Review pubs.acs.org/CR
Modulators of Protein−Protein Interactions Lech-Gustav Milroy,† Tom N. Grossmann,‡,§ Sven Hennig,‡ Luc Brunsveld,† and Christian Ottmann*,† †
Laboratory of Chemical Biology and Institute of Complex Molecular Systems, Department of Biomedical Engineering, Technische Universiteit Eindhoven, Den Dolech 2, 5612 AZ Eindhoven, The Netherlands ‡ Chemical Genomics Centre of the Max Planck Society, Otto-Hahn Straße 15, 44227 Dortmund, Germany § Department of Chemistry and Chemical Biology, Technical University Dortmund, Otto-Hahn-Strasse 6, 44227 Dortmund, Germany 3.4.2. Microarrays 3.4.3. One-Bead-Two-Compound Approach (OBTC) 3.4.4. Fragment-Based Drug Discovery (FBDD) 3.4.5. In Silico Screening 3.4.6. Multicomponent Reactions (MCRs) 4. Stabilization of PPIs 4.1. Cyclosporin A, FK506 and Rapamycin 4.2. Forskolin and Brefeldin A 4.3. Inositol Tetraphosphate 4.4. Phytohormones Auxin, Jasmonate, and Brassinolide 4.4.1. Auxin 4.4.2. Jasmonate 4.4.3. Brassinolide 4.5. Small-Molecule Stabilizers of OligomericState Homocomplexes 4.5.1. Tafamidis 4.5.2. Phenothiazines 4.5.3. Influenza Nucleoprotein (NP) 4.5.4. MDMX Homodimer Stabilizers 4.5.5. Topoisomerase II/ICRF-187 4.5.6. 1-EBIO Class Stabilizers of SK-ChannelCaM Interaction 4.5.7. RAR-NCoR 4.6. PPI Stabilizers of 14-3-3 Protein−Protein Interactions 4.6.1. Fusicoccin A 4.6.2. Cotylenin A 4.6.3. Pyrrolidone1 and Epibestatin 5. Concluding Remarks Author Information Corresponding Author Notes Biographies Acknowledgments References
CONTENTS 1. Introduction 2. Biochemical Techniques for Ligand Identification 2.1. Surface-Based Binding Assays, ELISA 2.2. Surface Plasmon Resonance, SPR 2.3. Fluorescence Polarization 2.4. Proximity-Based Methods 2.4.1. Fluorescence Resonance Energy Transfer (FRET) 2.4.2. Time-Resolved FRET (TR-FRET) 2.4.3. Bioluminescence Resonance Energy Transfer (BRET) 2.4.4. Amplified Luminescent Proximity Homogeneous Assay Screen (ALPHA Screen) 3. Approaches for Hit Identification of PPI Modulators 3.1. Structure-Based Design 3.1.1. Hotspot Residue Theory 3.1.2. Conformationally Constrained Peptides and Miniproteins 3.1.3. Peptidomimetics 3.1.4. Mimetics of Protein Secondary Structure 3.2. Natural-Product Inspired PPI Modulation 3.2.1. Natural Products 3.2.2. Approaches to Diversifying Natural Products 3.3. Supramolecular-Induced PPI Modulation 3.3.1. Supramolecular-Induced Protein Dimerization and Oligomerization 3.3.2. Supramolecular-Modulation of the Protein Surface 3.4. Compound Library Generation 3.4.1. Biological Techniques © 2014 American Chemical Society
4695 4696 4697 4697 4698 4699 4699 4699 4700
4700 4701 4701 4701 4703 4709 4712 4714 4714
4723 4723 4723 4725 4726 4727 4727 4728 4728 4728 4728 4729 4729 4730 4730 4730 4731 4731 4732 4733 4733 4734 4734 4735 4735 4736 4737 4737 4737 4737 4738 4738
1. INTRODUCTION Since Hedin’s characterization of trypsin and antitrypsin in 1906,1 arguably the first account of a regulatory protein− protein interaction (PPI), contemporary understanding of proteins and PPIs has been progressively transformed by
4715 4719 4719 4720 4721 4721
Special Issue: Chemical Biology of Protein-Protein Interactions Received: December 10, 2013 Published: April 15, 2014 4695
dx.doi.org/10.1021/cr400698c | Chem. Rev. 2014, 114, 4695−4748
Chemical Reviews
Review
compounds have been descended from a naturally occurring agent such as a protein, peptide or secondary metabolite. This review will emphasize actual and emerging approaches to designed molecules targeting binary PPIs with a well-defined interface. Work published from 2008 onward receives the highest priority as does work reporting high resolution protein X-ray crystallographic data as evidence of the interaction of a small molecule with its target protein−protein interface. Interfacial18 as well as allosteric PPI modulators are both covered in this review, whereas higher-order multiprotein complexes are not.19,20 Given the stringent selection criteria and despite our best attempts to comprehensively cover as many different aspects of PPI modulation as possible, we regret that some important pieces of work and topics may have been omitted, and that some of the topics covered may not have received as much focus as others. In this case, where possible we cite references to relevant and comprehensive review articles. This review begins with a up-to-date account of the principle biochemical techniques used to identify PPI modulators (section 2), e.g., ELISA and FRET based techniques, many of which are referred to in later sections. In the next section, a comprehensive account is provided of the different chemical methods currently used to access designed small molecule inhibitors of PPIs, including such details as the synthetic routes to the target molecules, as well as their molecular structures and PPI targeting properties (section 3). Finally, a highly detailed atomistic-level account of different established and new small molecule stabilizers of PPIs is given (section 4), to highlight the enormous potential of PPI stabilization as a drug discovery strategy, and yet it remains a conceptually underexplored mode of PPI modulation.21−24
landmark conceptual and technological advances in molecular cell biology, biochemistry and biophysics,2 not least, the sequencing of the human genome and the ensuing genomic technologies. Today, proteins can be viewed as the molecular smart phones of the cell, genetically programmed to enact specific cellular functions in response to external stimuli. Individually, proteins perform essential functions such as catalysis and the transport of molecules and ions. However, their effectiveness in the crowded cellular environment is only short-range and insufficient to sustain life without the involvement of other biomolecules such as other proteins or metabolites. Proteins manage long-range effectiveness through wider interactomes, highly organized and responsive protein networks, which relay protein function cell-wide via interactions between protein nodes: so-called protein−protein interactions (PPIs).3−6 This heightened awareness of a significant and ubiquitous role for PPIs and PPI networks in cellular physiology has created numerous new opportunities for drug discovery, as in principle every pathologically significant PPI becomes a potential drug target.7,8 The challenge though now is finding the small molecule modulators capable of performing the task.9,10 The major motivation behind campaigns in PPI drug discovery today is the urgent need for safer and more effective medicines for the benefit of patients. The targeting of small molecule modulators at conventional protein targets, for instance enzymes11 and ligand-activated transcription factors,12 has historically been a highly effective (and profitable) way to treat many types of diseases. However, a perceived crisis in research and development (R&D) productivity in the pharmaceutical industry today, which is connected with an over-reliance on the same drugs targeting the same protein targets, has called for innovative approaches to discover new small molecule modulators as a basis for new drugs.13,14 Furthermore, diseases common to old-age, such as cancer and Alzheimer’s are expected to become more prevalent in the future as average life expectancies in countries across the world continue to nudge upward, and yet, disturbingly, for many of these diseases there is currently no effective treatment, or current drugs suffer from issues of drug resistance, high toxicity or low efficacy. As this review seeks to clarify, PPI drug discovery has the potential to deliver on both fronts, given the ubiquitous importance of PPIs to cellular pathology, and their orthogonality to conventional drug targets. With reference to a recent review article on PPI drug discovery,15 we define a PPI as, “an interaction of two identical or dissimilar proteins at their domain interfaces that regulates the function of the protein complex (interactions involving enzyme active sites are not termed PPIs in drug discovery)”,15 and a small molecule modulator as, “a low-molecular weight natural product or synthetic agent with a significant degree of structural complexity, thus allowing target selectivity and good binding affinity, which can regulate a PPI through either direct or allosteric inhibition, or stabilization”.15 The review’s focus on drug-like small molecule modulators unfortunately means that no-less-relevant, larger biomolecules, for example nucleic acid as well as peptide aptamers16 and antibodies,17 have been excluded from consideration this time. The generality of the review’s title, “Modulators of Protein− Protein Interactions”, means that rational, structure-based approaches as well as random approaches based on highthroughput screening to identifying small molecule PPI modulators are equally considered. Furthermore, the modulator
2. BIOCHEMICAL TECHNIQUES FOR LIGAND IDENTIFICATION To interfere with a PPI of interest, compounds fit for the purpose must first be identified. Nowadays, various interdisciplinary techniques appear in the literature ranging from cell based high content screening and biochemical in vitro assays to bioinformatical methods. Cell based assays often address a certain cellular phenotype (proliferation, apoptosis, and morphology) rather than targeting a defined PPI. Therefore, these formats are not within the scope of this review. Bioinformatical methods usually require a high degree of insight into the PPI of interest including for example structural data. In silico screenings appear increasingly in the literature, and we describe some examples in section 3.4.5. However, the vast majority of compounds are still identified and validated with biochemical assays. Depending on the knowledge of the PPI, different sized collections of these molecules ranging from small arrays to large library collections of several hundred thousand compounds need to be tested in a reliable and fast manner (high-throughput screening, HTS). Accordingly, depending on the general principle of these screening methods and the biophysical characteristics that the proteins of interest have to possess, a differentiation can be made between surface based and proximity based assays. Hereafter, we will describe the general idea and the advantageous as well as disadvantageous features of these different assay types. New technologies and assay principles, claiming to be beneficial for the identification of small molecules in drug discovery, are constantly appearing in the literature. We have decided to 4696
dx.doi.org/10.1021/cr400698c | Chem. Rev. 2014, 114, 4695−4748
Chemical Reviews
Review
whereas the types of ELISAs vary from indirect, sandwich, to competitive setups. A general advantage of this method is the simplicity of instrumentation, as probes as well as plate readers are offered by numerous vendors. ELISAs show a high degree of flexibility and a rather high sensitivity. They are easy to set up using standard laboratory equipment and therefore suitable for small to medium sized libraries. On the other hand, assay development still requires specific antibodies and detection reagents. Moreover, the rather harsh washing conditions could affect weak and transient interaction, as they are common for PPIs, resulting in false hit detection (false positives or false negatives depending on the assay layout). Additionally, extensive blocking and washing steps as well as antibody incubations make these assay types a bigger effort for automated liquid handling systems during HTS. However, the ELISA technology is a very versatile tool in PPI modulation studies. This is illustrated by the fact that the inhibitors of protein−protein database (iPPI-DB) reports usage of ELISAs in about 20% of their collection.30 ELISAs performed during SAR studies are, for example, the LFA-1−ICAM interaction that plays a role in cell−cell adhesion of autoimmune diseases,31,32 acylpyrogallols as potent Bcl-2 inhibitors33 as well as the derivatives of an isoindolinone scaffold as MDM2−p53 inhibitors.34 Also, many examples from peptide ensembles of different size can be found for instance as antibiotics against the E. coli RNA polymerase σ(70)−core interaction35 or the modulation of the glycoprotein Ib−von Willebrand factor complex important during platelet-dependent thrombus formation.36 Another example is the PED/PEA15− D4a complex, which plays a crucial role in insulin resistance in type 2 diabetes and was successfully addressed in a screening of ∼20 000 peptides.37
focus on successful and well-established assay formats, with some examples in the context of PPI drug discovery. 2.1. Surface-Based Binding Assays, ELISA
Enzyme-linked immunosorbent assays (ELISAs) were first described in 1971 as a nonradioactive technology investigating antibody−antigen interactions.25 Decades later, these are still valid tools for the investigation of PPIs and their modulation. In brief: one binding partner is bound to a surface by nonspecific adsorption, covalent linkage, or affinity tags (streptavidin−
Figure 1. Assay principal for the ELISA. A surface bound PPI complex is exposed to either an inhibitory molecule (upper case, light red compound) or a noninhibitory molecule (lower case, light blue compound). Specific antibodies recognize the PPI complex and result in a signal derived from the activity of a conjugated enzyme (e.g., HRP, yellow circle). Differences in enzymatic activity are a measure of different efficacies of inhibitors.
2.2. Surface Plasmon Resonance, SPR
When a gold coated glass chip is illuminated with polarized light under conditions of total reflection, a proportion of the light enters the coated surface as a so-called evanescent wave. This is the underlying principle of surface plasmon resonance (SPR) experiments. The energy of this evanescent wave is missing in the reflected light beam and can be detected at a certain angle (SPR angle). Upon binding of macromolecules on the surface of the sensor, the SPR angle changes (Figure 2). These changes can be measured over time. SPR measurements have become popular in many fields over the years. The technique is also applicable for sensing changes in protein complexes on the surface of the chip which makes it a valuable tool for PPI investigations.38,39 SPR measurements are performed in a label-free manner; yet, one partner must be surface immobilized requiring the modification of involved proteins. Initial shortcomings of this method such as the limited number of flow chambers were substantially improved by more advanced microfluidic systems and the introduction of more advanced automated sample handling systems (e.g., 96- and 384-well systems). Additionally, the sensitivity of the machines increased, allowing the precise measurement of the direct binding of small molecule fragment libraries with weak affinities to an immobilized protein.40,41 Up to now, SPR systems are not applicable to initial compound screening in large formats such as HTS. This is due to the limited sample number these systems can handle in parallel. Also, the financial cost of purchasing such a machine is significant. Nevertheless, SPR offers an extensive pool of information as a secondary or downstream in depth validation
biotin, GST-GSH, and GFP-trap). Subsequently, residual binding spots on the surface are blocked and an incubation with the second binding partner is performed. Quantification of binding events is achieved by a specific antibody for the binding partner that is in turn recognized by an enzyme-coupled secondary antibody (Figure 1). Commonly used conjugated enzymes are horseradish peroxidase (HRP) and alkaline phosphatase.26,27 Miniaturization into microtiter plates of decreasing size as well as technological advances, such as automated liquid handling technologies, and increased sample throughput has enabled this method for HTS. The introduction of lanthanide based fluorophores instead of cross-linked enzymes enhanced the system, as it enabled time-resolved measurements.28,29 Dissociation-enhanced lanthanide fluorescence immunoassay (DELFIA, Perkin-Elmer) makes use of these fluorophores in an advanced manner. Here, the secondary antibody carries lanthanide ions (e.g., europium, samarium, or terbium). After extensive washing steps the ion is liberated. Once free in solution, the ions are incorporated into chelating complexes that enhance the fluorescent signal. These caged lanthanide fluorophores feature sharp, red-shifted emission peaks which allow detection in a spectral range where usual biological material or compounds do not interfere with the read out. Additionally, these chromophores feature long half-lifetimes that enable time-resolved measurements. Nowadays, the term ELISA is invariantly used to describe all sorts of surface binding based assays. Detection strategies vary from chromogenic, fluorescent to chemoluminescent read outs, 4697
dx.doi.org/10.1021/cr400698c | Chem. Rev. 2014, 114, 4695−4748
Chemical Reviews
Review
interleukin-2 to its receptor43 and the in vivo activation by MDM2 antagonists.44
Figure 3. Assay principle for a fluorescence polarization assay (FP). A fluorescently labeled molecule rotates slower when complexed to its protein binding partner (light gray body) compared to its rotation when free in solution upon disruption of the complex, e.g., by a competitor (light red compound). Therefore the FP signals differ, a principle which can be used to identify PPI modulating compounds.
