Regulated protein degradation is an essential aspect of cell signalling1. Cells must be able to respond immediately to environmental changes to maintain homeostasis or to undergo specified developmental decisions. Alterations in the transcriptome provide a means by which to buffer rapid shifts in extra- or intracellular signals, but post-translational modifications of the proteome provide a much faster mechanism for activation or inhibition of signalling pathways. Much of the control of signalling pathways occurs via protein phosphorylation or protein turnover, and in many cases these two mechanisms are interwoven. Protein phosphorylation is useful because it is reversible via protein phosphatases, thereby allowing a particular protein to be in two or more states and to interchange between them, depending on the signals being received. By contrast, protein degradation provides an irreversible means by which to alter the flux through a particular pathway2, 3, 4. The ubiquitin–proteasome system (UPS) is responsible for much of the regulated proteolysis in the cell, although it is now appreciated that ubiquitin and ubiquitin-like proteins (Ulps) can, under some circumstances, function in reversible systems, aided by the activity of de-ubiquitylating enzymes.
Our expanding understanding of the UPS and its role in human diseases has elicited significant interest in the development of small molecules that target specific components of the pathway. The role of the UPS in disease is really in its infancy, as only a small fraction (perhaps <20%) of the genes with potential links to the pathway based on known sequence motifs have been studied in any detail. Moreover, there has been no comprehensive analysis of mutations in the UPS in human diseases. Nevertheless, there is accumulating evidence of altered functions for components of the UPS in cancer, as well as neurodegenerative diseases. Examples of alterations in UPS genes will be elaborated below but it is clear that mutational inactivation of the machinery involved in linking ubiquitin to specific substrates (or removing it) can occur in tumour-suppressor proteins (for example, F-box and WD40 domain protein 7 (FBW7)), in proteins involved in DNA repair and genome integrity (for example, breast cancer type 1 susceptibility protein (BRCA1)), or in proteins that are important for proper neuronal homeostasis (for example, parkin)5, 6, 7, 8, 9, 10, 11, 12, 13. Moreover, it is also clear that alterations in the expression of UPS genes (most notably overexpression) can have a dramatic impact on the levels of their target genes, and this is sometimes linked to disease. This is most clearly understood for double minute 2 (MDM2) and S-phase kinase-associated protein 2 (SKP2), two proto-oncogenes that are often overexpressed in cancer and which promote the degradation of proteins that negatively regulate the cell-division cycle (p53 (Refs 14–21) and p27 (Refs 22–26)); however, other genes, such as the de-ubiquitylating enzymes CYLD, ubiquitin-specific peptidase 4 (USP4) and USP6, have also been linked to transformation27. Likewise, it is becoming appreciated that targets of the UPS are also frequently mutated in disease. In these cases, the target protein, normally destined for degradation, becomes mutated in residues that contribute to recognition by the UPS, making these proteins immune to the action of their cognate degradation machinery. This typically leads to aberrant expression and altered cellular properties, including transformation in the case of mutations in the c-MYC and 
-catenin transcription factors28, 29, 30.
Although there are potentially many different components of the UPS that might be targeted for inhibition in the context of a particular disease, the first success in the clinic has come from the inhibition of the proteasome itself. Bortezomib (Velcade; Millennium), a peptide boronic-acid inhibitor of the proteasome31, 32, 33, 34, 35, 36, is the first drug targeting the UPS to be approved for human applications, including treatment of relapsed or refractory multiple myeloma34, 37. The relative selectivity of proteasome inhibition for killing tumour cells as opposed to normal cells, coupled with the manageable side effects of the drug, were somewhat surprising initially. However, this selectivity can now be rationalized based on the idea that tumour cells generate higher concentrations of aberrant proteins as well as higher amounts of oncoproteins, making them more sensitive to the effects of proteasome inhibition. Nevertheless, the success with proteasome inhibition, at least for certain types of cancer, suggests that the development of more specific compounds targeting the UPS could be clinically important.
In considering the UPS for drug development, it is important to keep several issues in mind. First, what is the disease to be targeted and what is the nature of the alteration in the pathway that one is trying to modify with a small-molecule therapeutic? For example, it might be easier to inhibit a hyper-active or overexpressed component of the pathway than to resurrect activity from a mutant protein with reduced or negligible activity, such as occurs with familial mutations in the BRCA1 or parkin ubiquitin ligases. Second, it is important to consider how particular classes of proteins within the UPS can be targeted. Some components of the UPS — including the ubiquitin-activation machinery and de-ubiquitylating enzymes — are conventional enzymes and might therefore be more amenable to inhibition with small molecules than other components of the system, such as adaptor proteins, which typically lack active-site pockets. Moreover, because many components in the pathway are multi-gene families, some attention to specificity is required, as off-target effects can reduce the suitability of particular classes of inhibitors.
In this review, we describe the structures and functions of major components in the UPS and summarize our current state of knowledge for those UPS components that have been implicated in disease. We also discuss published efforts to identify inhibitors of the UPS pathway and suggest possible additional points of therapeutic intervention.
Targeting ubiquitin activation
Although most post-translational protein modifications involve attachment of relatively small functional tags, such as phosphates or methyl groups, the regulated protein-destruction pathway is different. Here, a protein targeted for degradation is marked with a chain of ubiquitin molecules38, 39, 40, 41, 42. Ubiquitin is a 76-amino-acid, evolutionarily conserved polypeptide that adopts a tightly packed, nearly globular conformation43. In the canonical pathway, a lysine residue in a protein is attached to the C-terminal carboxylate of ubiquitin through an isopeptide linkage [...-CO-NH2
-...]. This ubiquitin molecule then serves as the point of further chain extension via formation of additional isopeptide linkages with the C terminus of ubiquitin through lysine residues located on the ubiquitin molecule itself. Ubiquitin contains seven lysine residues. The most prominent site of ubiquitin chain extension is lysine-48 (K48); K48 polyubiquitin chains linked to a target protein are thought to be the primary form that is recognized by the proteasome44, 45, 46. However, other chain linkages, including K63-linked polymers, have also been identified and these chains seem to be regulatory as opposed to promoting degradation (Box 1). Figure 1 explains the five major steps of the protein polyubiquitylation pathway.
Figure 1 | Overview of the ubiquitin–proteasome system (UPS).
![Figure 1 : Overview of the ubiquitin|[ndash]|proteasome system (UPS). Unfortunately we are unable to provide accessible alternative text for this. If you require assistance to access this image, or to obtain a text description, please contact npg@nature.com](/nrd/journal/v5/n7/thumbs/nrd2056-f1.jpg)
Protein degradation through the UPS is a highly regulated process, involving several steps3, 38. The first step in the cascade is ubiquitin activation by E1 (ubiquitin-activating enzyme) followed by ubiquitin delivery to E2 (ubiquitin-conjugating enzyme) (a). The second step involves complex formation by E2-Cys
Ub, E3 (ubiquitin ligase) and the substrate (b). These initial steps involve formation of thiol esters between the active-site cysteines of E1 and E2 enzymes and the carboxy-terminal carboxylate of ubiquitin. The third step (c) comprises transfer of ubiquitins to the substrate lysine(s) to earmark the substrate with a polyubiquitin chain. In the fourth step of the pathway (d), a polyubiquitylated substrate is released from the E3. Proteasomes recognize the polyubiquitin chain as a signal to de-ubiquitylate and destroy the substrate. The fifth step seals the fate of a doomed protein (e). The proteasome unfolds the substrate in ATP-dependent manner, removes the ubiquitin chain through a proteasome-associated ubiquitin hydrolase activity, and threads the unfolded protein into the proteasome chamber, where the protease active sites are located. The ubiquitin molecules are recycled, and the peptides generated are used in major histocompatibility class I-coupled antigen presentation or degraded to amino acids that are recycled for new protein synthesis. MDM2, double minute 2; SCF, SKP1–Cullin–F-box; SKP, S-phase kinase-associated protein; RBX, RING-box protein.
The most upstream components of the ubiquitin (Ub) and ubiquitin-like protein (Ulp) conjugation machinery (generically referred to as E1 ubiquitin-activating enzymes) perform the activation step (Fig. 2), in which the Ub/Ulp first becomes adenylated on its C-terminal glycine residue and then becomes charged as a thiol ester, again at its C terminus. The Ub or Ulp is then transferred to one of several distinct E2 ubiquitin-conjugating enzymes, also through a thiol ester bond. Although the E1-activating enzyme for Ub was the first to be identified, we now know that there are several related enzymes, which serve to catalyse activation of different Ulps. Prominent among these are the E1 for the Ulps small ubiquitin modifier (SUMO; E1SUMO), NEDD8 (neural precursor cell expressed, developmentally down-regulated 8; E1NEDD8) and ISG15 (E1ISG15) (Box 1). The mechanism of action of these enzymes is thought to be very similar, although each of these Ulp E1s seems to be able to 'charge' only a single E2: ubiquitin-conjugating enzyme C9 (UBC9), UBC12 and UBC8, respectively. By contrast, E1Ub can charge a wide variety of E2s. In addition, these Ulps generally form mono-conjugates, and unlike ubiquitin do not generally form polymeric chains upon conjugation to targets.
