Identification of Novel Androgen Receptor Antagonists Using Structure

Jan 1, 2013 - Given the wealth of structural information on the AR in complex with a variety of ligands, we have applied an integrated structure- and ...
3 downloads 4 Views 2MB Size
Article pubs.acs.org/jcim

Identification of Novel Androgen Receptor Antagonists Using Structure- and Ligand-Based Methods Huifang Li,*,‡,† Xin Ren,‡,† Eric Leblanc,† Kate Frewin,† Paul S. Rennie,§,† and Artem Cherkasov§,† †

Vancouver Prostate Centre, University of British Columbia, 2660 Oak Street, Vancouver, British Columbia V6H 3Z6, Canada S Supporting Information *

ABSTRACT: Androgen receptor (AR) plays a critical role in the development and progression of prostate cancer (PCa). The AR hormone-binding site (HBS) is intensively studied and represents the target area for current antiandrogens including Bicalutamide and structurally related Enzalutamide. As resistance to antiandrogens invariably emerges in advanced prostate cancer, there exists a high medical need for the identification and development of novel AR antagonists of different chemotypes. Given the wealth of structural information on the AR in complex with a variety of ligands, we have applied an integrated structure- and ligand-based virtual screening methodology to identify novel AR antagonists. Virtual hits generated by a consensus voting approach were experimentally evaluated and resulted in the discovery of a number of structurally diverse submicromolar antagonists of the AR. In particular, one identified compound demonstrated anti-AR potency in vitro that is comparable to the clinically used Bicalutamide. These results set a ground for the development of novel classes of PCa drugs that are structurally different from current AR antagonists.



INTRODUCTION Prostate cancer (PCa) is the second leading cause of male cancer death.1 If the cancer is confined to the prostate, it is often curable by surgery or radiation treatment. However, in many cases, it has already spread to other tissues, particularly the bone, at diagnosis and usually requires some form of antimale sex hormone therapy to keep it in check. The antiandrogen such as Bicalutamide and Enzalutamide (initially called MDV3100) target the androgen receptor (AR), which is responsible for growth and progression of the disease. However, the emergence of recurrent, metastatic forms of castration-resistant/androgen-independent PCa continues to be a major challenge, with a median survival of only ∼18 months.2 The mechanism of resistance to current antiandrogens is unclear,3 but for the first-generation antiandrogens like Hydroxyflutamide and Bicalutamide, it is known that certain mutations in the hormone binding site (HBS) of the AR (including T877A and W741L/C) will convert the drugs from receptor’s antagonists to agonists.4,5 The recently approved antiandrogen Enzalutamide has been developed as a secondgeneration AR antagonist aiming to solve the problem of resistance.6 However, the resistance to Enzalutamide was already reported during the clinical trials.7 Part of the problem is that all known antiandrogens, including Enzalutamide, share the same structural motif responsible for effective AR HBS binding which may attribute to occurring resistance. Therefore, more efforts are expected to discover AR antagonists possessing novel chemical scaffolds that may partially address the problem of resistance in PCa. There is a wealth of structural information on a wild type AR complexed with diverse steroidal and nonsteroidal agonists © XXXX American Chemical Society

bound to the AR HBS, including testosterone, dihydrotesterone (DHT), and R1881 among others.8−10 The known antiandrogens such as Hydroxyflutamide and Bicalutamide (Figure S1) have only been cocrystallized with the mutated forms of the AR, interacting with the compounds in agonistic manner.4,5 These structures provide valuable insights into essential interactions between the AR HBS and agonistic ligands. However, there is no structural data for antagonistic interactions with the AR and the mechanism of exact action of antiandrogens is still not clear. This fact imposes limitations on structure-based discovery and rational design of novel AR HBS antagonists (and, perhaps, contributed to limited structural variability among clinically used AR antagonists). On the other hand, a large number of diverse AR binders have also been reported in the literature which provides important information for ligand-based models using methods of quantitative structure−activity relationship (QSAR). In this study, we have combined structure- and ligand-based strategies to identify novel AR antagonists of diverse chemical classes. First, the structure-based virtual screening was used to filter out compounds from public databases that are unlikely to have good binding affinity to the AR HBS. Although current docking programs can generate correct poses and capture essential interactions between ligands and the receptor, the corresponding scoring functions are not always sufficiently accurate. Thus, to complement the structure-based screening, we have developed a series of QSAR models that were subsequently used to select the final list of virtual hits. By taking Received: October 24, 2012