2.3. Fluorescence Polarization
The principle of fluorescence polarization (FP) is that fluorescent molecules emit light of a different plane as before when excited with linear polarized light. This effect is caused by the movement of the molecules in solution between excitation and emission. Since large molecules rotate slower than smaller entities, a difference in polarization can be observed for fluorescently labeled species of different size (Figure 3). The basic mechanism of this effect was initially described in 1926 by Perrin.45 FP can be quantified by measuring the contributions of the light intensity parallel and perpendicular to the plane of the excitation light beam. In the following decades, FP was used intensively in analyses of biological molecules in general and in drug discovery approaches including PPI investigations.46,47 Various ways of fluorescently labeling proteins are known. Some of these are not suitable for FP based assays as they increase the mass of the protein under study significantly (like fluorescent protein fusions e.g. GFP or fluorescently labeled antibodies). Chemical modifications with commercially available rhodamine, fluorescein or BODIPY dyes using thiol, amines, or acid reactive groups of proteins are currently a common choice. Once a protein of interest is labeled potential binding partners can be added and differences in polarization induced by compound binding can be observed (Figure 3). The assay essentially relies on a mass difference (preferably ∼10×) of the molecule carrying the fluorophore unbound vs bound to its protein partner. Therefore, the assay is often used in a competition format where the fluorescent part is represented by a peptide derived from the original binding partner. This peptide is usually dispatched from its unlabeled partner upon competitive compound binding, thus leading to a strong difference in FP. As the assay is not relying on surface decoration of macromolecules it is relatively straightforward to miniaturize into simple uncoated microtiter plates of all sizes (96-, 384-, and 1536-well plates), saving biological reagents as well as compound by simply reducing the assay volume to a few μL. Also, all components of the assay can be mixed prior to compound administration, dispensed into plates, and read out directly (“mix-and-read-assay”). In contrast to ELISA assays,
Figure 2. Principle of surface plasmon resonance (SPR). Under condition of total reflection, polarized light is penetrating the gold coated surface of a glass chip as an evanescent wave. The energy of the latter is missing in the detector in a certain angle (SPR angle). The SPR angle is measured over time. Molecules binding on the surface shift the angle and can therefore be detected. (a) Shows ligand 1 immobilized on the chip surface. (b) Upon application of a binding partner, the signal increases. (c) Application of an inhibitor lets the signal decreases.
tool. Since kinetic measurements are performed, quantitative binding characteristics such as on- and off-rates can be obtained in addition to merely qualitative affinities.41 The iPPI-DB currently lists SPR in about 5% of their biochemical tests as the method of choice. These involve the inhibition of BET recruitment to chromatin,42 the inhibition of the cytokine 4698
dx.doi.org/10.1021/cr400698c | Chem. Rev. 2014, 114, 4695−4748
Chemical Reviews
Review
2.4. Proximity-Based Methods
blocking, extended antibody incubation periods, and plate washing are not required during the screening process. Fluorescent probes are used in rather low concentrations (depending on the PPI complex) so that the total costs per well are comparably low. For larger libraries, automated liquid handling is beneficial, but as FP is a mix-and-read-assay, a simple liquid dispenser is sufficient. One disadvantage of FP assays is that the assay cannot be performed in a time-resolved manner, as the signal is to be measured instantly after excitation. Nonetheless, FP is the most frequently used assay format for the identification of PPI modulators. It is one of the most common primary assays on HTS platforms as its throughput can easily be scaled up. FP assays used for identification of PPI modulators can be found in various biological fields. The iPPI-DB reports FP assays in top ranking 51% of their cases.30 Additionally, there are several recent examples using FP assays targeting MDM2-MDMX interaction with the tumor suppressor p53,48 the hypoxia inducible factor-1α (Hif1α) in complex with p30049 that plays a role in activating VEGF and its pro-angiogenic function, or the oncogene Casitas B-lineage Lymphoma (Cbl).50 The modulation of pro-apoptotic regulators Bcl-x(L) and Mcl-151 as well as inhibiting the negative regulator Keap1 from its interaction with the anti-inflammatory and anticancerous transcription factor Nrf252 utilized successfully FP assays to identify functional peptides. Also, the 14-3-3 adaptor protein involved in many cellular functions was addressed several times using FP to find different classes of inhibitors like covalent binders, natural products from beetle extracts or phosphonates.53−55 FP played and will play a major role in drug discovery targeting PPIs. Nonetheless, FP is limited to PPI systems composed of partners whose size difference is large enough for a significant difference in FP signal. In practice, this often means that one binding partner has to be minimized to the essential interacting patches of peptide size to serve as a model system. To overcome these limitations, proximity based assay formats represent suitable alternative solutions (section 2.4).
2.4.1. Fluorescence Resonance Energy Transfer (FRET). Förster (or fluorescence) resonance energy transfer (FRET) is a phenomenon that occurs if the emission energy of a given excited donor molecule matches the excitation energy of an acceptor fluorophore which is in close proximity and can be excited (Figure 4). This acceptor molecule then emits the according energy as fluorescence. As each of these steps can be coupled with radiation-less relaxation of each state, energy is removed from the system so that the wavelength is red-shifted during FRET. This process was described by Theodor Förster in 1948.56 Since then this principle was used for many kinds of physical and biological analyses especially in imaging applications.57 FRET signals are strictly dependent on the distance (d) of the two fluorophores as FRET efficiency is proportional to d−6. Thus, energy transfer occurs within a range of 10−100 Å.58 For PPI measurements, two proteins need to be labeled with FRET-pairing fluorophores. Upon interaction the proteins bring the two fluorophores in proximity, within the suitable range, and FRET occurs.59 As this system is not size dependent, fluorescent labeling of the two proteins can be achieved either by chemical modification as described before in section 2.2 (using, e.g., dansyl-fluorescin, Cy3-Cy5, or Alexa488-Alexa594 as suitable FRET pairs) or by using specific fluorescently conjugated antibodies or fluorescent fusion proteins (e.g., CFPYFP and GFP-mCherry are suitable FRET pairs) or mixtures of both systems.60 Also, quenching molecules like dabsyl and dabcyl can be used in FRET assays. In these cases the fluorescence of the donor conjugated molecule can only be observed upon displacement of the quencher conjugated binding partner.61 Due to relatively low costs per well (depending on the complexity of the protein labeling strategy), FRET is highly applicable to compound screening. Also, it overcomes the size limitation problem of FP assays, its signal is robust, and these types of assays are straightforward to handle in terms of instrumentation. A typical FRET assay requires dispensing of protein mixtures and compound addition with subsequent fluorescent read out, which is supported by any plate reader. However, a downside of classical FRET assays in HTS is that compounds excited with the rather high energetic excitation beam often lead to false positive read outs. Several approaches have been taken to improve the assay performance of FRET for instance by substituting the initial excitation of the system with an enzymatic reaction, thereby utilizing bioluminescence as the energy source (BRET, section 2.4.3). Alternatively, donor fluorophores with long half-lifetimes where successfully introduced, shifting the measurement window in order to distinguish rapidly decaying signals of false positives from true hits (TR-FRET, section 2.4.2). As these improved formats have been used widely in PPI modulations, the iPPI-database only reports 0.3% of their cases as classical FRET based assays.62−64 The FRET technology has constantly been developed further and is still an ongoing field of research. Lately, miniaturization attempts were made for instance in a Lab-on-a-Chip manner where Benz and colleagues were able to measure PPIs in droplets of pL size.33 However, these technical novelties have yet to result in successful drug discoveries. 2.4.2. Time-Resolved FRET (TR-FRET). In order to eliminate the excitation and emission signals of biological as well as small chemical molecule reagents in the homogeneous assay mixture of FRET based assays, a modulation of the
Figure 4. Theoretical constraints for FRET. (a) Jablonski diagram showing the principal of FRET. A donor molecule becomes excited and instead of directly emitting fluorescence, the energy is transferred to an acceptor molecule that is excited and emits light. Radiation-less relaxation of excited states shifts the energy downhill and causes a redshift in the final fluorescence. (b) Schematic spectra of a FRET-pairing donor and acceptor molecule. Emission spectrum of the donor molecule (EmD) overlaps (gray area) with the exitation spectrum of the acceptor molecule (ExA) for FRET. Excitation wavelength λExD and emission wavelength λEmA of the FRET-pairing system are indicated (arrows). 4699
dx.doi.org/10.1021/cr400698c | Chem. Rev. 2014, 114, 4695−4748
Chemical Reviews
Review
parameters of TR-FRET assays are accountable for the rise in usage of these types of assays. In the field of PPI modulation TR-FRET systems play a role at all stages of drug development, ranging from primary assays for HTS, secondary assays and IC50 determinations. The iPPIDB quotes 20% of their content as derived from TR-FRET based assays.30 An example of successfully utilized TR-FRET is the inhibition of eph-kinase,69 where the authors used LANCE to validate and quantify their hits from an ELISA. The TWEAK-Fn14 complex plays a role in inflammation, autoimmune diseases, and cancer and was successfully evaluated in a 60 000 compound HTS using HTRF.70 Chung and Bamborough used time-resolved measurements in a fragment based setup identifying bromodomain inhibitors.71,72 Despite its slightly advanced instrumentation and costs, TR-FRET assays deliver many advantageous features and represent a valuable tool for PPI identification. 2.4.3. Bioluminescence Resonance Energy Transfer (BRET). In another derivative of FRET assays, the donor molecule is conjugated with an enzyme (e.g., Renilla luciferase, Rluc) that causes bioluminescence upon substrate addition (e.g., coelenterazine, coelenterazine 400a or DeepBlueC). The high energetic light pulse which leads to undesired excitation of biomolecules and small chemical compounds at the beginning of the measurement is lacking and the acceptor molecule is excited by the internal bioluminescence as light source.73,74 Suitable BRET-pairing systems are: Rluc-eYFP, Rluc-GFP2, Rluc8-GFP2, Firefly-DsRed, and Rluc/Rluc8-QDOT.75 BRET assays are most broadly applied to the analyses of membrane signaling such as GPCRs,76,77 whereas only few examples exist currently for the successful usage of BRET in PPI modulation: Hamdan and colleagues showed a β-arrestin recruitment to the chemokine receptor CCR5 using a cellular BRET screening of 26 000 compounds.78 Further examples are the CDK5−p25 interaction which is of crucial relevance to neurodegenerative diseases as well as the Bax-Bcl-x(L) interaction which is important for the regulation of apoptosis.79,80 2.4.4. Amplified Luminescent Proximity Homogeneous Assay Screen (ALPHA Screen). ALPHAScreen (amplified luminescent proximity homogeneous assay Screen, Perkin-Elmer) represents a remarkable advancement in the field of proximity based assays. It is a bead based proximity assay that has its origins in diagnostic assay technology termed LOCI (luminescent oxygen channeling immunoassay). The proximity principle is similar to FRET based systems, as the interaction of two labeled proteins result in the transfer of a signal from a donor to an acceptor, which generates a read out signal (Figure 6). The molecular details, however, are strikingly different: ALPHAScreen is a bead based assay, binding partners need to be tagged (e.g., via epitope recognition, GST, His6, HA, Flag, etc.). In ALPHAScreen, upon excitation of the donor bead with low energy red-shifted light (680 nm), the photosensitizing phthalocyanines release electronically excited singlet oxygen molecules that can diffuse in solution within a radius of 200 nm (compared to 1−10 nm with FRET) limited by their half lifetime. The acceptor beads contain three dyes, namely thioxene, anthracene, and rubrene. Once the excited oxygen molecule interacts with the thioxenes energy is released, which is subsequently transferred to the cascade of anthracene and rubrene and finally results in the emission of light at 520−620 nm.81 As one donor bead is able to release up to 60 000 singlet oxygen molecules the signal of one biological molecule bound
system was required. In TR-FRET lanthanide ions are used as fluorophores, exploiting their extended half-lifetimes in the right environment and in combination with suitable short lifetime acceptor molecules.28,65,66 Different vendors offer different solutions for coordination of the lanthanide ions (e.g., chelating methods: lanthanide chelate excite ‘LANCE’, Perkin-Elmer, or cryptates: homogeneous time-resolved fluorescence ‘HTRF’, CisBio).67 The general principle for time-
Figure 5. Assay principle of TR-FRET. (a) After donor excitation (blue arrow, λExD), PPI-dependent FRET occurs and red-shifted fluorescent light is emitted (red arrow, λEmA). The maximum distance for the resonance energy transfer is indicated (10−100 Å). Upon treatment with an appropriate inhibitor the FRET signal decreases. (b) Diagram indicates the long lifetime of caged lanthanide-ion fluorophores such as Terbium or Europium after excitation over time. These fluorophores are used in time-resolved measurements (gray area) in which the plate reader starts collecting photons after a short delay time to let the intrinsic fluorescence of matrix and compounds decay.68
resolved measurements is the same: long-lived donor fluorescence is used in combination with a delay time period of some μs in which unspecific excited molecules will rapidly decay their fluorescence. Subsequent read out of the acceptor signal is possible and enhances the specificity of the read out (Figure 5). Ratiometric correction (read out signal/donor fluorescence intensity) for differences in donor fluorescence from well to well are possible for a better signal comparison. Common TR-FRET-pairs are commercially available. Remarkably, Tb2+-cryptates (CisBio) can also be paired with green acceptors such as fluorescein or GFP enabling multiplex assays.68 Unlike FP assays TR-FRET assays are mass difference independent allowing different ways of labeling. Thus, vendors usually provide probes for classic chemical labeling of proteins as well as conjugated antibodies against common protein-tags (e.g., GST, His6, Myc, HA, Flag, etc.). TR-FRET assays find broad application in various fields of biology like receptor studies (GPCRs), kinase studies, protein−DNA interaction studies, cell surface protein interaction as well as intracellular PPIs.68 Compared to other assays like FP or ELISA, TR-FRET usually is more cost intensive per well. Additionally, the read out instrument of choice requires the implementation of a timeresolved method. Collectively, despite somewhat higher costs, the minimized false positive rate and the good statistical 4700
dx.doi.org/10.1021/cr400698c | Chem. Rev. 2014, 114, 4695−4748
Chemical Reviews
Review
medicinal chemistry methods which have been successful at addressing classical drug targets, have so far been less effective for PPIs. There is thus an urgent demand for innovative approaches to small molecule modulators designed specifically with the topologies of PPIs in mind. For this purpose, the field of synthetic organic chemistry is an unending source of new synthetic methodology and complex molecular architectures, which is proving highly beneficial for innovation in chemical biology91 and PPI drug discovery.92,93 This section has been subdivided into four inter-related topics − namely, structure-based design (section 3.1), natural product-inspired PPI modulation (section 3.2), supramolecularinduced PPI modulation (section 3.3) and compound library generation (section 3.4). Section 3.1 describes methods which use the peptide binding epitope of PPIs as a lead structure, including a discussion of hotspot residue theory, conformationally constrained peptides and miniproteins, peptidomimetics, and mimetics of protein secondary structure. In section 3.2, the discussion moves to chemical methods inspired by the guiding principles of biologically active natural products and ways to diversify natural products, including semisynthesis, diverted total synthesis, mutagenesis, and biology-oriented synthesis (BIOS). Section 3.3 introduces the exciting field of supramolecular-induced PPI modulation, in which small synthetic host molecules are designed to bind amino acid residues site-selectively at the target protein. Finally, section 3.4 discusses some of the powerful biological and chemical approaches to screening large compound libraries, including biological display techniques, microarrays, on-bead approaches, fragment-based drug discovery (FBDD), in silico screening and multicomponent reactions (MCRs). In line with the rest of the review, section 3 will emphasize recent work towards the development of PPI inhibitors, with priority given to work published from 2008 onwards. For an exclusive discussion of small molecule PPI stabilization, please refer to section 4.