Figure 2 | Biochemical steps in the ubiquitin-activation reaction.
This step of the ubiquitylation cascade requires ATP binding to the ATP-binding cleft of E1. This process can be divided into four major sub-steps: (a) adenylation of the carboxyl terminus of a free ubiquitin molecule by E1; (b) rapid cis transfer of the E1-bound ubiquitin molecule from AMP to the active-site cysteine in E1, with subsequent release of free AMP; (c) adenylation of another free ubiquitin residue by the same CysUb-loaded E1 molecule; and (d) recruitment of E2 (ubiquitin-conjugating enzyme) followed by transfer of the activated ubiquitin from the active cysteine of E1 to the catalytic cysteine of E2.
Although crystal structures of E1Ub have not been reported, detailed structural analysis of E1NEDD8 (Refs 47, 48) and E1SUMO (Ref. 49) have been reported and provide a framework for contemplating inhibitors of Ub and Ulp activation (Fig. 3). Both E1NEDD8 and E1SUMO are heterodimeric complexes in which each of the two subunits (amyloid- precursor protein binding protein 1 (APPBP1)/ubiquitin-activating enzyme 3 (UBA3) and SUMO-1-activating enzyme subunit 1 (SAE1)/SAE2, respectively) contains domains related to the bacterial thiamine biosynthesis protein ThiF. The ThiF domain contributes substantially to the ATP-binding site responsible for adenylation of ubiquitin (Fig. 3b,c). By contrast, E1Ub represents a fusion of these two proteins to form a single polypeptide which is expected to form a similar three-dimensional structure. In addition to the adenylate pocket, two additional functional domains are found in the complex. Both SAE2 and UBA3 share a domain containing a conserved cysteine residue, which is the site of thiol ester formation with the Ulp. This domain is referred to as the catalytic cysteine domain (CC) (Fig. 3b,c). In addition, the C terminus of these subunits also contains a small domain that forms a ubiquitin-like fold and is important for interaction with E2 (Fig. 3d). Structural analysis of E1NEDD8 with NEDD8 and with its cognate E2, UBC12, have provided insight into recognition of E2s by E1s47. The interaction of E1NEDD8 with UBC12 is bipartite in character — it involves interactions between two separate domains on each of these proteins. Insertion of 13 N-terminal residues of UBC12 into a surface groove on UBA3 is specific for the NEDD8 pathway, as there is no similarity between the N terminus of UBC12 and other E2s. However, the general mechanism of interaction between a ubiquitin-like fold of UBA3 and a core domain of UBC12 seems (Fig. 3b,d) to be widely conserved in various E1/E2 pairs, including those involved in ubiquitin activation and transfer47.
Figure 3 | Ubiquitin-activating enzymes as potential pharmacological targets.
a | The NEDD8-activating enzyme (E1NEDD8) is composed of UBA3 and APPBP1. In the presence of ATP, NEDD8 is activated through formation of a thiol ester, which is then transferred to UBC12. UBC12 transfers NEDD8 to a single lysine residue in the cullin proteins to form an isopeptide linkage. b | Organization of the APPBP1/UBA3 heterodimer and comparison of the domain structures of E1NEDD8 with E1Ub. c | Structure of the E1NEDD8 complex bound to NEDD8 (APPBP1 is shown in yellow, UBA3 in magenta, NEDD8 in pink). ATP (space-filled) is buried in the adenylation pocket and is located near the carboxy-terminal glycine residue of NEDD8. d | The ubiquitin-like domain of E1NEDD8 (magenta; see panel b) uses a -sheet surface common to all ubiquitin folds to bind to UBC12 (yellow). The catalytic cysteine of UBC12 is shown in a space-filled model. APPB1, amyloid- precursor protein binding protein 1; NEDD, neural precursor cell expressed, developmentally down-regulated; UBA, ubiquitin-activating enzyme.
Current models47, 48, 49 suggest that large conformational changes are necessary for delivery of the NEDD8 C terminus to the catalytic cysteine of UBC12. The ubiquitin-fold domain of E1NEDD8 was proposed to undergo a dramatic rotation around a flexible linker50 found between the Ub-fold domain and the NEDD8-binding site. This large conformational reorganization would bring the ubiquitin-fold domain of E1NEDD8 (together with bound UBC12) closer to the NEDD8-preloaded E1NEDD8 catalytic cysteine to facilitate NEDD8 transfer to UBC12. Importantly, the length and flexibility of this linker were shown to be crucial for transfer to UBC12 (Refs 48–50).
These and other studies suggest two possible approaches to inhibiting E1 proteins. The first involves the identification of inhibitors of Ub or Ulp adenylation by blocking either access of the Ub/Ulp to the adenylate site or by blocking access of ATP. There is a long history in the pharmaceutical industry of successful targeting of ATP-binding sites, most notably protein kinases. For instance, imatinib (Gleevec; Novartis), a potent inhibitor of the oncogenic BCR–ABL fusion kinase and several other tyrosine kinases, is approved for treatment of chronic myelogenous leukaemia51. Multiple clinical trials that address the potential role of imatinib in other malignancies, such as gastrointestinal stromal tumours, ovarian carcinomas and childhood gliomas, are currently under way, and there is some additional information suggesting that the scope of ATP pockets targeted by imatinib might be broader than previously anticipated. The experience gained in the identification of protein kinase inhibitors could have major benefits in efforts to identify inhibitors of E1s, although the aforementioned example of kinase inhibitors indicates that the small molecules targeting ATP-binding pockets are not always as specific as predicted.
The second potential site for drug development involves the E1–E2 interaction (Fig. 3d). It is important to keep in mind that such protein–protein interaction surfaces might make structure-based design of novel drugs challenging because of difficulties with identification of small-molecule-binding pockets52. Although it might be difficult to custom-tailor a small molecule that would bind the E1 linker and prevent its rotation, this domain of E1 is localized on the molecular surface and therefore would be available to small molecules. A functional or biophysical screen for small-molecule inhibitors of E1–E2 ubiquitin transfer could therefore possibly reveal stabilizers of the linker that could be utilized for non-specific silencing of ubiquitylation pathway(s).
Assuming that it is possible to develop a specific inhibitor of E1 enzymes, most likely through blocking access of ATP, the question arises as to the biological consequences of such inhibition. Previous studies in mammalian cells indicate that temperature-sensitive mutations in E1Ub arrest the cell cycle in late S phase and G2, indicating that E1Ub is required for cell proliferation. Similarly, mutations in E1NEDD8 effectively block cell division in Caenorhabditis elegans53, and both UBC12 and UBC9 are required for development in mammals. In principle, therefore, such E1 inhibitors might provoke cell-cycle arrest and therefore could be useful in hyperproliferative disorders such as cancer. However, it is likely that multiple pathways would be affected by inhibitors targeting the E1 machinery, and many of these pathways will be important for the functions of normal cells. One potential advantage of inhibiting E1NEDD8 over E1Ub is that the only validated function for NEDD8 is activation of the cullin class of ubiquitin ligases (Fig. 3a), several of which are candidate targets for drug discovery (see below); however, it is possible that other targets of NEDD8 conjugation (such as von Hippel-Lindau (VHL) tumour suppressor54 and p53 (Refs 55, 56)) exist as well, and this could potentially complicate clinical use of neddylation inhibitors. Nevertheless, it is hoped that inhibition of E1NEDD8 would inhibit cullin-based E3 conjugation of substrates but would presumably not affect the many hundreds of ubiquitylation reactions that are not cullin-dependent. Still, it remains to be determined whether a sufficient therapeutic window can be attained with an inhibitor of the E1 class of enzymes.
Targeting E3 ubiquitin ligases
Overview of E3s. The organization of the polyubiquitylation cascade is hierarchical. One common E1 enzyme activates ubiquitin for all cellular polyubiquitylation networks. Then, multiple combinations of several E2s and at least several hundred E3s are used to catalyse ubiquitylation of many more substrates in a target-specific manner (Fig. 4a). It is this carefully regulated pattern of interactions between E3s and their targets that provides the specificity necessary for appropriate degradation of ubiquitylation substrates.
