Targeting protein–protein interactions: Lessons from p53/MDM2

Abstract

The tremendous challenge of inhibiting therapeutically important protein–protein interactions has created the opportunity to extend traditional medicinal chemistry to a new class of targets and to explore nontraditional strategies. Here we review a widely studied system, the interaction between tumor suppressor p53 and its natural antagonist MDM2, for which both traditional and nontraditional approaches have been reported. This system has been a testing ground for novel proteomimetic scaffold-based strategies, i.e., for attempts to mimic the recognition surface displayed by a folded protein with unnatural oligomers. Retroinverso peptides, peptoids, terphenyls, β-hairpins, p-oligobenzamides, β-peptides, and miniproteins have all been explored as inhibitors of the p53/MDM2 interaction, and we focus on these oligomer-based efforts. Traditional approaches have been successful as well, and we briefly review small molecule inhibitors along with other strategies for reactivation of the p53 pathway, for comparison with oligomer- based approaches. We close with comments on an emerging dichotomy among protein–protein interaction targets. © 2007 Wiley Periodicals, Inc. Biopolymers (Pept Sci) 88: 657–686, 2007.

This article was originally published online as an accepted preprint. The “Published Online” date corresponds to the preprint version. You can request a copy of the preprint by emailing the Biopolymers editorial office at [e-mail]

PROTEIN–PROTEIN INTERACTIONS

Sequencing of the human genome has greatly expanded our knowledge of the primary structures of human proteins. This remarkable achievement, however, makes clear how little we truly understand about human physiology. Proteomic efforts have shown that regulation of protein function is as important as the function itself, adding another dimension of complexity to an already complicated picture. Proteins are produced, degraded, proteolytically processed, and post-translationally modified to varying extents and at different rates, all in response to environmental stimuli. Each of these events requires that proteins communicate with one another. Signals are continually sent and received via transient, specific physical contacts between different proteins. We are just beginning to learn about the roles of these noncovalent protein–protein interactions in living systems. However, we do know that disruption of these signaling pathways as a result of a small mutational change in one of the protein partners or deregulation of protein production can result in a disease state. Chemical intervention, in the form of new therapeutic agents, could potentially restore the delicate balance and help relieve human suffering. Unfortunately, it has not been easy to identify suitable inhibitors of protein–protein interactions.1, 2 Developing molecules that will bind tightly and specifically to a large, amphiphilic and often flexible surface displayed by a particular protein is a fundamentally different challenge from developing molecules that occupy a small, concave, and largely hydrophobic pocket on an enzyme or receptor.1a New strategies that depart from the accepted rules of medicinal chemistry may be required to develop useful inhibitors of specific protein–protein interactions.1b

p53/MDM2: A MODEL STUDY IN PROTEIN–PROTEIN INTERACTION INHIBITION

The biological importance of p53 (“p” for protein and “53” for its apparent molecular weight of 53 kDa)3 has been apparent for some time.4 The complexity of the roles played by p53 within the cell and the question of whether the p53/MDM2 interaction constituted a “druggable” target may have discouraged medicinal chemistry efforts until the 1996 publication of a cocrystal structure that elucidated this protein–protein interface.5 Since then, many reports on inhibitor development, from the efforts of both industrial and academic researchers, have emerged.6 Just as the development of protease inhibitors taught us about peptidomimetics for enzyme inhibition,7 p53/MDM2 has served in many respects as a model system for the inhibition of protein–protein interactions. It has recently become apparent that this system cannot be considered as broadly representative of all protein–protein interactions, because of the highly focused nature of the p53/MDM2 interface.8 This feature, however, has made the p53/MDM2 interaction well suited for initial efforts in this field, allowing the exploration of multiple strategies, at least some of which may be applicable to more challenging protein–protein interactions as well. As we discuss various approaches to this target later, we will offer comments on the advantages and disadvantages of each approach, highlight important achievements, and identify remaining challenges.

p53 Biology

The tumor suppressor protein p53 is a transcription factor that controls cellular response to stress (i.e., DNA damage, hypoxia, etc.) through the induction of cell cycle arrest (by activating transcription of the WAF1/Cip1 gene, which leads to expression of the cyclin-dependent kinase inhibitor p21)9 or apoptosis.10 The murine double minute 2 protein, or MDM2, downregulates p53 activity11 through a negative feedback loop12 by binding to the α-helical transactivation domain near the N-terminus of p53. This binding blocks the DNA-binding activity of p53.11a In addition, MDM2 exports p53 from the nucleus13 and acts as an E3 ubiquitin protein ligase,14 thereby targeting p53 for proteosomal degradation.15 (The human protein DM2 is referred to as HDM2 by some workers and MDM2 by others; we follow the latter convention.16)

In response to stress, p53 is phosphorylated on specific serine residues near the MDM2 binding domain, decreasing the affinity of p53 for MDM2 and activating p53 as a transcription factor.17 Overproduction of MDM2 inhibits activation of the p53 pathway, leading to uncontrolled cell proliferation. MDM2 amplification is observed in about 7% of human tumors and is most commonly found in soft-tissue tumors (20%), osteosarcomas (16%), and esophageal carcinomas (13%).18 This amplification makes tumors less susceptible to natural and chemotherapeutic signals to undergo programmed cell death, or apoptosis, and generally results in a poor patient prognosis. According to Vassilev,19 treatment of cells overproducing MDM2 with an inhibitor of the p53/MDM2 interaction should result in (1) the stabilization and accumulation of p53 resulting from the prevention of its export from the nucleus and degradation; (2) activation of MDM2 production; and (3) activation of other p53-regulated genes (i.e., the WAF1/Cip1 gene to produce p21) and the p53 pathway, thereby causing cell cycle arrest in the G₁ and G₂ phases and/or apoptosis. Disruption of the p53/MDM2 interaction is therefore a therapeutic target for the treatment of cancer.20 Extensive efforts toward this goal have eventually yielded tightly binding α-peptides, several α-helix mimetics, and potent bioactive small molecule antagonists of the p53/MDM2 interaction. However, cell cycle arrest and/or apoptosis resulting from treatment with a p53/MDM2 interaction inhibitor should only occur in cells with wild-type p53 and not in cells with transcriptionally inactive, mutant p53. Since p53 is mutated in ∼50% of human cancers,21 a strategy for reactivation of the p53 pathway in this context is desirable.

Early Efforts Toward Target Validation: Antibodies and Oligonucleotides

Before medicinal chemists became involved in targeting the p53/MDM2 interaction, two important techniques from molecular biology, antibodies and RNA interference (RNAi), were applied to the system resulting in validation of the target. In an early proof-of-principal experiment, Blaydes et al.22 found that microinjection of an antibody for the p53-binding domain of MDM223 caused a marked increase in p53-dependent transcription. While this strategy worked in cells, it is probably not a viable option for treatment in vivo. It should be noted here, however, that monoclonal antibodies are arguably the most successful class of protein–protein interaction inhibitors at present.24 Antibodies for a number of extracellular protein targets are either FDA-approved drugs or in clinical trials. Many more are in preclinical development. While highly potent and selective antibodies can be raised to many protein targets,25 the high costs and difficulties associated with manufacturing,26 the necessity of administration via injection,27 and the potential for immunogenicity remain significant disadvantages relative to orally bioavailable small molecules. When such a small molecule cannot be identified, then an antibody may be an acceptable therapeutic agent if the target protein is extracellular. It will be interesting to see in the years to come whether clinically used antibodies that block specific protein–protein interactions can be replaced by small molecules or other strategies.

Antisense oligonucleotides have been used to inhibit MDM2 expression, thereby reducing MDM2 protein levels, diminishing the MDM2 negative feedback inhibition of p53, and increasing the levels of functional p53.28 In addition, knock-out of the MDM2 gene results in p53 accumulation, p53-dependent gene expression, and growth inhibition of colon29 and prostate cancer cells.30 RNAi via antisense oligonucleotides or small interfering RNA has great potential for the treatment of diseases resulting from aberrant protein signaling.31 By targeting and degrading the mRNA that encodes the particular protein that is being overproduced, such as MDM2, the amount of the protein within the cell is decreased, thereby preventing its association with its partner, such as p53, and restoring the normal activity.32 Despite numerous modifications of the oligonucleotide backbone to improve stability in vivo,33 delivery of an oligonucleotide across both the cellular and nuclear membranes has proven to be a significant challenge, although very recently lipid nanoparticles have been used successfully for this purpose in nonhuman primates.34 Much remains to be done, including investigation of the threshold effect, or the actual amount by which transcription must be reduced to affect protein levels and signaling.35

Structural Characterization of the p53/MDM2 Interface

For chemists, intense interest in the p53/MDM2 interaction was sparked by a cocrystal structure of the complex between the MDM2 protein and a peptide corresponding to p53 residues 17–29.5 The structure revealed that the N-terminal transactivation domain of p53 adopts an amphipathic α-helical conformation and occupies a hydrophobic cleft on the surface of MDM2 (Figure 1). Three hydrophobic side chains from p53, those of Phe19, Trp23, and Leu26, are aligned along one face of the helix to make direct contacts deep in the MDM2 cleft. One can speculate that the bulk of the binding energy for this protein–protein interaction arises from van der Waals interactions of these three p53 side chains with the surface of MDM2. Other protein–protein complexes have been found to hinge upon a small number of “hot spot” interactions within the binding interface,36 but this situation is not universal.1b

Figure 1. Crystal structure of the p53/MDM2 protein–protein interaction (Ref.5, PDB, code 1YCR). MDM2 rendered as a surface using MOLCAD (Sybyl) and colored according to cavity depth, and the p53-peptide pictured as sticks.

