key: cord-0426050-srxijntn
authors: Guo, Yusong; Liu, Qi; Mallette, Evan; Caba, Cody; Hou, Feng; Fux, Julia; LaPlante, Gabriel; Dong, Aiping; Zhang, Qi; Zheng, Hui; Tong, Yufeng; Zhang, Wei
title: Structural and functional characterization of ubiquitin variant inhibitors for the JAMM-family deubiquitinases STAMBP and STAMBPL1
date: 2021-07-17
journal: bioRxiv
DOI: 10.1101/2021.07.16.452720
sha: 86a197804c81e430917d2e95ac792738c7d99535
doc_id: 426050
cord_uid: srxijntn

Ubiquitination is one of the most crucial post-translational protein modifications involved in a myriad of biological pathways. This process is reversed by deubiquitinases (DUBs) that deconjugate the ubiquitin (Ub) moiety or poly-Ub chains from substrates. In the past decade, tremendous efforts have been focused on targeting DUBs for drug discovery. However, most chemical compounds with inhibitory activity for DUBs suffer from mild potency and low selectivity. To overcome these obstacles, we developed a phage display-based protein engineering strategy for generating Ub variant (UbV) inhibitors and have previously successfully applied it to the Ub-specific protease (USP) family of cysteine proteases. In this work, we leveraged the UbV platform to target STAMBP, a member of the JAB1/MPN/MOV34 (JAMM) metalloprotease family of DUB enzymes. We identified two UbVs (UbVSP.1 and UbVSP.3) that bind to STAMBP with high affinity but differ in their selectivity for the closely related paralog STAMBPL1. We determined the STAMBPL1-UbVSP.1 complex structure by X-ray crystallography, revealing hotspots of the tight JAMM-UbV interaction. Finally, we show that UbVSP.1 and UbVSP.3 are potent inhibitors of the STAMBP isopeptidase activity, far exceeding the reported small-molecule inhibitor BC-1471. This work demonstrates that UbV technology is suitable to develop tool molecules for metalloproteases. These tools can be used to understand the cellular function of JAMM family DUBs.

Ubiquitin (Ub) is a small, highly conserved protein that plays a central role in the Ub-proteasome system (UPS) to tightly regulate numerous biological processes, including immune responses (1), DNA repair (2) , and the cell cycle (3) . During ubiquitination, the C-terminal carboxyl group of Ub is covalently attached to a substrate protein at the lysine ε-amine or the N-terminal primary amine through an isopeptide or an a-peptide bond, respectively. The conjugated Ub itself can be further ubiquitinated at its N-terminus or one of seven lysine residues, forming a poly-Ub chain (4) . The ubiquitination process is catalyzed by a three-enzyme cascade comprised of E1 activating enzymes, E2 conjugating enzymes, and E3 Ub ligases. Different linkages of poly-Ub chains serve distinct purposes. For example, the most abundant K48-linked poly-Ub chains target substrate proteins to proteasomal degradation (4), whereas K63-linked poly-Ub chains are mostly non-degradative and involved in modulating cellular signal transduction (5) .

The antagonists of ubiquitination are the deubiquitinase (DUB) enzymes that deconjugate Ub from modified protein substrates. The human genome encodes approximately 100 DUBs, dysregulation of which results in a variety of diseases (6) . DUBs are categorized into seven families based on structural folds: ubiquitin-specific proteases (USPs), JAB1/MPN/MOV34 metalloenzymes (JAMMs), ovarian tumor proteases (OTUs), ubiquitin C-terminal hydrolases (UCHs), Machado-Josephin domain-containing proteases (MJDs), motif interacting with ubiquitin-containing novel DUB family (MINDYs) (7, 8) , and newly discovered zinc finger containing ubiquitin peptidase (ZUP) (9) . DUBs have emerged as attractive targets for pharmacological intervention in various human diseases, particularly cancers (6) . However, most small-molecule DUB inhibitors developed have suffered from mild potency and poor selectivity (10) . One of the challenges in developing chemical probe quality (11) small-molecule DUB inhibitors is that the Ub-binding groove is usually large, shallow, and not suitably shaped for small-molecule binding (10) . The recent success in discovering potent and selective USP7 small-molecule inhibitors took a decade (12) (13) (14) (15) (16) and elegantly combined fragmentbased screens and structure-guided optimization. This strategy has yet to show general applicability to other members of the DUB superfamily. Here, we set out to develop potent and specific inhibitors for the JAMM-family DUBs.

