Specific targeting of the deubiquitinase and E3 ligase families with engineered ubiquitin variants

Abstract The ubiquitin proteasome system (UPS) has garnered much attention due to its potential for the development of therapeutics. Following a successful clinical application of general proteasome inhibitors much effort has been devoted to targeting individual UPS components including E3 enzymes and deubiquitinases that control specificity of ubiquitination. Our group has developed a novel approach for targeting the UPS proteins using engineered ubiquitin variants (Ubvs). These drug‐like proteins can serve as valuable tools to study biological function of UPS components and assist in the development of small molecules for clinical use. In this review, we summarize studies of Ubvs targeting members of three major families, including deubiquitinases, HECT E3 ligases, and CRL E3 ligases. In particular, we focus on Ubv binding mechanisms, structural studies, and effects on enzyme function. Furthermore, new insights gained from the Ubvs are discussed in the context of small molecule studies.


| I N T R O D U C T I O N
Ubiquitination plays a central role in controlling the stability and function of cellular proteins. Approximately a thousand of genes in the ubiquitin (Ub) proteasome system (UPS) are involved in controlling ubiquitination in a specific and timely manner by marking proteins for degradation by the proteasome or by regulating their function. The misregulation of UPS proteins has been increasingly linked to human diseases, in particular cancer, and as a result the UPS is considered important for therapeutic development. Unlike kinases, which have been targeted by numerous small molecule drugs, 1 development of small molecules targeting UPS components has lagged behind. To date, only general proteasome inhibitors 2 and thalidomide derivatives acting on CRBN 3 have been approved for treatment of haematologic malignancies. Additionally, a few compounds targeting specific UPS components 4,5 or families 6  Therapeutic targeting of UPS components can benefit greatly from the use of intracellular drug-like protein molecules, which are easier to generate than small molecule compounds, but like small molecules can be used to explore biological outcomes of targeting a specific protein site. Previous work demonstrated that Ub is amenable to genetic engineering, where de novo binders to cell-surface receptors were generated through ribosome display by varying several positions on the Ub surface. 16,17 However, engineered Ub variants (Ubvs) are particularly suited for targeting components of the UPS. 18 This is because virtually all UPS proteins already contain weak Ub-binding sites including active and regulatory sites. [19][20][21] While targeting active sites with high affinity Ubvs is expected to antagonize function, targeting regulatory sites may have a spectrum of outcomes ranging from antagonistic to agonistic.
Here we summarize the results and insights gained through targeting of three major UPS families with Ubvs. We describe the generation of Ubv inhibitors targeting DUBs, which was the first demonstration of the technology and laid the foundation for following studies. Next, we describe Ubvs generated against a family of HECT E3 ligases, which provided the first example of Ubvs acting as activators. Finally, we describe the development of Ubvs against CRLs including the SKP1-CUL1-F-box (SCF) family and the Anaphase Promoting Complex/ Cyclosome (APC/C) complex.

| IN H IB I TORS OF DUB P ROTE ASES
The human genome encodes 116 DUBs that are subdivided into five families based on the structures of their catalytic domains, including four families of cysteine proteases and a metalloprotease family 15 (Figure 1). DUBs constitute an important class of therapeutic targets with numerous family members implicated in a variety of diseases including cancer and neurodegenerative, infectious, and blood diseases. 22 Despite extensive efforts, potent small molecules have only been developed against a small number of DUBs, and most of these inhibitors exhibit low specificity and potency. 23 Furthermore, there are no published structures of human DUBs in complex with small molecule inhibitors, and this has prevented detailed understanding of the inhibition mechanisms to guide further design.
Motivated by the paucity of effective inhibitors of DUBs, our group developed an approach to generate potent and specific Ubvs as protein-based inhibitors. 18 The Ub specific protease (USP) subfamily contains a conserved binding site for the distal Ub that is conjugated through its C-terminal moiety to lysine in other Ub or protein substrates. Based on analysis of available USP-Ub structures, we targeted for combinatorial mutagenesis approximately 30 residues on Ub that interact with the USP Ub-binding site (Figure 2A). Resulting libraries containing billions of Ubvs were displayed on phage and subjected to binding selections for particular DUBs and other Ub-associated proteins. This strategy was successful in generating tight and specific Ubvs binding to USP2, USP8, or USP21 (Table 1). Additionally, this approach also generated specific Ubvs for the: DUBs OTUB1 and JAB1, members of the ovarian tumor protease (OTU) and JAB1/MPN/MOV34 (JAMM) subfamilies, respectively; E2 conjugating enzyme Cdc34; HECT E3 ligases NEDD4 and ITCH; and the non-catalytic UBD of USP37. These results showed that the common Ub epitope that engages many Ub-associated proteins can be fine-tuned for specific targeting of particular proteins.
