Drugging the undruggables: exploring the ubiquitin system for drug development

SCF^SKP2

The F-box protein SKP2 forms a complex with CUL1, SKP1, and a RING finger protein RBX1, together termed SCF^SKP234. SKP2 was first identified as a critical cell cycle regulator because it ubiquitinates several important cell cycle regulators, including p27^KIP1 and p21^CIP1, both are critical CDK inhibitors^35,36,37. SKP2 also plays a critical role in EGFR-mediated AKT ubiquitination and membrane recruitment³⁸. The oncogenic potential of SKP2 was suggested by its overexpression in a variety of human cancers^39,40. Importantly, this overexpression of SKP2 showed an inverse relationship with p27^KIP141,42. Furthermore, the protein levels of SKP2 could serve as a prognostic biomarker, with higher levels predicting poor patient survival^38,41,43.

Given the importance of SKP2 in regulating degradation of tumor suppressors and its clear oncogenic potential, inhibiting SKP2 may represent a unique opportunity for the treatment of different types of tumors. Unfortunately, unlike kinases, SCF^SKP2 is a large multi-subunit complex, and does not possess any obvious cavity for _targeting by small molecules. However, the success of developing GDC-199 (venetoclax), an inhibitor for a protein that does not have enzymatic activity (BCL2) has convinced many that the time for disrupting protein-protein interactions might finally be here⁴⁴. Indeed SKP2 does have several potential protein interaction interfaces that could be explored by small molecules to disrupt interaction with either p27 or SKP1 (Figure 2A).

Pocket 3, for example, is formed jointly by SKP2 and CKS1, which is essential for p27 binding and ubiquitination by SKP2 (Figure 2A). The pocket was interrogated in a virtual screen and 96 hits were confirmed in biochemical and biophysical studies⁴⁵. These compounds selectively inhibit SKP2-p27^KIP1 interaction, and therefore block the degradation of p27^KIP1.

In another in silico screening effort, compound 25 was identified to selectively suppress SCF^SKP2, but not other SCF E3 ligase activities⁴⁶. Mechanistically, compound 25 disrupts the interaction between SKP1-SKP2 and thus abrogates SCF^SKP2 ligase activity. Although no crystal structure is available, compound 25 presumably occupies pocket 1, but not pocket 2 of SKP2, both of which are critical for SKP1-SKP2 interaction (Figure 2A). However, careful analysis of pocket 1 and compound 25 suggests that the ligand might not fully occupy the pocket. Interestingly, the available structure of SKP2 lacks the N-terminal 96 amino acids⁴⁷. A potential explanation is that the missing N-terminal segment could fold back to buttress compound 25 to ensure a snug fit in the pocket. Despite this caveat, compound 25 exhibits potent antitumor activities in multiple animal models and synergistically inhibits tumor survival with chemotherapeutic agents (Figure 2B)⁴⁶, confirming that inhibiting SCF^SKP2 is potentially beneficial for cancer patients.

MDM2

As the guardian of the genome, p53 is arguably one of the most important tumor suppressors that controls the regulation and expression of many genes that mediate cell cycle arrest, DNA repair and apoptosis⁴⁸. Under physiological conditions, newly synthesized p53 quickly undergoes ubiquitination and degradation, resulting in a “futile cycle” and a very low “steady-state” level of protein. This is largely controlled by a RING finger E3 ligase, MDM2 (murine double minute 2, HDM2 in human)⁴⁹. In addition to being a transcriptional inhibitor of p53, MDM2 also tightly interacts with p53 protein itself by recognizing the N-terminal transactivation domain (TAD), allowing p53 to undergo ubiquitination and subsequent proteasomal degradation^50,51. As a negative regulator of p53, MDM2 is overexpressed in many cancers by either gene amplification or transcriptional up-regulation⁵². Furthermore, overexpression of MDM2 has been linked to worse prognosis in different types of tumors, correlating with altered p53 protein levels, although it has not been confirmed whether these tumors have wild-type or mutant p53^53,54.

