Structure Activity Relationship of USP5 Allosteric Inhibitors

USP5 is a deubiquitinase that has been implicated in a range of diseases, including cancer, but no USP5-targeting chemical probe has been reported to date. Here, we present the progression of a chemical series that occupies the C-terminal ubiquitin-binding site of a poorly characterized zinc-finger ubiquitin binding domain (ZnF-UBD) of USP5 and allosterically inhibits the catalytic activity of the enzyme. Systematic exploration of the structure-activity relationship, complemented with crystallographic characterization of the ZnF-UBD bound to multiple ligands, led to the identification of 64, which binds to the USP5 ZnF-UBD with a KD of 2.8 µM. 64 is selective over the structurally similar ZnF-UBD domain of HDAC6 and inhibits USP5 catalytic activity in vitro with an IC50 of 26 µM. This study provides a chemical and structural framework for the discovery of a chemical probe to delineate USP5 function in cells. Table of Contents Graphic


INTRODUCTION
Ubiquitin specific proteases (USP) are the largest subfamily of deubiquitinating enzymes (DUB), consisting of more than 50 proteases with distinct roles in ubiquitin (Ub) biology. USP5 (also known as isopeptidase T, IsoT) cleaves unanchored ubiquitin chains, leading to the regeneration of monoUb 1,2 and removes Ub chains to stabilize post-translationally modified proteins in cells [3][4][5][6][7][8][9][10][11][12] . Deletion of UBP14, the gene encoding a yeast USP5 orthologue, leads to the accumulation of free polyUb chains and proteasome inhibition 13 . In addition to the proper functioning of the proteasome, USP5 has been shown to be important in neuropathic pain 4,14-17 and cancer 3,7,12,[18][19][20][21][22][23][24] . A genome-scale CRISPR-Cas9 knockout screen across 804 cells lines and 28 cancer types from the BROAD Institute reveals that, with USP36, USP5 is the most essential USP 25,26 (Figure 1). Interestingly, independent studies show that knocking-down USP5 is toxic in pancreatic cancer cell lines but has no effect in non-cancerous HEK293 cells 19 and is well tolerated in mice 4 , suggesting that USP5 may be a valid therapeutic target in oncology. For each cancer type, the median CERES score across all associated cell lines was calculated. Genes are essential for cancer cell proliferation if the CERES score is lower than -1.
USP5 has multiple Ub binding modules: an N-terminal zinc finger ubiquitin-binding domain (ZnF-UBD) which recognizes the proximal free C-terminus of a broad subset of polyUb substrates 2,27-29 , a USP domain that forms the active site, and two ubiquitin-binding associated domains (UBA-1, UBA-2) 30 (Figure 2a).
The mechanism of polyUb processing by USP5 is not understood for all polyUb chain variants; site-4 directed mutagenesis experiments show USP5 processes Met1 and Lys48-linked polyUb chains sequentially from the proximal end of Ub chains 2,31 . The USP domain is highly conserved making it difficult to develop selective catalytic inhibitors; to date, out of 56 human USPs, selective inhibitors have been reported for USP1, USP7, USP9x, USP14 and USP30 [32][33][34][35][36][37][38] . The ZnF-UBD domain is present in a limited number of proteins in the human genome: USP3, USP5, USP13, USP16, USP20, USP22, USP33, USP39, USP44, USP45, USP49, USP51, HDAC6, a histone deacetylase protein, and BRAP, a BRCA1associated protein [39][40][41] . The role of the USP5 ZnF-UBD is not well understood. Hydrolysis of the artificial substrate, ubiquitin amidomethyl coumarin (Ub-AMC) by USP5 is activated when free monoUb occupies the ZnF-UBD, suggesting that this domain participates in the allosteric modulation of USP5 42 . However, in a separate study, deletion of the ZnF-UBD did not significantly affect hydrolysis of Ub-AMC 30 . It is therefore unclear whether targeting the ZnF-UBD is a valid strategy to antagonize the enzymatic activity of USP5.
We previously reported fragment-like molecules that bind to the USP5 ZnF-UBD with high micromolar affinity 43 . Here, we describe the progression of one of these compounds into a low micromolar USP5 inhibitor. We show that our ligand can displace Ub from the ZnF-UBD in the context of full-length (FL) USP5, inhibits the catalytic function of USP5 with a Lys48-linked di-ubiquitin (Ub2K48) substrate in vitro and is selective over HDAC6 ZnF-UBD. The accompanying crystal structures and structure activity relationship (SAR) provide a robust framework for the development of a chemical probe to interrogate the function of USP5 in cells and in vivo. 7 to confirm that the amide confers selectivity for USP5 over HDAC6 (Scheme 1). Both compounds exhibited comparable affinity for USP5 (KD= 34, 33 µM respectively) and selectivity over HDAC6 (11 and 16-fold respectively). Compound 4 was identified as a more favorable scaffold for developing potent and selective ligands and we next focused on further exploring the SAR around 4.  9

Structure Activity Relationship
Our initial SAR studies focused on varying substitutions to the carboxylic acid to determine if less acidic moieties could interact with Arg221 and Tyr261 (Table S4). Substituting the carboxylic acid to an ester (S25) or tetrazole (S26, S27) resulted in a complete loss of binding against USP5 indicating that the conserved network of direct and water-mediated interactions between the carboxylate and Arg221, Tyr261 and Val234 is essential for binding, as was previously observed with the corresponding binding pocket of HDAC6 49 . The addition of a (S)-methyl or dimethyl at the C2 position of the carboxylic chain and mono-N-methylation of the secondary amide were not tolerated (S28, S30, S31). A (R)-methyl at the C2 position of the carboxylic chain (S29) had weaker affinity for USP5 and improved affinity for HDAC6.
Lengthening and shortening of the carboxylic chain was also not tolerated (S32, S24).