Figure 4 | Ubiquitin ligases as ubiquitylation specificity modules.
a | The ubiquitin cascade is pyramidal in design. A single E1-activating enzyme transfers ubiquitin to roughly three dozen E2s, which function together with several hundred different E3 ubiquitin ligases to ubiquitylate thousands of substrates. b | Classes of ubiquitin ligases: single RING-finger E3s; HECT E3s; and multi-subunit RING-finger E3s, exemplified by the SCF complexes. Only HECT-domain E3s form a covalent bond with ubiquitin during polyubiquitylation of their target proteins. Specific E3s discussed in this review are indicated. c | SCF E3 complexes consists of a scaffold-like cullin molecule; a RING-finger-containing subunit (RBX1 or RBX2), which functions to bind E2s; and a substrate-specificity module, which binds substrates. d | Distinct cullins utilize structurally related specificity factors that are specific to each cullin. e | The crystal structure of SCFSKP2 reveals that the carboxy-terminal domain of CUL1 binds RBX1, while the amino-terminal recruits the substrate-specificity module (in this case, a SKP1/SKP2 heterodimer)58. -TRCP, -transducin repeat-containing protein; BTB, broad complex/tracktrack/bric-a-brac; DDB, DNA-damage binding protein; FBW, F-box and WD40 domain protein; HECT, homologous to E6-AP COOH-terminus; RBX, RING-box protein; SCF, SKP1–Cullin–F-box; SKP, S-phase kinase-associated protein; VHL, von Hippel-Lindau.
The exact composition of an E2–E3–substrate complex and the details of ubiquitylation biochemistry depend on the E3 class involved. E3s can be divided into three major classes: RING-finger E3s; HECT (homologous to E6-AP COOH-terminus)-domain E3s; and U-box E3s. Each of these three classes of E3s has a distinct protein interaction domain (RING-finger, HECT or U-box domain) to bind E2s, and other domains in the protein or protein complex function to recruit substrates57, 58, 59 (Fig. 4). E3s therefore serve to bring E2s into proximity of their substrates. As discussed below, the molecular interactions involved in the interaction of E2s with these three classes of E2 recruiters are highly conserved. Biochemically, the HECT class differs from the RING-finger class in that HECT domains function directly in ubiquitin transfer by forming a thiolester intermediate with ubiquitin, which is then transferred to substrate. By contrast, RING-finger E3s do not seem to function directly via a thiolester intermediate but simply function as adaptors60.
E3 specificity modules determine which substrate is to be ubiquitylated. It might therefore be feasible to regulate the activity of selected proteins by manipulating their specific ubiquitin ligases. Although bortezomib targets all ubiquitylated proteins destined for degradation by the proteasome, attacking a single ubiquitin ligase might allow for manipulation of distinct pathway components, leading to more selective stabilization of a subset of ubiquitylated proteins. This increase in the specificity of therapeutic intervention could potentially boost the effectiveness of the treatment and eliminate some nonspecific side effects at the same time. Here we will discuss a few ubiquitin ligases that are potentially attractive from a pharmacological or disease standpoint, and describe current strategies of manipulating a specific ubiquitin ligase activity, including the small-molecule inhibitors of ligase–target interactions.
RING-finger E3s as drug discovery targets. The RING-finger domain is a subtype of zinc-finger domain found in a large number of proteins in mammals (more than 300 RING-finger genes in humans)60, 61. RING-finger proteins constitute the largest class of ubiquitin ligases. There are two major sub-classes within the RING-finger class of E3s: simple RING-finger E3s in which the RING-finger and the substrate-binding domain are located on the same polypeptide; and cullin-based RING-finger E3s, which utilize RING-box protein 1 (RBX1) and RBX2 in complexes with modular cullin-dependent substrate receptor proteins (Fig. 4b–e). Because of the diversity of these substrate receptor families (Fig. 4d), there might be more than 200 different RBX1-/RBX2-dependent E3s in humans, as described below.
RING-fingers contain seven crucial cysteines and one histidine (C3HC4), or six cysteine and two histidine residues (C3H2C3), that hold two zinc atoms in a characteristic spatial conformation referred to as the cross-brace motif61 (Fig. 5). Although the sequence of all and length of some spacers separating the zinc-binding residues is quite diverse, there is significant three-dimensional similarity between different RING-finger domains, especially in the components known to interact with E2s (Fig. 5). Several RING-finger-containing E3s have been implicated in disease processes, including cancer and neurodegenerative diseases.
Figure 5 | RING-finger domain.
a | Schematic representation of a C3HC4 RING finger. Most RING fingers contain two zinc atoms (yellow) coordinated with cysteine or cysteine/histidine-rich clusters (red). The general consensus sequence is: C-X2-C-X9–39-C-X1–3-H-X2–3-C-X2-C-X4–48-C-X2-C, although some variations exist. b | Overlay of crystal structure of the RING-finger domains found in c-CBL (blue) and RBX1 (red) reveals a significant degree of structural similarity in their E2-binding components. RBX, RING-box protein.
In cases in which RING-finger E3s are overexpressed in disease (for example, cancer), development of inhibitory compounds might be warranted. There are two potential approaches for blocking the activity of RING-based E3s. One is to develop molecules that disrupt the interaction of the RING-finger with E2s. To date, such molecules have not been reported for any RING-finger E3. Alternatively, the identification of molecules that interrupt the interaction between the ubiquitylation substrate and the substrate interaction domain on the RING-finger protein might provide a means by which to selectively block degradation of one or a small number of proteins. As described below, recent results in the p53 arena indicate that this is possible, at least with the MDM2 RING-finger protein.
p53 and MDM2. Human cells contain intrinsic tumour-suppressor networks that dynamically respond to shifts in gene-expression patterns and monitor genome status to prevent neoplastic transformation. However, these mechanisms are not perfectly reliable, and their deterioration opens the way to tumorigenesis. The p53 tumour suppressor is amenable to experimental therapeutic approaches62 as a highly regulated node of multiple proliferation- and apoptosis-related networks63. Approximately 50% of all human tumours contain mutations in the p53 gene64, and the neoplasms that retain wild-type p53 frequently derange other elements of the p53 network — via promotion of p53 degradation, for instance65. Therefore, p53 is a well-recognized 'guardian of the genome' that prevents mutagenesis and, consequently, carcinogenesis by promoting cell-cycle arrest or apoptosis (Fig. 6a) (for a review see Ref. 66).
Figure 6 | p53–MDM2 interaction as a therapeutic target in human cancer.
![Figure 6 : p53|[ndash]|MDM2 interaction as a therapeutic target in human cancer. Unfortunately we are unable to provide accessible alternative text for this. If you require assistance to access this image, or to obtain a text description, please contact npg@nature.com](/nrd/journal/v5/n7/thumbs/nrd2056-f6.jpg)
a | Simplified view of p53 signalling pathway. Genotoxic stress activates a network of protein kinase pathways that converge on the p53 protein. Although unphosphorylated p53 is rapidly degraded via the ubiquitin–proteasome system, phosphorylated p53 cannot bind its primary E3 ligase (MDM2) and is stabilized. Stabilized p53 arrests the cell cycle, and promotes DNA repair and apoptosis, depending on the cellular context. Failure of the p53 response is the most common event in human cancers, and restoration of p53 stabilization is a well-established potential anticancer strategy. b | Structure of Nutlin (left) and RITA (right), the first small-molecule inhibitors of the p53–MDM2 interaction73, 75. c | Structure of the Nutlin–p53 complex19. Left: the sides of p53-binding groove in MDM2 (red) is limited by two 
-helices and a short -sheet while the bottom is formed by two shorter -helices perpendicular to the pocket sides. The amino-terminal domain of p53 (green) is stabilized in -helical conformation upon binding to MDM2 due to a network of hydrogen bonds trapping three crucial residues of p53, Phe19, Trp23 and Leu26. These crucial p53 residues point towards the bottom of the groove. Right: Nutlin prevents interaction between p53 and MDM2 by mimicking the conformation of Phe19, Trp23 and Leu26 of p53 to block the MDM2 p53-interacting domain73. MDM2, double minute 2.
There are multiple potential drug targets in this pathway62, the most attractive being the major ubiquitin ligase of p53, MDM2. MDM2 is an oncogenic RING-finger protein whose expression is transcriptionally induced by p53 to generate a negative feedback loop by degrading the p53 protein via the UPS (Fig. 6a). Logically, inhibition of the MDM2–p53 interaction or inhibition of the conjugation of ubiquitin to p53 might result in stimulation of the tumour-suppressor activity of p53. Moreover, inactivating MDM2 might prove beneficial not only in tumours carrying wild-type p53: MDM2 might work as a ubiquitin ligase for other anti-oncogenic proteins67.
