Article pubs.acs.org/jcim
Dihedral Angle-Based Sampling of Natural Product Polyketide Conformations: Application to Permeability Prediction Qin Wang,‡ Simone Sciabola,*,§ Gabriela Barreiro,§ Xinjun Hou,§ Guoyun Bai,⊥ Michael J. Shapiro,⊥ Frank Koehn,† Anabella Villalobos,§ and Matthew P. Jacobson‡ ‡
Department of Pharmaceutical Chemistry, University of California, San Francisco, California 94158, United States Neuroscience and Pain Medicinal Chemistry, Pfizer Worldwide Research and Development, Cambridge, Massachusetts 02139, United States ⊥ Discovery Sciences and †Oncology Medicinal Chemistry, Pfizer Worldwide Research and Development, Eastern Point Road, Groton, Connecticut 06340, United States §
S Supporting Information *
ABSTRACT: Macrocycles pose challenges for computeraided drug design due to their conformational complexity. One fundamental challenge is identifying all low-energy conformations of the macrocyclic ring, which is important for modeling target binding, passive membrane permeation, and other conformation-dependent properties. Macrocyclic polyketides are medically and biologically important natural products characterized by structural and functional diversity. Advances in synthetic biology and semisynthetic methods may enable creation of an even more diverse set of non-natural product polyketides for drug discovery and other applications. However, the conformational sampling of these flexible compounds remains demanding. We developed and optimized a dihedral angle-based macrocycle conformational sampling method for macrocycles of arbitrary structure, and here we apply it to diverse polyketide natural products. First, we evaluated its performance using a data set of 37 polyketides with available crystal structures, with 9−22 rotatable bonds in the macrocyclic ring. Our optimized protocol was able to reproduce the crystal structure of polyketides’ aglycone backbone within 0.50 Å RMSD for 31 out of 37 polyketides. Consistent with prior structural studies, our analysis suggests that polyketides tend to have multiple distinct low-energy structures, including the bioactive (target-bound) conformation as well as others of unknown significance. For this reason, we also introduce a strategy to improve both efficiency and accuracy of the conformational search by utilizing torsional restraints derived from NMR vicinal proton couplings to restrict the conformational search. Finally, as a first application of the method, we made blinded predictions of the passive membrane permeability of a diverse set of polyketides, based on their predicted structures in low- and high-dielectric media.
■
INTRODUCTION
These properties of macrocycles, especially their size and conformational complexity, also pose further challenges for computer-aided drug design. Although there is a wide range of conformational search algorithms available for small drug-like compounds,19 conformational sampling for larger, more flexible compounds is much more challenging because the search space grows exponentially with the number of rotatable bonds.20 Structure-based molecular docking relies on the ability to sample low energy conformations of the ligand,21,22 but available docking programs generally cannot efficiently identify all the relevant conformations of a macrocycle. Other molecular properties including ADME (absorption, distribution, metabolism, and excretion) properties also depend sensitively on threedimensional conformations. Previous work by Leung et al.23 has established a molecular mechanics-based model for passive
Polyketides, as one of the major classes of natural products, play important roles as therapeutic agents, functioning as antibiotics, 1−3 antifungals, 4,5 antiparasitics, 6 immunosuppressants,7 cholesterol-lowering,8 and antitumor agents.9 Recently, there has been a resurgence of interest in developing novel therapeutics through modification and manipulation of natural products, including polyketides.10−13 In particular, macrocycles are an attractive class of compounds for challenging targets such as protein−protein interactions,14,15 due to their larger size (generally >500 MW) and greater conformational complexity than most conventional smallmolecule drugs. At the same time, certain macrocycles can demonstrate drug-like pharmacokinetic properties such as good solubility, cell permeability, and oral bioavailability,16 despite the fact that they typically violate Lipinski’s rules of five17,18 and other rules-of-thumb for drug-like molecular properties. © XXXX American Chemical Society