Its oncogenic potential as well as being a negative regulator of p53 warrants consideration of MDM2 as an attractive drug _target. Among all the small molecules that inhibit MDM2, Nutlins, a family of cis-imidazoline analogues identified by high-throughput screening, holds the greatest potential and is currently being tested in clinical trials (Figure 3). Importantly, Nutlin treatment showed a dose-dependent anti-proliferative and cytotoxic activity that differed between cell lines depending on their p53 status⁵⁵. Furthermore, as anticipated, Nutlin treatment induced accumulation of wild-type, but not mutant p53 protein.

In addition to Nutlins, a couple of other small molecules have also been identified to disrupt MDM2-p53 interaction. For example, similar to Nutlins, MI-219 binds MDM2 and blocks its interaction with p53, leading to induction of cell cycle arrest and selective apoptosis in tumor cells⁵⁶. Another promising molecule, RITA (reactivation of p53 and induction of tumor cell apoptosis) has been shown to prevent the interaction of p53 and MDM2 and to induce p53 accumulation in tumor cells⁵⁷. In contrast to Nutlins, RITA binds p53 but not MDM2; therefore it might inhibit many other interactions of p53 that have little to do with ubiquitination of p53.

Although the small molecules disrupting MDM2-p53 interaction hold great potential in restoring p53 function, one caveat is that they are only efficacious in tumors harboring wild-type p53, as most p53 mutants are no longer subject to ubiquitination by MDM2 and become stabilized⁵⁸. Instead, molecules aiming to restore the native conformation of p53 mutants and reactivate their tumor-suppressor function may be of more benefit to a broader spectrum of cancers. For instance: PRIMA-1 and its analog APR-246 covalently modify p53 mutants through the alkylation of thiol groups, restoring wild-type conformation and function to mutant p53⁵⁹.

Inhibitor of apoptosis proteins

Inhibitors of apoptosis proteins (IAPs) are a family of anti-apoptosis proteins that function in part by inhibiting caspases. In humans, there are at least eight IAP family members⁶⁰. All IAP proteins have one to three baculoviral IAP repeat (BIR) domains that participate in binding caspases⁶⁰. Most IAPs have a RING domain at their C-terminus that is required for ubiquitination of their substrates as well as auto-ubiquitination of some members including c-IAP1, c-IAP2 and X-linked inhibitor of apoptosis (XIAP)^61,62. IAP proteins are implicated in various cancers and attempts are being made to _target them using small molecule inhibitors or antisense oligonucleotides⁶³.

SMAC/DIABLO is a mitochondrial protein that is released to bind and inhibit IAPs during apoptosis, thereby freeing caspases to activate apoptosis^64,65. Initial effort to generate IAP antagonists was aimed at mimicking the four amino-terminal residues of mature active SMAC that binds IAP. Several monovalent IAP antagonists containing a single SMAC AVPI-like motif were shown to effectively bind the BIR domains of IAP and induce apoptosis in cancer cells⁶⁶. Further optimized bivalent antagonists consist of two SMAC-mimetic motifs connected by a chemical linker, allowing simultaneous binding to the BIR2 and BIR3 domains of IAPs, leading to an even more robust activation of caspases and apoptosis⁶⁶. Interestingly, IAP antagonists induce a conformational change and formation of IAP dimers, leading to robust auto-ligase activity, auto-ubiquitination and ultimately degradation⁶⁷, consistent with IAP inhibitors behaving as “suicidal degraders”. Currently, a couple of IAP antagonists are under investigation in several phase I trials⁶³.

“Molecular switches”

Biological decisions are executed by signaling mechanisms that utilize molecular complexes assembled from cellular constituents (e.g., proteins, oligosaccharides, metabolites, etc.). “Molecular switches” in our context refers to small molecules that are able to bind cellular constituents and induce their assembly into protein complexes with altered activity. They are particularly useful in the ubiquitination cascade since they can modulate relatively weak interactions to produce agonists as well as inhibitors of ubiquitination.