Figure 6. Schematic overview of the ALPHAScreen principle. PPI that brings donor and acceptor beads in close proximity (minimally 200 nm) results in an ALPHAScreen signal (upper panel). Low energy red light (680 nm) excites photosensitizing phthalocyanines on the donor bead which release singlet oxygen. This results in an orchestrated excitation−emission cascade of three dyes on the acceptor bead if this is in close proximity. Rubrene finally emits higher energy light at 520− 620 nm. Upon inhibitor administration the signal is decreased, because the distance of the two beads has increased beyond the diffusion threshold of singlet oxygen.
to this bead is highly amplified, resulting in advantageous sensitivity with a remarkable dynamic range for this assay.82 Additionally, the energy of the photosensitizing reagent phthalocyanine results in a wavelength shift from low, during excitation, to higher energy during read out, resulting in very low background signal. The chemical probes needed for ALPHAScreen (PerkinElmer) are rather cost intensive and the plate reader of choice has to be ALPHAScreen enabled. Additionally, as the chemical reaction on the donor beads is excitable in the low energy range, undesired irradiation such as sunlight has to be minimized while preparing the measurement, complicating the handling. However, the assay statistics are outstanding, the beads do not interfere with automated liquid handling and the assay is titratable. Therefore, the assay is used for a broad spectrum of biological questions especially in PPI modulation involving primary screening, secondary validation as well as IC50 determination.81,83 Up to now just 1.6% of the iPPI-DB depict ALPHAScreen as their utilized assay, but the numbers are rising and remarkable examples have utilized ALPHAScreen so far: Examples published very recently involve the inhibition of Checkpoint Kinase 2 in a fragment based setup,84 Skp2− Cks1 anti- cancer target,85 the AF4−AF9 interaction as a novel target for MLL-R leukemia86 or the HIV-1−LEDGF interaction87,88 were studied in HTS or in combination with previous computational bioinformatics filters. Also, the PDE δ inhibitor impairing oncogenic KRAS signaling was found during a HTS using the ALPHAScreen technology.89
3.1. Structure-Based Design
The binding epitope of either one of the two protein partners has been a widely exploited starting point for the development of small molecule PPI inhibitors, in which a considerable emphasis has been placed on structure-based design and the guiding principle of hotspot residue theory. 3.1.1. Hotspot Residue Theory. Alanine scanning mutagenesis studies have concluded that PPI formation is predominantly driven by the need to maximize hydrophobic interactions between amino acid residues at the PPI interface (although electrostatic interactions do sometimes dominate, for instance in the case of phosphorylated peptide motifs), and that the free energy of binding is often focused at only a few key amino acid residues, termed hotspot residues.94 Hotspot residue theory has been ground-breaking for PPI drug discovery, as it implies that targeting hotspot residues with small molecules is an energetically feasible way to inhibit PPIs.15,95,96 Recent developments in this direction include new computational tools to better predict and characterize hotspot residues, in particular for PPI targets lacking structural information and for which alanine scanning mutagenesis is not an option. For example, cosolvent molecular dynamics (MD) simulations have recently been used to investigate protein−protein interacting surfaces.97,98 An alignment-free computational method termed “iPred” has recently been reported, which predicts PPIs and the location of hotspot residues.99 Large-scale screening and meta-analysis of all
3. APPROACHES FOR HIT IDENTIFICATION OF PPI MODULATORS PPIs are difficult, unconventional drug targets due to their unique molecular topologies.90 They operate via shallow extended solvent-exposed surfaces, compared to the drug pockets of conventional protein targets, such as enzyme active sites and transcription factors, which tend to be deeper and less accessible to bulk solvent. Consequently, the traditional 4701
dx.doi.org/10.1021/cr400698c | Chem. Rev. 2014, 114, 4695−4748
Chemical Reviews
Review
and “pocket-occupying” hotspot residues. The authors not only confirm that the most energetic hotspot residues are concentrated in a region of IKKβ previously known to bind at the NEMO surfaceW739, W741, and L742but that this druggable region extends to include residues L737 and F734 (Figure 7). Significantly, two previously unidentified hotspot regions on the IKK β surface, specifically residues L708/V709 and L719/I723, were also identified as potentially druggable pockets (Figure 7).
protein−protein complexes in the PDB and Molecular Modeling Database (MMDB) has recently been performed to identify potential small-molecule “multibinding sites”, which overlap with the protein−protein interface.100 The existence of “stability patches”, which sit in the vicinity of hotspot residues has also been postulated. Stability patches are well-defined clusters of highly immobile amino acids on the surface of the interacting proteins, which are believed to contribute to the high binding energy at protein interfaces. Evidence for the existence of stability patches has been put forward in studies on three different classes of PPIInterleukin 2 (IL2)-IL2-receptor (R) complex, mouse double minute protein 2 (MDM2)-p53, proliferating cell nuclear antigen (PNCA)-(Flap endonuclease
Figure 8. Synthetic amide-to-ester mutation of PDZ domaininteracting peptides derived from naturally occurring protein interacting partners. (a) X-ray cocrystal structure of a peptide bound to PSD-95 PDZ3. The backbone peptide−protein hydrogen bonds under investigation are indicated by gray and black dotted lines. Individual sites for hydrogen bond mutation are labeled (0, −1), (−1, −2), (−2, −3), and (−3, −4), from C-to-N terminus. (b) An outline of the synthesis of a dansyl-labeled amide-ester mutant peptide. Ester mutations were introduced though the coupling of α-hydroxy acids to the amide chain.103 Reprinted with permission from ref 103. Copyright 2013 American Chemical Society.
Figure 7. Mapping of the hotspot residues at the NEMO-IKKβ interface using FTMap. (a) An overlay of individual amino acid side chain groups with consensus cluster sites, which indicate binding hotspots on the NEMO protein surface, with a zoomed-in view of the identified hot-spot regions (NEMO protein removed for clarity). (b) Results from focused mapping of the same three hot-spot regions (colored blue, magenta, and yellow) within the search volumes (transparent gray) in the presence of the NEMO protein.102 Reprinted with permission from ref 102. Copyright 2013 American Chemical Society.
In addition to the amino acid side-chain residues, the amide backbone also contributes significant energy to the binding of protein partners. Recently, the role of hydrogen bonding between the amide backbone of the PPI binding epitopes of postsynaptic density 95 (PSD-95)/discs large/zonula occludens 1 (PDZ) domains was investigated using systematic synthetic amide-to-ester mutagenesis (Figure 8).103 For this work, four PDZ-peptide interactions were studies: the PDZ2 of SAP97, protein tyrosine phosphate-BL and the PDZ3 of PSD-95 and PTP-BL (Figure 8a). The synthesis of a focused library of depsideptides was performed (Figure 8b), corresponding to amide-to-ester mutants of short PDZ binding peptides derived
1) Fen-1 PPIsusing steered molecular dynamics (SMD) simulations.101 A computational fragment-based mapping approach, combined with alanine scanning mutagenesis studies, was used to identify the hotspot residues at the binding interface of nuclear factor kappa B (NF-κB) essential modulator (NEMO)-IKKβ PII − an important PPI for NF-κB signaling pathway (Figure 7).102 The energetic contribution of individual amino acids was quantified and a distinction made between “pocket-forming” 4702
dx.doi.org/10.1021/cr400698c | Chem. Rev. 2014, 114, 4695−4748
Chemical Reviews
Review
selection of the cyclic peptide cyclo-IYWNVSGW from a peptide library containing about 3.2 million members.113 After tagging the peptide with a cell permeable Tat sequence, inhibition of virus-like particle secretion was observed in cell culture. As another example, the heterodimeric transcription factor hypoxia inducible factor-1 (HIF-1) was targeted. HIF-1 regulates the cellular response to reduced oxygen levels playing an important role in tumor survival and progression. Consequently, inhibition of HIF-1 dimerization is considered an attractive therapeutic approach. The bacterial reverse twohybrid system was used to screen for HIF-1α−HIF-1β dimerization inhibitors.114 The selected peptide cylco-CLLFVY was again linked to the Tat sequence and showed inhibitory effects on HIF-1 dimerization in cell culture when used at low micromolar concentrations. Grossmann, Ottmann and coworkers reported a macrocyclization strategy that uses ringclosing olefin metathesis to conformationally constrain irregularly structured peptides.116 In this approach, a binding epitope of the virulence factor exoenzyme S was stabilized by replacing two lipophilic side chains with a hydrophobic crosslink. The resulting cyclic peptides proved efficient in inhibiting the interaction between a domain of exoenzyme S and the human target protein 14-3-3 in vitro. Another efficient approach that allows the generation of cyclic peptide libraries and their selection for protein binders is the phage-display based identification of bicyclic peptide ligands which has however not been used to select for PPI inhibitors so far.117 This technique provides access to large libraries of conformationally constrained peptides and can be expected to yield potent PPI inhibitors in the future. 3.1.2.2. Stabilized β-Sheets. β-Sheets are composed of at least two β-strand sequences that are aligned in a parallel or antiparallel fashion. Artificially stabilized β-sheets generally consist of a single amino acid sequence with a central β-hairpin mimetic that aligns the flanking sequences thereby nucleating the formation of an antiparallel β-sheet.118 Such hairpin-like structures can be further stabilized by a cross-linking of C- and N-terminus, using an approach developed by scientists at Polyphor termed peptide epitope mimetics (PEM)118 and by additional intrastrand bridges.119 The interaction between tumor suppressor p53 and its negative regulator proteins MDM2 or MDMX is a well-studied target in anticancer therapy and a commonly used model system for the evaluation of novel approaches for PPI inhibition. This interaction is mediated by a short α-helical peptide sequence of p53 binding to the globular domains of MDM2 or MDMX. Within the p53 helix phenylalanine (F19), tryptophan (W23), and leucine (L26) align along one face and were identified as hot spots for MDM2 binding (Figure 9a).120 Interestingly, phage display selections against MDM2 and/or MDMX tend to yield helical peptides.121,123 For example, Lu and co-workers performed a phage display with the synthetic D-enantiomer of MDM2 resulting in the selection of a peptide sequence that binds the protein in a helical conformation. Figure 9b shows the crystal structure of MDM2 in complex with the corresponding left handed helical D-peptide.121 Nonetheless, Robinson and co-workers were able to design a cyclic antiparallel β-sheet utilizing an L-/D-proline dimer as hairpin mimetic and incorporated the three hot spot residues at appropriate positions along one β-strand.118 The first macrocycle showed only weak inhibitory activity and was evolved into peptide 78A (Figure 9c) exhibiting submicromolar affinities for MDM2.118
from naturally occurring protein partners. A dansyl group was attached to the N-terminus to enable binding studies by a FP assay, and to measure binding kinetics by stopped-flow fluorimetry. The ester mutants bound with lower affinity than the amide analogs, though it is not yet clear to what extent the drop in affinity is influenced by changes in peptide conformation induced by the amide-to-ester mutation. While still in the early stages of evolution, hotspot residue theory in its present form has provided important guidelines for the design of PPI inhibitors, which, as the forthcoming subsections will demonstrate, have delivered some important designed molecules targeting a range of different PPI targets. 3.1.2. Conformationally Constrained Peptides and Miniproteins. Knowing that PPIs are mediated by constituting peptide sequences, the use of inhibitory peptides directly derived from binding epitopes of participating proteins was envisioned. However, inhibitory effects of such peptides highly depend on their structural characteristics both in solution and bound to the target protein.104 Generally, isolated peptides adopt a defined three-dimensional structure upon binding to a protein target contrasting with their flexible nature in the unbound state. This restriction in conformational freedom upon binding leads to an entropic penalty affecting the affinity for the target. Consequently, peptide modifications and nonnatural amino acids were developed that restrict the conformational freedom and induce a particular bioactive structure. In addition to increased target affinity, the constraint of conformational freedom was reported to support proteolytic stability, selectivity and in some cases cellular uptake of bioactive peptides.105 Different secondary structure motifs have been stabilized such as loops, β-sheets, and helices.106 This subsection focuses on recent reports of PPI inhibitors (for earlier examples see previous reviews)107,108 and does not give a general overview of conformationally constrained peptides. The peptide-derived inhibitors are grouped according to the type of stabilized secondary structure, starting with loops in cyclic peptides followed by β-sheets and helices. Finally, switchable secondary structure motifs are described. Often, more than one stabilization approach was utilized for the inhibition of a particular protein−protein interaction. To avoid redundancy, the figures in this subsection summarize different stabilization approaches for a given target protein or protein family: MDM2, BCL-2 family proteins, β-catenin, and estrogen receptors. 3.1.2.1. Cyclic Peptides. The term cyclic peptide refers to relatively short macrocyclic peptides that adopt looplike structures (in contrast to cross-links which stabilize β-sheets or helices). A variety of macrocyclization strategies have been developed either directly connecting the termini of a peptide or involving a cross-linking of side chains.109 Only a subset of these cyclization strategies has been used to generate inhibitors of PPIs. A very early example pointing toward an increased bioactivity of cylic peptides was reported by Ruoslahti and coworkers.110 Using phage display they identified a disulfide cross-linked peptide that acts as α5β1 integrin ligand and showed 10-fold higher affinity that its linear analog. Recently, Tavassoli and co-workers employed a bacterial reverse twohybrid system111 in combination with the split intein circular ligation of peptides.112 They identified various cyclic peptides113−115 targeting protein complexes such as Gag− TSG101113 or HIF-1α−HIF-1β.114 An important stage in the life cycle of HIV is the release of viral particles from the host cell. This step requires the virally encoded Gag protein to interact with host protein TSG101. The authors report the 4703
dx.doi.org/10.1021/cr400698c | Chem. Rev. 2014, 114, 4695−4748
Chemical Reviews
Review
Figure 10. Schematic drawing of different approaches for the covalent stabilization of α-helixes with relative spacing between cross-linked amino acids (i,i+n).
used examples. In another strategy the side chain macrocyclization of all-hydrocarbon α-methylated building blocks is used to enforce an α-helical conformation (stapled peptides).135 Finally, a particular N-cap, the so-called hydrogen bond surrogate (HBS), was reported to support nucleation of α-helices therefore increasing their conformational stability (Figure 10).136 In this subsection examples for stabilized αhelical interaction motifs are grouped based on the nature of the utilized stabilizing scaffold. Cross-Linked Side Chains. Covalent side chain tethers have been widely used for the stabilization of α-helical structures. A first proof of concept has been provided by the cross-linking of natural amino acids.105,108 Early examples involve the tethering of lysine and glutamate by lactam formation and disulfide bridges between two cysteine residues.126 Fairlie and coworkers showed that the simultaneous use of two lactam tethers provides a general approach for the stabilization of interfacial αhelical peptides.137 Disulfide bridges were evolved into thioether linkages to avoid reductive cleavage associated with a use of disulfides in biological systems.138 An alternative approach for the cross-linking of amino acid side chains was reported by Lin and co-workers, who use bisarylmethylene bromide for the covalent linkage of two cysteine residues at position i and i + 7, respectively (Figure 11a).139 This approach was utilized to stabilize an α-helical peptide known to compete with p53 for the binding of MDMX and MDM2 (Figure 9a). Compared to the unmodified sequence, the covalently crosslinked peptide showed modest increases in helicity and bioactivity and most importantly an enhancement in cell permeability. The cellular uptake of this inhibitory peptide was
Figure 9. Tumor suppressor p53 is negatively regulated by MDM2. p53 inhibition occurs via multiple mechanisms, including direct binding and thereby masking the p53 transactivation domain. Crystal structures of MDM2 (gray) in complex with different peptides are shown: (a) MDM2/MDMX binding sequence of p53 (blue, PDB 1ycr);120 (b) Left-handed helix DPMI-α (green, PDB 3lnj),121 (c) Cyclic antiparallel β-sheet 78A (orange, PDB 2axi),118 and (d) stapled peptide SAH-p53−8 (purple, PDB 3v3b),122 Panels b−d show superimposed helix of p53 in transparent blue (selected interacting residues are highlighted gray in the sequence and are shown explicitly in structures).
3.1.2.3. Stabilized α-Helices. α-Helices participate in many PPIs and can serve as starting point for the design of PPI inhibitors.124 A number of approaches utilize the helical binding epitope itself introducing moieties that stabilize the secondary structure.105,108 Early examples for helix stabilizing elements involve the cross-linking of side chains,125,126 the use of helix inducing natural127 and α-methylated amino acids,128,129 and the installation of N-terminal modifications (N-caps) capable of nucleating an α-helix by the formation of intramolecular hydrogen bonds.130 Recently, metal ion coordination,131,132 supramolecular interactions133 and multimer assembly134 were used to stabilize the α-helical conformation of peptides. In the context of PPI inhibition three major stabilization approaches were developed (Figure 10): One utilizes the covalent crosslinking of amino acids located along one face of an helix with lactam125 and disulfide bridges126 being the most frequently 4704
dx.doi.org/10.1021/cr400698c | Chem. Rev. 2014, 114, 4695−4748
Chemical Reviews
Review
alternative transition metal mediated cross-linking of amino acid side chains. Cross-Linked α-Methylated Amino Acids (Stapled Peptides). Verdine and co-workers have introduced the hydrocarbon stapling of α-helical peptides.135,145 This strategy combines two features previously employed to stabilize the αhelical structure of peptides: (I) the use of α-methylated amino acids128 and (II) the formation of olefin containing cross-links via ring-closing olefin metathesis introduced by Blackwell and Grubbs.146 So far, two cross-link architectures proved most useful for the generation of stapled PPI inhibitors: an eightcarbon tether with modified amino acids at positions i and i + 4, and an 11-carbon bridge with i,i + 7 spacing (Figure 10). In a ground-breaking study, the peptide stapling approach was used to develop an inhibitor that targets Bcl-2 protein family members and showed efficacy in a mouse model of Bcl-2 driven lymphoma.147 Bcl-2 family proteins represent key regulators of apoptosis and control the cellular survival via a complex network of interactions between pro- and antiapoptotic family members. Antiapoptotic BH (Bcl-2 homology) domain proteins (e.g., Bcl-2 and Mcl-1) promote cellular survival by trapping critical pro-apoptotic BH3 proteins (e.g., BID, BAD, and BIM) thereby inhibiting activation of pro-apoptotic multi BH domain proteins (e.g., BAX and BAK). Using different BH3 helices as starting point (BID,147,148 BAD,148,149 BIM,150 and Mcl-1143) a variety of “stabilized alpha-helices of Bcl-2 domains” (SAHBs) was designed by Walensky and co-workers. SAHB peptides proved to be helical, protease resistant and cellpermeable. However, not every stapled BH3 helix exhibits improved bioactivity143,151 which requires the synthesis and testing of a set of modified peptides to identify suitable candidates.143 For example, the generation of a potent Mcl-1 inhibitor involved the synthesis of about 20 different SAHB peptides finally yielding Mcl-1 SAHBD as the most active compound.143 The crystal structure of Mcl-1 SAHBD in complex with Mcl-1 (Figure 11b) reveals an α-helical conformation of the SAHB peptide and a direct involvement of the hydrocarbon cross-link in target binding.143 Stapled peptides were also used to generate inhibitors of the interaction between tumor suppressor p53 and its negative regulators MDMX and MDM2 (Figure 9a). The p53 derived i,i + 7 stapled peptide SAH-p53−8,152 developed by Walensky and co-workers, proved to have nanomolar affinities for both MDMX and MDM2 with a preference for MDMX.153 A crystal structure of SAH-p53−8 in complex with MDM2 (Figure 9d) verifies the α-helical conformation and reveals an intimate interaction between staple and target protein providing an explanation for the high stability of this peptide−protein complex.122 Further improved stapled peptides were derived from phage display selections154 and structure based sequence optimization aiming for the development of dual specific MDM2-MDMX inhibitors.155 In a xenograft cancer model the dual specific inhibitor ATSP-7041 was effective in the suppression of p53-dependent tumor growth verifying the potential of stapled peptides as therapeutically relevant inhibitors of PPIs. Estrogen receptors (ERs) belong to a class of transcription factors that can be activated by the binding of estrogen or peptide coactivators (Figure 12a). These activating peptide ligands can adopt an α-helical structure thereby presenting three conserved leucine residues along one face of the helix. This hydrophobic patch is recognized by a groove on the surface of ERs. The α-helical coactivator ligands served as
Figure 11. BCL-2 family proteins are key regulators in apoptosis. MCLl-1 and BcL-x(L) are antiapoptotic Bcl-2 family members that bind to α-helixes of antiapoptotic BH3 (Bcl-2 homology domain 3) proteins thereby inhibiting their apoptosis-inducing effect. An inhibition of these PPIs is considered an attractive anticancer strategy. Structures show antiapoptotic proteins (gray) bound to stabilized BH3 helices. Crystal structure of Mcl-1 (gray) with (a) bisaryl-bridged BphNOXA-2 (blue, PDB 4g35)141 and (b) stapled Mcl-1 SAHBD (orange, PDB 3mk8).143 (c) NMR-derived solution structure of Bcl-x(L) in complex with photoswitchable BAKI81Fi,i+11 (green, PDB 2lp8).144 Selected interacting residues are highlighted gray (in sequence) and shown explicitly (in structures).