According to Pavletich and coworkers,5 the pocket that accommodates Leu26 of p53 is defined by MDM2 residues Tyr100, Thr101, and Val53. The pocket for Trp23 of p53 is defined by Ser92, Val93, Leu54, Gly58, Tyr60, Val93, and Phe91 of MDM2. The pocket for Phe19 of p53 is composed by Arg65, Tyr67, Glu69, His73, Ile74, Val75, Met62, and Val93 of MDM2. The backbone of the p53 peptide forms 2.5 turns of α-helix (residues 18–26) with residues at either end adopting an extended conformation. The C-terminal end of the helix is less tightly wound, leading some to classify the segment between Trp23 and Leu26 as a type I β-turn rather than an α-helical turn. The p53/MDM2 interface buries 1498 Ų of surface area, 808 Ų on p53, and 690 Ų MDM2. The hydrophobic contacts are augmented by only two intermolecular hydrogen bonds: one between the Phe19 backbone amide NH of p53 and the carbonyl of the Gln72 side chain of MDM2 at one end of the cleft, and another between the p53 Trp23 indole NH and the MDM2 Leu54 backbone carbonyl deep inside the cleft.

The crystal structure revealed that p53/MDM2 is not broadly representative of protein–protein interactions, which often feature broad, flat interfaces that contain a few disperse hydrophobic patches. Instead, MDM2 has a continuous, narrow hydrophobic cleft that is almost deep enough to be called a “pocket,” and the complex can be reduced to a protein–peptide interaction. However, since the cleft is long and is located on the surface rather than within the interior of the MDM2 protein, blocking the p53/MDM2 interaction can be seen as a challenge distinct from that required for block substrate access to an enzyme active site. The structural data lead naturally to the hypothesis that a synthetic molecule displaying three hydrophobic groups in an orientation that mimics the presentation of the Phe19, Trp23, and Leu26 side chains by p53 should occupy the MDM2 cleft and thereby competitively inhibits the p53/MDM2 interaction.

Indeed, potent small molecule antagonists of p53/MDM2 binding have been developed via traditional structure-based design and/or combinatorial synthetic techniques. However, these results were somewhat slow to appear,37 and the delay fostered the development of alternative approaches to blocking the p53/MDM2 interaction, mostly from academic laboratories. These alternative approaches have focused on modular scaffolds intended to mimic the α-helical segment that forms the binding surface on p53. It is possible that such a scaffold could be broadly useful for mimicking of α-helices involved in a variety of interactions. Alternative strategies for creating protein/protein interaction inhibitors, such as fragment-based design, appear not to have been applied to the p53/MDM2 system, perhaps because of the success achieved with other approaches. It should be noted here, however, that creative fragment-based strategies have been successful for a number of challenging protein–protein interaction targets.38 Screening of very small organic molecules (average molecular weight of 100–300 Da) via “SAR by NMR”39 or “tethering”40 has identified low-affinity ligands for distinct sites on the targeted protein surface (hot spots). Subsequent chemical linkage of the fragments has generated potent ligands for such protein interaction partners as Bcl-x_L41 and interleukin-2.42

In Vitro Assays for the p53/MDM2 Interaction: ELISA, FP, and SPR

Several biochemical assays have been used to monitor inhibition of the p53/MDM2 interaction, including an enzyme-linked immunosorbent assay (ELISA), f luorescence polarization (FP), and surface plasmon resonance (SPR). In one ELISA format (Figure 2), recombinant MDM2 is adsorbed to the inner surface of the assay vessel (i.e., the well of a microtiter plate). Recombinant p53 and the potential antagonist are added to the well and allowed to bind to MDM2. Unbound protein and inhibitor are washed away, and an antibody for p53 is added. A second antibody that binds to the first antibody and is conjugated to an enzyme such as horseradish peroxidase is added next. An enzyme substrate with low absorbance of UV–visible light is added, and enzyme-catalyzed reaction converts this substrate to a highly absorbing product, the concentration of which is read out spectrophotometrically. If a candidate antagonist inhibits the interaction of p53 with MDM2, then p53 remains in solution and is washed away. The lack of bound p53 prevents binding of the antibodies, and limits or prevents the formation of the absorbing species. Complementary ELISA formats are possible, e.g., starting with immobilized p53 and detecting with an antibody for MDM2.

Figure 2. Schematic for p53/MDM2 ELISA. Figure adapted from Dr. H.-S. Lee.

The FP assay format is more amenable to high-throughput screening than is the ELISA format, because all species are in solution in the FP assay, and the FP format requires fewer steps than does ELISA.43 Recombinant MDM2 is mixed with a fluorescently labeled p53 peptide (probe) and the potential inhibitor. The sample is then irradiated with polarized light. If the labeled peptide binds to the protein, then it will tumble very slowly, and the fluorescence will be highly polarized. However, if the inhibitor binds to MDM2 and displaces the probe, then the fluorescent peptide will tumble very quickly in solution because of its low molecular weight, and the emission will be significantly depolarized. A larger decrease in FP indicates the presence of a stronger inhibitor. One limitation of the FP format is the requirement for a low-molecular weight but tightly binding probe. As a result, an FP assay typically reports on a peptide–protein interaction rather than a protein–protein interaction; in cases such as p53/MDM2, this situation is assumed to be acceptable because the peptide is thought to represent the entire binding epitope of one partner (p53).

SPR can also measure the affinity of ligand binding to a protein target in a competition format.44 Biotinylated p53 peptide is immobilized on a streptavidin-coated biosensor chip. A solution of MDM2 protein and the potential inhibitor is flowed over the surface. With increasing concentrations of the antagonist, the binding of MDM2 to the surface-bound peptide is increasingly inhibited, which leads to a decrease in the biosensor response. These experiments require a longer time (about 1 h per assay) than does FP, but throughput can be increased somewhat via introduction of parallel SPR channels. Modifying the format of the assay by adsorbing MDM2 to the chip and flowing over a solution of the inhibitor can reveal the kinetics of the binding event. Development of a reliable assay is critical, as screens for protein–protein interaction inhibitors are prone to false-positive results. For example, candidates can cause nonspecific inhibition at low micromolar concentrations via aggregation of the inhibitor and/or the protein target.45

α-PEPTIDE ANTAGONISTS OF THE p53/MDM2 INTERACTION

Inhibition of the p53/MDM2 interaction with a synthetic molecule was validated as a viable strategy by research conducted at Novartis. As part of their drug discovery program to establish a pharmacophore model, the Novartis team determined the specificity of the amino acid-binding pockets on MDM2 and developed a series of highly potent peptide antagonists.46 Even before publication of the crystal structure, they studied the p53/MDM2 complex with a series of monoclonal antibodies to identify the interacting domains on the respective proteins.47 The Novartis group mapped the MDM2-binding site on p53 using synthetic peptide libraries derived from the N-terminal region of p53. The active peptides defined the consensus MDM2-binding site on p53 to be Thr18-Phe-Ser-Asp-Leu-Trp23. This hexapeptide, however, was only a weak inhibitor, with an IC₅₀ value, or concentration required for the inhibition of 50% of p53 binding to MDM2, of 700 μM in an ELISA format.48 To find high-affinity ligands for MDM2 that are able to block the interaction with p53, the Novartis team screened peptide libraries via phage display.49 In this technique, a randomized set of peptides is displayed on the surface of filamentous phage, one peptide per phage particle. Phage displaying peptides that bind to the target protein are isolated and amplified. After several rounds of selection and amplification, the phage population should be highly enriched in phage displaying peptides that bind to the target. Sequencing the phage DNA should then reveal the sequences of tight-binding peptides. The most active peptide obtained in this way (2, Figure 3) showed a 28-fold greater inhibition of the p53/MDM2 interaction than did the wild-type p53-derived peptide (1). Peptide 2 was effective at inhibiting the p53/MDM2 interaction in cells50; in addition, this sequence was active when expressed either with a glutathione S-transferase tag51 or in the active-site loop of thioredoxin.52

Figure 3. IC₅₀ concentration of MDM2-binding peptides in a p53/MDM2 competition ELISA from Ref.46 (Aib = α-aminoisobutyric acid; Ac₃c = 1-aminocyclopropanecarboxylic acid; Pmp = phosphonomethylphenylalanine).

More detailed information on the amino acid requirements for a potent peptide inhibitor was obtained by synthesizing truncated versions of peptide 2 and testing these peptides in the ELISA. The truncation series included sequences containing 6 to 11 residues. While the 6- and 7-mer peptides were poor inhibitors of the p53/MDM2 interaction, an 8-mer peptide (3) was identified as the minimal sequence retaining micromolar affinity for MDM2. This peptide served as the starting point for optimization via structure-based design.

