Perspective pubs.acs.org/jmc
Fragment-Based Drug Discovery of Phosphodiesterase Inhibitors Miniperspective Fredrik Svensson,†,‡ Andreas Bender,‡ and David Bailey*,† †
IOTA Pharmaceuticals, St Johns Innovation Centre, Cowley Road, Cambridge CB4 0WS, U.K. Centre for Molecular Informatics, Department of Chemistry, University of Cambridge, Lensfield Road, Cambridge CB2 1EW, U.K.
‡
S Supporting Information *
ABSTRACT: Phosphodiesterases are proving to be fruitful targets for drug discovery. At the same time fragment-based drug discovery has matured into a powerful and widely applied technique. In this communication we review the application of fragment-based drug discovery for the successful identification of novel 3′,5′-cyclic nucleotide phosphodiesterase (PDE) inhibitors, concentrating on both experimental and computational strategies for fragment screening and hit-to-lead development. To this end, we also mine the open access databases ChEMBL and PDB for fragments showing PDE inhibitory activity, as well as SureChEMBL for recent PDE related patents, to provide a wider context for exploring fragment diversity. Together these approaches form an integrated experimental and computational platform to exploit fragmentbased drug discovery for this important gene superfamily.
■
PHOSPHODIESTERASES Cyclic nucleotide phosphodiesterases (PDEs) catalyze the hydrolysis of cyclic adenosine monophosphate (cAMP) and cyclic guanosine monophosphate (cGMP) to adenosine monophosphate (AMP) and guanosine monophosphate (GMP) (Scheme 1). Both cAMP and cGMP are essential components for cell signaling, and PDEs are important for regulating their intracellular concentrations, thereby maintaining cell function.1
gene encoding PDE12, cleaving 2′,5′-linked oligoadenylates, has been reported.3,4 Since the function and overall protein structure of this PDE are distinct from the other members of the PDE family, it will not be considered in depth in this review. Mammalian PDEs are broadly expressed, occurring in all tissues, although PDE6, PDE10, and PDE11 show high expression only in specific tissues, with PDE6 being found primarily in photoreceptors and the pineal gland, PDE10 in the brain and testes, and PDE11 in the prostate, testes, salivary gland, and the pituitary.5 Subcellular PDE localization is complex; recent studies of PDE10A expression in the brain show an elaborate topology, mediated by alternatively spliced protein variants,6 reflecting a highly regulated intracellular pharmacology.7 Crystallography enables a detailed understanding of the PDEs at a protein structural level. The first report of a crystallized PDE in 20008 revealed the structure of the catalytic domain of human PDE4B, and the PDB9,10 now includes structures of the conserved catalytic domains of all the human PDEs except PDE11. The PDE active site contains two divalent metal ions, one of these a Zn2+ and the other most often a Mg2+, held in a metal binding region at one end of the cyclic nucleotide binding pocket.11 The Zn2+ is essential in maintaining catalytic activity as it activates the water which catalyzes the hydrolysis of the
Scheme 1. Reaction Catalyzed by PDEs, Illustrated with cAMP as the Substrate
Mammals have at least 11 gene families encoding 3′,5′-cyclic nucleotide degrading PDEs (PDE1−PDE11), with additional intrasubtype diversity being obtained through different splice variants. Each PDE possesses a highly conserved catalytic domain but the superfamily as a whole shows large variation in its structural and regulatory domains. PDEs can be cAMP specific (PDE4, PDE7, and PDE8), cGMP specific (PDE5, PDE6, and PDE9) or catalyze the hydrolysis of both.2 A further © 2017 American Chemical Society
Received: March 14, 2017 Published: August 11, 2017 1415
DOI: 10.1021/acs.jmedchem.7b00404 J. Med. Chem. 2018, 61, 1415−1424
Journal of Medicinal Chemistry
■
cyclic phosphate ester bond of the substrate. Figure 1 shows the catalytic product AMP cocrystallized in PDE10A (PDB code
Perspective
PDE INHIBITION
The first PDE inhibitors to be discovered were in fact fragmentsized compounds (molecular weight of 80% inhibition at 200 μM.38,39 AstraZeneca have also reported fragment-assisted hit identification for PDE10A inhibitors.40 In a somewhat different approach, hits identified in a fragment screen were used to guide compound selection from the hits generated in a parallel HTS. This approach was chosen since fragment cocrystallization was not available for the project, making an entirely fragment based approach less attractive. Their fragment library 1417
DOI: 10.1021/acs.jmedchem.7b00404 J. Med. Chem. 2018, 61, 1415−1424
Journal of Medicinal Chemistry
Perspective
of fragment-based techniques and is an exciting development for the future. A number of different factors are presented as important for the decision to pursue fragment-based strategies in the studies shown in Table 1. In the early work by Card and Law an important consideration was the enhanced coverage of chemical space that fragment based strategies provide compared to conventional screening libraries.46 This advantage is amplified by the large size of the fragment libraries used in these studies. Other programs, in particular those targeting PDE10, have chosen a fragment based approach to exploit the more favorable physicochemical properties originating from specific fragments. The above examples show the versatility of fragment screening, both in terms of fragment library design and in the way in which information gained from the fragment screens can be used in drug discovery. FBDD approaches targeting the PDEs can be compared to earlier programs targeting the protein kinases,47 from which it is clear that similar factors, namely, amenability to X-ray crystallization, relatively high potency of starting fragments, as well as direct transferability of structural knowledge from one related target to another within the extended protein family, drive success in both programs.
