Discovery of small molecule antagonists of the USP5 zinc finger ubiquitin-binding domain

USP5 disassembles unanchored polyubiquitin chains to recycle free mono-ubiquitin, and is one of twelve ubiquitin-specific proteases featuring a zinc finger ubiquitin-binding domain (ZnF-UBD). This distinct structural module has been associated with substrate positioning or allosteric modulation of catalytic activity, but its cellular function remains unclear. We screened a chemical library focused on the ZnF-UBD of USP5, crystallized hits in complex with the protein, and generated a preliminary structure-activity relationship which enables the development of more potent and selective compounds. This work serves as a framework for the discovery of a chemical probe to delineate the function of USP5 ZnF-UBD in proteasomal degradation and other ubiquitin signalling pathways in health and disease.


Introduction
Ubiquitination is a reversible, post-translational modification involving the conjugation of a conserved 76 amino acid protein, ubiquitin (Ub), to substrate proteins for proteasomal targeting or the regulation of cell signaling [1][2][3] . Substrate proteins are generally ubiquitinated through an isopeptide bond between the C-terminal glycine residue of ubiquitin and the terminal nitrogen of lysine side chains [4][5][6] . Substrate proteins can be mono-or poly-ubiquitinated, where ubiquitin chains are formed through isopeptide linkages with any one of seven internal lysine surface residues or conjugation to the N-terminal amine group 7 . Deubiquitination, the removal of ubiquitin, is carried out by a family of deubiquitinase enzymes (DUBs) [8][9][10] . Ubiquitin specific proteases (USPs) are the largest sub-family of DUBs consisting of more than 50 cysteine-proteases with diverse roles in ubiquitin homeostasis 3 .
USP5 prevents accumulation of poly-ubiquitin chains which would otherwise overwhelm the proteasome by competing with ubiquitinated proteins targeted for degradation 13 . The domain architecture of USP5 consists of a zinc finger ubiquitin-binding domain (ZnF-UBD) spanning residues 173-283, a functional catalytic domain, two ubiquitin associated domains (UBA), and a cryptic zinc finger ubiquitin binding domain (nUBP) 12 . The ZnF-UBD recognizes the C-terminal di-glycine motif of ubiquitin 11 , and has been associated with allosteric modulation of USP5 activity 8,11,14 , and alternatively, with substrate recognition and positioning 12 .
In addition, USP5 plays a role in regulation of ubiquitin levels in DNA damage repair 25 and stress granules 26 . According to CRISPR-knockout screens against large panels of cancer cell lines, USP5 is essential to the survival of cells from most cancer types [27][28][29] . Inhibition of USP catalytic activity by small molecules has been reported for USP1, USP7 and USP14 [30][31][32][33][34] . Targeting the non-catalytic ZnF-UBD may be an alternative strategy to antagonize USP5 function, similar to the recent development of small molecule inhibitors against the non-catalytic domain of HDAC6 35,36 . Here, we report the structure-based discovery of the first USP5 ZnF-UBD inhibitors and their characterization by 19 F NMR spectroscopy, surface plasmon resonance (SPR) and X-ray crystallography. Experimental validation of four low-affinity ligands followed by a hit expansion campaign defined a preliminary structure activity relationship (SAR). Our work provides a framework for the development of potent USP5 ZnF-UBD inhibitors to investigate the function of this non-catalytic domain in cells.