Guided by the X-ray crystal structure of the p53/MDM2 complex, the Novartis researchers selected the residues on p53-derived peptide 3 that do not interact with the MDM2 protein in the bound conformation as sites for substitution with structure-promoting α,α-disubstituted residues.45 Examination of 3 by circular dichroism (CD) and nuclear magnetic resonance (NMR) spectroscopy revealed that this peptide adopts a random coil conformation in aqueous solution. Residues 20, 21, 24, and 25 (p53 numbering) were systematically replaced with α,α-disubstituted residues, which are known to promote α- and 3₁₀-helices in short peptide motifs.53 The Novartis team hypothesized that these substitutions would bias the peptide toward a helical conformation, similar to that required for binding to MDM2, thus reducing the entropic cost associated with binding and increasing the affinity of the peptide ligand for the protein. The best combination was achieved by replacing Asp21 with α-aminoisobutyric acid (Aib) and Gly25 with 1-aminopropanecarboxylic acid (Ac₃c) (peptide 4). Ac₃c was chosen to replace Gly25 because of the marked structural preference of Ac₃c for the i + 2 position in a type-I β-turn,54 which is similar to the local secondary structure observed for Trp23-Leu26 in p53. These modifications resulted in only a fourfold increase in MDM2-binding affinity of the peptide, but subsequent NMR experiments under physiological conditions confirmed that the conformational restrictions imposed by the α,α-disubstituted amino acids in the peptide sequence induce a conformation similar to that observed for the N-terminal segment of p53 in the MDM2-bound state.

The next round of optimization was focused on increasing the binding affinity of the peptide via chemical modification of the tyrosine side chain.45 Examination of a model of peptide 2 in complex with MDM2 revealed that the hydroxyl group of Tyr22 was positioned outside of the canonical p53-binding pocket and near the side chain of MDM2 residue Lys94. The tyrosine was replaced with phosphonomethylphenylalanine (peptide 5) in order to form a stabilizing salt bridge interaction with the ε-amino group of Lys94, resulting in a sevenfold increase in binding affinity. The phosphonate moiety also increased the water solubility of the otherwise highly hydrophobic sequence.

Modification of the Trp23 side chain resulted in a major increase in binding affinity, an important finding that has subsequently been used by numerous researchers. In the X-ray crystal structure of the p53/MDM2 complex, the side chain of p53 residue Trp23 is embedded in a deep hydrophobic pocket formed by MDM2 residues Leu57, Phe86, Ile99, and Ile103. Although the indole fits rather tightly, careful inspection revealed that the pocket is not completely filled. Specifically, an empty space near the sixth position of the indole ring remains at the bottom of the pocket. The space corresponds to the volume of a methyl group or a chlorine atom, and the possibility of increasing the binding affinity by establishing additional van der Waals interactions with the protein motivated the synthesis of peptides 6, 7, and 8. Replacement of Trp23 with 6-substituted tryptophans resulted in a substantial increase in binding affinity, which nicely correlated with the size of the substituent and therefore, presumably, the occupancy of the pocket. The most active compound, peptide 8, contained a 6-chloro-tryptophan and had an IC₅₀ value of 6 nM in their p53/MDM2 ELISA, a 63-fold increase in binding affinity relative to peptide 5 with the unsubstituted tryptophan. The discovery of this pocket has been exploited through the incorporation of p-chlorophenyl groups in a variety of small molecule inhibitors as will be discussed.

The seminal contributions of the Novartis researchers45 demonstrated that a number of the principles identified through the development of peptidomimetic enzyme inhibitors can be applied to the discovery of protein–protein interaction inhibitors. Structure-based design was employed to optimize the MDM2-binding affinity of short peptide motifs derived from the N-terminal domain of p53 by combining conformational constraints of the backbone with side chain functional group modifications that establish additional electrostatic and van der Waals interactions, both inside and outside the natural binding pocket. The Novartis effort was an excellent example of the integration of multiple techniques in drug discovery for the identification of compounds capable of inhibiting a therapeutically relevant protein–protein interaction. The refined pharmacophore model and other information emerging from this work undoubtedly assisted other groups in pursuit of p53/MDM2 interaction inhibitors. Such efforts have ultimately been rewarded with the discovery of small molecules that inhibit p53/MDM2 interaction.

NATURAL PRODUCT ANTAGONISTS OF THE p53/MDM2 INTERACTION

About 60% of drugs on the market today (excluding biologics) are of natural origin,55 but to date only three natural products have been reported as having any inhibitory activity against the p53/MDM2 interaction. Chalcone-based inhibitors were reported first and have been the most extensively studied.56–60 Chlorofusin was the second natural product inhibitor of p53/MDM2 to be reported.61–62 Efforts toward the total synthesis of chlorofusin are underway in a number of laboratories and will lead to further investigation of the structure/activity relationships among this molecule and its analogues.63–67 Hexylitaconic acid has also recently been reported to inhibit the p53/MDM2 interaction.68 Thus, natural products have not been a fertile source of lead compounds for the inhibition of the p53/MDM2 protein/protein interaction. While natural products are effective inhibitors for a few protein–protein interaction targets, such as paclitaxel, epothilones, and discodermolide for tubulin polymerization, it can be difficult to find natural product lead compounds for a new class of protein–protein interaction targets.

The Chalcones

The broad antitumor activities of natural product-inspired chalcones (1,3-diphenyl-2-propen-1-ones)56 led Holak and coworkers to examine the possible interaction of chalcones with the p53/MDM2 system in 2001.57 Compound 9 (Figure 4) had an ELISA IC₅₀ value of 206 μM and caused a shift in the pattern of the ¹H-¹⁵N HSQC NMR spectrum of ¹⁵N enriched MDM2, consistent with chalcone binding in the tryptophan pocket. The proposed binding model assumed that the extended π-system of the molecule is rigid and planar and that the monochlorophenyl group resides in the tryptophan binding site. Based on their NMR data, Holak and coworkers proposed that the second aromatic ring of the chalcone contacts residues Phe55 and Tyr56 of MDM2 outside of the canonical p53-binding pocket. In this orientation, the carboxylic acid and enone carbonyl are both solvent-exposed, but the carboxylic acid may be positioned correctly to engage MDM2 residue Lys51 in a salt-bridge interaction.

Figure 4. Claisen–Schmidt aldol condensation to form chalcone p53/MDM2 antagonists 9 (Ref.57) and 10 (Ref.59).

Despite their early identification as inhibitors of the p53/MDM2 interaction and their straightforward combinatorial synthesis using a Claisen–Schmidt aldol condensation protocol,58 chalcones have been the subject of only a few subsequent reports, perhaps because this class of compounds was not very selective for the target protein. Dichlorophenyl derivatives were equally toxic to both normal and malignant breast epithelial cells, possibly due to mechanisms independent of p53/MDM2.59 Other members of the original set of chalcones were observed to denature or facilitate the aggregation of the MDM2 protein during testing.57 A gel shift assay revealed that the p53 released from the p53/MDM2 complex by treatment with 9 was unable to bind DNA, suggesting an additional influence of the chalcone on the p53 protein. One could speculate on the potential of chalcones as Michael acceptors at the enone functionality, and that covalent modification resulting from attack by a protein nucleophile would affect p53 conformation and activity. A series of boronic chalcone analogues was prepared by Khan and coworkers to address the lack of specificity in the carboxylic acid-containing chalcones.59 They reported a modest improvement; compound 10 was 2.5- to 10-fold more toxic to human breast cancer cell lines than to a normal breast epithelial cell line at 10–40 μM. Isoliquiritigenin (4,2′,4′-trihydroxychalcone), a natural chalcone that is isolated from licorice root and shallot, has been shown to induce cell cycle arrest and apoptosis in liver cancer cells via the p53 pathway at 10–20 μg/mL, but its binding to MDM2 was not characterized.60

Chlorofusin

Williams and coworkers identified chlorofusin as an inhibitor of p53/MDM2 binding (11, Figure 5) after testing over 53,000 extracts from the fermentations of a diverse collection of microorganisms for this activity.61 This novel metabolite from the fungus, Microdochium caespitosum, had an IC₅₀ of 4.6 μM in a p53/MDM2 ELISA. Further studies using SPR confirmed that chlorofusin binds to the N-terminal region of MDM2 (K_d = 4.7 μM).62 Mass spectrometry, amino acid analysis, and NMR spectroscopy revealed that chlorofusin contained a densely functionalized chromophore with an azaphilone core linked through the terminal amine of an ornithine residue to a cyclic peptide of nine α-amino acid residues.60 Two of the amino acids contain nonstandard side chains (ornithine and 2-aminodecanoic acid), and four possess the D-configuration.

Figure 5. Structure of chlorofusin (Ref.61) and hexylitaconic acid (Ref.68).

The Boger and Searcey groups simultaneously reported syntheses of the cyclic peptide portion of chlorofusin. Boger and coworkers reported a convergent solution-phase synthesis of both the L-Asn3/D-Asn4 and D-Asn3/L-Asn4 diastereomers.63 NMR experiments identified the absolute stereochemistry of asparagine residues 3 and 4 as being L and D, respectively. Searcey and coworkers published a solid-phase synthesis via on-bead cyclization.64 Both groups reported that the cyclic peptide alone had no inhibitory effect on the p53/MDM2 interaction. A racemic synthesis of the azaphilone core of chlorofusin has been recently reported by Porco and coworkers.65 Preparation of the correct pure enantiomer followed by coupling to the cyclic peptide will afford chlorofusin. The future synthesis of chlorofusin analogues will help to elucidate structure/activity relationships of this class of natural products and may afford more potent ligands for MDM2. Williams and coworkers are investigating the biosynthesis of chlorofusin,66 which in the long term could potentially provide a number of analogues through manipulation of the polyketide and nonribosomal peptide biosynthetic machineries.67

Hexylitaconic Acid

Recently, another natural product inhibitor of the p53/MDM2 interaction, (−)-hexylitaconic acid (12, Figure 5), was reported.68 Isolated from the fermentation culture of a fungus, Arthrinium sp., which was separated from a marine sponge, (−)-hexylitaconic acid had an IC₅₀ of ∼230 μM in a p53/MDM2 ELISA. Both enantiomers of hexylitaconic acid were reported previously, but the absolute stereochemistry of the active compound is unknown.69 Four derivatives of 12, the monomethyl ester, a dihydro derivative, a dihydro derivative of the monomethyl ester, and itaconic acid, were prepared but showed no inhibitory activity. This compound is structurally distinct from other p53/MDM2 inhibitors, and more investigation of its mechanism of action is needed.