The best example of the concept of a “molecular switch” for ubiquitination is the plant hormone: auxin. Although auxin has been recognized to play a critical role in plant development, its mode of action has only become clear in the last decade. It turns out that auxin binds to an F-box component, TIR1, of the multi-subunit E3 ubiquitin ligase, SCF^TIR1, and initiates ubiquitination of key transcriptional repressors^94,95. Mechanistically, auxin enhances TIR1-substrate affinity by acting as a “molecular glue” (Figure 4A). The mechanism of auxin is not a special case. Another plant hormone, jasmonate, is sensed by the F-box protein CORONATINE INSENSITIVE 1 (COI1)⁹⁶.

In mammalian systems, equivalent molecules also play an important role to regulate protein ubiquitination. The RING finger E3 ubiquitin ligase RNF146 (also known as Iduna) is responsible for PARylation-dependent ubiquitination of both AXIN and tankyrases, and positively regulates Wnt signaling^97,98. Crystal structure suggests that when iso-ADP-ribose (iso-ADPr), the smallest internal poly (ADP-ribose; PAR) structural unit, binds RNF146 between its WWE and RING domains, it changes the conformation of RNF146 by switching the RING domain from an inactive state to an active one. Without PAR or iso-ADPr, the RING domain is unable to bind E2 conjugating enzymes efficiently⁹⁹. Therefore, PAR or iso-ADPr functions as “molecular switch” to change allosteric conformation of RNF146 and activate its ligase activity for ubiquitination and degradation of _target proteins.

Immunomodulatory drugs

Originally developed as a sedative, thalidomide was used to treat nausea and morning sickness in pregnant women before its teratogenic effect was revealed. Thalidomide and its derivative molecules (lenalidomide and pomalidomide, together are named immunomodulatory drugs or IMiDs) were later widely used for multiple myeloma patients due to their excellent efficacy, although their mechanism of action had not been clarified until recently^80,100. Taking advantage of high-performance affinity beads for thalidomide, Ito et al. showed that thalidomide binds to Cereblon (CRBN), a critical substrate receptor of the E3 ligase complex CRL4^CRBN101. Surprisingly, later studies suggested that lenalidomide-bound CRBN acquired the ability to _target for proteasomal degradation of two essential transcription factors in multiple myeloma: Ikaros family zinc finger proteins 1 and 3 (IKZF1 and IKZF3). Significantly these neo-substrates largely explained lenalidomide's therapeutic effect^102,103.

Although the E3 agonist theory satisfactorily explains the mechanisms of IMiDs in multiple myeloma patients, many outstanding questions remain to be addressed. For example, a decade of clinical trials have revealed that the seemingly antagonistic therapeutic combination of bortezomib (anti-proteolytic) and IMiDs (pro-proteolytic) confers a favorable prognosis compared with either alone^104,105. Furthermore, knockout of CRBN in zebrafish phenocopies the teratogenic effect of thalidomide, suggesting that thalidomide can block the E3 activity of CRL4^CRBN, at least for some substrates¹⁰¹. Both of these conundrums can be explained by a model where IMiDs behave as a switch to block ubiquitination of natural substrates of CRL4^CRBN, but promote ubiquitination of neo-substrates, such as IKZF1 and IKZF3 (Figure 4B). Indeed, MEIS2, an endogenous substrate binds CRBN, whereas the neo-substrates IKZF1, 3 are recruited to CRBN through the C4 amine of lenalidomide or pomalidomide, leading to their ubiquitination¹⁰⁶.

Protein-_targeting chimeric molecules

Like bi-specific antibodies, protein-_targeting chimeric molecules (PROTACs) are based on the concept of generating artificial molecules to recruit a specific ubiquitin ligase to a chosen _target protein. Therefore, PROTACs behave like bridging molecules that consist of a ligase-recruiting moiety linked through a short linker to a ligand that binds the _target protein¹⁰⁷ (Figure 5A). The first proof-of-concept PROTAC, PROTAC-1 was composed of an angiogenesis inhibitor, ovalicin, a covalent binder for the methionine aminopeptidase-2 (MetAP-2), and an IκB phospho-peptide that is recognized by the E3 ligase, SCF^β-TRCP. In the presence of PROTAC-1, MetAP-2 was recruited to SCF^β-TRCP, ubiquitinated and degraded in a PROTAC-1-dependent manner¹⁰⁷.