Virtual Screen & Hit Identification
Following our observation that compounds featuring a short carboxylic acid chain can mimic the C-terminal di-glycine carboxylate of ubiquitin and favorably exploit the ZnF-UBD of HDAC6 35,36 , we compiled a virtual library of 9480 in-house and commercially available compounds to screen against USP5 ZnF-UBD. The library was docked using ICM-Pro 37 (Molsoft, CA) and Glide 38 (Schrodinger, NY) to the X-ray crystal structure conformation of USP5 ZnF-UBD (PDB: 2G45) and an alternate conformational state modelled after the structure of HDAC6 ZnF-UBD in complex with inhibitors. In this alternate "stacked" conformation, the guanidinium plane of Arg221 is positioned parallel to the phenyl plane of Tyr259 ( Figure S1), which would allow favorable π-stacking interaction with aromatic inhibitors. 96 compounds were selected from ICM 37 and Glide 38 by visual inspection of the docked binding pose and physical samples of 33 (mostly fragments) were tested experimentally. Since one of the two tryptophan side-chains of USP5 ZnF-UBD is located at the targeted ubiquitin C-terminal binding pocket (Figure 1a), we decided to use a 19 F nuclear magnetic resonance (NMR) spectroscopy assay as a primary, qualitative screening method. We first verified that USP5 ZnF-UBD labeled with 5-fluorotryptophan (5FW) produced two distinct peaks in a 19 F spectrum. Titration of 5FW-USP5 ZnF-UBD with the C-terminal ubiquitin peptide LRLRGG resulted in broadening of the downfield peak ( Figure 1b), which we consequently assigned as Trp209 in the ubiquitin-binding site. Upon screening the 33 compounds, we found that 11 caused chemical shift perturbation and/or broadening of 19 F resonance, suggesting binding at the ubiquitin binding pocket ( Figure 1c, Table   S2). We next used SPR as an orthogonal binding assay to validate primary hits ( Figure 1d, Table   1). While the LRLRGG ubiquitin peptide bound USP5 ZnF-UBD with a KD of 52 ± 2 µM, 7 of the 11 NMR hits showed weak affinities in the high micromolar range (Table 1) 19 F NMR spectra of 5FW-USP5 ZnF-UBD before (blue) and after the addition of LRLRGG ubiquitin peptide (red) (at 6-fold excess) c) 19 F NMR spectra of 5FW-USP5 ZnF-UBD before (blue) and after the addition  To confirm that the screening hits bound at the C-terminal ubiquitin-binding pocket, and to guide future optimization efforts, we attempted soaking apo USP5 ZnF-UBD crystals with ligands but were unsuccessful. Co-crystallization attempts with the most potent compounds produced crystal structures with 1 (PDB: 6DXT), 5 (PDB: 6NFT) and 7 (PDB: 6DXH) ( Figure 2). Compound 1 was not fully resolved by electron density, and modeled for only one of the two USP5 ZnF-UBD molecules in the crystallographic asymmetric unit. The carboxylate group of the ligands recapitulates interactions observed with the ubiquitin C-terminal di-glycine 11 , and is engaged in three direct or water-mediated hydrogen bonds with surrounding residues. An aromatic ring πstacks with Tyr259, but we did not observe a conformational rearrangement of Arg221 that could have further increased stacking interactions, as seen with the corresponding HDAC6 ZnF-UBD arginine residue, Arg1155. The absence of π-stacking interactions may be due to the absence of an   of 80 ± 1 µM, suggesting an i-propyl at the para position of the benzene ring is preferred over a t-butyl moiety. The preliminary structure activity relationship emerging from this work will serve as a framework for improvement of the current chemical series and exploration of novel ligand scaffolds. In particular, we believe that fragment 21 and 37 would be good starting points for future optimization efforts.

Conclusion
Here, we report a combination of computational screens, binding assays, and crystal structures for the discovery of ligands targeting the ZnF-UBD of USP5. This work will serve as a platform to optimize the chemical series presented here, and potentially identify novel scaffolds, with the aim of developing potent inhibitors that are selective over ZnF-UBDs found in others USPs and HDAC6.

Primary Virtual Screening
Ligprep  MHz spectrometer equipped with a QCI cryoprobe. Each spectrum was acquired with 700-800 scans, an acquisition time of 90 ms, a recycle delay of 0.7 s and processed by applying an exponential window function (LB=10). The spectra was analyzed using software TopSpin (Bruker).

Surface Plasmon Resonance
Studies were performed using a Biacore T200 (GE Health Sciences  51 . The systems were solvated in an orthogonal box of SPC water molecules with buffer width of 5 Å for the complex and 10 Å for the solvent simulations. The full systems were relaxed and equilibrated using the default Desmond protocol, consisting of: (i) a minimization using a Brownian dynamics NVT integrator for 100ps with the solute molecules restrained (50 kcal/mol/Å2), (ii) 12 ps simulation in the NVT ensemble, keeping the restraints and temperature at 10 K, (iii) 12 ps simulation in the NPT ensemble, keeping restraints and temperature at 10 K, (iv) 24 ps simulation in the NPT simulation with solute heavy atom restraints at 300 K, and (v) 240 ps simulation in the NPT ensemble at 300 K without restraints. Production simulations in the NPT ensemble lasted 5 ns for both the complex and the solvent systems. A total of 12 λ windows were used for all calculations. The free energy differences between the initial and final states were calculated using the Bennett acceptance ratio (BAR) method 52 , and the errors estimated using bootstrapping 53 .

Notes:
The authors declare no competing financial interest.