SMALL MOLECULE INHIBITORS OF p53/MDM2

Syc-7

The first de novo designed nonpeptidic small molecule inhibitor of the p53/MDM2 interaction was reported by Zhao et al. in 2002.70 Using computer-aided design and their knowledge of the p53/MDM2 cocrystal structure, these workers appended a variety of aromatic rings on a [2.2.1]bicyclic scaffold (Figure 6) in hopes of mimicking the presentation of p53 residues Phe19 and Trp23 within the binding cleft of MDM2. Five of their 23 compounds showed activity in a p53/MDM2 ELISA at micromolar concentrations. The most active compound, 13, was tested in a variety of cell-based assays (10 μM concentration), and Zhao et al. observed a time-dependent accumulation of the p53, MDM2, and p21 proteins and cytotoxicity via induction of apoptosis. While the potency of 13 is rather low, this molecule provided inspiration for the designs of other researchers. Future compounds removed the flexibility of the chains bearing the aromatic groups in order to reduce the entropic cost associated with binding and increase affinity.

Figure 6. Structure of the first nonpeptidic small molecule inhibitor of the p53/MDM2 interaction (Ref.70).

Sulfonamides

Galatin and Abraham used a pharmacophore model and insilico screening to identify sulfonamide 14 (Figure 7) as a potential ligand for the p53-binding cleft of MDM2.71 Compound 14 had an IC₅₀ of 32 μM in vitro and yielded a 20% increase in p53-dependent transcription at 100 μM in a cell-based assay. The authors claimed that the molecular weight of the compound could likely be reduced by removing one of the phenyl substituents from the pyrazolidindione ring without affecting binding affinity, but no further investigations have been reported.

Figure 7. Structure of sulfonamide 14 (Ref.71).

Nutlins (cis-Imidazolines)

The first potent and selective small molecule antagonists of the p53/MDM2 interaction with both in vitro and in vivo activity were reported by scientists at Roche in 2004.72 While historically it had been difficult to identify small molecule inhibitors of protein–protein interactions, the focused nature of the p53/MDM2 binding interface and the presence of deep, well-defined hydrophobic pockets on the surface of MDM2 raised expectations that a small molecule could be found to do the job. A high-throughput screen of the Roche compound collection produced several lead structures. One class was a series of cis-imidazoline analogues that were named Nutlins (for Nutley, NJ inhibitors). As racemic mixtures, compounds 15, 16, and 17 had IC₅₀ values between 100 and 300 nM by SPR in a competition format. The racemic mixture of compound 17 was resolved on a chiral column, and the enantiomers were assayed individually. The enantiomer shown in Figure 8 had an IC₅₀ of 90 nM, while the other enantiomer was 150 times less potent (IC₅₀ = 13.6 μM).

Figure 8. IC₅₀ concentration of Nutlins analyzed by Biacore's surface plasmon resonance technology in a solution competition format (Ref.72).

The binding mode of the Nutlins was revealed by the cocrystal structure of 16 with MDM2 (Figure 9).72 The small molecule mimics the interactions of the p53-peptide, with one bromophenyl ring sitting deeply in the tryptophan binding pocket, the other bromophenyl substituent occupying the leucine binding site, and the ethyl ether side chain directed toward the phenylalanine binding pocket. The imidazoline scaffold replaces the α-helical backbone and is able to direct, in a fairly rigid fashion, the projection of the three hydrophobic substituents into their respective binding pockets. Most of the small molecule inhibitors discussed in the Small Molecule Inhibitors of p53/MDM2 section of this review share this common theme of a rigid heterocyclic scaffold with p-halophenyl appendages.

Figure 9. Crystal structure of 16 in complex with MDM2 (Ref.73, PDB code 1RV1).

To test whether inhibition of p53 binding to MDM2 would translate into activation of the p53 pathway, a battery of cell-based and animal experiments was performed.72 The Nutlins inhibited the growth and viability of cultured cancer cells with IC₅₀ values between 1.4 and 1.8 μM. As expected, treatment of cells having functional p53 with the Nutlins led to increased levels of the p53, MDM2, and p21 proteins, but this treatment had no effect on cells with mutant p53. Additionally, the Roche workers showed that accumulation of p53 was due to reduced degradation rather than increased production via measurement of the mRNA levels with PCR. Cell cycle analysis revealed that treatment with the Nutlins arrested cells in the G₁ and G₂ phases, preventing mitosis. The Roche group found that the Nutlins could induce apoptosis via caspase activation. Treatment of cells with the inactive enantiomer of 17 showed that the effects were specific to p53/MDM2 inhibition and that the compound class is not intrinsically cytotoxic. Finally, this group found that oral administration of 17 to nude mice with an established tumor xenograft resulted in 90% inhibition of tumor growth, relative to controls. These small molecule antagonists are being used to investigate p53 signaling (i.e., the role of phosphorylation of p53 in transcriptional activation)73 and have been advanced for treatment of acute myeloid leukemia74 and multiple myeloma.75

Despite these exciting results, the Roche researchers were only cautiously optimistic about the future of p53/MDM2 inhibitors. While the Nutlins have excellent in vitro activity, they are about 100-fold less active in vivo. It is possible that the Roche effort has not yet identified the optimal molecule or that the signaling pathway downstream of p53 is not completely intact, making p53 activation ineffective.76 A more likely explanation for limited Nutlin activity is obtained through careful consideration of p53 regulation. In cancer cells containing wild-type p53 and overproducing MDM2, inhibition of the p53/MDM2 interaction leads to an increase in cellular levels of p53. Increased levels of transcriptionally active p53 lead to even higher levels of production of MDM2. Therefore, because of the autoregulatory nature of the feedback loop between p53 and MDM2, the inhibitor works against itself by raising the intracellular levels of its target protein as a direct result of its mechanism of action. Thus a relatively high steady-state concentration of the inhibitor is required to have a significant sustained effect. It will be interesting to see how this complex issue is resolved in the future.

The researchers at Roche have devised a clever strategy to expand the potential use of Nutlin p53/MDM2 inhibitors from treatment of the 7% of human cancers with aberrant production of MDM2 and functional p53 to a therapy for the 50% of cancers with an inactive p53 pathway due to p53 mutation. Pretreatment with the Nutlin induces cell cycle arrest in normal proliferating cells but does not affect the cancer cells with mutant p53. Subsequent treatment with a mitotic inhibitor, such as paclitaxel, causes mitotic arrest and apoptosis of the cancer cells (because they are still proliferating), but does not cause any cytotoxic effects in the normal cells.77 The clinical application of this method for selective chemotherapy deserves further investigation.

Benzodiazepinediones

Grasberger et al. at Johnson & Johnson Pharmaceuticals developed a benzodiazepinedione class of small molecule p53/MDM2 inhibitors.78 More than 338,000 compounds were screened for MDM2 binding using the ThermoFluor® assay.79 This assay uses fluorescent dyes to monitor protein unfolding as a function of temperature, with the idea that compounds that bind to the target protein are expected to increase the protein's thermal stability. The hits included 116 compounds from a benzodiazepinedione combinatorial library that had been synthesized via a highly efficient Ugi four-component condensation (Figure 10).80 An anthranilic acid, an α-amino ester, an aldehyde, and 1-isocyanocyclohexene81 were combined, followed by acid-catalyzed cyclization, producing the desired benzodiazepinediones in good yield and purity.82 The discovery of this class of inhibitors is an excellent example of a successful application of combinatorial chemistry using a multicomponent reaction. After careful investigation of the structure/activity relationship of this compound class,80a an optimized inhibitor, 18 (IC₅₀ = 80 nM by FP and 30 μM in cells), was cocrystallized with the MDM2 protein (Figure 11).78 In the bound state, the compound projects the p-chlorophenyl groups into the Trp23 and Leu26 binding pockets on MDM2. The iodobenzene portion of the benzodiazepinedione occupies the phenylalanine binding site.

Figure 10. Ugi four component condensation to form benzodiazepinedione 18 (Ref.78).

Figure 11. Crystal structure of 18 in complex with MDM2 (Ref.78, PDB code 1T4E).

To address the observed 375-fold lower potency in cells relative to protein-based measurements, 18 was optimized for greater cellular permeability and then applied in vivo as a sensitizing agent in conjunction with doxorubcin.83 The carboxylic acid functionality of 18 was replaced with a methyl group, and a valeryl acid (19) or piperazine (20) solubilizing moiety was added (Figure 12), yielding compounds with similar potency in vitro and improved cell-based activity. Compound 19 was threefold more potent, on average, than 18 in cells, but 20 was ∼22-fold more potent than 19, with an average IC₅₀ of 700 nM in cells expressing wild-type p53. The Johnson & Johnson team then tested whether activating p53 in combination with chemotherapy would result in enhanced antitumor activity. They found that administration of doxorubicin alone resulted in a four- to six-fold increase in p53 levels, but a synergistic effect was achieved by treating cells with a combination of 20 and doxorubicin, resulting in an 18-fold increase in p53 levels. These workers found that treating mice having established tumor xenografts with 1.5 mg/kg doxorubicin and 100 mg/kg 20 had a similar effect on tumor size as a larger dose of doxorubicin alone (3 mg/kg). However, the mice treated with the combination therapy experienced a lesser degree of weight loss than those on doxorubicin alone, suggesting a reduction in the side effects of the chemotherapy. The prospect of p53/MDM2 inhibitors as sensitizing agents during chemotherapy needs to be investigated in the clinic. The primary concern is the degree of selectivity that can be achieved between normal and cancer tissue in order to reduce side effects.