MDM2 disarms p53 by at least three complementary mechanisms: physically blocking the N-terminal transactivation domain on p53; promoting nuclear export of p53 to keep this transcription factor away from its target genes21; and ubiquitin-dependent p53 degradation15, 16, 21, 63. It has therefore been proposed that the p53 protein could be reactivated in p53-positive cancer cells via disabling MDM2. The resurrected p53 would then promote cell-cycle arrest, enhance apoptosis and synergize with conventional anticancer treatment. This concept is being addressed experimentally with increasing success.
Initial attempts to crack the MDM2–p53 regulatory loop included the generation of anti-MDM2 antisense oligonucleotides68, 69, 70, scaffold-attached peptides71 and proteins72 that were useful as a proof of principle. But the major challenge was to discover bioavailable small molecules that could block the MDM2–p53 interaction and potentially find their way into the clinic. These efforts were supported by an understanding of the anatomy and physiology of the p53–MDM2 interaction.
From a structural standpoint, the MDM2–p53 interface19 looks inviting for experimental drug design: MDM2 contains an open p53-binding pocket that might be accessible for small molecules (Fig. 6c). The p53-binding pocket on MDM2 is approximately 25 Å long and 10 Å wide. Two -helices supported by a short -sheet mark the boundaries of the groove; the bottom is formed by two shorter, antiparallel -helices that are roughly perpendicular to the pocket margins19. Most amino-acid residues lining the interior surface of the groove are hydrophobic. The MDM2-binding domain of p53, which overlaps the p53 transactivation domain, is flexible in solution but adopts a stable -helical conformation upon binding to MDM2 as a result of generating a network of hydrogen bonds between the hydrophobic MDM2 pocket and a few crucial amino-acid side chains on the p53 N terminus, including Phe19, Trp23 and Leu26 (Fig. 6c)
The first small-molecule MDM2 inhibitors — Nutlins (cis-imidazoline derivatives) — were identified in a chemical library screen73 for small molecules able to block the interaction between p53 and MDM2. Based on their structural, biochemical and pharmacodynamical parameters, Nutlins seem to be bona fide anti-MDM2 compounds. Structurally, they mimic the spatial conformation of the crucial MDM2-interacting residues on p53 and therefore are able to occupy the p53-binding pocket and displace p53 from it73 (Fig. 6c). As expected, Nutlins activate p53-dependent cell-cycle arrest and apoptosis in cancer cell lines. Importantly, Nutlins are effective in vivo upon oral administration: they halt the growth of nude mouse tumour xenografts without noticeable toxicity to healthy tissues73. These findings indicate reasonable bioavailability of these small-molecule MDM2 inhibitors.
Another p53-stabilizing small molecule (RITA; 2,5-bis(5-hydroxymethyl-2-thienyl)furan) was initially found to have anticancer activity in the National Cancer Institute Anticancer Drug Screen74 and recently scored as a hit in another functional chemical library screen designed to identify compounds that specifically arrest growth of a p53-positive cancer cell line75. RITA is unrelated to the Nutlins both structurally and functionally (Fig. 6b). In contrast to the Nutlins73, RITA does not seem to bind MDM2. Instead, it binds the N terminus of p53 and, although no definitive structural data are currently available, it seems to either prevent the recognition of p53 by MDM2, stabilize the N-terminal -helix domain of p53 in the MDM2 groove, or both75. Interestingly, RITA seems to prevent p53 from interacting with other regulatory proteins75, such as p53-associated parkin-like cytoplasmic protein (PARC, the p53 cytoplasmic anchor) and p300, which, in addition to acetylating p53 (Refs 76, 77), can promote p53 polyubiquitylation by MDM2 as p53's 'E4'78, 79. The RITA-stabilized p53 is transcriptionally active as it can promote expression of endogenous p53-target genes75. Most excitingly, RITA slows down the growth of mice tumour xenografts in a dose-dependent manner75, although oral bioavailability of RITA was not tested in this initial study.
One question that has not been addressed is whether RITA and Nutlins might show synergistic antitumour activity in the mouse tumour xenograft model, because they attack different facets of the MDM2–p53 recognition process73, 75. Furthermore, the interaction between RITA and p53 should be looked at more closely by NMR or crystallography to help understand the mode of RITA-mediated p53 activation, especially given that the initial attempts to detect RITA–p53 interaction by NMR produced ambiguous results80, 81. It is also necessary to understand why administration of RITA causes cell-cycle perturbations and slight increase of apoptosis in p53-negative cells75. Does this mean that RITA targets other cellular proteins — for example, other p53-related proteins such as p63 or p73? Finally, possible side effects of long-term Nutlin/RITA administration should be addressed in the animal model, especially because furan derivatives were previously claimed to be mutagenic82, 83, 84, 85. So although it is clear that small molecules can be generated that have the desired effect of activating p53 via suppression of its degradation, further study is required to determine whether these molecules will be useful in the treatment of human cancer.
Parkin. In Parkinson's disease, the dopaminergic neurons of substantia nigra degenerate and die. Loss of these neurons leads to partial denervation of the striatum. The clinical consequence is impaired extrapyramidal control of skeletal muscle movement on multiple levels, which manifests principally as rest tremor, muscular rigidity and bradykinesia. Although some of the Parkinson's symptoms are initially controlled by dopamine precursors and agonists that boost the activity of remaining dopaminergic neurons in the substantia nigra, the disease eventually progresses towards a patient's incapacitation.
The malfunction of the ubiquitin–proteasome pathway leading to improper handling of misfolded proteins likely contributes to the pathogenesis of Parkinson's disease. Lewy bodies, the histopathological hallmark of Parkinson's disease, contain ubiquitin and filamentous aggregates of denatured proteins including -synuclein. There is ongoing debate as to whether this collapse of the UPS in patients' brains is a consequence of genetic insults, environmental influences such as oxidative stress or toxin exposure, or both. Since Parkinson's disease most probably represents a constellation of diverse neurodegenerative syndromes that result in clinically similar outcomes rather than a single pathological entity, both hypotheses seem to be true, at least in some cases. (For more a detailed discussion of diagnosis, pathogenesis and genetics of Parkinson's disease, see Refs 9–13, 86–92.)
Approximately 50% of the patients with a specific type of juvenile Parkinson's disease carry mutations in a gene named parkin93, which encodes a RING-finger E3. The parkin molecule contains an N-terminal Ubl (ubiquitin-like) domain (which has been postulated to mediate parkin's interaction with the proteasome) and two C-terminal RING-finger domains that flank an in-between-RING (IBR) domain. The C-terminal parkin domain binds putative ubiquitylation substrates as well as other interacting molecules, such as U-box-containing E3 ligase, carboxyl terminus of the HSC70-interacting protein (CHIP) and the chaperone heat-shock protein 70 (HSP70).
It seems that parkin protects neurons and glial cells from excitotoxic insults, presumably by acting as an E3 for as yet unknown proteins. Parkin-knockout mice show some evidence of neurotoxicity, although in a manner that is not reminiscent of Parkinson's disease patients. parkin-mutant Drosophila show muscle degeneration and decreased lifespan, which is possibly due to mitochondrial failure that results in oxidative damage of the tissues94, 95; the Drosophila phenotype can be rescued by expression of human parkin96. Interestingly, parkin gene therapy has also been shown to rescue the rat model of Parkinson's disease (-synucleinopathy)97. Although it is not known whether and how abnormal parkin function contributes to the pathogenesis of sporadic Parkinson's disease in humans, the hope is that research on parkin could at least pinpoint the affected pathways and therefore provide targets for therapeutic manipulation.
What is the mechanism of parkin's neuroprotective function in the central nervous system? A major limitation in our understanding of parkin is that we currently do not know what the relevant substrates of its E3 activity are. Several candidate proteins — including Parkin-associated endothelin receptor-like receptor (PAEL-R, an endoplasmic reticulum-associated, misfolding-prone protein of unknown function), -synuclein98, misfolded dopamine transporters and polyglutamine-repeat proteins, cyclin E, synaptotagmin XI and tubulin — have been suggested to be substrates, but the relevance of these proteins is in question because the abundance of none of the candidate substrates is increased in neurons of mice lacking parkin. Point mutations in both the first and second RING-finger domains of parkin, as well as mutations in the ubiquitin-like domain, are found in patient families. This genetic clue suggests a role for ubiquitin ligase activity in the function of parkin that promotes neuronal survival. As such, any small-molecule therapeutics targeting parkin would need to be capable of promoting activity of an otherwise non-functional or partially functional allele. With the existing understanding of RING-domains and their interaction with E2s, it is not clear how this would be accomplished, other than by gene therapy, which would be extremely challenging in the case of Parkinson's disease. In principle, it might be possible to identify small molecules that facilitate recruitment of E2s to a defective parkin RING-domain, but such molecules would probably be allele-specific and would therefore be of limited utility. This exemplifies the situation for many E3s whose loss of function promotes disease, including BRCA1.