Figure 4. One of the fragment hits (Ki = 210 μM) reported by Recht et al.37 for PDE10A (PDB code 4MSH).
■
SCREENING METHODS Fragments typically bind to their targets with a much lower affinity than hits identified in HTS, which imposes restrictions on the screening methods that can be used for their detection. On the other hand, the low number of compounds typically screened in an FBDD screening campaign allows the use of highly informative methods with a much lower throughput, such as NMR and X-ray crystallography.48 PDEs are multidomain effectors, with discrete, highly conserved catalytic domains.15 Screens can use both full-length enzymes or isolated catalytic domains,49 although crystallization studies are usually carried out using the catalytic domain. One structure of a near full length PDE is available in the PDB,50 with the associated report illustrating a mechanism for allosteric regulation of PDE activity that might lead to differences in assay results between full-length protein and isolated catalytic domains.51 Few studies have reported using both full length and catalytic domain with the same set of compounds, although in the study of Orrling et al.,42 results from inhibition data obtained using the catalytic domain alone appear to translate well to the full-length enzyme. There are reports in the literature of differences in inhibition profile stemming from binding of compounds to sites other than the catalytic domain, only detectable using the full length enzyme.49 Of the other studies analyzed in this review, most used catalytic domain constructs for the primary fragment screening with the exceptions being Krier et al.,34 who used purified PDE from bovine aorta, and Raheem et al.39 and Shipe et al.38 who used transiently expressed full-length enzymes. An understanding of the impact of additional domains, especially the nucleotide-binding GAF domains,51 on full-length PDE enzyme function and inhibitor design is an important question that requires further comparative studies. Surface plasmon resonance (SPR) is a popular and informative method for fragment screening, and its use has been reported in the development of inhibitors for PDE4,52 PDE5,53 and TbrPDEB1.52 SPR readily identifies fragment hits in the mM to μM range but has the additional benefit of measuring the kinetics of binding, yielding both “on” and “off”
Figure 5. Crystal structure (PDB code 5C1W) including one of the fragment hits (Ki = 8.7 μM) reported from the Merck PDE10A program.38
included 3000 fragments and produced a hit rate of 14% when screened at 100 μM. The use of fragment screening data to aid HTS campaigns has also been reported by researchers from Novartis.41 Not all reported PDE inhibitors come from targeting human PDEs, and recently the use of PDE inhibitors to tackle parasitic diseases has attracted some attention,22 with the Phosphodiesterases for Neglected Parasitic Diseases (PDE4NPD) project aiming to develop selective inhibitors of parasite PDEs. To this end, Orrling et al. report the development of PDE inhibitors targeting Trypanosoma brucei PDEB1, starting from fragments.42 Fragment screening for both Trypanosoma brucei PDEB1 and human PDE4 inhibitors has also been performed.43 A library of 1040 fragments was screened at 200 μM against the parasite PDE and at 100 μM against human PDE4, identifying 26 and 41 hits, respectively. Fragments that were promising hits in biochemical assays were then screened in parasitic phenotypic screens to identify antiparasitic activity against a range of different parasites. This study nicely illustrates the point that fragments can be potent enough inhibitors to display activity in cell-based phenotypic screens, as further demonstrated in a recent study by Parker et al. employing fragmentbased screening in human cells.44 The advent of cell-based mechanism-informed phenotypic screening45 broadens the uses 1418
DOI: 10.1021/acs.jmedchem.7b00404 J. Med. Chem. 2018, 61, 1415−1424
Journal of Medicinal Chemistry
Perspective
Scheme 2. Optimization of the Initial Hit Compound in the Merck PDE10A Project
methods that have successfully been applied to fragment screening against the PDEs, encompassing many of the commonly used methods in FBDD.48,60 However, it is difficult to compare the success rates of different screening strategies due to the varying fragment library composition, screening concentrations, and activity cut-offs employed. Given the relative ease of PDE domain crystallization, it is fair to say that design-led screening strategies for PDE inhibitors should include early stage crystallization of hit compounds, as early structural information has been shown to greatly facilitate lead optimization.