While the other six families of human DUBs are cysteine proteases, the JAMM family is unique in that they are metalloproteases. The JAMM domain (17) contains a catalytic zinc ion coordinated by two histidines, one aspartate/glutamate, and one water molecule that is hydrogen-bonded to an adjacent glutamate (18) . The JAMM family comprises a total of 14 DUBs, seven of which are predicted to be catalytically inactive (pseudo-DUBs) because of substitutions of essential Zn 2+ -coordinating residues (19) . Among the seven other JAMM members, STAM-binding protein (STAMBP, also known as AMSH), STAMBP-like 1 (STAMBPL1, also known as AMSH-LP), and MYSM1 (20) can function as isopeptidases independently, whereas BRCA1/BRCA2 containing complex subunit 36 (BRCC36), proteasome 26S subunit, non-ATPase 14 (PSMD14, also known as RPN11), and COP9 signalosome subunit 5 (CSN5) act in macromolecular assemblies (21) , and finally, the DUB activity and function of MPND is poorly understood (22) .

All known broad-spectrum JAMMfamily inhibitors are chelating agents (e.g., 1,10-phenanthroline and thiolutin), which have a high affinity for divalent metal ions and chelate the active site Zn 2+ ion (23, 24) . Notably, a potent and moderately specific RPN11 inhibitor capzimin was developed from a chelating agent-like small molecule through medicinal chemistry optimization (25) . CSN5i-3, an orally available inhibitor, has been discovered by Novartis targeting CSN5's deneddylation activity and confer high specificity towards other metalloproteinases (26) . For this work, we chose STAMBP as the representative JAMM DUB for targeted inhibition.

STAMBP consists of three distinct regions, including an N-terminal microtubule interacting and trafficking (MIT) domain, a central SH3-binding motif (SBM), and a C-terminal catalytic JAMM domain (18) . STAMBP plays an essential role in endocytosis and endosomal-lysosomal sorting of cell-surface receptors by deubiquitinating and rescuing ubiquitinated cargo proteins from lysosomal degradation. The MIT and SBM domains mediate critical protein-protein interactions (17) to recruit STAMBP to the endosomal sorting complexes required for transport (ESCRT), which regulate the multivesicular body (MVB) biogenesis of ubiquitinated cellreceptors (27) . Recessive mutations in the STAMBP gene were found to cause a severe developmental disorder, microcephaly capillary malformation syndrome (MIC-CAP), due to elevated Ub-conjugate aggregation and resulted progressive apoptosis (28) .

At the biochemical level, STAMBP cleaves K63-linked poly-Ub chains specifically, and the cleavage efficiency is increased when binding to STAM proteins (STAM1 or STAM2) (29) . STAMBPL1 shares 56% overall sequence identity to STAMBP and 68% identity in the catalytic domains (30) . Both STAMBP and STAMBPL1 localize to early endosomes by binding to clathrin (31); however, STAMBPL1 fails to bind to STAM due to residue substitutions in the SBM-like motif, suggesting it may function in different signaling pathway from that of STAMBP (30) .

The crystal structure of STAMBPL1 JAMM catalytic domain in complex with K63-linked diUb revealed its linkage specificity towards K63-linked poly-Ub chains involves the catalytic groove and two insertions that are conserved in STAMBP subfamily JAMMs: ins-1 (aa 314-339) and ins-2 (aa 393-415) (32) . The distal Ub interacts with ins-1 and the catalytic groove, and the proximal Ub interacts with ins-2 and the catalytic site. Recently, Bednash et al. used in silico molecular modeling and virtual screening and discovered a compound, BC-1471, that inhibits STAMBP and decreases protein level of its inflammasome substrate NALP7. However, the in vitro DUB assay showed that BC-1471 could not fully inhibit STAMBP activity even at 100 µM and no cocrystal structure was provided to support the mechanism of action (33) .

In this work, we conducted phagedisplay selections and identified Ub variant (UbV) inhibitors for STAMBP with high binding affinity. We solved the crystal structure of STAMBPL1 in complex with one of two UbVs and identified the hotspots responsible for tight binding. In addition, in vitro functional characterization confirmed specific and potent inhibition of STAMBP by UbVs. Importantly, the UbVs exhibited an inhibitory effect superior to BC-1471.