SCF ubiquitin ligases. The SCF (SKP1–Cullin–F-box) complex is a multi-subunit ubiquitin ligase that uses interchangeable specificity factors (F-box proteins) to recognize specific substrates99. These complexes contain four core subunits (Fig. 4c,e). Cullin 1 (CUL1) serves as a scaffold for assembling the ubiquitin-conjugating machinery and the substrate. The C terminus of CUL1 interacts with the RING-finger protein RBX1 (also called ROC1), which in turn interacts with E2 ubiquitin-conjugating enzymes. The N terminus of CUL1 binds to SKP1, which in turn binds to F-box proteins. The human genome contains at least 68 F-box proteins, and each of them is likely to target multiple substrates for degradation100. Other cullins use different classes of substrate specific adaptors, including BTB proteins for CUL3 and suppressor of cytokine signalling (SOCS)-box proteins for CUL5 (Fig. 4d)87, 88. There are more than 170 BTB proteins and more than 45 SOCS proteins encoded by the human genome. However, we are only beginning to decipher the complicated molecular networks linking SCF E3s to their substrates. Even for the most thoroughly studied CUL1-based SCF complexes, only a small number of F-box protein–substrate pairs have been identified. Most prominent among these are SCFSKP2, SCF
-TRCP and SCFFBW7. For these widely studied pathways, it is clear that a single F-box protein can have multiple substrates, and in some cases such substrates can actually act antagonistically to each other. For example, -transducin repeat-containing proteins (-TRCPs) promote the degradation of distinct proteins that either activate proliferation and survival of mammalian cells (such as -catenin101, 102), or inhibit such proliferation and survival (such as inhibitor of NF-
B (IkB)103). Clearly, more research is needed to understand which proteins are targeted by specific SCF E3s and, consequently, to determine which SCF complexes deserve more attention from a therapeutic standpoint.
The majority of F-box proteins interact with their targets in a manner that depends on post-translational modification of the target, most commonly phosphorylation44, 86. F-box proteins contain C-terminal protein-interaction domains that interact specifically with modified targets. These interaction domains include WD40 domains and leucine-rich repeats (LRRs) which are thought, in many cases, to represent phosphopeptide-interaction motifs. Indeed, WD40 motifs derived from F-box proteins have been shown to interact directly with small phosphopeptides referred to as phosphodegrons104, 105, 106. Within each class of WD40 or LRRs, the specificity for phosphopeptides differ, so these domains can be thought of as scaffolds that are used during evolution to adapt new phosphopeptide-binding surfaces with distinct specificities. Because the interaction surface of substrate and the F-box protein provides the most specific point for therapeutic intervention, understanding how substrates and F-box proteins interact has been an active area of research.
SCFSKP2. SKP2 is an F-box protein that specializes in the degradation of several negative cell-cycle regulators, such as the cyclin-dependent kinase inhibitor p27 (Refs 22, 26). p27 is phosphorylated on threonine-187 by cyclin-dependent kinases (CDKs) to generate a phosphodegron that is recognized by SCFSKP2. Interestingly, SKP2-knockout mice are viable but have significantly decreased body mass25 (which is a reverse phenocopy of p27 loss104, 105, 106), and the phenotype seen in SKP2-/- hepatocytes is reversed by p27 disruption24. SKP2-/- cell lines show a reduced growth rate and a tendency to undergo endoreduplication of their genetic material in concordance with stabilization of p27 (Ref. 25). Interestingly, the small CDK-interacting protein CKS1 is a subunit of the SCFSKP2 complex. CKS1, like SKP2, is required for p27 degradation, and CKS1 seems to be important for the interaction between SKP2 and p27 (Ref. 60). Mice and cells lacking CKS1 have phenotypes quite similar to those produced by loss of SKP2. These findings implicate a crucial role of this CDK-binding protein in the SCFSKP2-dependent pathways.
SKP2 has also been implicated in ubiquitin-dependent degradation of other cell-cycle regulatory molecules, such as a retinoblastoma-family protein p130 (Refs 107, 108), the CDK inhibitors p21 (Ref. 109) and p57 (Ref. 110), and the cell-cycle inhibitory transcription factor forkhead box protein O (FoxO)111, 112. As expected from its role in the destruction of antiproliferatory molecules, SKP2 is a putative proto-oncogene113. Indeed, SKP2 is overexpressed in many human cancer types36. Forced overexpression of SKP2 in mouse prostate gland114 and T-lymphocyte progenitors115 stimulates tumour formation, confirming the oncogenic potential of this F-box protein. In all cancers analysed to date, overexpression of SKP2 correlates with loss of p27 expression. SKP2 therefore seems to be a valid therapeutic target and several pharmaceutical companies have active drug discovery programmes targeting this ubiquitin ligase.
Downregulating SKP2 activity might prove useful in the therapy of cancers that gain a malignant phenotype resulting, in part, from SKP2 overexpression (Fig. 7a). On the ex vivo level, it has been shown that inhibiting SKP2 by RNA interference116, 117, 118 or intracellular injection of anti-SKP2 antibodies slows down proliferation of cancer cell lines. These findings have been confirmed in vivo in a tumour xenograft model116, 117, 119, further reinforcing the possibility that SKP2 is a potential pharmacological target.
Figure 7 | Recognition of phosphorylated substrates by SCF ubiquitin ligases.
a | SKP2 targets cell-cycle inhibitors (p27 and p130) for ubiquitin-dependent degradation and therefore can work as an oncogene when hyperactive. b | Structure of the SCFSKP2 complex bound to the coactivator protein CKS1 and a phosphodegron from the p27 protein, which is targeted for destruction by the SCFSKP2 complex. SKP1 is shown in yellow, SKP2 in magenta, CKS1 in pink, and p27 phosphodegron in green. The position of phosphothreonine 187 in the p27 phosphodegron is shown in red (stick model). The phospho-T187 moiety interacts predominantly with a basic pocket in the CKS1 protein. CKS1, in turn, interacts with the concave surface of the leucine-rich repeats in the carboxy-terminus of SKP2. c | Sequences of doubly phosphorylated phosphodegrons found in targets of the SCF-TRCP ubiquitin ligase. d | Structure of the -catenin phosphodegron bound to the WD40 repeats of -TRCP123. The surface of -TRCP is displayed in an electrostatic view with basic surfaces in blue and acidic surfaces in red. The -catenin phosphodegron (orange) is bound to a highly basic surface of -TRCP and the phosphate groups (green and red) interact directly with R285 and R431. -TRCP, -transducin repeat-containing protein; CDK, cyclin-dependent kinase; CKS, CDC kinase subunit; SCF, SKP1–Cullin–F-box; SKP, S-phase kinase-associated protein.
The recent crystal structure of the SKP1–SKP2–CKS1–phospho-p27 peptide complex120 provides insight into structural features in the complex that might be exploited for the development of inhibitors of p27 degradation. SKP2 contains an N-terminal F-box motif (three-helix bundle) that interacts with an antiparallel helical cluster in SKP1 (Fig. 7b). SKP2 also has C-terminal leucine-rich repeats, which function in CKS1 and substrate recognition (Fig. 7b). Major contacts between phosphorylated T187 in the p27 peptide and the SCF complex occur via CKS1, which binds, in turn, to the concave surface of the SKP2 leucine-rich repeats. Multiple interactions maintain the association of CKS1 with SKP2, and the p27 peptide with CKS1. There are three potential targets for small molecules within the SCFSKP2/CKS1 complexes: the SKP1–[F-box]SKP2 interaction surface58; the cluster of CKS1 residues involved in substrate recognition; and the cluster of SKP2 residues that interact with CKS1. Although disabling the SKP1–SKP2 interaction might be possible in theory, sequence conservation within the F-box might limit specificity. The interaction of the F-box with SKP1 involves interactions between multiple helical bundles, and many of the key conserved residues in the F-box are hydrophobic and interact with a hydrophobic surface in SKP1. It is unclear whether molecules could be designed that would specifically bind to one F-box but not the dozens of other F-box proteins in the cell. Perhaps it is more likely that molecules that disrupt the interaction of either p27 or CKS1 with SKP2 would be identified. However, the interaction between CKS1 and SKP2 occurs over a relatively flat surface and involves 1,200 Å2 of buried surface area. The identification of molecules blocking the SKP2–CKS1 interaction could therefore be challenging. Nevertheless, the biological rationale for developing an SCFSKP2 inhibitor is clear, making attempts to drug this unconventional target potentially worthwhile.