Figure 12. Structures of benzodiazepinediones optimized for cell-based activity (Ref.83).

Spiro-Oxindoles

Wang and coworkers used structure-based design to develop a new class of small molecule p53/MDM2 antagonists based on a spiro-oxindole core.84 These workers focused on the indole ring of p53 residue Trp23 not only because it is buried deep within a hydrophobic pocket on the surface of MDM2 but also because the indole NH forms a hydrogen bond with the backbone carbonyl of MDM2 residue Leu54. Wang and coworkers hypothesized that an oxindole could mimic the tryptophan side chain and performed a substructure search for natural products containing an oxindole ring. Several natural alkaloids that possessed a spiro-oxindole core structure were identified, but modeling suggested that none would be a suitable inhibitor because of predicted steric clashes with the MDM2 protein. Nevertheless, Wang and coworkers adopted the spiro(oxindole-3,3′-pyrrolidine) core structure for their design, reasoning that the spiropyrrolidine ring provided a rigid scaffold from which two hydrophobic substituents could project into the Phe19 and Leu26 binding sites on MDM2. A chlorine atom was added at the 6-position of the oxindole, based on the earlier peptide work.46 Several analogs were rapidly constructed using an asymmetric 1,3 dipolar reaction85 as the key step. Compound 21 had an IC₅₀ of 8.46 μM in a competition FP assay (Figure 13). Modeling suggested that additional space remained in the phenylalanine and leucine binding pockets, and so a chlorine atom was incorporated at the meta-position of the phenyl ring, and the isobutyl group was replaced with a 2,2-dimethylpropyl group resulting in compound 22, which was 98 times more potent than the original lead compound. Furthermore, 22 was highly effective against human prostate cancer cells, with an IC₅₀ of 0.86 μM. Compound 22 was 13 times more toxic to these cancer cells than to normal cells with wild-type p53.84

Figure 13. Synthesis and structure of spiro-oxindole small molecule p53/MDM2 inhibitors (Ref.84).

Isoindolinones

Hardcastle et al. have reported inhibitors of the p53/MDM2 interaction that are based on an isoindolinone scaffold.86 Weak inhibitors, identified by compound screening via ELISA, were improved based on computationally guided variations in peripheral substituents. One of the most effective compounds was 23 (Figure 14), which had an IC₅₀ of 16 μM and which induced expression of p53-dependent genes in a cancer cell line with high MDM2 levels (10–40 μM).

Figure 14. Structure of isoindolinone p53/MDM2 inhibitor (Ref.86).

Summary of Small Molecule Inhibitors

Potent and selective small molecule inhibitors of the p53/MDM2 interaction have now been developed. Taking advantage of the pharmacophore established by the published cocrystal structure and earlier peptide work, several groups used a combination of structure-based design and combinatorial chemistry to identify a lead compound. The designs have typically incorporated a heterocyclic core that projects p-haloaromatic appendages toward the Phe19 and Trp23 binding pockets on the surface of MDM2. It took several years for the first reports to emerge,70–72 but once it was clear that small molecules could block p53/MDM2 binding, multiple reports appeared. The current challenge is to obtain molecules that are as potent in vivo as they are in vitro, perhaps avoiding the feedback loop by which p53 induces MDM2 expression. New strategies for using p53/MDM2 inhibitors as sensitizing agents for malignant cells or selective protection of normal cells during chemotherapy may emerge as important therapies for the clinical treatment of cancer.77, 83

OLIGOMERIC SCAFFOLDS FOR α-HELIX MIMICRY AND p53/MDM2 INHIBITION

During the 8 years between publication of the p53/MDM2 cocrystal structure in 1996 and report of the Nutlins in 2004, many researchers initiated investigations of nontraditional strategies for the inhibition of the p53/MDM2 interaction, seeking inhibitors that do not conform to the “small molecules” preferred by medicinal chemists. These efforts were motivated by the lack of reports of small molecule p53/MDM2 inhibitors, and by a general sense that protein–protein interactions might not be amenable to inhibition by small molecules. The unifying theme of these nontraditional approaches was to develop an oligomeric scaffold capable of mimicking the three-dimensional display of side chains from an α-helix. Such a proteomimetic scaffold could be applied not only to p53/MDM2 but also to other protein–protein interactions that feature an α-helical binding epitope on at least one partner, such as Bcl-2 family/BH3 domain interactions or α-helical coiled-coil interactions. A number of successful scaffolds for inhibition of p53/MDM2 are described later. Although the potency of these oligomers is not usually as high as has subsequently been obtained with the most effective small molecule antagonists, especially in vivo, the principles developed in the course of these studies may yet prove useful in other contexts. A number of oligomer-based strategies were aimed at improving upon the disadvantages of peptide therapeutics, such as their proteolytic and metabolic instability, and this goal was achieved.

Retroinverso Peptides

Kahne and coworkers tested a series of peptides to understand how backbone modifications (i.e., chirality and sequence reversal) of the p53 α-helical segment would affect its interaction with MDM2.87 Since the p53 backbone makes only one hydrogen bonded contact with MDM2, Kahne and coworkers hypothesized that the natural left-handed α-helix merely provides a scaffold for the correct spacing and orientation of the interacting side chains. These workers synthesized a number of analogues of the natural L-α-peptide 25 (residues 15–29 of p53), including the enantiomer (D-α-peptide 25), the L-α-retropeptide 26 (a peptide with the opposite order of residues), and the D-α-retropeptide or “retroinverso” peptide 27 (Figure 15). The side chains of peptides 25 and 26 are not superimposable with those of 24 because of their mirror image and N→C vs. C→N relationships, respectively. In the preferred left-handed helical conformation formed by oligopeptides of D-amino acids, the side chains of peptide 27 also do not align with those displayed by the natural peptide in the right-handed α-helical form. However, if retroinverso peptide 27 could adopt a right-handed helix, then its side chains would overlay with those presented by the natural peptide. While peptides 25 and 26 had no measurable affinity for MDM2 in an ELISA, the retroinverso isomer 27 interacted with MDM2 with potency comparable to that of the natural sequence 24. In spite of the reversed positions of the nitrogen and oxygen atoms in the backbone, the switched orientation of the backbone hydrogen bonds, and the inverted chirality of the side chains, the retroinverso peptide is still able to mimic the natural peptide. These workers speculated that the retroinverso peptide is being forced to adopt a disfavored right-handed helical conformation, but no structural evidence was provided. Because of their proteolytic stability, this type of retroinverso peptide might have some utility if active against extracellular protein targets.

Figure 15. Development of retroinverso peptide p53/MDM2 antagonists (Ref.87).

Peptoids

The development of peptoids (oligomers of N-alkyl glycine) as ligands for MDM2 has been pursued by both Appella and coworkers and Kodadek and coworkers95, 97 research groups. Peptoid monomers differ from natural α-amino acids in that the side chain is attached to the backbone nitrogen instead of the α-carbon.88 Peptoids can be considered as a type of “foldamer,” unnatural oligomers that adopt a predictable conformation89; the peptoid backbone can be sterically biased to form a helical conformation by incorporating chiral side chains.90 This feature facilitates the structure-based design of peptoids that display side chains in a predictable three-dimensional array.91 Through the submonomer synthetic approach,92 a wide variety of side chains can easily be incorporated.93 Peptoids are resistant to proteolytic degradation, and several biological applications of peptoids have been reported.94

Structure-Based Design Approach.

Appella and coworkers attempted to mimic the display of p53 side chains Phe19, Trp23, and Leu26 on a helical peptoid scaffold.95 The helical conformation of chiral peptoids is similar to the type-I polyproline helix, in which the amide bonds are cis and the carbonyl oxygens point toward the N-terminus, resulting in a helix macrodipole that is opposite that of an α-helix, relative to the backbone orientation. To develop an α-helix mimic, the peptoid helix was superimposed upon the α-helix of the p53 peptide via molecular modeling. Since the peptoid helix has only three residues per turn, the interacting residues were displayed along one face at the i, i + 3, and i + 6 positions. It was noticed that the side chains project from the peptoid and α-helical scaffolds at different angles. To compensate, an additional methylene group was inserted in the side chain of the peptoid monomers between the backbone and hydrophobic group. To ensure aqueous solubility of the peptoid, three achiral peptoid glutamic acid analogue residues were incorporated at positions expected to be solvent-exposed. Since the conformational stability of the peptoid helix used for the structure-based design relies upon steric constraint of the backbone via incorporation of chiral monomers, four such residues were included in the design, resulting in 28 (Figure 16). The CD spectrum of 28 in water showed the minima at 219 and 205 nm and maximum at 193 nm that have been reported to be characteristic of the peptoid helix. However, this peptoid showed no ability to disrupt protein–protein interaction in a p53/MDM2 FP assay.