To develop potent and specific inhibitors for STAMBP, we decided to employ a structure-based combinatorial Ub engineering strategy to inhibit the Ub-STAMBP protein-protein interaction (34) selectively. Large hydrophobic surfaces on Ub (~2000 Å 2 ) mediate interaction with DUBs, including the Ile36 patch (Ile36, Leu71, and Leu73), the Ile44 patch (Leu8, Ile44, His68, and Val70), and the Phe4 patch (Gln2, Phe4, Thr14) (35) . Ub binds to DUBs with low affinity but high specificity, and thus mutations can be introduced to improve binding affinity without compromising the specificity. Indeed, UbVs for USP-family (34, 36) and OTU-family DUBs (34) can block the binding of substrate Ub to inhibit DUB activity.

We conducted five rounds of selections from an M13 bacteriophage pool representing the UbV library -Library 2 described previously (34) -for phage clones that bind to the biotinylated human full-length STAMBP protein (residues 1-424) (Fig. 1A) . A total of 96 clones (48 from Round 4 and 48 from Round 5) were tested for binding activity to STAMBP using phage enzyme-linked immunosorbent assay (ELISA). Among them, 53 clones that displayed specific binding with STAMBP were subjected to DNA sequencing, which returned 14 unique UbV sequences (Fig. S1) . We then selected the three most potent UbVs (denoted UbV SP.1 , UbV SP.2 , and UbV SP.3 , where "SP" stands for STAMBP) for follow-up characterization (sequences shown in Fig. 1B) . Finally, the binding specificity of these UbVs against a panel of 7 human DUBs of different families was assessed by phage ELISA (Fig. 1C) . All three UbVs were confirmed to bind both the full-length and the JAMM domain of STAMBP (STAMBP JAMM ), suggesting that the binding was mediated by the JAMM domain. In addition, UbV SP.1 exhibited cross-reactivity with STAMBPL1, while the other two variants showed significant preferential binding to STAMBP over STAMBPL1. Importantly, all three UbVs showed no binding to two USP-family DUBs (USP7 and USP14), a JAMM-family DUB complex BRISC, or an OTU-family DUB OTUD1.

Structural characterization of STAMBPL1 in complex with UbV SP. 1 We then conducted crystallization trials for STAMBP JAMM or STAMBPL1 JAMM in complex with the UbVs. Diffractionquality crystals could only be obtained for STAMBPL1 JAMM in complex with the dualspecific UbV SP.1 despite a sequence identity of 68% between the two catalytic domains ( Fig. 2A) . We solved the crystal structure of the UbV SP.1 /STAMBPL1 JAMM complex at 2.0 Å resolution (PDB:7L97, Fig. 2B occupies the distal Ub position in the structure. A total of 11 mutated residues differentiate UbV SP.1 from Ub.wt (Fig. 1B) .

The last two C-terminal residues (Ala77 and Ser78) of UbV SP.1 were not observed in the electron density map and were not modelled. Based on the crystal structure, it is evident that seven of the mutated residues (sites 2, 6, 12, 47, 63, 64, and 68) are distant from STAMBPL1 and unlikely to contribute to the tight binding observed. (Fig. 2B) The primary interaction interface was observed in the C-terminal tail of UbV SP.1 , which includes three mutations, L73H, G75H, and G76S (Fig. 2B) . PISA analysis (37) of the interface of the complex structure reveals the UbV has a buried surface area (BSA) of 1239 Å 2 . In contrast, the seven tail residues of the UbV (residues 70-76) contribute almost half of the BSA with 556 Å 2 .

The overall structure and mode of UbV SP.1 binding are similar to that of the distal Ub observed in the K63-diUb cocrystal structure. The RMSD of the backbone Ca atoms of STAMBPL1 in the two structures is only 0.29 Å over 146 residues, whereas the RMSD of UbV SP.1 compared to the distal Ub is 0.34 Å over 66 residues. Superposition of the complex structures in their entirety presents an overall RMSD of 0.56 Å over 212 residues, larger than the RMSDs of the individual molecules, suggesting a relative conformational movement. The conformational changes are most evident for three regions: a loop N-terminal to ins-1, the loops around the catalytic site, and the loops chelating the structural zinc ion. A Ca RMSD-based sausage representation of the STAMBPL1 structure visualizes the regions that have the largest conformational change upon UbV SP.1 binding (Fig. 2C) .