further improved by conjugation of spermine to the Cterminus.140 Later, the same group reported the use of bisphenylmethylene bromide for a stabilization of the NOXAderived α-helical peptide targeting the BCL-2 family protein MCL-1.141 In this study, D-cysteine is introduced at position i and the natural L-enantiomer at position i + 7. The crystal structure of the resulting peptide inhibitor Bph-NOXA-2 with MCL-1 verifies the α-helical conformation of the peptide in the bound state (Figure 11a). To achieve activity in cell-culture, the peptide sequence was further optimized by replacing polar residues with alanine thereby increasing the overall hydrophobic character of the peptide which supported its tendency to cross cell membranes and reach its intracellular target.141 Focusing on cysteine residues at positions i and i + 4, Greenbaum and co-workers describe a screening of 24 different cross-linkers regarding their ability to stabilize an α-helical peptide.142 This study indicates that structurally rigid linkers appear to provide the strongest stabilizing effect, with dibromom-xylene representing the most suitable cross-linking agent. Using glutamate side chains as chelate ligands, Ball and coworkers report p53-derived α-helical dirhodium metallopeptides that bind MDM2.131 This approach represents an 4705
dx.doi.org/10.1021/cr400698c | Chem. Rev. 2014, 114, 4695−4748
Chemical Reviews
Review
Figure 13. β-Catenin is a central hub in the Wnt signaling pathway. A binding of β-catenin to transcription factors of the LEF/TCF family such as TCF4 is a prerequisite for the activation of canonical Wnt signaling. For expression of a subset of Wnt target genes a binding of BCL9 is necessary. Middle: Crystal structure of β-catenin (gray) bound to the β-catenin binding domains (CBDs) of TCF4 (blue) and BCL9 (green, PDB ID: 2gl7),166 Top: crystal structure of the StAx35−β-catenin complex (PDB ID: 4djs) with the sequence of StAx-35 and −35R, respectively.161 Bottom: sequence of stapled peptide SAHBCL9B164 (selected interacting residues are highlighted gray in sequences and shown explicitly in structures).
Figure 12. (a) Estrogen receptor (ER) α (gray) bound to coactivator peptide NRCA (blue, PDB 2qgt). (b) ERα (gray) in complex with disulfide cross-linked PERM-1 (green, PDB 1pcg).156 (c) ERβ (gray) bound to stapled SP1 (orange, PDB 2yjd).157 (d) ERα (gray) in complex with stapled SP2 (purple, PDB 2yja).157 (b−d) Superimposed helix of NRCA is show in transparent blue (selected interacting residues are highlighted gray in the sequences and are shown explicitly in structures).
model system for the evaluation of various stabilizing scaffolds. Early examples involve the use of lactam156,158 and disulfide tethers (PERM-1, Figure 12b).156 Phillips et al. describe the use of i,i+4 staples for the design of stabilized ER ligands and report the crystal structure of two stapled peptides (SP1 and SP2) bound to ER (Figure 12c and 12d).157 These structures show direct interactions between the staple and surface residues of ER. In the case of SP1 (Figure 12c), these interactions induce a rotation of the helix thereby shifting the register by one amino acid relative to the starting sequence (Figure 12a). This dramatic rearrangement indicates a significant contribution of the hydrocarbon staple to the binding affinity. The inhibition of certain transcriptional activator complexes represents an appealing strategy for the inactivation of signaling pathways that appear essential for the survival of cancer cells. Due to the involvement of numerous PPIs, a targeting of these complexes proved to be particularly challenging.159 Notably, the stapling of α-helical interaction motifs provided inhibitors of the Notch and Wnt signaling pathway, respectively. The heterotrimeric NOTCH1 transcription factor complex was directly inhibited by the stapled peptide SAHM1 that triggers reduced expression of NOTCH1 target genes in cell culture
and shows antiproliferative effects in a mouse model of NOTCH1-dependent lymphoblastic leukemia.160 The activation of Wnt signaling depends on the formation of a complex between coactivator protein β-catenin (gray) and transcription factors of the LEF/TCF family such as TCF4 (blue, Figure 13). Verdine and co-workers report the stapling of the α-helical βcatenin binding domain (CBD) of Axin sharing a binding site with TCF4 and the subsequent affinity optimization of this peptide using phage display.161 Cell penetration and localization properties of β-catenin targeting peptides are crucial for the desired biological activity.161,162 Therefore, an introduction of arginine residues, known to promote cellular uptake, was performed. The resulting active peptides StAx-35 and −35R compete with TCF4 for β-catenin binding and selectively reduce the expression of Wnt target genes. A crystal structure of StAx-35 bound to β-catenin (Figure 13) verified occupancy of the expected binding site163 overlaying with the one of TCF4.161 Walensky and co-workers used the CBD of BCL9 (green, Figure 13) as starting point for the design of stapled peptide SAH-BCL9B.164 The same sequence was previously stabilized by triazole containing cross-links yielding peptides 4706
dx.doi.org/10.1021/cr400698c | Chem. Rev. 2014, 114, 4695−4748
Chemical Reviews
Review
Figure 14. aPP-derived miniprotein library targeting the p53-MDM2 PPI (grafted p53 residues underlined and highlighted in bold).178 Adapted with permission from ref 178. Copyright 2006 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim.
Figure 15. (a) Structural comparison of the α-helical domain of BmBKTx1 and the helical segment of p53.180 Adapted with permission from ref 180. Copyright 2008 American Chemical Society. (b) Overlay of the X-ray cocrystal structures of Stingin 1 (green) and phage peptide, PMI (pink), bound to MDM2.189 Reprinted with permission from ref 189. Copyright 2009 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim.
with increased affinity for β-catenin.165 SAH-BCL9B competes with β-catenin−BCL9 dimerization in vitro and induces the reduction of Wnt reporter activity in cell culture. In addition the authors report the peptide to inhibit proliferation and cellular migration of tumor cells.164 Waldmann, Grossmann and co-workers reported the generation of stapled peptides capable of binding Rab proteins,.167 Rab proteins are master regulators of intracellular vesicular transport and trafficking and belong to the family Ras-related small GTPases that proove to be extremely difficult targets. In this study, the authors use a stapled peptide to inhibit the interaction between a Rab-protein and a corresponding effector protein in vitro. Another set of PPIs targeted by stapled peptides aims for the inhibition of HIV. Debnath and co-workers report the inhibition of viral capsid formation using a stapled version of a peptide sequence (CAI) that was earlier shown to inhibit the viral Gap poly protein essentially involved in this process.168 Later, Walensky and co-workers used a double stapling approach for the stabilization of the gp41 peptide, a known HIV-1 fusion inhibitor.169 Remarkably, stapled peptide SAHgp41(626−662) exhibits increased protease resistance and oral availability resulting in improved antiviral activity.169 Hydrogen Bond Surrogates (HBS). N-terminal caps are stabilizing scaffolds that do not require a modification of amino acid side chains.130 Their helix inducing effect originates from the formation of hydrogen bonds at the N-terminus of a peptide thereby nucleating helix formation.170,171 Hydrogen bond surrogates (HBS) replace the very first hydrogen bond in an α-helical peptide by a covalent linkage, thereby representing
a special type of N-caps. An early HBS example was reported by Cabezas and Satterthwait utilizing a hydrazone bridge.172 Later, Arora and co-workers designed all hydrocarbon surrogates that were cyclized by ring-closing olefin metathesis.136 HBS stabilized peptides proved useful for the generation of various peptide-derived PPI inhibitors.172,173 Arora and co-workers reported the stabilization of a gp41 peptide capable of inhibiting the gp41 mediated cell fusion.173 In addition, an HBS stabilized p53 helix proved useful for the inhibition of the p53−MDM2 interaction.174 Arora and co-workers generated hypoxia-inducible factor 1α (HIF-1α) derived HBS stabilized peptides that inhibit the interaction between HIF-1α and basal transcriptional coactivator p300/CBP.175 The HBS helix binds p300/CBP thereby inhibiting the expression of hypoxiainducible genes leading to suppressed tumor growth in a murine xenograft model of renal cell carcinoma.176 Finally, the same lab reported the design of an orthosteric inhibitor of the interaction between the small GTPase Ras and its nucleotide exchange factor Son Of Sevenless (SOS).177 Using an SOSderived α-helix it was possible to generate HBS stabilized peptides capable of inhibiting the nucleotide exchange in vitro and in cell culture. Taking into consideration that small GTPases are considered as particularly challenging intracellular targets, these results highlight the potential of the HBS approach for the stabilization of α-helical peptides and their use as PPI inhibitors. Miniproteins. A viable alternative to short cross-linked peptides is to graft the biologically active peptide epitope or hot spot residues to naturally occurring miniature (mini)proteins. 4707
dx.doi.org/10.1021/cr400698c | Chem. Rev. 2014, 114, 4695−4748
Chemical Reviews
Review
Cyclotides. Cyclotides are plant-derived disulfide-rich miniproteins with an intriguing circular cystine knot (CCK) topology, which enhances the metabolic stability and cell penetrating properties of these peptide sequences.190 Just as for natural miniproteins, the tertiary structure of cyclotides
Miniproteins are promising drug candidates due to their stable, well-defined secondary structures,178−183 synthetic tractability, and cell permeability.184,185 Miniprotein grafting has been extensively used for directed protein targeting, for example to ion channels.186,187 Early work on miniprotein grafting targeting PPIs has been reported for p53-MDM2.188 Schepartz and co-workers developed inhibitors of the p53MDM2 PPI using the miniprotein avian pancreatic polypeptide (aPP, 1) (Figure 14).178 In contrast to other disulfide-rich miniproteins (see below), the tertiary structure of aPP is stabilized via multiple hydrophobic contacts between the eighteen-residue α-helix and an eight-residue PPII helix, which are linked via a type I β-turn. Here, the hotspot residues of p53 were grafted to the solvent exposed face of the miniprotein’s α-helix and phage display screening performed to optimize the folding and MDM2 binding properties. The most promising miniproteins were synthesized by solid-phase peptide synthesis, and labeled with 5-iodoacetamidofluorescein to enable analysis of MDM2 affinity in an FP assay. These efforts resulted in potent, low-nanomolar affinity inhibitors of the p53-MDM2 PPI (e.g., 2). Subsequent remodeling of the miniprotein’s fold via strategic replacement of a helix breaking proline residue disfavored dimer formation, which enabled PPI studies across a broader miniprotein concentration range.179 A similar approach was used by Lu et al. starting from a different disulfide-rich miniprotein − the potassium ion channel toxin BmBKTx1 (Figure 15a),180 which afforded a series of modified miniprotein-based inhibitors with submicromolar affinity to both MDM2 and MDMX. The same group also grafted the hotspot residues of a phage peptide, PMI, to the disulfide-rich 18-mer miniprotein, apamin (Figure 15b).189
Figure 16. Miniproteins 3−6 targeting the androgen receptor (AR)coactivator interaction.183 Adapted with permission from ref 183. Copyright 2009 Royal Society of Chemistry.
The choice of starting miniprotein is an important consideration as it predetermines the topology of the protein interacting surface. To highlight this point, four structurally different disulfide-rich α-helical miniproteinsapamin, κhefutoxin1, a scyllatoxin analog, CD4M3, and Om-toxin3, each consisting of different length α-heliceswere compared as scaffold structures for the development of androgen receptor (AR)-coactivator inhibitors (3−6, Figure 16).183 AR binds to coactivator proteins via a highly conserved helical FXXLF motif. The phenylalanine and leucine hotspot residues were grafted onto the different miniprotein scaffolds guided by computational design, and the target miniproteins synthesized by solid phase peptide synthesis.182 Significant differences in affinity were observed between the different miniprotein, including a 10-fold improvement in activity compared to competitior phage peptides. Miniprotein-derived inhibitors of the estrogen receptor (ER) have also been developed by using phage display screening, starting from the apamin miniprotein scaffold.181
Figure 17. (a) Solution (NMR) structure of cyclotide MCo-PMI (purple) in complex with hDM2 (cyan blue) represented in ribbon format. b) Backbone superposition of the cocrystal structures of PMI peptide (green) (PDB id: 3EQS) and cyclotide (purple) in complex with MDM2 (blue) and hDM2 (light blue), respectively. The key pocket-forming side chains represented in stick format.191 Reprinted with permission from ref 191. Copyright 2013 American Chemical Society.
tolerates substantial sequence variation, and can therefore function as a scaffold structure for the grafting of biologically active peptide epitopes. Li and co-workers engineered a cyclotide to specifically inhibit the MDM2/MDMX/p53−PPI by grafting a nanomolar potent α-helical phage peptide, PMI (bearing conserved p53 hot spot residues Phe19, Trp23, and Leu26) into the peptide sequence of the cyclotide Momordica cochinchinensis trypsin inhibitor I (MCoTI-I, Figure 17).191 4708
dx.doi.org/10.1021/cr400698c | Chem. Rev. 2014, 114, 4695−4748
Chemical Reviews
Review
Figure 18. Molecular model of a p53-derived peptide bound to MDM2, and MDM2 bound to a self-assembled nanoparticle bearing p53-derived peptide grafted to the surface.192 Reprinted with permission from ref 192. Copyright 2013 American Chemical Society.