Received: April 27, 2016 Published: October 12, 2016 A
DOI: 10.1021/acs.jcim.6b00237 J. Chem. Inf. Model. XXXX, XXX, XXX−XXX
Article
Journal of Chemical Information and Modeling Table 1. Dataset of 37 Polyketides with Available Crystal Structures from CSD/PDBa
a
compound name
molecular weight
backbone length
rotatable bonds
backbone rotatable bonds
Nodusmicin Methymycin Spinosyn A Brefeldin A 5,6-dihydrocineromycin B Cethromycin Clarithromycin Erythromycin A Erythromycin D Lankamycin Telithromycin analog 6-deoxyerythronolide B Oleandomycin Roxithromycin Dirithromycin Latrunculin B Troleandomycin Azithromycin Avermectin B1a Avermectin B2a aglycone chalcomycin Epothilone A Epothilone B Epothilone D Bafilomycin Latrunculin A Lankacidin C Tubocurarine Geldanamycin Herbimycin A FK506 Cytovaricin Virginiamycin M1 Rifapentine Rifampin Rifaximin Rapamycin
422.5 469.6 731.9 280.4 296.4 765.9 747.9 733.9 703.9 831.0 797.0 386.5 687.8 837.0 835.1 395.5 814.0 749.0 873.1 602.7 700.8 493.6 507.7 491.7 622.8 421.5 459.5 609.7 560.6 574.7 804.0 901.1 525.6 877.0 822.9 785.9 914.2
10 12 12 13 14 14 14 14 14 14 14 14 14 14 14 14 14 15 16 16 16 16 16 16 16 16 17 18 19 19 21 22 23 25 25 25 29
14 17 21 13 14 23 26 26 25 26 25 18 23 32 30 14 23 27 23 18 25 19 20 18 21 14 15 14 16 16 29 34 18 23 22 19 31
9 11 12 11 12 14 14 14 14 14 14 14 14 14 14 12 14 15 12 12 14 16 16 15 10 12 11 10 11 11 19 20 16 13 13 13 22
CSD reference code
PDB ligand code
NDMSCN 3FWO:MT9* XOLHUB BREFEL DOZWUL ATOFIZ NAVTAF
1S9D:AFB 1NWX:773* 1J5A:CTY* 3FRQ:ERY 2JJO:EY5 3PIO:LMA*
GOPGAT
KAHWAT POBKOG
GEGJAD BASVAS BASTUK PVVBDM01 TIPFON FORTUB ZUKZEK TUBCUR10 HRBMCN FINWEE10 BOKWIH VIRGFM MAFLAI OWELOS XOHZIE RAPMCN11
1Z8O:DEB 2IYA:ZIO 1JZZ:ROX* 2V52:LAB 3I56:TAO* 1YHQ:ZIT
4I50:EP 1Q5D:EPB 1PKF:EPD 1SQK:LAR 3JQ4:LC2* 3PMZ:TUB 1YET:GDM 3IHZ:FK5 4HUS:VIR 2A69:RPT 2HW2:RFP 3KZ7:RAP
Asterisks indicate that resolution is worse than 2.5 Å or R-free value greater than 0.32.
numbers of rotatable bonds in the macrocyclic ring (9−22 in the test set considered here). Therefore, we have adapted the PLOP loop prediction algorithm to macrocycle sampling. Previously, this algorithm was used to predict the conformations of cyclic peptides, with modifications introduced to treat D-amino acids, nonstandard side chains, and one type of backbone modification, N-methylation.29,30 In this work, we generalize this method to treat a much wider array of macrocycles, including most naturally occurring and synthetic macrocycles, and we benchmark the algorithm using a data set of macrocyclic polyketides. Our sampling protocol consists of two stages: initial sampling of the full conformational space without any constraints, followed by more focused sampling around low-energy conformations. A validation set consisting of 37 polyketides that have crystal structures available in either the Cambridge Structural Database (CSD) 31 or Protein Data Bank (PDB)32,33 has been constructed for benchmarking conformational sampling. For each polyketide, the predicted conformations were compared to its crystal structures to evaluate sampling performance. We also investigated, in a preliminary manner, the use of torsional
membrane permeability, and the cornerstone of this model is thorough conformational sampling of compounds in silico. The importance of three-dimensional structure in understanding and optimizing the properties of polyketides has been emphasized in prior work.24,25 A primary challenge in sampling macrocycles is to identify all low-energy conformations that are consistent with ring closure; that is, the macrocyclic ring imposes complex and highly nonlinear constraints on the accessible conformational space. Similarly, the major challenge in protein loop prediction is identifying all conformations consistent with the ends of the loops being restrained by the rest of the protein, as well as stereochemical and geometric restraints. The loop prediction method in the Protein Local Optimization Program (PLOP, distributed as part of the Prime software package by Schrodinger LLC),26 among many other conformational search algorithms, has been shown to be capable of effectively sampling protein loop conformations, with subangstrom accuracy achievable for many loops ranging between 4 and 20 residues,26−28 corresponding to 8−40 restricted rotatable bonds along the backbone. Macrocyclic polyketides have similar B