A comparison of the conformation of the UbV SP.1 and the distal Ub in the STAMBPL1/K63-diUb complex structure when the STAMBPL1 JAMM domain of the two structures are aligned reveals a significant conformational difference of the tails of the two bound Ub molecules (residues Arg72 and onwards; Fig. 2D -G). The backbone of the C-terminal tail of UbV SP.1 has the largest conformational movement compared to Ub.wt (Fig. 2D ) and is excluded from the catalytic groove and projects in the direction of ins-2 while interacting with ins-1 (Fig. 2E) . In contrast, the isopeptide bond of the physiological K63-diUb substrate is, as expected, buried in the catalytic groove and spans the catalytic site in an orientation that is primed for hydrolysis (Fig. 2E, 2H) . Sequence comparison of UbV SP.1 with Ub.wt (Fig. 1B) and structural analysis (Fig. 2F) suggests the G75H mutation creates a steric hindrance that prevents the access of the tail to the catalytic groove, thereby providing a structural basis for the observed conformation. Instead, His75 forms a Tshaped π-stacking interaction with Tyr322 on ins-1 to stabilize the interaction. Meanwhile, the side chain of His73 in UbV SP.1 (equivalent to Leu73 in Ub.wt) forms a hydrogen bond with the hydroxyl group of Tyr367 and is buried in a hydrophobic pocket formed by Tyr367 and Trp345 of STAMBPL1. PISA analysis (37) also reveals that His73 has the largest BSA of 168 Å 2 among all UbV SP.1 residues. (Fig.  2F) .

Despite the ins-1 helix being a key region interacting with the distal Ub, it shows a relatively small conformational change compared to the rest of the molecule except for Glu326, a loop residue Nterminal to ins-1. Glu326 has the largest movement of Ca among the whole STAMBPL1 JAMM molecule at 4.4 Å (Fig.  2C) . This prompted us to look further into the detailed interactions in this region. UbV SP.1 has a glycine at residue 42 ( Fig.  2G) , and the equivalent site on Ub.wt is occupied by an arginine (Arg42), with its side chain extending towards ins-1, effectively expelling the side chain of Glu326 away from the Ub.wt (Fig. 2H) . In UbV SP.1 , the R42G mutation creates space to accommodate the side chain of Arg72 in the C-terminal tail from UbV SP.1 (Fig. 2G ) to flip towards this vacancy (Fig. 2H) . A strong hydrogen bond network is then formed between the Arg72 guanidinium and the backbone carbonyl group of Glu326 in ins-1, and between the amide group of Gln49 of UbV SP.1 with the backbone carbonyl and amide of Val328 and Glu326, thus stabilizing the interaction between the STAMBPL1 and UbV SP.1 (Fig. 2G) . In the STAMBPL1/K63-diUb complex, the side chain of Arg72 is oriented towards ins-2 but does not directly contact any STAMBPL1 residues (Fig. 2H) . The coordinated mutation of R42G and flip of Arg72 side chain on UbV SP.1 with the movement of Glu326 on STAMBPL1 ( Fig. 2G-2H ) creates another hotspot for the tight interaction between UbV SP.1 and STAMBPL1.

To quantify the interaction of the UbVs with STAMBP and STAMBPL1, and to validate the key residues responsible for the tight binding, we used isothermal titration calorimetry (ITC) to measure the interaction thermodynamics (Fig. 3) . All three UbVs evaluated showed a sub-µM affinity with STAMBP JAMM , and only UbV SP.1 showed a sub-µM affinity for STAMBPL1; Ub.wt does not show observable binding with STAMBPL1. While other UbVs had at least 14-fold higher affinity for STAMBP over STAMBPL1, UbV SP.1 has only 3.6-fold greater affinity towards STAMBP over STAMBPL1, indicating that it is a dual-specific binder. This is consistent with the result from phage ELISA (Fig. 1C) .