The peptide was inserted into the flexible loop 6 region of MCoTI-I, between residues Ser31and Gly33, to minimize steric interference with the cyclotide scaffold. The N-terminal region of the apamin miniprotein was inserted between the Nterminus of the PMI peptide and the cyclotide to favor the biologically active α-helical conformation. The linear MCoTIderived sequence was prepared by either chemical synthesis (SPPS) or bacterial recombinant expression (modified Mxe gyrase A intein and a TEV protease recognition sequence) and protein folding induced by reduced glutathione (GSH). The resulting cyclotides bound to MDM2 and MDMX with low nanomolar affinities, and, importantly, induced cytotoxicity in p53 wild-type human cancer cells in a p53-dependent manner both in vitro and in vivo. Self-Assembled Peptide Nanostructures. Self-assembling protein-like nanostructures, which combine peptide grafting with supramolecular chemistry, are a viable alternative to the natural scaffold structures described previously. Recently, a p53derived peptide targeting MDM2 was inserted into a macrocyclic peptide scaffold capable of forming self-assembled peptide nanostructures also known as αSSPNs (Figure 18).192 The driving force for self-assembly in this case is β-sheet formation, and the self-assembled state is thought to stabilize the helical secondary structure of the p53 peptide, thus mimicking the influence of protein folding on secondary structure formation. The p53-grafted αSSPN inhibited binding of a fluorescently labeled p53 peptide to MDM2 with low micromolar affinity, and demonstrated improved stability compared to the unmodified p53 peptide. 3.1.2.4. Switchable Secondary Structures. The design of stabilizing scaffolds that allow the temporal control over peptide secondary structures can provide access to novel applications. So far, different stimuli were implemented
including metal sensitive193 and photo-144,194,195 or redoxswitchable moieties.132,196 One example involves the photoinduced and irreversible cycloaddition of a p53-derived peptide generating a cell-permeable dual inhibitor of MDM2MDMX.197 The reversible photocontrol of an α-helix was achieved by Woolley and co-workers who installed an azobenzene based cross-linker designed to stabilize the helical structure in its trans conformation.194 A similar approach was used by Allemann and co-workers who reported the photocontrol of a BAK BH3 helix by installation of an azobenzene linker.195 The solution structure of the controllable peptide BAKI81Fi,i+11 in its helical conformation was determined in complex with the target protein Bcl-x(L) (Figure 11c) using NMR.144 The structure clearly shows the azobenzene linker in its trans-conformation supporting the bioactive α-helical conformation of the peptide. 3.1.3. Peptidomimetics. An important and frequently challenging step in structure-based design is the translation of theoretical and structural information about protein−protein interfaces into bona fida small molecule inhibitors. Arguably the most straightforward approach is to optimize the amino acid sequence of a peptide derived from the natural binding epitope or a phage-derived peptide (section 3.4.1). This approach is frequently the first validatation of PPI targetability, provides useful structural indicators for the design of nonpeptidic inhibitors, and can result in useful biochemical and cellular probes for chemical biology studies. Recent examples of this include a structure-based design of p53 transactivation domains (TAD) mimetics,198 and inhibitors of ubiquitin E3 ligase SCFFbx4.199 However, while the opportunities for amino acid sequence diversity are great, side-chain diversity is limited to the twenty standard natural amino acids, which places a ceiling on the lead 4709
dx.doi.org/10.1021/cr400698c | Chem. Rev. 2014, 114, 4695−4748
Chemical Reviews
Review
Figure 19. Development of peptiodmimetic inhibitors (7−9) of the VHL-HIF-1α interaction using a rational design approach by Ciulli and coworkers.201
optimization process. Furthermore, peptide inhibitors composed of natural amino acids are prone to proteolytic degradation and exhibit poor cell penetrating properties. The use of non-natural amino acids and other artificial modifications dramatically increases the opportunities for structural diversity, and can be used to favor the binding properties and metabolic stability of the peptide-derived modulators. Ciulli and co-workers adopted a rational design approach to develop peptidomimetic small molecule inhibitors of the VHLHIF-1α PPI. Hypoxia-inducible factor 1α (HIF-1α) is a transcription factor which regulates the expression of a host of gene products in the hypoxic state, for instance in chronic anemia associated with chronic kidney disease and cancer. Under normoxic conditions, the intracellular concentration of HIF-1α is kept low by VHL-mediated ubiquitination. Critical to this regulatory process is hydroxylation at position P564 of HIF-1α by prolyl hydroxylase domain (PHD) enzymes, which is essential for the binding of HIF-1α to the VHL complex. As PHD inhibition strategies often suffer from off target effects, inhibition of VHL-HIF-1α is a highly attractive alternative approach to restore the transcriptional activity of HIF-1α.200,201 For this work, hydroxy-proline (Hyp) was used as the chemical starting point (the minimal recognition unit), considering its importance for HIF-1α binding to VHL, and de novo design software used to guide the design of Hyp analogs. A competitive FP assay was used to evaluate the compounds ability to bind to the VHL complex, and verified using WaterLOGSY NMR spectroscopy. A structure-guided medicinal chemistry approach was adopted, in which analogs were synthesized using an efficient solid-phase synthesis strategy in an effort to improve on the binding affinity. Finally, X-ray cocrystallographic data verified that the small molecule indeed binds to VHL at the HIF-1α binding site. The active compound from the initial studies was 7 (Figure 19), with an IC50 = 117 ± 10 μM. After structure-guided medicinal chemistry optimization, the activity of the compound was initially increased to 4.1 ± 0.4 μM (compound 8, Figure 19). More extensive structure− activity relationship (SAR) studies were performed to expand on the promising initial activity, culminating in the first submicromolar inhibitor of the VHL-HIF-1α interaction (9, Figure 19), with a binding mode distinct from that of compound 8.4 This work is an important example of the strength of using rational design instead of a fragment-based drug discovery approach (section 3.4.4).202
Wang and co-workers recently developed peptidomimetic inhibitors of the MLL1-WDR5 interaction.203 Mixed lineage leukemia 1 (MLL1) is a histone H3 lysine 4 (H3K4) methyltransferase (HMT), responsible for the mono-, di-, and trimethylation of histones. The H3K4 HMT activity of MLL1 is regulated by the complex between MLL1, WDR5, RbBP5, and ASH2L. Disruption of the MLL1-WDR5 PPI through point mutations at WDR5 caused complex dissociation and a loss of MLL1 H3K4 HMT activity. This makes small molecule inhibition of the MLL1-WDR5 PPI a highly promising strategy for the treatment of certain forms of leukemia. In previous studies the Ala-Arg-Ala motif of MLL1 corresponding to residues 3764−3766 was shown to be the minimum binding epitope, and tripeptide Ac-ARA-NH2 identified as a potent binder of WDR5 (Ki = 120 nM). The authors systematically
Figure 20. Potent inhibitors of the MLL1-WDR5 interaction reported by Wang and colleagues.203
evaluated the binding of Ac-ARA-NH2 at the five subpockets of the WDR5 binding site focusing on the N and C-terminus as well as the three amino acid residues, through a combination of natural and unnatural modifications. For peptide generation, a solid-phase peptide synthesis-based approach was employed and analogs were compared initially by FP. An important outcome was a more than 100-fold gain in binding affinity (Figure 20). The cocrystal structure was solved for compounds MM-101 and MM-102. Both compounds bind to the central channel of the WD40 propeller, precisely where the MLL1 peptide binds. The increase in activity compared to the MLL1 WIN peptide could be explained by additional hydrophobic 4710
dx.doi.org/10.1021/cr400698c | Chem. Rev. 2014, 114, 4695−4748
Chemical Reviews
Review
interactions at the WDR5 surface not made by the MLL1 WIN peptide. Importantly, MM-102 − the most promising compound of the series − effectively reduced the expression of two MLL1 targeted genes, HoxA9 and Meis-1, which are essential for MLL1 mediated leukemogenesis. Furthermore, the same compound was effective at inhibiting cell growth and inducing apoptosis in leukemia cells bearing MLL1 fusion proteins. This outcome highlights the potential of structure-based approaches for the development of small molecule PPI inhibitors. Macrocyclic inhibitors of the menin-MLL1 PPI have also been developed, starting from an octapeptide derived from the menin-binding motif of MLL1.204 3.1.3.1. Bioisostere Replacement. An important property of drug molecules is their cell permeability given that protein targets are frequently intracellular. Amino acid phosphorylation is an ubiquitous post-translational modification at protein− protein interfaces, with the phosphate group contributing a significant portion of the binding energy. However, phosphorylated peptides are challenging lead structures given the instability of the phosphate ester group towards phosphatases and due to the negative charge of the phosphate ester group at physiological pH, which prevents cellular uptake.205 A molecularly efficient solution to these problems is bioisostere replacement. A bioisostere is a substituent or functional group with similar physical or chemical properties, which elicits the same or improved affinity/selectivity, ideally with reduced toxicity, increased metabolic stability and improved bioavailability.205 There is currently considerable interest in the development of bioisostere replacements for the phosphate ester group.205 A recent example of this applied to PPI inhibtion can be seen in the combined work of Yao206 and Arrendale207 on small molecule inhibitors of 14-3-3-mediated PPIs. Yao and co-workers laid the important foundations by adopting a small molecule micro array to discover the first reported nonpeptidic inhibitor of 14-3-3 mediated PPIs.206 In a similar way to Ciulli, Crews et al.,201 in their work on the VHLHIF-1α PPI (Figure 19), Yao uses a phosphorylated peptide sequence, the minimal recognition motif for binding to 14-3-3 (Figure 21, 10), as the chemical starting point for the construction of the peptidomimetic library, leading to compound 11 as the most potent inhibitor molecule (IC50 = 2.6 μM, in a competitive fluorescence polarization assay). Arrendale and colleagues used 10, and combined bioisostere replacement with a pro-drug strategy to improve metabolic stability and increase cellular uptake (Figure 21, 12), with an IC50 of 5.0 μM against human DG75 leukemia cells.207 Ji et al. used bioisostere replacement to design inhibitors of the β-catenin/T-cell factor (Tcf) PPI.208 Preliminary alanine scanning and surface plasmon resonance (SPR) studies quantified the contribution of different hotspot residues of the β-catenin protein surface. The side chain residues of K435/ K508 and the backbone amide nitrogen of N430 and their interaction with the D16/E17 residues of Tcf4 (through charge and H-bonding interactions) were discovered to be the most important elements of β-catenin/Tcf binding. Importantly, though, close inspection of the cocrystal structures of β-catenin with Tcfs, suggested further room for optimization by targeting additional charge and H-bonding bonding elements in the vicinity of the binding interface. Virtual docking studies were performed and imdazole and tretrazole ring systems chosen to mimic the pKa of the carboxylic acid groups of Glu-Asp (pKa 4.5−5.6), which in addition to interactions with Lys435/
Figure 21. Peptidomimetic inhibitors of the 14-3-3 protein−protein interactions reported by Yao206 and Arrendale.207
Figure 22. Development of small molecule inhibitors by bioisostere replacement targeting the β-catenin/T-cell factor PPI.208
Lys508/Asn430 could make favorable cation-π interactions with R469 not seen between Tcf4 and β-catenin (Figure 22, 13). While the Glu-Asp dipeptide, 14, measured a Ki = 370 μM, the inhibitor molecule, 15, was a satisfying 100-fold more active toward inhibition of Tcf4 binding to β-catenin (Figure 22). Site-directed mutagenisis and SAR studies confirmed the mode of action, though the activity of 15 could not be improved and 4711
dx.doi.org/10.1021/cr400698c | Chem. Rev. 2014, 114, 4695−4748
Chemical Reviews
Review
Figure 23. Lead compound, piperidinone, 16 (Pip-1) and the discovery of AM-8553 (compound 17), a potent and selective inhibitor of the p53MDM2 PPI with superior in vitro and in vivo properties and a promising drug profile.210 Piperidinone analog 18 (Pip-2) and Nutlin 3a (19). (a) Xray cocrystal structure of 18 bound to human MDM2.211 (b) A novel ligand-induced β-strand structure at the N-terminus region of the protein is highlighted yellow. Reprinted with permission from ref 211. Copyright 2012 American Chemical Society.
3.1.4. Mimetics of Protein Secondary Structure. All small molecule inhibitors of PPIs are to some degree mimetics of protein secondary structure. In contrast to conformationally constrained peptides and miniproteins (3.1.2) and the previous section, peptidomimietic (3.1.3), this section focuses on the development of artifical oligomeric small molecule mimetics, which try to capture the backbone and side chains topology of the native α-helical secondary structure. For accounts on small molecule mimetics of other protein secondary structures, please refer to the following reviews: β-strand/β-sheets,219−223 βhairpin,117,224 α-turn,225 and β turn.226,227 3.1.4.1. Foldamers. Foldamers are artificial oligomers, which, under the guidance of noncovalent interactions, fold into stable secondary structures with a well-defined display of side chain residues.228−231 Foldamer is a general term encompassing a range of different molecular classes. Foldamers have been designed to mimic a number of different protein secondary structures, including α helices,105,107,232−236 β-strand/βsheets, 107,233,235 which are exhaustively reviewed elsewhere.105,107,232−236 β-Peptide Foldamers. β-peptide foldamers mimic a range of different protein secondary structures,237,238 some of which have been successfully targeted to PPIs.239 For example, βpeptide inhibitors of p53-MDM2 were developed in which the p53 hotspot residuesF19, W23, and L26were introduced at three residues intervals along one face of the β3-peptide 14-
the important issue of cell permeability is the subject of ongoing research. A related structure-based design approach was used targeting two proximal hotspot regions at the CD4gp120 interface to develop small molecule inhibitors of HIV-1 entry.209 Nonpeptidic Peptidomimetics. A structure-based design approach, combining chemical synthesis, X-ray crystallography, and conformational analysis, was used to identify novel nonpeptidic inhibitors of the p53-MDM2 interaction.210 A rigorous optimization of the lead compound, piperidone 16, was performed, focusing on the piperidone ring conformation and side-chain groups, in particular, the N-alkyl substituent (Figure 23). Most notably, the introduction of an α-methyl substituent, stabilized a favorable trans-diaryl conformation (Figure 23, compound 17). Interestingly, in-depth biophysical and structural analysis of 16 (also known as Pip-1) and its structural analog 18 (a.k.a. Pip-2) identified a ligand-dependent ordering of MDM2 (Figure 23). Direct contacts between the m-chlorophenyl substituent of piperidinone 18 and amino acid residues Val14 and Thr16 assist in stabilizing a β-strand structure in the N-terminal region (Figure 23),211 which is otherwise unstructured in the apo form, or in complex with p53 or Nutlin-3a (19, Figure 23). Thio-benzodiazepines have also been recently reported as inhibitors of p53-MDM2.212 Other nonpeptidic peptidomimetic PPI inhibitors include those for the nuclear receptor (NR)-coactivator interaction.213−218 4712
dx.doi.org/10.1021/cr400698c | Chem. Rev. 2014, 114, 4695−4748
Chemical Reviews
Review
Figure 24. α/β-peptide foldamer 20; (a) comparison of the different binding modes of a foldamer (green) and the BimBH3 peptide (blue) bound to the Bcl-x(L) surface.250 Reprinted with permission from ref 250. Copyright 2009 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim. (b) overlay of the foldamer:Bcl-x(L) cocrystal structure (navy:white) with the cocrystal structure of PUMA:Mcl-1 (PDB ID: 2ROC; green:red).251 Reprinted with permission from ref 251. Copyright 2011 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim.
