatment of dyslipidaemia, type II diabetes and artherosclerosis, while their effects on the regulation of cell proliferation are under investigation. Several natural compounds have been identified that activate PPARs, including the tetrahydrocannabinol (THC) metabolite THC-11-oic acid, carnosic acid and carnosol, and resveratol. These can provide a starting point for the combinatorial exploration of natural-product-derived compounds for lead discovery and development, with the aim of substituting existing PPAR agonists with potentially safer drugs containing novel scaffolds. Here, we present a virtual screening protocol that led to a PPARg agonist from a combinatorial compound library that was derived from the scaffold structure of a-santonin, a natural sesquiterpene lactone found in mugwort. We demonstrate that it is possible to find lead candidates in small combinatorial compound collections with minimal experimental effort by “fuzzy” pharmacophore screening. For the generation of a pharmacophore query, we superimposed four high-resolution X-ray structures of the PPARg ligand binding domain in complex with agonists (PDB ID: 1nyx with Ragaglitazar, 1knu with YPA, 1i7i with Tesaglitazar, 1zgy with Rosiglitazone, Figure 1 A). The resulting ligand alignment served as the basis for pharmacophoric point assignment by our software LIQUID, as described previously. Briefly, LIQUID represents potential pharmacophoric points (lipophilic, hydrogen-bonding) in a molecule by Gaussian densities. These densities are converted to probabilities for the pairwise matching of compounds. As a result of LIQUID matching and scoring, a screening library is sorted so that the best matching compounds appear at the top of the ranked list. From this list, the most promising screening candidates are picked. For the generation of the LIQUID descriptors from 3D molecular conformations, we used cluster radii of 1 for lipophilic centres and 2 for hydrogen-bonding centres (donors, acceptors, and donor + acceptor). No other pharmacophoric features were considered to obtain a coarse-grained model that allowed for scaffold hopping to occur. The resulting pharmacophore query is depicted in Figure 1 B. This procedure was performed in order to obtain a “receptor-relevant” pharmacophore model of PPARg agonists instead of using a ligandbased spatial alignment of artificially generated conformers. This structure-based alignment of multiple ligands allowed us to compute “fuzzy” pharmacophoric feature points, so that we obtained a probability-weighted consensus model. It is noteworthy that the explicit consideration of “voids” or “forbidden regions” is not required, as the probabilities for the presence of a pharmacophoric feature adopt values close to zero in the vicinity of the model. We then searched the AnalytiCon Discovery collection of natural-product-derived combinatorial compounds (version 01/ 2007, 15 590 entries) for hits matching the LIQUID pharmacophore query. A single 3D conformation was computed for each compound by using Corina v3.2 (Molecular Networks GmbH, Erlangen). This concept was shown to be sufficient for firstFigure 1. A) Structural superimposition of four PPARg-agonist complexes; B) The alignment-derived LIQUID pharmacophore query. In the pharmacophore model, lipophilic centres are shown in green, potential hydrogenbond donor sites are shown in blue, and potential hydrogen-bond acceptor sites are shown in red. Approximate locations of helix 3 and the AF2-helix are indicated. The trivariate Gaussians are shown with widths of one standard deviation in each direction.