The structures of USP2, USP21, USP8, and OTUB1 complexes revealed that the Ubvs bound to the distal Ub-binding site and acted as inhibitors of diUb cleavage. However, the Ubv interaction mode FIG URE 1 The ubiquitin proteasome system. The cartoon shows an overview of the ubiquitination process and the major components of the UPS. In the first step of Ub conjugation, E1 consumes ATP to form a high energy thioester bond between the Ub C-terminal carboxyl group and the E1 active site cysteine thiol group. In the next step, the activated Ub is transferred to the E2 active site cysteine. In the final step, E3 mediates conjugation of the Ub C-terminal carboxyl group to an amino group of a lysine residue in a substrate protein or another Ub molecule. In this step, the Ub is either transferred through an E3 active site cysteine (HECT E3 ligases) or directly from an E2 to the receiving lysine. DUBs catalyze the cleavage of the Ub C-terminal carboxyl from substrate proteins or from Ub chains to reverse ubiquitination. The structure of Ub (PDB: 1UBQ) is shown (top right), indicating the location of the C-terminal carboxyl conjugation site and the seven acceptor lysine residues, which are shown as red sticks. The number of members in each UPS family is indicated in parenthesis and was taken from the following studies: 1, 7 2, 8 3, 9 4, 10 5, 11 6, 12 7, 13 8, 14 9 15 varied between different complexes. In the USP2-Ubv.2.3 and USP21-Ubv.21.4 complexes, the Ubv bound in an orientation that was virtually identical to that of wild-type Ub (Ub.wt) ( Figure 2B,C). Notably, these Ubvs contained only three substitutions relative to the Ub.wt, demonstrating that strengthening just a few key interactions can lead to dramatic improvements in affinity and specificity (Table 1). In the case of OTUB1, although the Ubv bound in a similar orientation, it was slightly rotated relative to Ub.wt ( Figure 2E). However, still only a small number of substitutions relative to Ub.wt were sufficient to generate high affinity and specificity (Table 1). Ub.wt is known to allosterically activate the binding of OTUB1 to UbcH5b-Ub complex 26 and the Ubv also re-capitulated this property, albeit less efficiently despite much higher affinity. This highlights the fine-tuned nature of Ub-substrate interactions, where even small changes in binding mechanism can lead to altered function. Finally, in the case of USP8, there is no structure available for the complex with Ub.wt, but the Ubv contained many mutations and bound in an orientation that was drastically different from that expected for an Ub substrate, as the tail of Ubv.8.2 pointed away from the active site cleft ( Figure 2D) 18 Thus, many substitutions in Ubvs can work together to produce drastic changes in the binding mode, resulting in high specificity and affinity.
Ubvs targeting USP8 and USP21 were validated in cellular assays and, consistent with high affinity and specificity, they co-immunoprecipitated with their cognate USPs, blocked ubiquitination of endogenous protein substrates, and modulated the activity of the pathways regulated by the USPs ( Table 1).
The high affinities of Ubvs generated against DUBs and other UPS proteins make them valuable tools to explore functional details of Ub interaction, which is otherwise difficult to investigate due to the low affinity and promiscuity native interactions. For example, we took advantage of the virtually identical binding modes between Ub.wt and  25 The Ub is shown as surface and the residues of the identified core functional epitope are labeled and colored red or yellow if they were the same or different, respectively, in Ubv.2 and Ubv.21 GORELIK AND SIDHU | 33 and Ubv.21, and these were involved in conserved interactions between the Ubvs and USPs. In contrast, the three residues that differed in Ubv.2 or Ubv.21 relative to Ub.wt clustered together and mediated different interactions with USPs that could be exploited to generate high specificity inhibitors ( Figure 2F). Similar analyses could be extended to dissect other Ub interactions with diverse members of the UPS and should prove useful for structure-guided design of specific inhibitors.
Others have also generated Ubv inhibitors of other USPs. 27,28 In contrast to our approach which targeted surface exposed Ub residues involved in USP binding, these studies randomized computationally selected residues predicted to affect the conformation of the b1-b2 loop that interacts with USPs. Phage display was used to generate Ubvs with submicromolar affinities for USP14 and reduced affinities for the UCH DUB subfamily. 27 NMR analysis demonstrated that substitutions in these Ubvs do not cause detectable changes in the b1-b2 loop conformational state, but rather, slow down its conformational motions, which highlights the importance of conformational dynamics for Ub interactions. The Ubv library designed to alter the b1-b2 loop conformation was also used to generate binders to USP7 and affinity was further improved with additional surface mutations. 28 The Ubv inhibited catalytic activity, but the structure of the Ubv-USP7 complex was not solved, preventing characterization of the binding mechanism.