DOI: 10.1021/acs.jcim.6b00237 J. Chem. Inf. Model. XXXX, XXX, XXX−XXX
Article
Journal of Chemical Information and Modeling
Figure 1. Flowchart representing the conformational sampling protocol (left), illustrated with results obtained for herbimycin A (right). (B) Superposition of the random initial conformation (orange) with the crystal structure (cyan), backbone RMSD 0.93 Å. (C) Superposition of all energy-minimized conformations from the initial sampling. (D) Energies and backbone RMSDs for all conformations from the initial sampling, relative to the crystal structure, and superposition of the sampled conformation with the lowest backbone RMSD (0.38 Å) with the crystal structure. (E) Relative energies and backbone RMSDs for all conformations from the optional refinement stage, and superposition of the sampled conformation with the lowest backbone RMSD (0.22 Å) with the crystal structure.
to other tools for computer-aided drug design. As the first application, we used the generated low-energy conformations in low- and high-dielectric implicit solvent to predict the absolute passive permeabilities of a diverse set of polyketide macrocycles (distinct from those in the crystal structure sets), which were measured using the Madin−Darby Canine Kidney Low Efflux transporter (MDCK-LE) cell monolayer assay.35 While we have previously reported extensive benchmarking of the membrane permeability model for small molecules and cyclic peptide macrocycles,23,36 we have not previously investigated membrane permeability of polyketides. In this prospective study, the average absolute error in predicting in vitro passive permeability (log Pe) was 0.5 log units. Of interest, over half (20/36) of the polyketides in this data set had MDCK-LE permeability
restraints on the aglycone backbone atoms of polyketides derived from vicinal proton coupling interactions measured by NMR spectroscopy. Our initial results suggest that the use of sparse experimental structural restraints is a facile and efficient way to restrict the conformational search and achieve more accurate results. The same data set of 37 polyketides was used as benchmark to compare PLOP performances against another macrocycle conformational sampling method which uses MacroModel.34 While this was not the main objective of our work, we believed it was important to assess the performance of our workflow against another well accepted methodology for macrocycle conformer generation. The results from conformational sampling of polyketides and other classes of macrocycles can be used as starting structures C
DOI: 10.1021/acs.jcim.6b00237 J. Chem. Inf. Model. XXXX, XXX, XXX−XXX
Article
Journal of Chemical Information and Modeling exceeding 3 × 10−6 cm/s, including 10 which have molecular weights exceeding 500 Da (largest is 847 MW), providing further evidence that macrocycles including polyketides can achieve drug-like passive permeability, including in some cases where they greatly exceed common rules-of-thumb for drug-like properties.
exhaustive search of backbone dihedral angles (i.e., those in the macrocyclic ring), at a given angular resolution. In contrast to previous work on protein loop and cyclic peptide sampling, no “rotamer libraries” for backbone aglycone dihedral angles were used for the macrocycle sampling process. Although in principle it might be possible to exclude certain dihedral angles as implausible, since they would lead to high energies, we have not found that to be necessary (and it would be challenging to implement for each new macrocycle); instead, we simply sample all possible values between −180° and 180°. The only exceptions are conjugated bonds, including double bonds and peptide bonds, for which a dihedral angle of either 0° or 180° can be assigned. The search process employs an adaptive resolution from 120° to as fine as 5°. Specifically, the dihedral angle resolution is gradually decreased (finer sampling) until more than 100 distinct macrocycle conformations have been generated. For efficiency, the backbone aglycone ring is separated into two independent halves during the dihedral angle sampling. After independent sampling of each half, pairs of conformations from each half that close the ring are identified. Specifically, each independently sampled half-macrocycle ends with the same atom, i.e., the atom where the ring was “broken”, and we consider the loop “closed” if the distance between the two copies of this atom (on each half) is 10 × 10−6 cm/s,