helix.240 In this case, a significant improvement in MDM2 binding affinity was achieved through the introduction of nonnatural α-amino acids.241 Promising activity toward the related MDMX protein was also reported,242 which raises the potential of β-peptide foldamers as dual MDM2/MDMX inhibitors for the treatment of tumors overexpressing one or both of these negative suppressors. The cellular uptake efficiency of β-peptide foldamers has also been improved through judicious modification of the side chain residues.243−245 High-resolution cocrystallographic data of β-peptide foldamers bound to target protein surfaces are urgently needed though before structureguided rational design-based approaches to PPI modulation can begin in ernest.246 α/β-Peptide Foldamers. In contrast to β-peptide foldamers, α/β-peptide foldamers are composed of mixtures of α- and βamino acids residues, which means that their secondary structures more closely mimics that of native α-helical peptides.247,248 Therefore, α/β-peptides foldamers more closely match the binding profile of α-helical peptides and can be rationally modified to fine-tune target affinity and selectivity. While the presence of natural α-amino acid residues does make foldamers more susceptible to enzymatic degradation, this tendency can be mitigated through a careful distribution of the “right” β-content along the α/β-oligomer backbone.249 A series of high resolution cocrystal structures of α/β-peptide bound to different protein surfaces have paved the way for the structurebased rational design of foldamer modulators of PPIs. The high-resolution crystal structure of a 15-mer α/β-peptide foldamer (20), bound to the antiapoptotic protein, Bcl-x(L) was reported by Fairlie and Gellman and colleagues (Figure 24).250 The N-terminal region of foldamer 20, modeled on the α-helical region of the BH3 domain of pro-apoptotic Bim and Bak, consists of α amino acids in alternation with β residues bearing five-membered (APC) ring constraints, while the Cterminal region contains exclusively α amino acids. The βamino acid residues promote helical folding of the foldamer, while the α amino acids residues contain side chain groups
important for protein recognition at the Bcl-x(L) surface (Figure 24a). A foldameric α/β-peptide mimetic of the PUMA BH3 domain has also been cocrystallized with Bcl-x(L) (Figure 24b),251 and a rational design approach used to improve affinity toward Bcl-x(L) and the pro-related apoptotic protein Mcl-1.252 Similarly, inhibition of HIV-cell fusion was achieved using gp41-mimetic α/β-peptides.253 Taken together, this work provides convincing evidence of the capacity of foldamers to not only mimic the key interactions of native α-helical structures, but also to contribute to protein binding through additional non-native protein interactions. Attempts have been made to develop α/β-peptide inhibitors of topologically irregular protein interfaces, such as the recognition surface of vascular endothelial growth factor (VEGF).254 Aside from introducing ring constraints into the α/β-peptide peptide backbone, the use of Coulombic interactions between acidic and basic side chain residues has proved to be another effective way to stabilize the helical secondary structure in work on the CHR domain of HIV protein gp41.255 A systematic evaluation of binding to two different protein partners Bcl-x(L) and Mcl-1 using structurally diverse α/β-peptide foldamers derived from Bim BH3 domain was also performed.256 3.1.4.2. Peptoids. Peptoid inhibitors of PPIs share many of the characteristics and much of the potential of their β-peptide and α/β-peptide foldameric counterparts.257 As oligomers of unnatural, N-substituted glycine bulding blocks, they are resistant to proteolytic degradation and, though intrinsically more flexible than β-peptide and α/β-peptides due to the loss of H-honding capacity, their backbones can be induced to fold into well-defined helical structures through the careful choice of N-substituent groups. Peptoid inhibitors of p53-MDM2,258 19S RP,259 VEGF−VEGFR2,260 and Apaf1261 have been reported, and are all discussed in detail as part of a dedicated and comprehensive review on the structure−function of peptoids.257 4713
dx.doi.org/10.1021/cr400698c | Chem. Rev. 2014, 114, 4695−4748
Chemical Reviews
Review
3.1.4.3. Proteomimetic Oligomeric Scaffolds. In contrast to foldamers, whose structures are more closely resembling the native polypeptide chain, proteomimetic oligomeric scaffolds are instead composed of conformationally rigid artificial oligomeric structures such a terphenyls and oligobenzamides. Proteomimetic oligomeric scaffolds targeting a range of different PPIs have been extensively reviewed elsewhere,105,107,232−235 including most recently for oligamides262,263 targeting p53-MDM2264−269 and Bcl-x(L)-mediated PPIs.270,271 3.2. Natural-Product Inspired PPI Modulation
Self-medication is commonplace throughout the animal kingdom.272−274 Just as organisms instinctively search the surrounding flora and fauna for food in reponse to malnourishment, a state of disease caused by imbalances in homeostatis, so they search for the most effective pharmaceutical agent to treat other diseases. Recent data suggests that the Neanderthals, who died out over 24 000 years ago, self-medicated with plant extracts through an awareness of the nutritional and medicinal value of the surrounding vegetation.275 Thus, by instinct or by learning, man has traditionally turned to nature as a source of medication. The pharmaceutically active agent in natural extracts is often a single secondary metabolite, a natural product, which is typically biosynthesized in minute quantities by the host organism (e.g., plant or bacteria). In the context of this review, we thus define a natural product as a nonpeptidic low molecular weight metabolite of biosynthetic origins. Natural products have proved to be a tremendously rich and profitable source of drug molecules for the pharmaceutical industry. Despite this though, interest in natural product drug development waned at the end of the last century, owing to the technical difficulties of screening natural product extracts and the rise of combinatorial chemistry (combichem). In the intervening period, many new synthetic chemotypes were discovered and applied to drugs, through combichem did not fully live up to the heightened expectations and with accompanying technological advances in HTS,276 there is a renewed interest in natural products.277−279 However, the supply from their natural sources is often unsustainable, and notwithstanding some impressive total synthesis efforts,280−282 the chemical complexity of natural products frequently prohibits chemical access to sufficient quantities for advanced biological testing. Nevertheless, their unrivaled structural complexity, diversity, and biological relevance means that natural products will continue to provide a rich source of lead structures and important guiding principles for the design of compound libraries for screening campaigns. 3.2.1. Natural Products. This section of the review provides an overview of natural products characterized as PPI modulators in the time period from 2008 to the present day, as an illustration of their rich structural and functional diversity. For a comprehensive review of natural products as a source of new drug molecules in the period 1981−2010, including many targeting PPIs, the readers are encouraged to read Newman and Cragg’s excellent review,283 and all previous reviews of theirs cited therein. Where possible, natural products are grouped according to a common PPI target. 3.2.1.1. Pro-Apoptotic Proteins. Current inhibitors of the Bcl-2 family of antiapoptotic proteins bind to Bcl-2 and Bclx(L), but target Mcl-1 with only low affinity, rendering them less effective for the treatment of cancer types overexpressing Mcl-1. Thus, there is a need for Mc1−1-selective small
Figure 25. Natural products targeting pro-apoptotic PPIs, Mcl-1-BIMBak and p53-MDM2.
molecule inhibitors. The marine natural product, marinopyrrole A, 21 (Figure 25), was recently reported to inhibit BIM binding to Mcl-1 resulting in degradation of Mcl-1 via the proteasome system. The mode of action of 21 differes from other known inhibitors, such as the pan-Bcl-2 inhibitor obatodax, and while active towards Mcl-1-dependent leukemia cells, 21 could not induce cell death in Bcl-2- or Bcl-x(L)-dependent leukemia cells.284 Oxy-polyhalogenated diphenyl ethers such as 22 (Figure 25) recently identified by Crews et al. also exhibited moderate inhibitory effects toward the Mcl-1 mediated PPI (Mcl-1/Bak) in a FRET based screen of marine extracts.285 Rigorous studies on the total synthesis of p53-MDM2 PPI inhibitor, chlorofusin (23, Figure 25), have been key to determining the absolute stereochemistry, and to enable important structure−activity relationship studies286 in the absence of high resolution X-ray cocrystallography data. 3.2.1.2. Phosphorylation-Mediated PPIs. Phosphorylationdependent protein−protein interactions are important intracellular drug targets given their involvement in many intracellular signaling pathways. Very few natural products are known which selectively inhibit phosphorylation-dependent PPIs. Noteworthy, therefore, are reports by Berg and coworkers of salvianolic acid A (24) and B (25), components of the roots of Salvia miltiorrhiza (Danshen), as inhibitors of SH2 domain binding to phosphotyrosine-containing peptides (Figure 26).287 A small panel of disease-relevant SH2 domainsLck, Src, STAT1, STAT2, and STATwere evaluated in a FP assay against a small library of natural products isolated from plant extracts, and 24 and 25 both emerged as the most active compounds against Lck and Src SH2 domains, with midmicromolar IC50 values. Molecular modeling studies (Lck SH2 domain/pYEEI cocrystal structure) suggested above all that the catechol group A (Figure 26) can function as a bioisostere for the phosphotyrosine functional group, forming hydrogen bonds with Arg134 and Glu157. This result may lead to more potent lead compounds targeting Src4714
dx.doi.org/10.1021/cr400698c | Chem. Rev. 2014, 114, 4695−4748
Chemical Reviews
Review
but instead targets a site remote from the ATP binding pocket of the Hsp90 protein.290 A high-throughput screening of a 123,599-strong natural product library of marine natural products, microbial metabolites and plant extracts against three different PPIs, proteasome assembling chaperone 3 (PAC3) homodimerization, T-cell factor-7/β-catenin, and PAC1-PAC2, identified the fungal metabolite JBIR-22 (31) to be a submicromolar inhibitor of PAC3 homodimerization (Figure 27).291 Molecular docking studies suggested that 31 inhibits protein homodimerization through binding at the PAC3 dimerization surface. Pateamine, 32, a marine metabolite from Mycale sp., and simplified analog 33 (Figure 27), desmethyl-des-amino pateamine (DMDA-PatA) have recently been reported to inhibit translation initiation through targeting of eIF4AI and eIF4AII, two isoforms of the RNA-dependent ATPase and ATP-dependent helicase, eIF4A, a component of the eIF4F complex.140 X-ray crystallography data are still lacking, though 32 and 33 are both believed to exert their effects by inducing global changes in the eIF4A protein. Compound 33 was later reported to prevent cachexia-induced muscle wasting in mice caused by the cytokines interferon γ and tumor necrosis factor α or by C26-adenocarcinoma tumors.292,293 Prieurianin/endosidin 1, 34 (Figure 27) − a secondary metabolite isolated from tropical trees − was identified by Coupland, Waldmann and colleagues from the screening of a natural product library against the model plant Arabidopsis thaliana in search of circadian clock effectors.294 While the precise target of 34 remains elusive, it is known to stabilize the actin cytoskeleton and affect endosome trafficking in vivo but not in vitro, indicating the need for actin-associated proteins. Furthermore, early indications are that 34 would appear to interact with actin filaments via a unique molecular mechanism, potentially involving actin binding proteins such as profilins, formins, Actin Related Protein 2/3s (ARP2/3s), ADF/cofilin, Actin Interacting Protein 1 (AIP1) or gelsolin/ villins. A common difficulty with targeting PPIs is the lack of control over selectivity, as one of the two proteins at the interfaces commonly recognizes a number of different protein partners. Two different compound libraries were screened for inhibition of transcriptional activator, MLL, binding to the GACKIX domain of coactivator CBP/p300.295 Whereas a 50 000-strong commercial library composed of small drug-like molecules did not produce any hit compounds in this assay, a structurally diverse library of over 15 000 marine-derived natural product extracts did. Lobaric acid, 35, and sekikaic acid, 36 (Figure 27), were characterized as a novel class of GACKIX inhibitors, capable of modulating two distinct protein activator binding sites through allosteric binding to a dynamic surface within the CBP/p300 coactiavtor protein. 3.2.2. Approaches to Diversifying Natural Products. Natural products are excellent lead structures for PPI drug discovery in view of their diverse and biologically relevant structures.279,283 However their usefulness is restricted by the fact that natural products are commonly available in only minute quantities from often unsustainable natural sources. Furthermore, natural products typically require a degree of chemical optimization before arriving at a druggable entity. Therefore, sustainable chemical or genetic methods are needed, which can deliver biologically active natural products to scale, or which can produce structurally diverse analogs libraries of natural products or natural-product-like compounds in a time efficient manner.
Figure 26. Natural products targeting phosphorylation-dependent PPIs.
family SH2 domains. Berg and colleagues combine the powerful directing effect of the phosphate group with the biological relevance of natural products to identify inhibitors of the STAT5b SH2 domain (26, Figure 26) and the Pin1 substrate binding domain (27, Figure 26).288 Poloxin (28) and thymoquinone A (29) were recently reported as the first nonpeptidic inhibitors of polo-box domain (PBD) of serine/threonine kinase Polo-like kinase 1 (Plk1) binding to phosphoserine/phosphothreonine-containing peptides.289 PBD functions as a site for intracellular anchoring of Plk1, and it is thought that inhibiting the PPI formed between PDB and target phosphorylated peptide motifs can disrupt intracellular localization and would represent a viable alternative to inhibiting kinase activity rather than targeting the ATP binding site of Plk1. Gambogic acid, 30 (Figure 27), inhibits the chaperone protein Hsp90 (90 kDa heat shock protein). According to SPR spectroscopic analysis, 30 binds to the N-terminal domain of Hsp90 with a Kd in the micromolar range, and does not compete with geldanamycin, an established Hsp90 inhibitor, 4715
dx.doi.org/10.1021/cr400698c | Chem. Rev. 2014, 114, 4695−4748
Chemical Reviews
Review
Figure 27. Natural product modulators of diverse PPIs.
3.2.2.1. Total Synthesis. Total synthesis in the broader sense is the art of synthesizing complex molecular structures from simple chemical building blocks.280 The total synthesis of an unknown natural product is frequently an important confirmation of its predicted structure296 and can sometimes deliver sufficient quantities of material for initial target validation studies. Once successfully charted, the synthetic route opens up new synthetic opportunities including access to structural analogs for structure−activity relationship studies, and the diverted total synthesis of chemical biology probes (refer to section 3.2.2.3).297 With a few most notable exceptions,281,282 the scalability of a natural product synthesis is inversely proportional to its structural complexity, which therefore puts greater emphasis on other chemical approaches to access scalable quantities of compound, such as semisynthesis (section 3.2.2.2) or mutagenesis (section 3.2.2.4), or approaches which recapitulate the bioactivity of complex natural products in a more molecularly efficient manner, for instance the biology-oriented synthesis (BIOS) of small compound libraries (section 3.2.2.5). 3.2.2.2. Semisynthesis. A typical semisynthesis campaign begins with an advanced intermediate or a mature natural product isolated from a natural source, and makes use of chemical methods to complete the targeted synthesis of a biologically active compound or functional probe.298 Thus, much or all of the compound’s structural complexity is assembled biosynthetically by the host organism, in stark contrast to total synthesis, which builds up molecular
complexity in a stepwise manner starting from simple chemical building blocks. For PPI drug discovery semisynthesis can be a significantly more efficient way to access a target natural product than competing methods (e.g., the semisynthesis of microtubule stabilizing paclitaxel from 10-deacetylbaccatin III)299 and can lead to clinically useful analogs, which would be difficult to prepare via other synthetic or genetic routes (e.g. semisynthesis of rapamycin analogs targeting mTOR).300 Where natural product semisynthesis has proved particularly useful is in the molecular interrogation of stabilizing protein interfaces. Ohkanda and colleagues prepared semisynthetic analogs of fusicoccin A capable of labeling the 14-3-3 adapter protein in a ligand and site-selective manner.301 Ottmann and Kato et al. also rationally designed a semisynthetic fusicoccin analog, which selectivity stabilized the 14-3-3/TASK3 PPI.302 3.2.2.3. Diverted Total Synthesis. The expression “diverted total synthesis” implies a rationale-driven diversion from the original total synthesis route to a target natural product, with the purpose of validating a specific chemical or biological hypothesis.297 Analog studies for structure activity relationship studies is one example of diverted total synthesis. Danishefsky’s work on the diverted total synthesis of migrastatin analogs as potent inhibitors of cell migration,303,304 and Fürstner’s on actin-targeting latrunculin analogs305 are early demonstrations of the approach. Recently, a diverted total synthesis approach was adopted by Waldmann, Arndt and co-workers to target molecularly efficient cell-permeable molecular probes of intracellular actin filaments 4716
dx.doi.org/10.1021/cr400698c | Chem. Rev. 2014, 114, 4695−4748
Chemical Reviews
Review
In later work, results from structure−activity relationship studies of natural analogs of 37 and 38 suggested that the depsipeptide macrocycle would tolerate structural variation of the L-Ala methyl substituent, as it was hypothesized that the methyl substituent points away from the actin binding site into solvent.307 The alanine residue was replaced with a lysine (40) and was found to retain the potent actin-targeting properties of the parent compound. Various dye molecules were attached to 40 via the lysine side-chain, and the conjugates used to stain actin filaments in fixed cells. A BODIPY analog of 40 was found to be cell permeable and could be used to stain static actin filaments in live cells (BODIPY-40, Figure 28). 3.2.2.4. Mutagenesis. An alternative approach to the chemical synthesis of natural products has been to elucidate the biosynthetic gene cluster encoding the enzymes responsible for the assembly of the target natural product in the host organism. Information of this kind can then be used to reprogramme the dedicated biosynthetic pathways, or engineer artificial cellular synthetic production systems in a plug-and-
Figure 29. Chemical structure of rapamycin, 41, and rationally designed rapalogs 42 and 43 generated by re-engineering the rapamycin polyketide synthase.315
Figure 28. Diverted total synthesis of 39 and 40, actin-targeting analogs of bipartite natural products jasplakinolide/jaspamide (37) and chondramide C (38).306 Overlay of fluorescence microscopy images of live U2OS cells: (left), expressing mCherry-actin, which visualizes monomeric G-actin; (middle) cell-permeable BODIPYlabeled jasplakinolide/chondramide C analog, BODIPY-40 which selectively labels actin filaments (F-actin); (right), overlay of the left and middle images, which distinguishes dynamic from static actin structures. Reprinted with permission from ref 306. Copyright 2012 American Chemical Society.
play fashion.308 In this way the natural product and associated analogs can be synthesized to scale,,309−313 either in the native host or in a stable heterologous host.314 The potential advantages of this approach include a significantly reduced environmental impact, increased source sustainability and the enormous unlocked potential for structural diversity. Wilkinson and co-workers rationally optimized the anticancer properties of the complex polyketide natural product rapamycin (41, Figure 29), produced by Streptomyces rapamycinicus NRRL5491, via a re-engineering of the polyketide synthase responsible for its production and by referral to the X-ray cocrystal structure of FKBP12−rapamycin−FRAP.315 In particular, compounds rapalog 42, lacking the methyl substituent at C17 and rapalog 43, lacking the OMe and methyl substituents at C16 and C17, respectively. By inspection of the X-ray crystal structure, it was reasoned that replacing the methoxy substituent with a secondary alcohol functionality at C16 might lead to a favorable hydrogen-bond interaction with the protein polyamide backbone while at the same time mitigating potentially destabilizing steric interactions at the protein surface (43, Figure 29). Removal of the methyl group at the C17 position might also facilitate improved binding of the triene region (C17−C22) of 42 at the rapamycin binding domain of mTOR. Importantly, these rapalogs displayed
based on the bipartite cyclodepsipeptide natural products, jasplakinolide, 37, and chondramide C, 38 (Figure 28).306 Both natural products bind to three actin monomers simultaneously. At low concentrations this leads to stabilization of actin filaments, while at higher concentrations it causes mass disruption of actin dynamics and a breakdown of the actin cytoskeleton. Earlier work had established a unified synthetic route to both 37 and 38: a novel and efficient solid-phase-based synthesis using ring-closing metathesis to forge the depsipeptide macrocycle.307 An analog library was synthesized to probe the actin pharmacophore, which gave rise to simplified analog 39 (Figure 28), lacking a methyl substituent in the polyketide region and the bromo-substituent on the indole ring of the Dtryptophan. Analog 39 was found to be as cytotoxic as 37 and 38 in cell culture. The modifications also improved the overall efficiency of the synthesis and paved the way to gram quantities of the compound and the preparation of molecular probes. 4717
dx.doi.org/10.1021/cr400698c | Chem. Rev. 2014, 114, 4695−4748
Chemical Reviews
Review
Figure 30. Tailoring of the aglycon317 and gluoside moieties318 of fusicoccin A, 44, and the structurally related brassicicene C help to define a common biosynthetic logic, which connects cotylenin A, 45 (biosynthetic modifications elucidated highlighted in bold red, responsible enzymes and host organism highlighted in gray bold italics).