SCF
-TRCP. -transducin repeat-containing proteins 1 and 2 (-TRCP1 and -TRCP2) are members of the F-box/WD40 subfamily of F-box proteins and are arguably the best-understood mammalian F-box proteins. -TRCP1 and 2 are the products of distinct genes but are 85% identical with each other and are thought to be largely redundant in function43. Initial interest in SCF-TRCP as a drug target came from its linkage with the nuclear factor-B (NF-B) pathway, which is crucial to both cellular survival functions as well as the response of cells to inflammatory agents. NF-B is held in an inactive form within the cytoplasm through association with IB. In response to cytokines and other extracellular signals, the IB kinase complex phosphorylates IB, thereby promoting its degradation through the UPS; this allows re-localization of NF-B into the nucleus, where it activates the expression of genes important for cytokine and survival responses. The identification of SCF-TRCP as the E3 for IB60, 103, 121 suggested that it might be a target for molecules that act as anti-inflammatory agents by blocking IB degradation.
Early studies indicated that the interaction of IB with -TRCP occurred directly through a phosphodegron generated on IB upon phosphorylation of S32 and S36 by IB kinase, and this interaction involved C-terminal WD40 repeats in -TRCP102, 121. This phosphodegron conforms to a canonical phosphodegron sequence recognized by -TRCP (DpSG
XpS, where is a hydrophobic residue) (Fig. 7c).
The potential use of -TRCP as a target for anti-inflammatory agents is complicated by the diverse biological settings in which it functions. It is now clear that -TRCP promotes the ubiquitylation of numerous other proteins, including the oncogenic transcription factor -catenin, the cell-cycle regulatory proteins Early mitotic inhibitor 1 (EMI1) and cell division cycle protein 25A (CDC25A), and the progesterone receptor60. Each of these proteins contains a phosphodegron that is quite similar to that found in IB (Fig. 7c). Previous studies have demonstrated that an inability to destroy -catenin can promote the transcription of genes such as c-MYC and cyclin D1 that lead to cellular transformation (reviewed in Ref. 60). Indeed, mutations in the phosphodegron of -catenin have been identified in various types of human cancer and these proteins are highly stable, leading to inappropriate transcription of -catenin target genes. So the question arises of how -TRCP recognizes its substrates and whether it is possible to identify small molecules that would inhibit ubiquitylation of IB but not -catenin. Given the conservation of the phosphodegron sequence in various -TRCP substrates, it seems likely that the binding site for -TRCP will be similar for different substrates. Indeed, previous work has revealed the structure of -TRCP bound to the phosphodegron from -catenin102 (Fig. 7d). Both phosphoserine residues in the phosphodegron, as well as the conserved aspartate residue at position 1, make crucial contacts with a basic surface of the WD40 repeats; furthermore, the hydrophobic residue at position 4 in the phosphodegron (Fig. 7c) is buried in a hydrophobic pocket at the central cavity of the WD40 repeat. The residue at position 5 in the phosphodegron (Fig. 7c) is the most variable, reflecting the fact that it is directed away from the -TRCP surface. Consistent with an overlapping binding site, mutations in key residues in -TRCP that contact the -catenin phosphodegron block the ability of SCF-TRCP to promote IB ubiquitylation in vitro102. Given the architecture of the -TRCP–-catenin phosphodegron complex and the structural relationship between phosphodegrons in diverse -TRCP substrates, the development of inhibitors that selectively affect one substrate and not one of several others is likely to be a challenge.
SCFFBW7. Like -TRCP, FBW7 is a WD40-containing F-box protein that functions within the SCF complex to promote the degradation of proteins primarily involved in cell proliferation, including oncogenes. However, unlike -TRCP, FBW7 seems to be quite susceptible to mutation during transformation. The substrates of FBW7 include cyclin E, c-MYC, c-JUN, Notch and sterol regulatory element-binding proteins (SREBPs)99. Each of these proteins contains a doubly phosphorylated phosphodegron with the consensus sequence pTPXXpS, which is thought to bind to the surface of FBW7 in a manner quite similar to that seen with the -TRCP–-catenin complex. This interface is central to FBW7 function and is quite revealing in what it tells us about transformation mechanisms. First, arginine residues in the FBW7 WD40 repeats make up the bulk of the interaction surface with the phosphodegron, based on structural analysis of the yeast orthologue of FBW7, CDC4, the structure of which has been solved bound to the cyclin E phosphodegron (reviewed in Refs 60, 99). Importantly, the arginines that make up this binding pocket are targeted for mutation in several types of human cancers5, 122, providing genetic evidence of the crucial nature of this binding site in promoting the degradation of these oncogenic proteins99. Likewise, several targets of FBW7 have been shown to be mutated within the phosphodegron that is recognized by FBW7. These mutations either abolish the capacity of these sequences to be phosphorylated or block the capacity of the phosphorylated proteins to be recognized by FBW7. For example, the most frequent mutation of c-MYC in cancer is seen in the T58 residue, which serves as the first phosphoacceptor residue in the c-MYC phosphodegron that is bound by FBW7 (reviewed in Ref. 99). Recent experiments in mice indicate that mutation of T58 in c-MYC leads to increased stabilization of c-MYC, as expected based on its resistance to destruction by FBW7, but also acquires new activity which reduces its apoptotic tendencies123. This mutation in c-MYC therefore becomes highly transforming. Similarly, the v-JUN mutation represents mutations in the viral JUN protein that block its capacity to be phosphorylated in the degron, leading to stabilization and higher transforming activity30.
As with parkin, the loss of FBW7 seems to be detrimental to cells, leading to genomic instability122. Despite our understanding of the biochemical requirements for its recognition of substrates and its role in the degradation of oncoproteins, it will remain a challenge to develop approaches that reactivate mutant alleles. Given that many oncoproteins, including c-MYC, accumulate in cells containing FBW7 mutations, it seems plausible that drugs that are synthetically lethal with loss of FBW7 might be identified. However, because FBW7 mutations are relatively rare, it is not clear how much of an impact such molecules would have in the clinic.
CUL2–VHL. The key role of angiogenesis in malignant tumour growth has been firmly established, and it is reasonable to expect that an anti-angiogenic protein such as VHL would have tumour-suppressor activity. Indeed, the VHL ubiquitin ligase, which inhibits angiogenesis under normoxic conditions, is mutated in a familial cancer susceptibility syndrome, von Hippel-Lindau syndrome. Moreover, somatic mutations of VHL genes are found in a fraction of some human cancers, such as sporadic clear-cell renal carcinomas124. It is thought that loss of VHL activity forces normoxic cells to behave as if they were exposed to hypoxia, and the burst of vascular endothelial growth factor (VEGF) synthesis resulting from VHL mutations is thought to stimulate the formation of new blood vessels. Not surprisingly, an abnormally dense meshwork of capillaries are a feature of tumours found in the VHL patients, and sporadic renal-cell carcinomas are highly vascular as well.
The VHL protein is a specificity module of the CUL2-based SCF complex, and is analogous to F-box proteins in its function. CUL2-based SCF complexes are composed of CUL2; two adaptors (elongin B and elongin C) that recruit the specificity module to the rigid CUL2 core; and the VHL subunit125 (Fig. 4d).
The transcription factor hypoxia-inducible factor-1 (HIF1) is the best-known target of VHL. The physical interaction between these proteins is oxygen-dependent125. Under normoxic conditions, proline-564 in HIF1 is oxidized by oxygen-dependent prolyl-4 hydroxylases. This allows the VHL complex to recognize and polyubiquitylate HIF1125,126. However, hypoxia decreases the capacity of prolyl hydroxylases to hydroxylate HIF1, and this in turn increases the half-life of HIF1 by disrupting its interaction with VHL, which promotes expression of hypoxia-inducible genes such as VEGF124.
The structures of VHL in complex with elongin B/elongin C heterodimer and a 20-amino-acid-long, hydroxyproline-containing HIF1 peptide have been solved127, 128. The hydroxyproline 564 (Hyp564) residue is buried in the substrate-binding pocket of VHL, which explains the requirement of this modification for HIF1 ubiquitylation. Interestingly, the tumour-associated mutations in VHL are precisely clustered around the Hyp564-binding site. These findings suggest that VHL mutations might render the E3 ligase inactive against hydroxylated HIF1, which leads to activation of hypoxia-responsive genes in the presence of oxygen.