These results further highlight the amenability of the Ub scaffold for generation of Ubvs targeting the DUB family.
In summary, tight and specific Ubv inhibitors have been generated against several DUBs and have been shown to be useful tools for exploring molecular details of DUB interactions and for investigating biological consequences of inhibition. While all structurally characterized Ubvs bound to the distal Ub-binding site and blocked substrate binding, it is intriguing to speculate that other Ub-binding sites on DUBs 29 may be targeted to generate modulators that alter rather than block function. The same may be true for members of other UPS families and, as described below, additional studies have focused on HECT and CRL families of E3 ligases with the goal of generating Ubvs against known and previously uncharacterized Ub binding sites.

| M O D U LA T O R S O F H E C T E L I G A S E S
Of the 600 E3 ligases encoded by the human genome, 28 belong to the extensively characterized HECT family, and these contribute to many essential cell processes and have been linked to numerous diseases. 30 HECT family members share a conserved C-terminal catalytic HECT domain that is composed of the flexibly tethered N-lobe and Clobe. The N-lobe binds to an E2 enzyme charged with Ub, while the Clobe receives the Ub transferred from the E2 enzyme ( Figure 3A). In in the case of ITCH, 33 or both in the case of WWP1 and WWP2. 34,35 In addition to the C-lobe active site that interacts with Ub, members of the NEDD4 family also contain a weak Ub-binding N-lobe exosite. 36 Several studies suggested that binding of the substrate-linked Ub to the N-lobe exosite enhances polyubiquitination by orienting the distal end of growing polyubiquitin chains [37][38][39] and by enhancing the processive ubiquitination mode. 40 Additionally, the N-lobe exosite was shown to directly overlap with the C2 binding surface on SMURF2 and NEDD4, suggesting that Ub binding to this site might also contribute to relieving C2 mediated autoinhibition. 41 To date only a handful of small molecules targeting HECT E3 ligases have been reported, including a general HECT inhibitor 42 and an ITCH inhibitor that target active sites, 43 a SMURF inhibitor that disrupts substrate binding, 44 and a NEDD4 inhibitor that binds the N-lobe exosite and inhibits processive ubiquitination. 40 While these small molecules provide some insight into different mechanisms that can modulate HECT function, the small number of available molecules, lack of structural information (a structure is only available for the NEDD4 inhibitor), and lack of potency and specificity in some cases suggest that alternative methods are required to systematically investigate the mechanisms and biological outcomes of specifically targeting HECT E3 ligases across the whole family.
We used the phage-displayed Ubv library (Figure 2A) to target HECT domains from 19 human proteins (including all members of the NEDD4 family) and from Rsp5, the yeast homologue of human NEDD4. 18,45 These selections yielded a total of 69 unique Ubvs, which displayed high affinity and specificity for their targets, although some cross-reactivity was observed for closely related homologs ( Table 2).
The Ubvs acted as inhibitors, activators or modulators of ubiquitination  Ub.wt in complex with NEDD4 or Rsp5 ( Figure 3E-H) (Table 2). Additionally, all Ubvs generated against HECT3 ligases were pooled and screened for their effects on cell migration. This screen identified several Ubvs whose targets (HACE1, SMURF2, and WWP1/2) were known to be involved in cell migration and which affected cell migration in accordance to their in vitro properties ( Table 2). The screen also recovered Ubv.NL.1 targeting NEDD4L as a strong inhibitor of cell migration, identifying NEDD4L as a novel regulator of this process (Table 2).
In summary, members of the HECT3 ligase family are amenable to targeting with Ubvs that are specific, biologically active and possess a range of effects on enzyme function ( Figure 3I). In most cases, only one type of Ubv (inhibitor, activator or modulator) was obtained against a single HECT E3 ligase. However, additional phage selections with tailored Ubv libraries and using existing Ubvs to block unwanted interactions may produce Ubvs with different action modes. This tool kit of Ubvs should prove invaluable for exploring HECT E3 ligase biology and for facilitating structure-guided design of small molecule inhibitors.