enzyme responsible for the biosynthesis of 46 could be expressed equally in both prokaryotic E. coli as well as eukaryotic A. nidulans and S. cerevisiae cells, for heterologous fermentation. The yeast strain produced the highest yields of the desired metabolite in this case. 3.2.2.5. Biology-Oriented Synthesis (BIOS). Past successes of natural products as drug molecules has sustained interest in their use as lead structures for PPI drug discovery. However, their prohibitively complex structures have motivated the search for alternative sources of leads for PPI drug discovery campaigns. By virtue of their biosynthetic origins, natural products occupy a chemical space, which is quite unique from all other classes of small molecule.320−322 The concept of ‘natural product chemical space’ has inspired new approaches to the design of small compound libraries, with as their aim the recapitulation of the structural complexity and biological relevance common to natural products.323−325 The approach which comes most to the fore is biology-oriented synthesis (BIOS),326−328 which aims to generate focused collections of structurally diverse compounds built around biologically relevant and prevalidated scaffold structures, such as those found recurrently in drug molecules or as fragments of
enhanced inhibition of cancer cell lines as compared to rapamycin 1. An impressive 1 g/L biosynthesis of taxadiene was recently reported (about 15 000-fold improvement over the wild-type) by Stephanopoulos et al.. Taxadiene is noted as the first committed intermediate along the biosynthetic route to Taxol, a potent anticancer natural product first isolated from the Taxus brevifolia Pacific yew tree.316 Significant progress has been made toward elucidating the biosynthesis gene cluster of the diterpene glucosides, fusicoccin A (44) and cotylenin A (45), produced by Phomopsis amygdali and Cladosporium sp. 501−7W (Figure 30). Analogs 44 and 45 both selectively stabilize different 14-3-3-mediated PPIs, which would suggest that selectivity in 14-3-3 PPI stabilization is encoded in subtle structural differences between fussicocin analogs, which warrants further investigations into their biosynthetic gene cluster. Key enzymes responsible for the molecular tailoring of the aglycon317 and sugar moieties of fusicoccin A318 have recently been characterized. A preparative scale fermentative production of the shared 5-8-5 fusicoccane tricyclic diterpene precursor has recently been described (46, Figure 30).319 The bimodular fusicoccadiene synthase, the 4718
dx.doi.org/10.1021/cr400698c | Chem. Rev. 2014, 114, 4695−4748
Chemical Reviews
Review
differences in the spatial arrangement of side chain functionality between analogous spirooxindole scaffolds as the reason for their contrasting biological profiles. A focused BIOS library based on the oxepane scaffold structure yielded Wntepane 1 (50, Figure 32) − an activator of the Wnt signaling via reversible binding to van-Gogh-like receptor protein 1 (Vangl1).332 Tetrahydroisoquinoline inhibitors of tubulin polymerization have also been reported (51, Figure 32), which target a site distinct from the colchicine and the vinca alkaloid binding sites.333 Finally, tubulexin A, a tetrahydropyran derivative (52, Figure 32), was characterized as an inhibitor of mitosis through targeting of the CSE1L protein and the vinca alkaloid binding site of tubulin.334 3.3. Supramolecular-Induced PPI Modulation
Supramolecular chemistry is the study of chemical systems governed by reversible noncovalent interactions and inspired by the assembly of biomolecules in nature.335,336 A supramolecular PPI modulator disrupts its protein target via noncovalent interactions. In this sense, all small molecules discussed in this review are to some extent supramolecular PPI modulators. However, in this section 3.3, supramolecular-induced PPI modulation comes to signify a role reversal compared to classical small molecule host-interactions, where the protein (more specifically a side chain residue of the protein) plays the role of the guest, and a synthetic small molecule the role of the host. Modern drug discovery and supramolecular chemistry are overlapping research fields, which apply many of the same molecular design principles. In modern drug discovery, drug molecules disrupt their protein targets via a lock-and-key mechanism in which the protein lock plays host to the small molecule key. For classical drug targets, such as enzymatic active sites or ligand-activated transcription factors, the energy of binding is typically channelled through noncovalent interactions between a single low molecular weight molecule and protein side chain residues buried deep within a solventexcluded pocket. For PPIs, the lock-and-key mechanism still applies, though the binding energy is this time focused at multiple hotspot residues, which are heterogeneously distributed across a more expansive protein interface. In modern supramolecular chemistry, supramolecules are assembled via noncovalent interactions between host and guest molecules, which function via a similar lock-and-key mechanism, and through which much of the binding energy is channelled, similar to hotspot residues at protein−protein interfaces. 3.3.1. Supramolecular-Induced Protein Dimerization and Oligomerization. Protein dimerization is a ubiquitous class of PPI, which adds stability and function to proteins in the cell. Chemical methods for controlling protein dimerization are therefore of benefit to the chemical biology and drug discovery fields.337 Supramolecular host−guest chemistry is one method for inducing dimerization in intrinsically nondimeric or dimeric proteins. The host−guest elements are either genetically encoded or chemically inserted via expressed protein ligation, and dimerization controlled in a spatiotemporal fashion by varying the concentration of host−guest constructs. The advantage of this approach is its orthogonality to other more direct chemical methods. Supramolecular induced protein assembly mediated by highaffinity host−guest binding to β-cyclodextrin338,339 have been reported.340−344 Cucurbit[8]uril (Q8)345−349 forms 1:2 ternary complexes with two Phe-Gly-Gly or Trp-Gly-Gly tripeptides350
Figure 31. Small molecule BIOS library targeting the SH2-domain functions of STAT.
Figure 32. Different bioactive small molecules (49−52) identified from biological screening of focused BIOS compound libraries.
bioactive natural products. Waldmann and colleagues used a BIOS approach to develop inhibitors of the SH2-domain functions of STAT, with activity in vitro and in cells (46−48, Figure 31).329 The same research group also identified 3,3′-pyrrolidinylspirooxindoles as disruptors of microtubule polymerization (49, Figure 32).330 Spirooxindoles are a recurrent chemical motif in bioactive natural products, for example the tubulin polymerizing spirotryprostatin B1. Non-natural spirooxindoles have also been reported as potent peptidomimetic inhibitors of p53MDM2.331 For the preparation of 49, a metal-catalyzed cross coupling methodology was used, which generated structurally diverse analogs with excellent enantiopurity.330 Perhaps the most intriguing aspect of this BIOS library is the identification of a novel mode-of-action for this compound class: namely, the interference with microtubule polymerization and the formation of multipolar spindles and more than one microtubule organizing center per cell, which contrasts with the biological profile of spirooxindoles reported previously. The authors cite 4719
dx.doi.org/10.1021/cr400698c | Chem. Rev. 2014, 114, 4695−4748
Chemical Reviews
Review
example, these molecules were used to enhance the functionally important tetramerization of the tumor suppressor p53357 or a potassium channel.358 In 2012, the crystal structure of cytochrome c in complex with a Calixarene was reported (Figure 33).361 In the asymmetric unit of the protein crystal two copies of cytochrome c and three Calixarene molecules could be identified. One Calixarene molecule bound to each Lys4 and Lys22 of one cytochrome c and to Lys89 of the second cytochrome. The lysine side-chains are enclosed by the hydrophobic interior of the Calixarene molecules with the sulfonate-bearing rings establishing electrostatic contacts with the γ-amino group of the enclosed lysine as well as additional polar parts of the protein. The Calixarene molecule binding to the Lys89 site for example establishes further polar contacts of its sulfonate rings to the backbone amide of Lys89 and a saltbridge to the side-chain of Lys5. The existence of multiple lysine binding sites of Calixarene on the cytochrome c surface observed in the crystal structure and related NMR mapping studies inspired the authors to define a model of the Calixarene/cytochrome c interaction where the supramolecular ligand continuously explores and camouflages the surface of the protein by sequentially binding to up to five high-probability binding sites. The reported promotion of cytochrome c assembly together with the activities toward tetramerization of p53 and potassium channels suggests a valuable role for Calixarenes in PPI modulation in the future. A third class of supramolecular ligands whose binding to proteins have been structurally elucidated are Molecular Tweezers, that in contrast to the two former ring structures constitute an open belt-like organization with a molecular cavity formed by alternating norbornadiene and benzene rings.362 In enzymatic assays the Molecular Tweezers were shown to inhibit several hydrolases in a manner reversible by the addition of free lysine derivatives363,364 which prompted the assumption (supported by in silico analyses) that the Molecular Tweezers preferentially bind to exposed lysines to confer their activity. In 2013, the crystal structure of a Molecular Tweezer bound to the adapter protein 14-3-3 was published (Figure 33).365 Here, corresponding electron density for the supramolecular ligand could only be observerd associated with one surface exposed lysine, Lys214. In addition to direct binding to 14-3-3 as measured by SPR and ITC, the Molecular Tweezer inhibits the binding of fluorescently labeled peptides derived from the protein kinase C-RAF and the pathogenic protein ExoS. This PPI inhibiting activity can be explained by the location of Lys214 which is situated at the rim of the central binding cleft of 14-3-3, the primary site of interaction with their partner proteins including C-RAF and ExoS. Judging by the excellent initial electron density the molecular tweezer wraps around the side chain of Lys214 and embraces it almost completely, electrostatically engaging the terminal NH2 with one of its phosphate groups while the other points away from the amino acid. One explanation for the fact that only Lys214 was observed bound to the Molecular Tweezer could lie in its unique environment in the protein. By binding to Lys214, the hydrophobic corpus of the Tweezer is shielded from the aqueous environment by a ring of nonpolar residues (Tyr213, Thr217, and Leu218). This “binding preference” of the Molecular Tweezer to certain lysines with a rather hydrophobic direct environment is in perfect agreement with QM/MM calculations of possible interactions of the supramolecular ligands with all 17 surface-exposed lysines in 14-3-3. These
or 1:1:1 ternary complexes with methylviologen and dihydroxynapthalene, thus enabling supramolecular-induced homo-351 and heterodimerization of intrinsically nondimeric proteins.352 Q8 has also been used to induce tetramerization of dimeric proteins,353 and significantly augment (50-fold) the catalytic activity of dimeric caspases.354 Hou and co-workers also used Q8:FGG binding to generate nanowires of glutathione Stransferase (GST).355 3.3.2. Supramolecular-Modulation of the Protein Surface. A seven-membered cucurbituril (Q7) has recently
Figure 33. Supramolecular ligands whose interactions with proteins have been solved by X-ray crystallography. For details, see main text.
been shown by isothernal titration calorimetry and protein Xray crystallography to bind to the N-terminal phenylalanine of insulin.356 The Ka of the interaction is in the same range as many physiologically relevant protein−protein interactions and should in principle suffice to confer PPI modulating activity. In the Q7-insulin crystal structure the aromatic side chain of the N-terminal phenylalanine of insulin is accommodated in the cavity of Q7 (Figure 33), covering around 200 Å2 solventaccessible surface area of insulin. A considerable number of polar contacts is established between Q7′s carbonyl oxygens and main-chain nitrogens of Phe1, Val2, and Asn3 also including the side-chain nitrogen of the latter. Whereas the overall conformation of insulin shows no change upon Q7 binding, the first four N-terminal amino acids unfolded and moved away from the proteińs main body to interact with the supramolecular ligand. Importantly, Q7 showed no measurable affinity for insulin lacking the N-terminal phenylalanine, impressivly demonstrating the specificity of this compound. Another family of supramolecular ligands that have been shown to interact with proteins are the Calixarenes.357−360 For 4720
dx.doi.org/10.1021/cr400698c | Chem. Rev. 2014, 114, 4695−4748
Chemical Reviews
Review
results could help in the design of Molecular Tweezers with more specific binding modes. 3.4. Compound Library Generation
The high-throughput screening (HTS) of compound libraries is a highly effective way to discover new PPI modulators.11,12 In the case of well-defined PPIs with established modulator compounds, HTS is an opportunity to serendipitously discover new classes of modulator with different modes of interaction. For new or less well-defined PPIs, which lack endogenous lead structures (e.g., a peptide binding epitope), compound screening is arguably the only reliable way to make the important breakthrough. The dynamic and heterogeneous behavior of PPIs makes them more challenging drug targets than classical enzymatic drug targets. Therefore, success with PPI drug discovery is particularly reliant on methods capable of generating and screening large high-quality (diverse, complex and biological relevant) compound libraries. Increasing the library size is challenging from the perspective of compound handling, scale, and the method of hit detection. Increasing the diversity, complexity and biological relevance of compound libraries challenges the synthetic chemist to devise new and more efficient synthetic methodologies fit for the purpose. The methods for generating compound libraries discussed herein meet these challenges in elegant and molecularly efficient ways. 3.4.1. Biological Techniques. Phage display366−368 and ribosome display369−371 are robust and highly efficient techniques for screening large peptide (and protein) libraries for novel binders of protein surfaces or to optimize the affinity and selectivity of known peptide binders. In the case of phage display, the peptide sequence is expressed on the surface of the phage particle; phage libraries of diverse peptide sequences are then simultaneously screened against a surface-immobilized protein target during an initial panning phase. The optimal phage binders are then amplified and sequenced. The size and diversity of phage libraries are restricted though by the tranformation efficiency of the bacterial cell. In the case of ribosome display the transcription and translation processes are all performed in vitro, which means the technique is in principle capable of handling larger and more structurally diverse peptide libraries. The advantage of both techniques lies with their exquisite efficiency: compound library synthesis (translation) and compound screening are combined within the same technique, while repeated panning, leads to a Darwinian selection of the best binding peptide sequences. 3.4.1.1. Yeast “N” Hybrid. Classical yeast two-hybrid (Y2H) systems372,373 have proven to be highly effective tools for the discovery and characterization of PPIs. Recent adaptations to Y2H, termed “yeast three-hybrid systems’’, are showing some applications for the charaterizing of small molecule-protein interactions and PPI drug discovery.374 Phage Display. Phage display has been extensively used to dissect the protein interface of PPIs. Important examples include the nuclear receptor-coactivator interaction375−377 and p53-MDM2.378 The rational optimization of phage-derived peptides guided by high-resolution X-ray cocrystallography is a highly effective way to engineer potent and selective PPI inhibitors. A phage display screen of the p53-MDM2 and p53MDMX PPIs identified pDI − LTFEHYWAQLTS as a potent dual MDM2/MDMX inhibitor.123 The cocrystal structures of pDI bound to MDM2 and MDMX helped to steer the design of more potent inhibitors (Figure 34), culminating in pDIQ − ETFEHWWSQLLS, a pDI analog with four mutations in the
Figure 34. X-ray cocrystal structure of pDIQ (orange) bound to MDM2 (top) and MDMX (bottom).123 The C-terminal region of pDIQ is helical when bound to MDM2, but unwinds partially when bound to MDMX. Reprinted with permission from ref 123. Copyright 2010 American Society of Biochemistry and Molecular Biology.