A

dx.doi.org/10.1021/ci300514v | J. Chem. Inf. Model. XXXX, XXX, XXX−XXX

Journal of Chemical Information and Modeling

Article

Descriptor Selection. INDUCTIVE descriptors, 25−27 DRAGON 5.5 descriptors,28 and MOE descriptors were calculated for ZINC compounds to be screened. To avoid possible cross-correlation among independent variables, the pairwise correlation among all the QSAR parameters were computed, and descriptors with either low variability (value −8). The active site was defined from the coordination of crystallographic ligand (PDB code: 2PNU.pdb), using the default settings. To avoid any bias in docking programs, compounds with favorable Glide scores were subsequently docked by eHiTs, and compounds with eHiTs score higher than a cutoff value (−3) were removed. The root-mean-square deviation (RMSD) of the docked poses generated by Glide and eHiTs were then calculated to keep compounds with consistent docked poses. Compounds with high RMSD values (>2 Å) were removed, and Lipinski’s rule16 was applied to the remaining compounds, which were advanced to the next stage of analysis. 2. Ligand-Based Virtual Screening. Data Set. The success of any QSAR model depends on accurate and clean training data,17 proper representative descriptor selection methods,18 suitable statistical methods,17 and, most critically, both internal and external validation of resulting models.19 A total of 1220 chemical structures were taken from the literature and public databases, and reported activities were converted into binary format. Part of the set consisted of 625 known AR binders reported previously,20 including 394 chemicals measured by a Danish lab with activities in binary form, and 231 chemicals with IC25 values which were converted into binary format. In particular, a given chemical exhibiting IC25 value lower than the test concentration (10 μM) was classified as active or inactive if the IC25 value was above 10 μM or the cytotoxicity (IC50) was over 3 μM. Another set of 595 known AR binders was assembled from public databases such as CHEMBL,21 BindingDB,22 DrugBank,23 and ChemSpider.24 Among those, chemicals with an IC50 value lower than 20 μM were defined as active AR inhibitors. Additionally, the external set (89 molecules) of known AR inhibitors previously used in the literature20 was also considered to validate the developed QSAR models. The structures of the studied chemicals were optimized by the MMFF94x force field implemented in MOE program. B

dx.doi.org/10.1021/ci300514v | J. Chem. Inf. Model. XXXX, XXX, XXX−XXX

Journal of Chemical Information and Modeling

Article

Figure 1. The pipeline of the discovery process for AR antagonists.

Table 1. RMSD Values between the Docked and Bound Conformations

a

Glide

2AM9

3L3X

2PNU

1Z95

eHiTs

2AM9

3L3X

2PNU

1Z95

testosterone DHT EM5744 Bica

0.79 0.66 NAb 4.81

0.22 0.27 NAb NAb

0.48 0.55 0.28 1.29

3.14 1.08 1.5 0.34

testosterone DHT EM5744 Bica

0.70 0.92 7.56 3.08

1.04 0.75 4.63 4.15

3.35 0.56 1.00 0.65

1.11 1.11 1.85 2.21

Represents Bicalutamide. bNA: ligands were skipped by glide sp docking program.

Table 2. Enrichment Factor (EF) Values of the Screening of DUD Using Two Crystal Structures PDB code 2AM9 EF1% EF20% a

3L3X

1Z95

2PNU

Glidea

eHiTsa

Glidea

eHiTsa

Glidea

eHiTsa

Glidea

eHiTsa

26 3.6

34.7 4.7

26 3.6

32.2 4.7

11 2.8

35.9 4.6

26 4.1

35.9 5.0

Docking program.

Bicalutamide could not be accurately fitted into smaller 2AM9 and 3L3X. However, the self- and cross-docking of these ligands into 2PNU and 1Z95 were accurate and reproducible in both docking programs (see Table 1). Prescreening with the Directory of Useful Decoys (DUD). The four crystal complexes were also evaluated by prescreening with the directory of useful decoys (DUD) data set.50 The docking enrichment factor (EF) established for known AR binders from the DUD were similar and suitable in cases of 2AM9 and 3L3X structures (Table 2), while docking with 1Z95 target resulted in low enrichment. The AR HBS structure corresponding to 2PNU protein database entry yielded the highest EF indicating its better ability to differentiate AR binders. Therefore, consistent with the self- and cross-docking evaluations, the crystal structure 2PNU was selected for virtual screening. Structure-Based Virtual Screening. Molecular docking was initially performed with the Glide SP program using a subset of 3 million compounds from ZINC database. The Glide SP docking score cutoff (>−8) was used to identify low- or no affinity ligands to be discarded. The compounds left (702,151) were further redocked into 2PNU using eHiTs, and only compounds with eHiTs score less than −3 (571,666) were kept. The docked poses were compared with those from the Glide experiment, and compounds with low RMSD values (