| I N H IB I T O RS O F S C F E3 L I G A S E S
The SCF family of E3 ligases is one of the largest and best characterized of the CRL families ( Figure 1). The SCF complex is composed of the invariant RBX1, CUL1, and SKP1 subunits, and one of the 69 F-box proteins that determine substrate specificity ( Figure 4A). The cullin protein CUL1 brings together the RING protein RBX1, which recruits the E2 enzyme, and the adaptor SKP1 in complex with the F-box protein, which recruits substrate. All F-box proteins are defined by the presence of a small F-box domain that interacts with SKP1. The family is further subdivided according to the nature of the substrate-binding domain including WD40, LRR and "other domains," which are referred to as the FBW, FBL, and FBO subfamilies, respectively ( Figure 4A). Numerous F-box proteins control essential cell processes including cell cycle, DNA repair and apoptosis. 57 Consequently, many F-box proteins are attractive targets for treatment of cancer and other diseases. However, only a few F-box proteins have been targeted with small molecules, with the majority of the effort devoted to SKP2, which drives cell cycle progression and is the best validated cancer target within the family ( Figure 4B). Most reported inhibitors function by disrupting the interaction between an F-box protein and its substrate. Additionally, inhibitors that disrupt the interaction of an F-box domain with SKP1 were generated for SKP2 and the yeast protein Met30 ( Figure 4B).
To gauge the potential of targeting F-box proteins with Ubvs, we used the phage-displayed Ubv library (Figure 2A) against the wellcharacterized F-box protein FBW7 in complex with SKP1. 56 Unlike DUBs and HECT E3 ligases, Ub-binding sites have not been characterized on most F-box proteins, the exception being FBW family members which interact with Ub through the WD40 substrate-binding domain that is thought to control auto-ubiquitination. 58 Unexpectedly, an Ubv (Fw7.1) generated against FBW7 in complex with SKP1 did not target its WD40 domain, but instead bound at the interface of the SKP1 protein and the F-box domain ( Figure 4C).  Table   3). The Ubvs generated against FBW7 and FBW11 were tested in cells and, consistent with their in vitro properties, disrupted the interactions of their cognate F-box targets with CUL1 and inhibited degradation of known F-box substrates (Table 3).
Ubvs generated against F-box proteins show that members of this family can be inhibited in a systematic manner by targeting the CUL1binding surface on the SKP1-F-box complex and that highly specific Ubvs can be obtained. This inhibitory site was not previously discovered by small molecule studies and offers several advantages ( Figure   4B)

| I N H IB I T O RS O F AP C /C
The APC/C complex contains at least 15 different core subunits and is the most elaborate of the CRL E3 ligases. 60 The organization of the catalytic core resembles that of the SCF E3 ligases, as it contains the cullin subunit APC2, the RING protein APC11, the adaptor protein APC10, and two interchangeable substrate binding subunits CDC20 and CDH1. APC/C plays a central role during cell cycle, where APC/ C CDC20 is responsible for driving the anaphase transition and mitotic exit, while APC/C CDH1 is mainly involved in governing transition through the G1 phase. 61 Considering the central role of APC/C in cell cycle progression, it represents an attractive target for cancer therapy especially in the case of the APC/C CDC20 complex that is required for mitotic exit. 62 To date, two small molecule inhibitors that either block CDC20 and CDH1 interaction with APC/C 63 or disrupt substrate binding to CDC20 64 have been generated, and they provide some insight into the mechanisms and outcomes of APC/C inhibition. However, given the central role of the APC/C complex in cell biology and its immense complexity, development of additional reagents would be highly beneficial for investigating APC/C function and assessing the consequences of targeting different sites.
Phage-displayed libraries ( Figure 2A) were used to generate Ubvs targeting APC11, the RING subunit of APC/C (Table 3). 65 APC11 contains an Ub-binding exosite, which presumably serves to capture substrate-linked Ub in proximity to the E2 active site and contributes to chain elongation mediated by the UBE2S E2 enzyme. 66 Analysis of the APC11-Ubv complex coupled with NMR and enzyme assays demonstrated that the Ubv binds through the same interface and targets the same surface on APC11 as Ub.wt ( Figure 4E). Accordingly, the Ubv impeded in vitro multiubiquitination mediated by the UBE2C E2 enzyme and chain elongation mediated by the UBE2S E2 enzyme in the same manner as the mutations to the APC11 Ub-binding exosite ( Figure 4F). The inhibitory effect of Ubv on APC/C function was also observed in a Xenopus egg system (  The APC11-binding Ubv provides yet another example of Ubvs mimicking the interaction of Ub.wt and demonstrates the utility of using high affinity Ubvs for structural studies. While the Ubv was used to help define the architectures of the APC/C complexes, it may also prove useful for exploring the biological consequences of inhibiting the APC11 Ub-binding exosite. Furthermore, given the complexity of the APC/C complex, it would be interesting to explore whether Ubv inhibitors or modulators can be generated against other subunits of the APC/C complex.