19-25)
showed that a phenyl ring (4) or a N-methyl-piperazine (25) were preferred at this position. The piperidine substituted with a phenyl (4) is preferred over other groups for efficient positioning of the terminal phenyl for π-stacking interactions with Tyr223, as seen with the complex structure of 1 (PDB: 10 7MS5). A spiro group (27) was found to be slightly more potent and selective than 4, but chemical tractability considerations led us to further explore substitutions on the phenyl ring of 4.  Figure S3). The fluorine group of 48 makes hydrophobic interactions with Trp209, which may account for the 4-fold increase in potency for USP5 ZnF-UBD in comparison to 4 (Figure 3). However, while the added fluorine increases affinity for USP5, it also results in a four-fold decrease in selectivity against HDAC6.

17
To build upon the improvement afforded by the fluoro-phenyl of 48 and the pyridine of 51, four derivatives of these compounds were synthesized and tested (Compounds 61-64; Scheme 3; Table 5). While no apparent benefit was observed in replacing the terminal phenyl ring of 48, we were pleased to observe further improvement in potency and selectivity when chlorinating this ring in 51 (compound 64: USP5

Characterization of Compound 64
We used SPR to show that lead compound 64 binds to the full-length (FL) USP5 with a KD of 8 ± 2 µM,  Figure S7). Importantly, we observed dose-dependent inhibition of USP5 catalytic activity by 64 in an internally quenched fluorophore (IQF) assay with a native K48-linked di-ubiquitin (Ub2K48) substrate, with an IC50 of 26 ± 9 µM in vitro (Figure 5d). Taken together, these data suggest that targeting the ZnF-UBD is a valid strategy for the selective inhibition of USP5, a putative cancer target.
The potency of the compounds presented in this work needs to be further improved before inhibitors can be tested in cellular systems or animal models. To evaluate the development potential of this chemical series and support further optimization efforts, we verified that our most advanced compound was stable in mouse and human liver microsome stability assays (>90% compound remaining after 60 min, Table   S5). substrate. An IC50 of 26 ± 9 µM was obtained from the average of three independent measurements.
Fluorescence signal is normalized to control (no compound, DMSO only).

CONCLUSION
In summary, we optimized a preliminary chemical scaffold through structural analysis, docking and SAR studies to identify a small molecule antagonist targeting the USP5 ZnF-UBD with low micromolar affinity that is selective over a structurally similar ZnF-UBD in HDAC6. We show that compound 64 occupies the USP5 ZnF-UBD, can displace Ub in the context of FL USP5 and inhibits USP5 deubiquitinase activity.

22
The ZnF-UBD has been hypothesized to recognize and position substrates or allosterically modulate USP5 catalytic activity 30,31 . Our antagonists inhibit USP5-mediated cleavage of a Ub2K48 substrate showing that when the ZnF-UBD is blocked, Ub2K48 cannot be properly positioned for catalytic cleavage. USP5 was reported to cleave a diverse array of differently branched polyubiquitin substrates, and it remains to be seen whether the inhibitory activity of ZnF-UBD ligands is substrate-specific. The chemical inhibitors and accompanying crystal structures reported here will enable further optimization into a potent and selective chemical probe to investigate the cellular function of USP5.

Hit Expansion & Docking
Substructure search was run against ~1.2 billion compounds from the Enamine REAL database (Enamine Ltd., 2019) and SciFinder. Ligprep (Schrodinger, New York) was used to prepare the ligands using default settings. The X-ray structures of USP5 ZnF-UBD (PDB ID: 6DXH 43 , 7MS5) were prepared with PrepWizard 50 (Schrodinger, New York) for the assignment of bond order, assessment of correct protonation states and optimization of hydrogen-bond assignment at pH 7.3, and restrained minimization using the OPLS3e force field. Receptor grids were calculated at the centroid of the co-crystallized ligand for each respective complex. The library was docked to the USP5 ZnF-UBD structure for each respective complex using Glide SP 44 (Schrodinger, New York) with hydrogen-bond constraints with the NH of Arg221 backbone, NH of Arg221 side chain and OH of Tyr261. Docked compounds were clustered in ICM 51 and compounds were selected to be purchased based on docking pose, score, visual inspection, and cost.

Fluorescence Competition Assay
All Fluorescence was measured using a Biotek Synergy H1 microplate reader (Biotek) at emission and excitation of 528 nm and 485 nm, respectively. The data were analyzed with GraphPad Prism 8.4.2.

Surface Plasmon Resonance Assay
Studies were performed with a Biacore T200 (

Data Collection, Structure Determination and Refinement
X-ray diffraction data for USP5 171-290 co-crystals with 1, 48 were collected at 100 K on RIGAKU FR-E X-ray generator with a RIGAKU Saturn A200 CCD detector. Diffraction data for 64 was collected at APS beamline 24-ID-E. Images were processed with HKL3000 52 or Xia2 53 and scaled with AIMLESS 54 .
Graphics program COOT 55 was used for model building and visualization, REFMAC 56 for restrained refinement and MOLPROBITY 57 for analysis of model geometry.

Internally Quenched Fluorophore Assay
Experiments were performed in a total volume of 60 µL in 384-well black polypropylene microplates (Greiner

Author Contributions
The