Based on structural analysis, we hypothesized that R42G and L73H, among the 11 mutations of UbV SP.1 , are the key residues that enhance the binding affinity of the UbV SP.1 . To test this hypothesis, we constructed three mutants of Ub.wt, with either a single mutation Ub R42G , Ub L73H , or a double mutation Ub R42GL73H . We compared the binding affinity of these mutants with both STAMBP JAMM and STAMBPL1 JAMM using ITC ( Fig. 3A-C) . Both Ub.wt and Ub R42G showed no binding to STAMBPL1 JAMM , while Ub L73H showed only weak binding to STAMBPL1 JAMM . The double mutant Ub R42GL73H , however, showed a sub-µM affinity (KD of 0.88 µM) only 5fold weaker than that of UbV SP.1 , suggesting a synergistic effect of the two mutations. The binding of the mutants with STAMBP JAMM show a similar trend; the R42G mutation reduced the binding of the Ub to STAMBP JAMM and the L73H mutation only slightly increased the affinity. The double mutant Ub R42GL73H , however, resulted in a sub-µM affinity at 0.15 µM, only 3-fold weaker than the UbV SP.1 . These results suggest the mutations at sites 42 and 73 are the hotspots for UbV SP.1 for its high binding affinity towards both STAMBP JAMM and STAMBPL1 JAMM .

UbV SP.1 and UbV SP. 3 

To investigate the inhibitory effects of UbVs on STAMBP in vitro, we first performed a fluorescence resonance energy transfer (FRET)-based K63-diUb substrate cleavage assay. We found that UbV SP.1 and UbV SP.3 are potent inhibitors for the isopeptidase activity of STAMBP JAMM . The half-maximal inhibition (IC50) was 8.4 nM for UbV SP.1 and 9.8 nM for UbV SP.3 both having a Hill slope of -0.8 ( Fig. 4A-B) . To benchmark the inhibitory potency of our UbV inhibitors, we obtained previously reported STAMBP small-molecule inhibitors having two different chemistries each reported as BC-1471 (33): consisting of a core 2-6-morpholino-4-oxo-3-phenethyl-3,4-dihydroquinazolin-2-yl-thio-and an Nlinked tetrahydrofuran-2-yl-methyl acetamide (CAS 896683-84-4, racemate) or a furan-2-yl-methyl-acetamide (CAS 896683-78-6). In both cases, we did not observe the inhibitory activity reported for BC-1471 as there was no difference in the diUb deconjugation reaction rate of STAMBP JAMM with or without the compounds ( Table 2) .

Next, we compared the K63-diUb (FRET) cleavage inhibition by the UbVs with STAMBP JAMM to the inhibition of fulllength STAMBP activated by STAM protein and full-length STAMBPL1. STAM contains three Ub-binding domains: VPS27/Hrs/STAM (VHS), Ub-interacting motif (UIM), and SH3 domains. The SH3 domain mediates the interaction of STAM with the SBM motif N-terminal to the JAMM domain of STAMBP. The VHS domain shifts the cleavage preference of STAMBP to longer poly-Ub chains by binding to a Ub that is spatially distant from the JAMM domain and helps position the cleavage site (38) , while the UIM domain is proposed to bind to the proximal Ub when the JAMM domain of STAMBP recognizes the distal Ub of diUb (39, 40) . As expected from IC50 assays, UbV SP.1 resulted in complete inhibition of STAMBP JAMM and UbV SP.3 reduced the activity to 10% of the uninhibited enzyme ( Table 2 , UbV concentration=1 µM). Full-length STAMBP activated with STAM1 was inhibited by both UbV SP.1 and UbV SP.3 , and the activity was reduced to 7% and 13%, respectively. The activity of the paralog STAMBPL1 with the K63-diUb FRET substrate was inhibited to 44% by UbV SP.1 ; however, there was no observable inhibition by UbV SP.3 . This is consistent with the binding affinity data shown in Fig. 3 .

Next, we assessed the effect of UbV SP.1 and UbV SP.3 using a K63-linkage poly-Ub (Ub2-Ub7) cleavage assay. In this assay, the isopeptidase activity of STAMBP can be visualized by monitoring the appearance of mono-Ub (Ub1) and the disappearance of the Ub2-Ub7 bands (Fig.  5A) . STAMBP in complex with different UbV inhibitors were incubated with K63linkage poly-Ub, and STAMBP JAMM samples in the presence and absence of Ub.wt were used as controls. Both UbV SP.1 and UbV SP.3 potently inhibited the isopeptidase activity of STAMBP JAMM (Fig.  5B ) and full-length STAMBP (Fig. 5C ) towards the cleavage of K63-linked poly-Ub chains. In addition, UbV SP.1 appeared to inhibit STAMBPL1 to some extent, whereas UbV SP.3 showed no inhibitory effect with STAMBPL1 (Fig. S2A) .