Figure 16. Structures of peptoid ligands for MDM2.

Replacement of the side chain carboxylates with phosphonates reduced helical stability (as judged by CD), increased water solubility, and gave rise to weak inhibition of p53/MDM2 binding (29, IC₅₀ = 188 μM). Appella and coworkers proposed that increased binding affinity for MDM2 resulted from greater electrostatic attraction of the phosphonates, relative to the carboxylates of 28, to the side chains of MDM2 residues Lys94, His96, and Lys51, which are positioned around the periphery of the p53-binding pocket. In a continued effort to make the peptoid more hydrophilic, p-nitro groups were added to the phenyl rings of the chiral α-methylbenzyl subunits. One of the phosphonates was replaced with a sulfonamide to reduce the net charge of the peptoid. These changes generated 30, which did not display a helical signature in the CD spectrum but showed improved ability to inhibit p53/MDM2 binding (IC₅₀ = 17.8 μM). Since the reason for this improvement was unclear, a series of peptoids was prepared in which the side chains presumed to interact with MDM2 were removed (31, 32, and 33 in Figure 17). The IC₅₀ values for these compounds were only 1.4- to 3.4-fold higher than that of 30, which seems to suggest that none of the side chains intended to play a key role in binding to MDM2 is crucial for this interaction.

Figure 17. Structure and IC₅₀ values for peptoids with N-methyl replacement of the hydrophobic residues.

Scrambling the relative positions of the hydrophobic achiral side chains had a similarly small effect (34 and 35, Figure 18). Preparation of the choroindole derivative at position 5 (36) reduced the IC₅₀ by only 2-fold, to 9.9 μM. In contrast, as discussed earlier, replacing Trp with 6-chlorotryptophan in the peptide series had reduced the IC₅₀ value by 63-fold.45

Figure 18. Structure and IC₅₀ values of peptoids with scrambled sequence of hydrophobic residues (34 and 35) and 6-chloroindole side chain at position 5 (36).

Appella and coworkers investigated how varying the conformational stability of the backbone would affect the ability of the peptoid to interfere with p53/MDM2 binding (Figure 19). Scrambling the chirality of the helix-promoting N-(p-nitrophenylethyl)-glycine residues (37) and even replacing them with achiral N-(p-nitrobenzyl)-glycine residues (38) had virtually no effect on activity. Introducing a 6-chloroindole side chain at position 5 of the achiral backbone resulted in the best inhibitor in this series (39), with an IC₅₀ of 6.6 μM. A final peptoid was prepared to determine the contribution of the p-nitrobenzyl group at position 3 by replacing it with hydroxyethyl side chain (40). The resulting twofold decrease in activity (IC₅₀ = 12.7 μM) suggested that the p-nitrobenzyl group contributes only modestly to binding.

Figure 19. Structure and IC₅₀ values of scrambled chirality (37), achiral backbone (38), most potent (39), and p-nitrobenzyl → hydroxyl ethyl (40) peptoid ligands.

In summary, Appella and coworkers began with a peptoid that was designed to adopt a helical conformation and mimic the display of p53 side chains Phe19, Trp23, and Leu26 on the α-helix. However, this peptoid had no ability to inhibit p53/MDM2 interaction. Removing the backbone conformational constraint by replacing chiral with achiral residues increased binding affinity, which suggests that the canonical peptoid helix is not quite correct for p53 mimicry; however, a helical (or distorted helical) conformation may nevertheless be required. Information on the structure of the peptoid in complex with MDM2 would facilitate future designs. The modification of the hydrophobic side chains intended to interact with the MDM2 surface resulted in very small changes in binding affinity, suggesting that these groups are not positioned deeply in their proposed binding sites. The greatest incremental increases in peptoid binding were obtained through replacement of the carboxylates in the original design with phosphonates and the addition of a p-nitro group to the phenylethyl side chains. Observing such dramatic effects from manipulation of the hydrophilic residues is unusual. In the end, low micromolar binders were prepared.

Combinatorial Approach.

Bachhawat-Sikder and Kodadek were the first to examine peptoids in this system, and they took a completely different approach in both purpose and method to the development of peptoid ligands for MDM2. Recognizing the potential of proteomics for medicine and biology,96 Bachhawat-Sikder and Kodadek have utilized peptoids as protein capture reagents.97 Using microwave-assisted98 combinatorial chemistry and on-bead screening to construct and assay large peptoid libraries, they identified a peptoid (41, Figure 20) that binds to a MDM2-maltose binding protein (MBP) fusion protein. To validate the hit, a resynthesized and purified sample was tested by isothermal titration calorimetry and found to have an IC₅₀ of 37 μM. Appella and coworkers found that 41 does not disrupt p53/MDM2 binding, according to an FP assay,95 which suggests that this peptoid binds to MDM2 at a site distinct from the p53-recognition cleft (control studies showed that 41 does not bind to MBP). While attached to TentaGel macrobeads, weakly binding ligand 41 could selectively capture MDM2 from a complex mixture of proteins.96

Figure 20. Peptoid (41) and chimeric peptoid-chalcone(42) ligands for MDM2 (Refs.96 and98).

In a slightly different approach, a bivalent high-affinity ligand for MDM2 was developed by capping a different peptoid library with a chalcone.99 The resulting chimeric chalcone-peptoid ligand 42 showed greater affinity for the MDM2 protein (K_d = 1.3 μM) than did either of its constituents (peptoid or chalcone alone). Immobilization of 42 on a surface created a protein-detecting array that was able to bind MDM2 selectively. These results suggest that it may be possible to quantify the amount of MDM2 in a biological sample and determine whether MDM2 is overproduced in a particular cancer, which would be helpful in “personalizing” the clinical use of p53/MDM2 inhibitors.

Terphenyls

Hamilton and coworkers used a terphenyl derivative to mimic the crucial α-helical region of p53 and disrupt p53/MDM2 complexation.100 These workers proposed that placing appropriate alkyl or aryl substituents at the three ortho positions of the terphenyl scaffold would cause these substituents to be projected with a similar distance and angular relationship to the i, i + 4, and i + 7 side chains along one face of an α-helix, if the terphenyl backbone adopts the expected staggered conformation.101 A series of potential terphenyl inhibitors was prepared by sequential Suzuki couplings. Screening in a p53/MDM2 FP assay identified 43 (Figure 21) as the most potent inhibitor, with K_i = 182 nM (deduced from IC₅₀). Further studies revealed that the choice of side chain, relative positions of the side chains on the scaffold, and length of the scaffold were all critical to the interaction. The binding mode of the terphenyls to MDM2 was investigated using ¹H-¹⁵N HSQC NMR; these compounds consistently produced chemical shift changes for the residues within the p53-binding pocket on MDM2. Compound 43 caused a significant change in the chemical shift of MDM2 residue L85, which lies deep within the tryptophan binding pocket, possibly indicating that the 2-naphthylmethylene side chain inserts into this pocket. One concern about the terphenyl scaffold is its extreme hydrophobicity, which may result in nonspecific interactions. To address this potential problem, Hamilton and coworkers tested an underivatized terphenyl and found that it did not measurably interact with MDM2. The selectivity of compound 43 was assessed by comparison with 44, a potent inhibitor of the Bcl-x_L/Bak protein–protein interaction.102 Compound 43 was found to bind to MDM2 100-fold more tightly than did 44, and 43 was 14- and 82-fold selective for MDM2 relative to Bcl-x_L and Bcl-2, respectively. However, compound 43 was inactive in a cell-based assay.103 Another terphenyl derivative, 45, had modest activity in the FP assay and similar activity in a p53/MDM2 ELISA (IC₅₀ = 20 μM). This compound was able to activate transcription by p53 in cells at a 30 μM concentration.

Figure 21. Terphenyl inhibitors of the p53/MDM2 interaction.

The terphenyl scaffold has been used to mimic α-helical recognition surfaces of other proteins, including the C-helix region of gp41, which is important for HIV viral fusion.104 This scaffold has recently been improved through the incorporation of a 1,6-disubustitued indane at the central position of the terphenyl, which allows incorporation of a side chain that could mimic the i + 3 position of an α-helix, in addition to the i, i + 4, and i + 7 positions (46, Figure 22).105 The solubility of the scaffold has been increased by preparation of a terpyridine derivative (47).106 Other similar scaffolds have been proposed (48)107 and utilized (49)108 as α-helix mimics.

Figure 22. Structures of α-helix mimetics: diphenylindane (46, Ref.104), terpyridine (47, Ref.105), biphenyl (48, Ref.107), and terephthalamide (49, Ref.108).