PROTAC technology had remained largely dormant for over a decade due to complex synthetic chemistry, as well as issues of cell permeability. In fact, the first generation of PROTAC was mainly dependent on peptides as the binding moiety for either E3 ligase or _target of interest. The development of specific ligands for E3 ligases has significantly improved the technology by providing more drug-like molecules. For example, as the chemical moiety interacting with the E3 enzyme CRBN^106,108, the phthalimide ring was used as the E3 recruiter, whereas a competitive BET bromodomain inhibitor, JQ1 was used to bind the bromodomain protein BRD4¹⁰⁹ (Figure 5B), a transcriptional co-activator that regulates the expression of genes that promote cancer cell proliferation and survival¹¹⁰. A short linker brings the phthalimide ring together with JQ1 and the hybrid molecule, named dBET1, was confirmed to interact with BRD4 in a mode similar to JQ1. dBET1 treatment induced robust BRD4 degradation, resulting in strong apoptosis in lymphoma cell lines as well as primary blasts from patients with leukemia¹¹¹. Interestingly, dBET1 showed a better efficacy compared with JQ1 in both in vitro and in vivo models, suggesting that _target degradation might be preferable to _target inhibition¹¹¹. Another PROTAC aiming to remove BRD4, ARV-825 also recruits BRD4 to the CRL4^CRBN, leading to significant degradation of BRD4 in all Burkitt's lymphoma cell lines tested¹¹². Similar to dBET1, ARV-825 showed better pharmacodynamics than small-molecule BRD4 inhibitors, with more effective suppression of c-MYC transcript levels and downstream signaling, resulting in more effective cell proliferation inhibition and apoptosis induction¹¹².

PROTAC offers an excellent opportunity to degrade many difficult oncoprotein _targets. Furthermore, recent discoveries suggest that the earlier bottleneck of PROTAC, its relatively low efficacy, is no longer an issue, with reported EC50s in the nanomolar range^111,112. In addition to CRBN, several other PROTACs have been developed employing different E3 enzymes such as CRL2^{VHL113,114,115}, suggesting that PROTAC could be developed as a platform technology. As a new technology, PROTAC also faces several important challenges. First, as complex large molecules, the major potential drawback is their likely poor oral bioavailability and fickle pharmacokinetic properties. In addition, PROTAC technology relies on identification of ligands for E3 ligases and oncoproteins, which by itself, could be a daunting task. Furthermore, diagnostics could be a challenge for many PROTAC molecules since _target E3 ligases might not necessarily co-express with _target proteins in all tissues. Furthermore, most PROTAC molecules may suffer the pharmacological “hook effect”, with higher concentrations of PROTACs preventing substrate degradation due to univalent saturation inhibiting bivalent bridging. This would make dosing of PROTACs in patients a tricky challenge. Lastly, a linker is required to connect the two chemical moieties; although both shorter and longer linkers have been reported, it still seems empirical and optimization is required for each pair of ligase and _target protein, since it is likely that optimal linker identity will vary for different proteins _targeted.

Hydrophobic tagging

Correct three-dimensional structure is essential for the function of proteins¹¹⁶. A major driving force for protein folding is to minimize the number of hydrophobic side-chains exposed to water¹¹⁷. Exposure of hydrophobic residues might cause protein misfolding and engage cellular quality control machinery to induce protein degradation¹¹⁸ (Figure 6). This machinery could be deployed to remove unwanted oncoproteins by “tagging” them with a synthetic hydrophobic ligand, thereby recruiting quality control machinery to initiate proteasomal degradation.

The concept of “hydrophobic tagging (HyT)” is similar to that of PROTAC, but instead of using a ligand to recruit a specific E3 ligase, a synthetic hydrophobic group, such as adamantane, linked to a chemical moiety that specifically recognizes the protein of interest, assumes the role of “recruiter” for the degradation machinery¹¹⁹ (Figure 6). Upon binding to the _target protein, the hydrophobic tag mimics or induces a misfolded state. Although the precise mechanism of action of HyT is not completely clear, the general consensus is that modification of _target proteins with a bulky hydrophobic side-group attracts the chaperone machinery, the primary goal of which is to help refold misfolded proteins¹²⁰. Since the covalent modification cannot be easily removed, the _target protein remains unfolded and is eventually cleared by ubiquitin-proteasome mediated degradation.