rates for ligand binding. SPR methods typically use immobilized catalytic domains of the PDEs, but other assay configurations can also be used, such as the displacement assays using platebased optical sensor technology reported by scientists at AstraZeneca for their PDE10A program.54 An inhibitor is immobilized on the chip, to which a solution containing the PDE protein is added, which coats the immobilized inhibitor. Compounds can then be added in solution, and the rate at which they displace the protein can be measured. This approach removes the need to modify the protein for attachment to the chip and has a very broad sensitivity range, identifying inhibitors with a KD in the range 40 nM to 500 μM. Recht et al. have reported fragment screening against PDE4A 36 and PDE10A, 37 applying a combination of calorimetry and X-ray screening. Nanoscale isothermal titration calorimetry (ITC) screening was conducted at a concentration of 2 mM, and compounds displaying promising inhibition were advanced to X-ray screening. Calorimetric approaches are attractive since they do not require compound prelabeling, and readout is independent of the spectroscopic properties of the screening collection. Not only biophysical but also biochemical approaches have been used for fragment screening against PDEs. Examples include the use of SPA to measure PDE4 activity,33 and immobilized metal ion affinity-based fluorescence polarization assay (IMAP) to measure PDE10A activity.55 SPA measures levels of protein-bound radiolabeled cAMP and is usually carried out in a 384-well plate format, while IMAP is routinely configured for ultrahigh throughput screening in 1536-well microplates. Computational methods are an essential component in FBDD, both in general29,56 and for efforts targeting individual PDEs,35 helping to design the fragment library and expand the identified hits. This is of particular value when the project does not have ready access to crystallography to determine fragment binding modes. Virtual screening of fragments is generally considered a more difficult task compared to that for drug-sized molecules, mainly due to the low binding affinity of fragments and their high number of potential poses.57 The use of molecular interaction fingerprints has been shown to improve the success of fragment virtual screening and docking, exemplified among other targets by PDE4D.58 Ultimately, the choice of screening method (biophysical or biochemical) will reflect the nature of the targets (soluble domains, membrane-bound protein) and the throughput required. The range of assays available gives plenty of scope for precise tailoring to the situation in hand and allows for the use of multiple techniques in parallel, a practice that has gained increasing traction.59 The examples above illustrate the many
■
FRAGMENT TO LEAD STRATEGIES Once a fragment hit has been identified, different strategies can be applied to develop the fragment into a lead structure.61 Even though fragment hits generally have a lower affinity than hits generated in high throughput screening (HTS) campaigns, the hits often have a high binding affinity per heavy atom. This is referred to as ligand efficiency (LE),62 calculated as binding energy per non-hydrogen atom. Similar metrics, such as binding affinity in relation to lipophilicity of the fragment (lipophilic ligand efficiency, LLE)63 can also be derived. However, LLE does not relate directly to the size of the molecule and therefore might not be the best choice for fragment prioritization.64 Although LE and LLE metrics are widely applied, they have also been the focus of some criticism.65 Both size and lipophilicity tend to increase during the hit optimization process, and LE and related measures can serve as useful milestones for monitoring physicochemical properties during lead development. The PDE10A program reported by Merck38,39 describes how an initial fragment hit was developed through iterative rounds of testing and X-ray crystallographic optimization. The study also illustrates some of the opportunities facilitated specifically by fragment screening. Lead compounds deemed to suffer from off-target liabilities were immediately improved by swapping their cores for another fragment hit from the initial screen. Key structures from the Merck optimization program are shown in Scheme 2; initial fragment hits were developed into a highly selective final inhibitor while maintaining high LE and improved physicochemical properties. Selectivity toward PDE10 was obtained by focusing design around two key amino acids in the active site. Tyr683 is only present in PDE2 and PDE10, and by interacting with this residue while also exploiting the difference in site accessibility resulting from a change from Gly715 in PDE10A to Leu715 in PDE2, highly selective inhibitors were produced.38 1419
DOI: 10.1021/acs.jmedchem.7b00404 J. Med. Chem. 2018, 61, 1415−1424
Journal of Medicinal Chemistry
Perspective
Orrling et al. describe a fragment growing strategy using iterative docking into a homology model to guide optimization that was applied to develop potent inhibitors against TbrPDEB1.42 Starting from a fragment-sized inhibitor derived by combining features from known PDE inhibitors, an improvement in potency from an initial 12 μM to a final 0.049 μM was achieved, albeit at relatively low LE (Scheme 3). Figure 7. Overlay of the 70 PDB structures containing a fragmentsized inhibitor (44 unique fragments) crystallized with a PDE, superpositioned on the PDE10A active site. Binding modes of crystallized fragments are very similar across the PDE superfamily, interacting with the hydrophobic clamp and the invariant glutamine.
Scheme 3. Orrling et al. Report Fragment Growing as a Method of Improving Potency
preferred mode of binding for PDE inhibition. This has implications for the structure of active fragments. In order to form the interactions described, fragments have to contain an aromatic ring system as well as a hydrogen bond donor and/or acceptor in close proximity. These features contribute to defining a core pharmacophore for PDE inhibition. Guidelines for idealized fragment properties have been formulated as the “rule of 3”,67,68 which generalizes fragment hits for optimal lead development as having a molecular weight of