In principle, small-molecule activators of mutant VHL might be able to prevent tumour angiogenesis, but development of such molecules has not yet been attempted, and it is unclear precisely how such molecules might be identified. It is more likely that secondary pathways that are crucial for the survival of cells carrying VHL mutations, which might be exploited by compounds that are synthetically lethal with VHL mutations, will be pursued. VHL-dependent protein degradation relies on additional signalling networks, which might affect drug discovery. For instance, it was found that HSP90 prevents some VHL client proteins from ubiquitylation and degradation129. Consistent with this finding, an HSP90 inhibitor can stimulate proteasomal degradation of HIF1 in a VHL-independent manner and is currently being evaluated in clinical trials of clear-cell renal carcinoma patients130.
The U-box as a divergent RING ubiquitin ligase. U-box domain-containing proteins represent the third major class of ubiquitin ligases57, although the hypothesis that all U-box-containing proteins are bona fide ubiquitin ligases is controversial131. The U-box motif is structurally similar to the RING-finger domain and, like RING-domains, the U-box binds to E2s. However, the U-box does not bind metal ions; instead, the spatial fold of the U-box is held in place by a network of intra-molecular hydrogen bonds132. The family of human U-box proteins is much smaller than other ubiquitin ligase types; so far, only about a dozen members of this family have been identified in humans.
The U-box family of proteins has been identified relatively recently131 and therefore information regarding the possible role of these E3s in human disease is sparse133, although a CHIP U-box ubiquitin ligase was proposed to degrade the cancer-related protein ERBB2 (Refs 134, 135) and the cystic fibrosis transmembrane conductance regulator136 (CFTR), and to interact with parkin, possibly to stimulate parkin's activity137. From a structural point of view, targeting the U-box for inhibition has the same limitations as targeting RING-finger domains, in that structural features useful for inhibitor binding and specificity are generally not present.
Overview of HECT-domain ubiquitin ligases. The HECT (homologous to E6-AP COOH-terminus138) domain was first recognized in a p53 ubiquitin ligase, E6-AP, which associates with p53. Since this time, approximately three dozen human HECT-domain-containing proteins have been identified in the human genome, yet very little is known about this family of genes. Indeed, to date only a small number of these genes have been linked with human disease59, 139, 140, 141, 142. Again, the majority of mutations are loss-of-function mutations, which limits their utility as drug targets. Structurally, a HECT domain is bipartite59: its N terminus has an -helical/globular conformation and contains the catalytic cysteine, and its C terminus is more elongated and interfaces with E2 (Fig. 8b). The N and C termini are connected via a short, flexible linker143. Interestingly, the interface between E6-AP and its E2 enzyme is very similar to the interface between c-CBL (a RING-finger E3) and its E2 (Fig. 8a). Such a degree of structural conservation implies that both HECT-domain ubiquitin ligases and RING-finger ubiquitin ligases might utilize analogous mechanisms of interaction with their ubiquitin-conjugating enzymes. Molecules that interact with the surface of E2s would therefore be expected to block a large number of different ubiquitylation pathways. By contrast, the actual surfaces of HECT and RING domains, while similar in shape, differ significantly in their makeup (Fig. 8). Again, however, these surfaces are quite shallow and are not well adapted for targeting by small molecules.
Figure 8 | Structure of HECT-domain ubiquitin ligases.
a | The E2 UBC7 (red) uses the same structural motif to interact with two different classes of E3s: the CBL RING-finger (green) and the E6-AP HECT domain (blue). b | Comparison of the structures of the HECT domains of WWP1 and E6-AP indicate a possible conformational change important for ubiquitin transfer. An unstructured hinge region (red) links the catalytic domain (blue) containing the active site cysteine (yellow) to the UBC-binding domain (green). UBC7 bound to E6-AP is shown in pink with its catalytic cysteine in yellow143. HECT, homologous to E6-AP COOH-terminus; UBC, ubiquitin C.
E6-AP. E6-AP is a HECT domain-containing p53 ubiquitin ligase144 that was discovered as a protein associated with the E6 human papillomavirus (HPV) oncoprotein145. E6-AP targets p53 for degradation in HPV-E6-expressing cells but not in uninfected cells146. Several other targets of E6-AP have been identified, including the proto-oncogene c-MYC147 and two novel regulators of telomerase (hTERT) transcription148, but their contribution to E6-AP-associated pathologies remains elusive.
Interestingly, mutations of E6-AP have been implicated in Angelman's syndrome (also known as 'happy puppet' syndrome)142, 149. Patients suffering from this disorder exhibit a variety of symptoms, including mental retardation, neurodegeneration and puppet-like movements150. It is not yet known whether accumulation of any specific E6-AP substrates underlies the pathogenesis of the disease, but the clustering of Angelman's syndrome-associated point mutations around the E6-AP active site59, together with the finding that these mutations compromise enzymatic activity128, is consistent with this possibility.
The role of E6-AP in HPV-induced cervical carcinoma is well understood151. Certain sexually transmitted HPV serotypes, such as HPV-16 and HPV-18, are highly oncogenic due to degradation of p53 and contribute to malignant transformation in uterine cervical epithelial cells. E6-AP and HPV itself are therefore considered as pharmacological targets for cervical cancer. It was demonstrated that knocking down E6-AP expression by in vitro-selected ribozymes in uterine epithelium-derived, HPV-positive HeLa cells induces apoptosis upon exposure to genotoxic compounds152. Additionally, pharmacological targeting of the viral oncoprotein E6 with zinc-ejecting compounds inactivates HPV E6 and therefore silences aberrant E6-AP activity. Zinc-ejecting compounds are able to penetrate HPV-infected cells and specifically remove zinc atoms from the E6 protein, thereby inactivating E6 by disrupting its functionally crucial zinc fingers. Such small-molecule zinc ejectors could potentially be applicable for clinical use153 — in the topical treatment of cervical carcinomas and cervical papillomas, for instance — if their in vivo safety and bioavailability are more thoroughly addressed. Nevertheless, any E6-AP inhibitors identified for use against HPV infection will need to be evaluated for their effects on normal E6-AP function in other cells, including neurons, where E6-AP inhibition could be detrimental.
In principle, inhibitors of the HECT domain of E6 might prove useful in the treatment of cervical cancer. Although it is hoped that the introduction of HPV vaccines154, 155, 156 into the clinic will significantly decrease the number of patients suffering from this disease, it is currently unclear whether these immunization strategies will provide benefit to persons previously exposed to HPV, which indicates an ongoing need to develop parallel therapeutic modalities against cervical cancer. Such HECT domain inhibitors might take advantage of the extra step involved in HECT-dependent protein ubiquitylation that is not required for other classes of E3s — namely, transfer of ubiquitin to a cysteine residue in the HECT protein itself, followed by transfer to the substrate. Because this step requires a large conformational change in the HECT domain (Fig. 8b), it might be possible to identify molecules that block this transition or stabilize the HECT domain in a non-functional conformation. Clearly, identification of other HECT-domain-containing proteins whose gain of function is important for human disease will boost efforts to discover pharmacological inhibitors of HECT domains.
Artificial recruitment of substrates to ubiquitin ligases. The generation of Protacs (protein-targeting chimeric molecules) is based on the concept of generating artificial ubiquitin ligases for chosen target proteins. Protacs are heterobifunctional molecules that consist of a ubiquitin-ligase-targeting subdomain linked to a target protein-binding site157. Such artificial molecules have been used to artificially recruit an SCF complex and degrade specific target proteins (including breast- and prostate cancer-related steroid hormone receptors158). These factors — if their bioavailability is improved — could theoretically be used to fine-tune the expression of specific proteins in cancers and other diseases that result from the accumulation of abnormal proteasome substrates. Similarly, F-box-based SCF target receptors with novel specificities have been developed. Such artificial F-boxes can ubiquitylate proteins that are not normally targeted by the SCF, which provides a means by which to induce the degradation of a particular protein for therapeutic purposes159. A major issue with this approach is that the novel substrate receptor is likely to illicit an unwanted immune response, therefore limiting its application to human diseases. However, the further development of this approach might facilitate the degradation of proteins for specific research purposes.