The first proof of concept study was carried out by hydrophobic tagging of the bacterial dehalogenase (HaloTag) with an adamantane moiety. The HyT molecules induced the degradation of HaloTag fusion proteins from different cellular compartments¹¹⁹. Using a similar system, ER-localized HaloTag (ERHT) protein was conditionally destabilized using a small hydrophobic tag molecule (HyT36) to reveal the mechanism of ER stress response¹²¹. Aiming to degrade a pseudo-kinase HER3, Gray and colleagues synthesized a bivalent molecule connecting hydrophobic groups with a non-functional HER3 binder. The resultant HyT molecule TX2-121-1 induced degradation of HER3 and blocked HER3-dependent signaling¹²².

An interesting question regarding HyT molecules is whether a covalent ligand for a protein of interest is essential, since most successful HyT molecules are all covalent binders. This could limit the potential of this technology. To this end, a HyT molecule linking the hydrophobic Boc₃Arg group (a modified arginine amino acid) to trimethoprim (TMP), a non-covalent inhibitor of eDHFR has been generated. The resultant TMP-Boc₃Arg HyT molecule led to 60% degradation of eDHFR in lysates and 30% in intact cells, although a high concentration was needed¹²³, suggesting that non-covalent binders might not be as effective as covalent ones.

Similar to PROTACs, hydrophobic tagging technology offers great potential to degrade disease-related proteins. The main limiting factor is probably their large size (almost invariably greater than 500 Daltons in mass), and thus likely poor pharmacokinetic properties that may limit their use to systemic administration. Compared with PROTACs, HyT molecules offer one less layer of complexity by avoiding ligands for E3 ligases. It would be interesting to test this platform by appending hydrophobic moieties (such as adamantane) with available ligands for _targets of interest, and to test their ability to degrade the _target proteins.

Selective estrogen receptor degraders

Estrogen receptors are a family of nuclear receptors activated by the hormone estrogen¹²⁴. It is estimated that around 70% of breast cancer patients have over-expressed estrogen receptors. Tamoxifen, and some other ER antagonists in breast tissue, together with aromatase inhibitors, have been widely used for ER-positive breast tumors¹²⁴. But it is almost inevitable that resistance to endocrine therapy eventually arises. Next generation sequencing has revealed that point mutations in ESR1 are drivers for resistance to the existing therapies^125,126. Furthermore, ER antagonists can have agonist properties in certain tissues, like the uterus, complicating their use. Not surprisingly, therefore, there has been considerable interest in developing selective ER downregulators (SERDs) to _target ER for degradation¹²⁷. Currently, fulvestrant, an anti-estrogen with pure antagonist activity, is the only SERD molecule approved for the treatment of breast cancer¹²⁸. However, the poor pharmaceutical properties of this injectable drug and its lack of superiority over second line aromatase inhibitors in late stage breast cancer have negatively impacted its clinical use.

Recently, a series of orally bioavailable SERDs were identified and optimized by Genentech/Seragon. The lead compound GDC-0810 showed robust activity in degrading ER-α in MCF-7 cells, and demonstrated anti-cancer activity in both tamoxifen-sensitive and tamoxifen-resistant breast cancer models¹²⁹. Importantly, in phase I/IIa studies that include 41 postmenopausal patients, all of whom had advanced ER-positive breast cancer disease that had progressed after at least 6 months of endocrine therapy, GDC-0810 competed effectively with tracer to occupy the ER at all doses tested, even in patients with ESR1 mutations, suggesting robust _target engagement (http://www.aacr.org/Newsroom/pages/News-Release-Detail.aspx?ItemID=711#.VmCWoOOfrRY). Although the mechanism of action by GDC-0810 is not completely clear at this point, it is suspected that certain E3 enzymes are activated upon ER-α engagement by the GDC compound, leading to its ubiquitination and degradation. Alternatively, SERD molecules might also behave as their SERM (Selective ER Modulators) counterparts, by changing the conformation of the α12 helix of ER, to expose hydrophobic residues that engage the quality control machinery to degrade the ER^130,131,132.