amino acid sequence (underlined) which was 5-fold more potent than pDI. pDIQ unwinds at the C-terminus of the αhelix on binding to MDMX, and makes contacts with a hydrophobic patch, which is unique to MDMX. A detailed atomistic picture of peptide−protein interactions of this sort is fundamental to the development of new and more potent inhibitors targeting the p53-MDM2/MDMX PPI. The structural diversity of phage display has until recently been limited to the screening of structurally diverse linear peptide libraries based on standard natural amino acids. The coupling of site-specific unnatural amino acid mutagenesis (section 3.4.1.4) or the post-translation chemical modification of amino acids side-chains creates opportunities to enhance the structural diversity of compound libraries. Macrocyclic peptides are a recurring theme in bioactive natural products and PPI drug discovery. In general, macrocyclization correlates with increased potency as it lowers the entropic cost of protein binding by preorganizing the peptide into its bioactive conformation. Macrocyclization can also increase cell permeability and resistance to enzymatic degradation (section 3.1.2). An elegant phage display-based strategy was used to generate and screen artificial bimacrocyclic peptides.379 A random linear phage peptide library (>4 × 109 members) bearing three reactive cysteine residues was chemically postmodified with a mesitylene linker to induce peptide bicycle formation (Figure 4721
dx.doi.org/10.1021/cr400698c | Chem. Rev. 2014, 114, 4695−4748
Chemical Reviews
Review
Figure 35. General strategy for the generation and enrichment of bicyclic peptide PPI modulators via post-translational chemical modification of phage peptides with mesitylene.379
35). This approach has so far yielded a number of enzyme inhibitors, including a potent inhibitor of human plasma kallikrein. The broad applicability of phage display to the study of PPIs coupled with the promise of screening macrobicyclic peptide libraries (see section 3.4.3), means that phage-selected bicyclic peptides116 will be a valuable tool in future PPI drug discovery campaigns. For Haberkorn and colleagues phage display screening of a combinatorial peptide library based on the Min-23 miniprotein scaffold was essential for identifying SFTI-1 inhibitors of the delta-like ligand 4 (Dll4)-NOTCH1 PPI.380 The chemical synthesis of Min-23 inhibitors of Dll4-NOTCH1 and the phage display screening on the SFTI-I scaffold were either unsuccessful or yielded unsatisfactory results. The binding epitope of a Min-23 phage peptide was successfully grafted to the variable loop region of the SFTI-I using a solid-phase based synthesis, with retention of specificity for Dll4 binding and tumor-targeting activity in vitro and in cell culture. Importantly, the grafted SFTI-I peptides also showed proteolytic stability. Phage display screening is also useful for optimizing the binding properties of rationally designed PPI inhibitors. For example, Schepartz and Brunsveld both used phage display screening to optimize modified α-helical miniproteins targeting p53MDM2178 and the ER-coactivator PPI,181 respectively. Other cell-based display techniques not covered in this review, which have been used to study PPIs, include yeast display381−383 and bacterial cell surface display.384 3.4.1.3. Ribosome Display. The diversity of compound libraries generated by cell-based techniques such as phage is currently limited by cellular transformation efficiency, and the substrate scope of the transcription/translation machinery. In noncellular display techniques, however, such as ribosome display and mRNA display, all steps are performed in vitro.26 In principle therefore ribosome display facilitates the identification of highly evolved peptide binders beyond the reach of phage libraries.385,386 Initially developed by Hanes and Plückthun for use as a protein evolution tool,369 ribosome display was first applied to the study of PPIs by Taussig and colleagues to characterize the interaction between lymphocyte signaling protein Grb2 and the N-terminal SH3 domain of Vav1.387 Recently, Brunsveld and colleagues used ribosome display to re-evaluate the AF-2 of the
Figure 36. Ribosome display screening identifies proline derived peptides as potent ER binders (Top). High resolution X-ray cocrystallography demonstrates a clear role for the flanking proline residues in precisely determining the helix length (Bottom).388 Reprinted with permission from ref 388. Copyright 2013 American Chemical Society.
estrogen receptor (ER) in search of new coactivator peptide consensus motifs.388 Earlier phage display studies375 had concluded that both the α- and β-isoform of ER preferentially bind short helical LXXLL peptides at the ER coactivator binding site (AF-2), where L = Leu and X = any other amino acid. Important for potent peptide binding are surface Glu and Lys residues (Figure 36), which align with the helix dipole. Eight panning rounds of ribsome display screening were performed against both ERα and ERβ. The canonical LXXLL consensus motif emerged on sequencing and cluster analysis of peptides enriched over four panning rounds. Intriguingly, the LXXLL motif was seen to ‘mature’ into PXLXXLLXXP by the eighth round. Subsequent biochemical and biophysical studies of a series of prolinyl-peptides identified them as potent, low nanomolar inhibitors of ER-coactiavator binding, and that the flanking proline residues in particular were important for determining the precise length of the α-helix and the strength of the surface binding through additional conformational constraints (e.g., His N-terminal to the flanking proline Pro2, Figure 36). Other recent applications of ribosome display screening have been in the optimization of the binding affinity of a human IL13-neutralizing antibody for the potential treatment of asthma,389 the generation of protein nanoarrays,390 and the selection of photoresponsive peptide aptamers.391 A related in vitro display technique useful for the screening of peptide and protein binders is mRNA-display.392,393 In contrast to ribosome display, where the nascent peptide (or protein) and progenitor mRNA are held in a metastable noncovalent assembly with the ribosome, the mRNA and peptide molecules are covalently linked via a puromycin linker.394 This approach has been used to screen for binders of calmodulin.395 A more time-efficient variant of mRNA display, termed transcription− 4722
dx.doi.org/10.1021/cr400698c | Chem. Rev. 2014, 114, 4695−4748
Chemical Reviews
Review
approach has been applied with success to identify potent inhibitors of Tumor Necrosis Factor-α (TNF-α), for instance Anticachexin C1 (53, Figure 37). In this case, a mesitylene linker was introduced to induce bicycle formation. Noteworthy for this method is the ability to topologically separate inner and outer regions of each bead through the careful choice of reaction conditions, which enables the cosynthesis of linear and bicyclic forms of each peptide on the same bead, thus aiding in the decoding of active compounds by MS analysis. 3.4.4. Fragment-Based Drug Discovery (FBDD). Fragment-based drug discovery (FBDD) also referred to as fragment-based lead discovery (FBLD) uses a combination of biophysical techniques to detect the weak binding of small fragment molecules, typically with a molecular weight (MW) < 300 Da.93,420−422 The result is a highly efficient screening of accessible chemical space of a protein target, which leaves sufficient room for the optimization of binding affinity through chemical modifications. The most success until now has been targeting enzyme active sites. But, more and more examples are beginning to appear in the literature of PPI inhibitors
translation coupled with association of puromycin linker (TRAP) display, has recently been reported.396 3.4.1.4. Flexible tRNA-Acylation Ribozymes (Flexizymes). Phage and ribosome display techniques are limited by the substrate scope of the transcription/translation machinery. Sitespecific unnatural amino acid mutagenesis of proteins in vitro and in cells397 is possible using orthogonal tRNA/aminoacyltRNA synthetase (tRNA/aaRS) pairs.398 The genetic code is rapidly expanding,399 which so far includes post-translational modifications400 multisite-specific mutagenesis,401 and the incorporation of nonproteinogenic amino acids into whole organisms.402 The coupling of unnatural amino acid mutagenesis to in vitro display techniques represents a powerful way to expand the structural diversity of peptide-based libraries. Suga et al. engineered flexible tRNA-acylation ribozymes (flexizymes)403,404 coupled to a flexible in vitro translation (FIT) system integrated with mRNA display for in vitro selection. Referred to as random nonstandard peptide integrated discovery (RaPID), this system has been used to generate and screen large libraries of natural product-like macrocyclic peptides incorporating unnatural nonproteinogenic amino acids.405−407 The approach was recently used to identify potent macrocyclic peptide inhibitors of E6AP408 Akt2,409 and VEGFR2 activity.410 3.4.2. Microarrays. Peptide microarrays have been more extensively used to study PPIs, for instance in epitope mapping of protein surfaces,411−414 rather than as a tool to identify PPI modulators. Similar to on-bead strategies, microarrays are more effective than biological display techniques for assembling and screening large, structurally diverse nonpeptidic compound libraries. For example, a microarray approach was used to improve the binding affinity of a six residue peptoid, to the KIX domain of transcription coactivator CREB-binding protein (CBP).415 3.4.3. One-Bead-Two-Compound Approach (OBTC). A neat chemical alternative to the generation and screening of bicylic peptides by phage display379 is to use the one-bead-twocompound (OBTC) approach, as illustrated by Pei and colleagues.416 The advantage of OBTC compared to biological
Figure 38. Inhibitors of the BH3-Mcl-1 PPI (56) derived from fragments 54 and 55 using a fragment-based drug discovery and rational structure-guided design approach.423
discovered by FBDD, as the field gets to grips with more difficult and less tractable molecular targets. In this regard, work by Ciulli, Crews and colleagues on the pVHL-HIF-1α interaction has been highly instructive (section 3.1.3).202 Friberg combined fragment-based methods with a structurebased design approach to discover potent inhibitors of BH3Mcl-1 (Figure 38).423 An important outcome of this work is the strong coherency of the SAR data for both the fragment hits and the merged compounds. In a separate example, FDA-approved drug molecules were employed as fragment molecules for the structure-based design of novel scaffold structures targeting Bcl-2 and Bcl-x(L) PPIs.424 A pharmacophore model was formulated, based on the cocrystal structure of BAD BH3 binding to Bcl-x(L), against which 1410 FDA-approved drugs were screened. Whereas compound 57 bound Bcl-2 and Bcl-x(L) weakly (Figure 39), covalent attachment to fragment molecules targeting a proximal binding site yielded subnanmolar affinity binders of both Bclx(L) and Bcl-2. Here the choice of linker was key to the identification of high affinity binders. Compound 58 was cytotoxic toward H146 and H1417 cancer cell lines in the low nanomolar range. The use of approved drug moelcules ensures good pharmacological and toxicological properties early on in the drug discovery program.
Figure 37. Anticachexin C1 (53), a potent, nonprotein inhibitor of Tumor Necrosis Factor-α (TNF-α), identified by the OBTC approach.416
display methods is the control over structural diversity: in fact the one-bead-one-compound approach also works well for the generation and screening of nonpeptidic or peptidemimetic compound libraries.417−419 Where OBTC falls down though (and where phage display really excels) is in the evolutionary optimization of active compounds. Nevertheless, the OBTC 4723
dx.doi.org/10.1021/cr400698c | Chem. Rev. 2014, 114, 4695−4748
Chemical Reviews
Review
3.4.4.2. Fragments Prepared by Diversity-Oriented Synthesis (DOS). Aside from biological relevance, another important criterion for the design of high quality fragment/ compound libraries is structural diversity. The structural diversity of a compound library is expressed at different molecular levels: from the nature of the side chain functionality, to the configurational arrangement of stereogenic centers, and ultimately the size and shape of the molecular framework or scaffold around which side chain functionaility is oriented (i.e scaffold diversity). Side chain variation has for all intents and purposes been mastered.429 How to maximize stereochemical and skeletal variation within a compound library in a molecularly efficient manner has proved to be more challenging. On the subject of structural diversity, BIOS430 and diversity oriented synthesis (DOS)431 are similar approaches. Where they differ though is in the methodological approach to library design and preparation. While BIOS focuses more on the classification of scaffold structures of known natural products and drug molecules as a means to generating guiding priniciples for the design of BIOS compound/fragment libraries,327,432,427 DOS aims to generate entirely new non-natural scaffold structures, which mimic the structural diversity of natural
Figure 39. Lead compound 57, derived from the same core scaffold as FDA-approved drugs Lipitor and Celecoxib, and the structure base design of potent Bcl-x(L) and Bcl-2 inhibitor 58.424
NMR spectroscopy was used to screen fragment binders of the GTPase K-Ras as inhibitors of Soc-mediated and K-Ras activation and signaling.425 A fragment-based approach combining ITC, NMR spectroscopic and X-ray crystallographic techniques has also been used to identify novel inhibitors of BRCA2−RAD51.426 3.4.4.1. Natural Product Fragments. The biological relevance of natural products is equally well serving as inspiration for the design of high quality fragment libraries for PPI FBDD.427 Fragment libraries are typically dominated by structurally diverse, yet flat aromatic (sp2-rich) ring structures, which means that the chemical space surveyed by such libraries is relatively conservative. Natural product fragments of a similar size and molecular weight as regular FBDD fragments (e.g., commercially available and analogous libraries) by comparison occupy an entirely different region of chemical space. The hypothesis is that just as natural products are considered structurally diverse, then so must the fragment molecules that collectively make up complex natural products (at least compared to regular fragment libraries). Structural diversity is assured by the increase in sp3-hybridized character and stereogenicity of natural product fragments compared to regular fragments. The potential of natural product fragment libraries for PPI drug discovery was demonstrated in recent work by Waldmann and colleagues. 427 They used a deconstructive chemoinformatic approach to analyze >180 000 library of natural product structures to identify a series of fragment molecules targeting p38α MAP kinase and inhibitors of several phosphatases, which were validated by protein X-ray cocrystallography. Amino acids or dipeptides would also appear to fit the mold of natural product fragments, as work on the structure-based design of small molecule inhibitors of the VHLHIF-1α PPI would indicate (see section 3.1.3 on peptidomimetics). In this case, Ciulli and colleagues effectively transform a small yet structurally complex amino acid fragment with weak initial activity into a potent PPI inhibitor through the introduction of aromatic side-chain groups.428
Figure 40. Generation of over eighty structurally different scaffold structures within one compounds library using a diversity-oriented synthesis (DOS) strategy by Nelson and colleagues using the build/ couple/pair DOS strategy.443 Adapted with permission from ref 447. Copyright 2009 Nature Publishing Group.
products in a molecularly efficient manner.433 The truly innovative element of DOS thus lies in the methodological processes for generating especially skeletal diversity in small compound libraries using as few synthetic steps as possible.434−439 Where PPI drug discovery stands to gain from DOS fragment libraries is in the degree of chemical space coverage not seen in other fragment libraries.440−442 The potential of this approach is evident in work by Nelson and colleagues443 on the synthesis of over eighty different scaffold structures in a single DOS library (Figure 40).443−447 A number of other examples of DOS libraries have recently been reported.440,448−452 A number of examples of PPI inhibition using small molecules DOS libraries have been reported.453,454 Schreiber and co-workers identified Robotnikin a small molecular binder of Sonic hedgehog (Shh), as modulator of Shh signaling.444 4724
dx.doi.org/10.1021/cr400698c | Chem. Rev. 2014, 114, 4695−4748
Chemical Reviews
Review
Other small molecule modulators of Shh signaling resulting from DOS compound libraries have recently been disclosed.455 DOS libraries have also been designed targeting the antiapoptotic protein Bcl-x(L)456 DOS is thus an excellent strategy for maximizing structural diversity within compound libraries and is set to play a leading role alongside BIOS in delivering high quality small molecule and fragment libraries for FBDD campaigns.440 3.4.4.3. Target-Guided Fragment Assembly. FBDD is an efficient way to identify weakly active PPI modulators for PPI drug discovery. An often difficult step in this process though is in the optimization of these weak fragments to make stronger and more selective modulators. An efficient approach which elegantly rolls interative screening and optimization into one step is target-guided synthesis, which can be either under kinetic or thermodynamic control. Kinetic Target Guided Synthesis. In this case, small molecule fragments are assembled in the presence of the protein target using irreversible bond-forming reactions such as the in situ 1,3-dipolar cycloaddition reaction between organic azides and alkynes.457,458 Recent applications of this approach include to develop inhibitors of mycobacterial transcriptional regulator, EthR,459 and Abl tyrosine kinase.460 PPIs are
A random library of nine different thio acids and nine different sulfonyl azides were screened in the presence of Bclx(L), leading to exclusive formation of sulfonamides SZ4TA2, SZ7TA2, SZ9TA1, and SZ9TA5 (Figure 41) as judged by LC/ MS-SIM analysis.462 Control experiments indicated that the template-assembled compounds were the most active of all the eighty-one possible acylsulfonamide combinations. Investiagtions into the Bcl-x(L) templation effect indicated that assembly occurred at the BH3 binding site of Bcl-x(L). In a related approach, semisynthetic epoxy-analogs of the 14-3-3stabilizing natural product Fusicoccin were subjected to in vitro epoxide ring-opening by thiol-containing pentapeptides, templated by 14-3-3ζ.463 Dynamic Combinatorial Chemistry. The second targetguided strategy, termed dynamic combinatorial chemistry, works under thermodynamic control.464,465 Compound libraries formed under dynamic reversible conditions respond to changes in the free energy of the system, for example on addition of a template, through either amplification or abbreviation of the individual components. In proteintemplated synthesis, the protein is intimately involved in the Darwinian assembly of its own best binder from a structurally diverse pool of fragment molecules facilitated by reversible chemical transformations.465 Just as for kinetic target-guided synthesis, dynamic combinatorial library (DCL) synthesis is a useful method for targeting enzymatic active sites. Methodological breakthroughs have brought less tractable targets such as PPIs within reach. Greaney, Campopiano and colleagues recently demonstrated the use of aniline-catalysis for the generation of a DCL of acylhydrazones targeting two isozymes of glutathione S-transferase (GST).466 Crucial to the success of this approach was the ability of the DCL to reach equilibration within a reasonable time frame (