Targeting UPS downstream of ubiquitin ligases
Small-molecule proteasome inhibitors. Bortezomib, a small-molecule inhibitor of the 20S proteasome particle, has been reviewed in detail elsewhere31, 32. Importantly (and surprisingly, given the pleiotropic role of proteasomes in a cell), bortezomib showed selective cytotoxicity to cancer cells compared with normal cells in both in vitro and in vivo assays. It is currently the only proteasome inhibitor approved for the clinical treatment of human cancer: a large multicentre single-arm trial of bortezomib in relapsed, refractory multiple myeloma patients (the SUMMIT trial34), formed the basis for the accelerated approval of bortezomib by the FDA in May 2003 and the EMEA in April 2004. Bortezomib more recently gained full approval from the FDA in multiple myeloma patients who have received one prior therapy35 and continues to be studied in a variety of haematological malignancies and solid tumours, including non-Hodgkin's lymphoma, prostate, breast and non-small-cell lung cancers. One challenge for the future will be the development of additional proteasome inhibitors, possibly displaying either better potency or fewer side effects.
Metalloisopeptidase targets in the ubiquitin–proteasome cascade. A highly diverse family of mammalian metalloproteinases (MMPs) controls multiple physiological and pathological processes, including inflammatory-associated tissue remodelling and production of biologically active peptides from precursor proteins. More recently, a role for a small sub-family of MMPs in the ubiquitin–proteasome pathway has been uncovered. The MMP-based regulation of the ubiquitylation cascade occurs on two levels. First, the metalloisopeptidase RPN11, a subunit of the proteasome lid, de-ubiquitylates UPS targets immediately prior to their destruction in the proteasome160. The de-ubiquitylase activity of RPN11 is necessary for the maintenance of proteasome activity160. For some time, it was assumed that proteasomal targets are de-ubiquitylated merely to prevent futile degradation of 'recyclable' ubiquitin residues. However, it seems plausible that long polyubiquitin chain attached to a UPS target might choke the proteasome channel due to steric hindrance. If this hypothesis is true, arresting RPN11-mediated de-ubiquitylation would inhibit degradation of perhaps all polyubiquitylated proteins due to obstruction of available proteasomes by degradation-resistant polyubiquitin chains. Therefore, RPN11 inhibitors could serve as a novel subclass of general antiproteasome chemotherapeutics.
Second, a metalloisopeptidase component of the COP9–signalosome complex functions to control the activity status of cullin-based E3s. All known cullin proteins are modified on a single lysine residue with the ubiquitin-like protein NEDD8 through an isopeptide linkage with the C-terminal glycine of NEDD8. This modification is thought to dramatically enhance the activity of SCF complexes, probably by facilitating binding of activated E2 ubiquitin-conjugating enzymes (reviewed in Ref. 60). Cullin neddylation is a highly dynamic process, and cycles of neddylation and deneddylation are required in vivo for proper control of SCF function. The CSN5 subunit of the COP9–signalosome complex is the metalloisopeptidase responsible for removing NEDD8 from cullins to facilitate this process.
In the absence of neddylation and association with substrate receptors, cullin–RBX1 complexes are held in an inactive form by cullin-associated and neddylation-dissociated protein 1(CAND1). CAND1 associates with multiple structural components of the cullin, as exemplified by the structure of the CAND1–CUL1–RBX1 complex161, and this association is anticipated to block both neddylation and association with SKP1–F-box protein complexes. In response to an unknown signal, CAND1 is removed, allowing assembly of Cullins with their substrate receptors and allowing attack by the neddylation machinery (UBC12 and APP-BP1/UBA3) to produce an active cullin-based E3. Ultimately, these active cullin complexes are thought to be disassembled to allow the cullin–RBX1 complex to be recycled, an event that is thought to be crucial for proper in vivo function. This disassembly step is performed in large part by the signalosome complex, with removal of NEDD8 by the CSN5 subunit being a central step.
As CSN5 does not seem to regulate the activity of non-cullin-based ubiquitin ligases161, 162, silencing CSN5 activity via small molecules could provide a means by which to downregulate the UPS network in a more specific manner than general proteasome inhibition. Such drugs could fill the gap between more general inhibitors of proteasome activity, such as bortezomib, and inhibitors of specific ubiquitin ligases. However, because CSN5 exists as a subunit of the COP9–signalosome complex163, 164, which has additional functions, it is conceivable that inhibition of CSN5 could affect other processes. At this point, the only known enzymatic role of CSN5 within the signalosome is deneddylation of cullins163, 164. Clearly, a better understanding of the signalosome is needed to determine whether any subunits other than CSN5 deserve attention as potential drug targets.
Importantly, RPN11 and CSN5 are closely related to each other. The isopeptidase activity of both these enzymes depends on a conserved, metalloprotease-related sequence (EXnHS/THX7SXXD) referred to as a JAMM motif165 and is sensitive to metal chelators. Proteasomes containing a JAMM-deficient RPN11 subunit fail to de-ubiquitylate and degrade their substrates160. Similarly, point mutations of the JAMM motif of CSN5 disrupt CSN-dependent deneddylation of Cul1 in Schizosaccharomyces pombe and its orthologue, Cdc53, in Saccharomyces cerevisiae166. JAMM motifs therefore represent novel targets for pharmacological antiproteasomal manipulation. However, several other JAMM-motif-containing proteins can be found in the human genome, and precisely how similar the active sites of these enzymes are to CSN5 and RPN11 remains to be determined. The functions of these JAMM-motif-containing proteins are unknown and therefore specificity will be an issue. Nevertheless, previous extensive work in the development of inhibitors of therapeutically important metalloproteases will provide an extensive history upon which to initiate the search for JAMM-motif inhibitors.
Modulators of polyubiquitin chain recognition. Recognition of polyubiquitylated targets by a 19S proteasomal regulatory complex represents a newly discovered approach for inhibition of UPS. In the vast majority of cases, binding of ubiquitin chains to the proteasome is an obligatory step in target protein degradation167. This step would ordinarily be regarded as therapeutically unattractive because there is no discrete enzymatic activity related to polyubiquitin recognition by the 19S complex, and the anticipation is that the interaction between ubiquitin and the 19S complex would present a large surface area that would not be amenable to small-molecule development.
Although this might still be true, recent innovative work from Verma et al.168 has demonstrated that large-scale, unbiased functional screens can provided novel insight into such problems by revealing unexpected biochemical mechanisms with therapeutic potential. Their biochemical screen of more than 100,000 compounds for inhibitors of cyclin B ubiquitin-dependent degradation resulted in identification of relatively simple small molecules, named ubistatins168. In an in vitro system, ubistatins inactivated turnover of both cyclin B and SIC1, which are substrates of two distinct E3s (APC and SCFCDC4, respectively), but the destruction of ornithine decarboxylase (ODC) was unaffected168. Since ODC is one of only a few substrates that can be degraded by proteasomes in an ubiquitin-independent manner, this finding suggested that ubistatins are not proteasome inhibitors but target another component of the UPS. Indeed, it was shown that ubistatins specifically bind interfaces between K48-linked ubiquitin molecules168. This leads to the hypothesis that ubistatin binding results in a change of the multi-ubiquitin chain conformation towards a conformation that cannot be bound by polyubiquitin receptors of the proteasome or that physically blocks recognition.
One pharmacological disadvantage of ubistatins is that these compounds have strong negative charges and therefore are not cell-permeable168. Although they worked in an ex vivo system, they had to be microinjected into cells168, which clearly precludes their administration to laboratory animals. Nevertheless, ubistatins could become a starting point from which to develop other small-molecule polyubiquitin chain modifiers with better bioavailability features. Minimally, these results provide evidence of a new therapeutically attractive step in the ubiquitin–proteasome pathway.
Concluding remarks
The UPS represents a major system controlling many cellular processes. It is composed of five major steps: activation; conjugation; ubiquitin-conjugate recognition; ubiquitin removal; and substrate degradation by the proteasome. Because the greatest amount of specificity is present in the conjugation step, it is anticipated that drugs targeting particular E3s, possibly by blocking substrate binding, are likely to provide the greatest level of selectivity. However, because E3s are unconventional enzymes, the development of specific inhibitors represents a significant challenge. Additional structural biology, coupled with a better understanding of the roles of the large number of E3s that exist in the human genome, will no doubt aid in this effort. Perhaps the most therapeutically attractive points in the ubiquitin cascade are the ubiquitin-activation step and the final step represented by proteasomal degradation. Both of these steps involve what might be considered more classical drug targets: enzymes that use ATP in the ubiquitin-activation step and proteolytic enzymes functioning in the degradation step. Somewhat problematic is that each of these steps are used by hundreds of different processes in the cell. Inhibition of either of these steps is therefore likely to interrupt many diverse pathways. However, the finding that cancer cells are more sensitive to defects in protein degradation and that inhibitors of the proteasome are useful in certain types of cancer therapy suggests that further emphasis on this pathway could provide new therapeutic strategies to attack proliferative disorders.