β-Hairpin Protein Epitope Mimetics

Robinson and coworkers have successfully mimicked the α-helix of p53 with a cyclic β-hairpin to generate inhibitors of the p53/MDM2 interaction.109 This work is remarkable because it shows that very different types of protein secondary structure can lead to very similar surfaces. Robinson and coworkers noticed that the distance between the C_α atoms of the Phe19 and Trp23 residues on one face of the MDM2-bound p53 α-helix is similar to the distance between the C_α atoms of two residues (i and i + 2) along one strand of a β-hairpin. A cyclic β-hairpin (50, Figure 23) was designed as a scaffold to hold the side chains of a phenylalanine and a tryptophan residue in the correct relative positions, so that they could interact with their respective binding sites on MDM2. These workers used a D-Pro-L-Pro dipeptide turn unit to stabilize the β-hairpin conformation.110 Compound 50 was identified as a weak lead with an IC₅₀ of 125 μM in a solution-phase competition SPR assay. Using parallel synthesis,111 Robinson and coworkers elucidated the structure/activity relationships among cyclic peptides related to 50. The optimized inhibitor 51, with an IC₅₀ of 140 nM, was nearly 900-fold more potent than the lead. A β-hairpin conformation was observed for the free ligand by ²H NMR spectroscopy, and the ¹H-¹⁵N HSQC NMR data for the MDM2 complex showed that the β-hairpin binds in the p53-binding site on MDM2. A crystal structure of the MDM2/51 complex was obtained (Figure 24).112 Residues Phe1, Trp(6-Cl)3, and Leu4 of 51 were observed to bind in the hydrophobic p53-binding cleft on MDM2. The side chain of the Trp(6-Cl)3 residue inserts deeply into the Trp23 binding pocket on MDM2. These three side chains are found on one face of the β-hairpin together with Trp6 and Phe8; this cluster of hydrophobic residues stabilizes the β-hairpin conformation. Interestingly, the side chains of Trp6 and Phe8 in the second β-strand stack on both sides of MDM2 residue Phe55 on the side of the cleft. Thus, these residues appear to make favorable van der Waals contacts with portions of the MDM2 surface that are not utilized in binding to the N-terminal α-helix of p53, a factor that helps explain the high potency of 51. This intriguing structural result suggests that future efforts to block specific protein–protein interaction might benefit from rational efforts to make contacts beyond the natural binding site.

Figure 23. β-Hairpin inhibitors of the p53/MDM2 interaction (Ref.109).

Figure 24. Crystal structure of 51 in complex with MDM2 (Ref.112, PDB code 2AX1).

The cyclic β-hairpin peptides of Robinson and coworkers have been successfully applied to several biomolecular targets, demonstrating the utility of these protein epitope mimetics.113 It will now be interesting to see whether such peptides display activity in vivo against an intracellular target such as p53/MDM2, because these hairpin peptides will probably exhibit poor cell permeability. It is truly remarkable that α-helix and β-sheet protein architectures are found to be interchangeable in this context, lending credence to the general strategy of using unnatural but well-folded oligomeric scaffolds to mimic protein surfaces.

p-Oligobenzamides

Guy and coworkers used computational methods to design de novo a modular scaffold (52, Figure 25) that mimics the projection of the i, i+ 4, and i+ 7 positions of an α-helix.114 These researchers synthesized a combinatorial library in which the size and nature of the side chains were varied. Compound 53 was identified as the most potent molecule, with a K_d of 12 μM. Binding of 53 in the p53 pocket on the surface of MDM2 was observed by NMR. Of note is the fact that 53 only contains two of the side chains from the original design, and mimics containing an additional monomer unit were much less active.

Figure 25. Structure of p-oligobenzamide scaffold.

β-Peptides

Inhibitors of the p53/MDM2 protein–protein interaction based on the β-peptide scaffold have been investigated in both the Schepartz and Gellman115 laboratories. β-Peptides (oligomers of β-amino acids) are a well-characterized class of foldamers that can adopt a wide variety of discrete secondary structures.89 It is interesting to note that our understanding of β-peptides folding rules arose from fundamental studies of oligomer conformational behavior rather than from a desire to develop a scaffold for α-helix mimicry. The most intensively studied β-peptide secondary structure is the 14-helix, which is defined by 14-membered ring NH_i→O = C_i+2 hydrogen bonds between backbone amide groups. Seebach et al. discovered that β-peptides composed exclusively of β³-residues (Figure 26) can form the 14-helix,116 and Gellman and coworkers showed simultaneously that use of β-amino acids with a six-membered ring constraint, such as trans-2-aminocyclohexanecarboxylic acid or trans-4-aminopiperidine-3-carboxylic acid, lead to a dramatic enhancement in 14-helix stability relative to β³-amino acids.117 In contrast, five-membered ring constraint [as in trans-2-aminocyclopentanecarboxylic acid (ACPC) or trans-3-aminopyrrolidine-4-carboxylic acid (APC)] leads to a completely different secondary structure, the 12-helix, which is defined by 12-membered ring C = O_i→HN_i+3 hydrogen bonds.118 Combining constrained and acyclic residues allows one to prepare β-peptides that display specific arrays of diverse side chains on a stable three-dimensional scaffold.119 Derivatization of β-peptide monomers gives access to more diversity than is available with α-amino acids, since the position of the substituent can be varied between the α and β carbons of the β-amino acid residue, to generate β²- or β³-amino acids, respectively. The predictable relationship between β-amino acid sequence and folding at short oligomer lengths raises the prospect of endowing β-peptides with useful functions. A number of applications have been reported for β-peptides.120 Proteolytic121 and metabolic122 stability and the prospect of intracellular delivery120i make β-peptides very attractive from a biomedical perspective. Both 12- and 14-helical β-peptide scaffolds (Figure 27) have been explored for the development of p53/MDM2 inhibitors.

Figure 26. Structures of β-amino acids.

Figure 27. Structures of α- and β-peptide helices.

12-Helical β-Peptides.

Gellman and coworkers sought to employ the β-peptide 12-helix for development of p53 mimics that would inhibit the p53/MDM2 interaction.123 They believed that the 12-helix was particularly well suited to mimicking the α-helix.87 Both have an N→C helical dipole. The two helices also have a similar pitch despite a difference in the number of residues per turn, 5.4 Å and 3.6 for the α-helix and 5.3 Å and 2.5 for the β-peptide 12-helix, respectively. The internal diameter of the β-peptide 12-helix is only slightly larger than that of the α-helix, 2.3 Å and 2.2 Å, respectively. However, incorporation of the 12-helix-promoting ACPC residues makes the external diameter of the β-peptide 12-helix considerably wider, so that its fit into the cleft on MDM2 could be less favorable than for an α-peptide α-helix. Using a helical wheel diagram and molecular modeling with the crystal structure, these workers considered both right- and left-handed helical designs, corresponding to the (S,S) and (R,R) configurations of ACPC, respectively. After several iterations of the design, synthesis, and testing process, the most active derivative (54, Figure 28) had an IC₅₀ of ∼250 μM in a p53/MDM2 ELISA. These results suggest that the β-peptide 12-helix is not well suited for mimicry of the α-helical domain on p53 that is bound by MDM2.

Figure 28. 12-Helical β-peptide inhibitor.

14-Helical β-Peptides.

Schepartz and coworkers explored the β-peptide 14-helix as a scaffold for displaying a set of side chains in a protein-mimetic manner.124 β-Peptides were designed to mimic the projection of the three hydrophobic side chains (Phe19, Trp23, Leu26) from the α-helical segment of p53.125 Since there are three residues per turn of 14-helix, β³-hPhe, β³-hTrp, and β³-hLeu were incorporated at positions 3, 6, and 9 of deca-β-peptide 55 (Figure 29) in order to be arrayed along one helical face in the folded conformation. The 14-helical conformation was stabilized by β³-hOrn/β³-hGlu side chain/side chain salt bridges along the solvent-exposed second face of the helix.126 The third face, which was not designed to interact with the protein in the bound state, was composed of β³-hVal residues. The overall design is unusual from a structure-based design standpoint, because a left-handed β-peptide helix is being used to mimic a right-handed α-helical structure. Furthermore, β-peptide 55 is designed to bind in an opposite N→C orientation than the p53 α-peptide. In spite of these and other structural differences between the β-peptide 14-helix and the α-helix, these compounds were shown to inhibit the p53/MDM2 interaction with moderate efficacy. β-Peptides containing alternative arrangements of the β³-hPhe, β³-hTrp, and β³-hLeu residues or β³-hAla-mutants thereof were not active.

Figure 29. 14-Helical β-peptide inhibitors of the p53/MDM2 interaction.

Deca-β-peptide 55 was reported to inhibit the interaction between MDM2 and a fluorescently labeled α-peptide corresponding to residues 15–31 of wild type p53 with an IC₅₀ of ∼94 μM in a FP competition assay.124 Molecular modeling suggested a potential steric clash between the C-terminus of the β-peptide decamer with the MDM2 protein. Truncation of the original design produced octa-β-peptide 56, which was more potent (IC₅₀ = 80 μM) but less selective. Investigation by 2D 1H NMR spectroscopy suggested that β-peptide 55 adopts a slightly distorted 14-helical conformation in methanol.125 Schepartz and coworkers reasoned that β-peptides composed entirely of β³-residues can depart slightly from an idealized 14-helical conformation, as expected based on the well-known flexibility and low intrinsic 14-helical propensity of β³-residues,127 and they have speculated that such distortion is necessary for optimal binding to the MDM2 cleft. Further efforts to optimize the binding affinity of β-peptide 55 using a combinatorial approach identified 58 as a sevenfold more potent inhibitor (IC₅₀ = 13 μM), approaching the activity of the p53-derived α-peptide.128 β-Peptides 55 and 58 differ from one another in that three β³-hVal residues on the former are mutated to β³-hIle in the latter. In the 14-helical conformation, these three residues would align along one side of the helix. Since this triple substitution improves the binding affinity, it seems possible that the β³-Ile side chains of 58 make contacts with MDM2 at positions outside the p53-binding cleft. Such interactions would be analogous to the “extra” contacts observed by Robinson and coworkers in the complex between β-hairpin peptide 51 and MDM2.112

These proof-of-principle experiments demonstrate the potential biological applications of β-peptides.129 In addition, these studies show that structure-based design is of only limited utility for the initial discovery of β-peptide inhibitors of protein–protein interactions. The difference in activities of the 12- and 14-helical designs demonstrates the importance of scaffold choice for a particular target, and hence, the need for a large set of foldamer backbones with distinct conformations, which offer distinct ways to arrange sets of side chains in space.

Miniproteins

Schepartz and coworkers have also developed a miniature protein capable of inhibiting the p53/MDM2 interaction.130 The three critical MDM2 contact residues from p53 (Phe, Trp and Leu) were grafted onto the α-helical segment of the avian pancreatic polypeptide (aPP).131 aPP is a well-folded 37-residue polypeptide consisting of an 8-residue polyproline II helix linked through a type I β-turn to an 18-residue α-helix.132 Screening a phage library that varied five presumed noncontact residues in the α-helix of aPP identified peptide 59 with a low micromolar IC₅₀ value in an FP assay (Table I). Further optimization was achieved by varying the residues designed to contact MDM2 in the bound state (60) but not through an attempt to increase the conformational stability of the structure (61). Since these sequences are genetically encodable, they could be used as probes of the p53 pathway in cells.

Table I. Sequence and Activity of Miniature Protein p53/MDM2 Inhibitors

No.	Sequence	IC50 (μM)
aPP	GPSQPTYP GDDAPVEDLIRF YNDLQQYLNVVTRHRYC
59	GPSQPTYP GDDAPVEDLIRF KFLLQWYLLALTRHRYAAAC	3.2
60	GPSQPTYP GDDAPVEDLIRF KFLLQWYLLALSLRNYAAAC	1.6
61	GKSWMTVP GDDAPVEDLIRF KFLLQWYLLALTRHRYAAAC	>150

Summary of Proteomimetic Inhibitors

A number of oligomeric scaffolds have been successfully developed for structural mimicry of the α-helix. These structures are typically much larger than the small molecule inhibitors discussed previously, which allows the oligomeric molecules to access spatially separated binding pockets on the protein surface. The larger size of the proteomimetics, relative to small molecule inhibitors, allows the oligomers to bury more surface area, which offers the prospect of increased binding affinity. However, despite the potential advantages relative to small molecules, most of the oligomers examined to date have not yielded p53/MDM2 interaction inhibitors that are as effective as the best small molecules (e.g., the Nutlins). Only the β-hairpin peptides of Robinson et al. approach this level of efficacy in inhibiting the protein–protein association. Once small molecule p53/MDM2 inhibitors were identified, efforts toward the development of proteomimetic inhibitors were significantly curtailed.

Another common shortcoming of the oligomeric scaffolds is likely to be poor cell permeability;120i indeed, the oligomeric approach may be generally unsuitable for intracellular targets, unless new delivery strategies are developed. Nevertheless, important advances have been made that will be applicable to a large number of protein–protein interaction targets that, unlike p53/MDM2, have been intractable to small molecule approaches. Many of these protein–protein interactions exist on the cell surface, obviating the need for proteomimetic inhibitors to be cell-permeable.

OTHER STRATEGIES FOR ACTIVATION OF THE p53 PATHWAY

While a majority of the research effort has focused on inhibiting the p53/MDM2 interaction, a variety of other approaches to activation of the p53 pathway have been reported. Molecules that reactivate mutant p53 through stabilization of the wild-type conformation have been proposed.133–144 A ligand for p53 that acts as a potential allosteric inhibitor of the p53/MDM2 interaction has also been reported.145 Another common strategy, inhibition of homooligomerization, would not be therapeutically useful in this case because p53 is only active as the tetramer. While these ideas demonstrate creativity, a number of conflicting reports and poorly characterized mechanisms of action have hindered progress.

Reactivation of Mutant p53

Strategies for the reactivation of mutant p53 in vivo would be useful for treatment of ∼50% of cancers. For a tumor to escape the consequences of functional p53, the protein may be mutated at a number of positions to prevent signaling either through destabilization of the protein's tertiary structure or the protein–protein or DNA-protein signaling complex.133 A variety of molecules that bind to mutant p53 and stabilize its wild-type conformation, thus restoring its function, have been described but not elaborated. A short peptide (REDEDEIEW-NH₂) that binds to and stabilizes the core domain of p53 has been proposed to act as a chaperone to maintain existing or newly synthesized destabilized p53 mutants in a native conformation.134 The p53 protein is active when it is tetrameric,135 so Giralt and coworkers developed a tetraguanadinium ligand (62, Figure 30) that binds to the surface of the tetramerization domain to stabilize the active oligomeric state of the protein.136

Figure 30. Structures of molecules designed to activate p53.

A small molecule named PRIMA-1 (p53 reactivation and induction of massive apoptosis, 63) was reported by Selivanova and coworkers to be capable of restoring wild-type function to mutant p53.137 However, more recent investigations have shown that while PRIMA-1 and derivatives thereof (65) do selectively eliminate cells expressing mutant p53, there is no evidence of restoration of wild-type p53 properties.138 In fact, a more potent derivative named PRIMA-1MET (64) did not affect mutant p53 protein levels but did upregulate the expression of PUMA,139 a BH3-only proapoptotic factor that is regulated in both p53-dependent and -independent manners.140 While the idea continues to be investigated with small molecules (66),141 and peptides,142 the molecular mechanism by which a small organic molecule can induce refolding of a mutant protein remains unclear. Efforts to date have been reminiscent of the chemical genetics approach,143 where a small molecule that produces the desired phenotype is identified (i.e., apoptosis via the p53 pathway), but the actual molecular target is difficult to discern.144

Small Molecule Ligand for p53

Selivanova and coworkers have also identified a small molecule named RITA (reactivation of p53 and induction of tumor cell apoptosis, 67), which they believed to bind to p53 rather than MDM2, thereby inhibiting the p53/MDM2 interaction, perhaps by altering the conformation of p53.145 They studied the interaction of RITA with p53 using fluorescence correlation spectroscopy and measured a K_d of 1.5 nM. RITA also showed activation of the p53 pathway in cells at a 10 μM concentration. The authors proposed that RITA binds to p53 and induces a conformational change in the protein that disrupts its interaction with MDM2, stabilizing p53 and causing its accumulation in cells. However, Holak and coworkers investigated the binding of RITA to p53, MDM2, and the complex by NMR and found no evidence to support the idea that RITA inhibits the p53/MDM2 interaction.146 The true mechanism of action remains to be elucidated.147

STRUCTURAL INSIGHTS

Recent elucidation of the uncomplexed MDM2 protein structure by NMR spectroscopy has revealed a number of conformational adjustments that accompany the binding of p53 (Figures 31 and 32).148 As had been suggested previously,149 MDM2 residues 18–24 form a flexible “lid” over part of the binding site that must be displaced upon binding of the p53 peptide. The two MDM2 subdomains that form the walls of the hydrophobic cleft must swing apart to open the groove between them and expose the deeper hydrophobic pockets for the p53 residues Phe19, Trp23, and Leu26. Conformational plasticity of the binding groove is a common theme among protein–protein interactions that makes them difficult targets.150 In this case, the other reported crystal structures of MDM2 in complex with peptides and small molecules are very similar, and the most productive conformation to target has been the bound state.

Figure 31. NMR structure of free MDM2 (Ref.148, PDB code 1Z1M).

Figure 32. Comparison of MDM2 structure in the free (left) and bound (right) states.

CONCLUSIONS

Selective disruption of protein–protein interactions, in general, is regarded as a very challenging goal by scientists seeking to devise therapeutically useful molecules. The interaction between p53 and MDM2 was originally viewed by many workers as exemplary of this challenge. As we have tried to show in this review, quite a number of diverse efforts have been made to develop inhibitors of p53/MDM2 complexation, inspired by the biomedical significance of this interaction and the availability of high-resolution structural characterization of the p53/MDM2 interface. Some of these efforts have involved nontraditional approaches, mostly based on the use of a variety of oligomeric backbones as scaffolds to create molecular surfaces that mimic an α-helical epitope on p53. These oligomer-based approaches tend to generate compounds that differ from typical drug molecules in terms of physical characteristics such as size and polarity (i.e., the oligomers are large and polar). Traditional medicinal chemistry efforts have recently been successful in generating low-molecular-weight compounds that inhibit binding of p53 to MDM2 and that appear to show the expected effects in vivo. Thus, it now appears that nontraditional approaches may not be necessary to block this particular protein–protein interaction. However, these nontraditional strategies retain appeal for a subset of protein–protein interactions. Recent developments suggest a dichotomy among protein–protein interactions, with one class amenable to disruption by more-or-less traditional methods and another class requiring alternative strategies.8 It now appears that the p53/MDM2 interaction is exemplary of the first class. In this case, contact between the two proteins centers on just three side chains from p53, which bind into three complementary pockets that are closely spaced within a cleft on MDM2. This situation can be contrasted with the interfaces frequently observed between intercellular signaling proteins and their membrane-bound receptors, in which the contacting surfaces are relatively large and relatively flat. Such surfaces are found, for example, in the complexes between vascular endothelial growth factor (VEGF) and the extracellular portions of its cell-surface receptors.151 Interactions between VEGF and its receptors can be blocked with clinically useful efficacy, but an antibody (bevacizumab) is required to do this job.152 It remains to be seen whether traditional or nontraditional approaches will generate chemically synthesizable agents that can block this type of protein–protein interaction.

Acknowledgements

The authors are grateful to Dr. Dan Appella for editorial assistance with this manuscript.