We then compared the inhibitory effects to the small-molecule inhibitor BC-1471 (33), UbV SP.1 exhibited a much greater inhibitory effect to STAMBP than BC-1471 ( Fig. 5D) . Importantly, the inhibition by UbV SP.1 is dose-dependent, but this is not the case for BC-1471 (Fig. S2B) , further confirming that BC-1471 is unlikely a potent inhibitor of STAMBP. In addition, we observed inhibition of STAM1-activated full-length STAMBP by both UbV SP.1 and UbV SP.3 in this chain cleavage assay (Fig.  S2C) . Finally, neither UbV SP.1 and UbV SP. 3 showed inhibitory effects to other JAMM DUBs, such as BRISC (Fig. S2D) , and MYSM1 (Fig. S2E) , which indicated the inhibitory specificity of UbV SP.1 and UbV SP.3 .

Over the past few years, the phagedisplay based protein engineering platform targeting Ub-mediated protein-protein interactions (41) has been utilized to develop potent and specific modulators for a repertoire of UPS components, including E2 conjugating enzymes (42) , E3 ligases (43) (44) (45) , and DUBs (36, (46) (47) (48) . These DUBs include seven USPs (USP2, USP7, USP8, USP10, USP15, USP21, and USP37); one OTU (OTUB1); and two viral papain-like proteases (MERS-CoV PLpro and CCHFV-L, which have USP and OTU structural fold, respectively) (36, 41, (46) (47) (48) .

The UbVs described herein for STAMBP and STAMBPL1 are the first examples of protein-based potent and specific deubiquitination inhibitor developed for the JAMM family DUBs. Previously, a moderately selective (>5-fold towards several JAMM DUBs) RPN11 inhibitor capzimin was generated by optimization of a non-specific small molecule inhibitor 8-TQ, which has structural similarity with chelating agent 8-hydroxyquinoline (25) . Moreover, a small-molecule inhibitor CSN5i-3 was developed to inhibit CSN5, demonstrating the feasibility of targeted inhibition of JAMM DUBs (26) . CSN5 is unique in that it needs the whole CSN complex to elicit deneddylation activity and has NEDD8-conjugated cullin-RING E3 ligases as specific substrates. While BC-1471 was recently reported as a STAMBP chemical inhibitor (33), we were not able to detect its inhibitory activity in our in vitro assay. We note that BC-1471 was identified from an in silico screening method with no confirmation using a cocrystal structure or an orthogonal biophysical technique to validate a direct binding (33) . In addition, previous in vitro assays did not show complete inhibition of STAMBP activity by BC-1471 at high concentrations (100 µM) or under a native cellular phenotype such that the observed inhibition could be the result of off-target effects (33) .

Similar to other UbVs targeting USPs, including UbV core for USP37, UbV 7.2 for USP7, and UbV 10.1 for USP10 (36, 46) , UbV SP.1 has enhanced affinity primarily due to amino-acid substitutions at the C-terminal variable region (N60-G76+). Structural characterization of the STAMBPL1 JAMM -UbV SP.1 complex, along with mutation analysis of amino-acid substitutions, identified hotspot residues contributing to the enhanced affinity of the variants for the catalytic domains of STAMBP and STAMBPL1. With a single residue substitution (L73H) in the C-terminus alone, the affinity of Ub.wt for STAMBP was enhanced nearly 7-fold. Subsequently, a double mutation (R42G, L73H) enhanced the affinity of Ub.wt for STAMBP by 150fold.

The simplicity of the phage display selection process opens the opportunity for rapidly developing high-affinity and selective DUB inhibitors. Another advantage of the UbV platform over the peptide or small molecule compound screening strategies comes from the fact that Ub.wt is a low-affinity natural substrate for the target DUBs, and the variable C-terminal region are topographically placed in the proximity of the catalytic groove (49) . Theoretically, this confers a significant reduction of the search space to engage binding during the selection process, further improving the efficiency in selecting highaffinity binders. In addition, the large Ub binding surface on the DUBs provides opportunities for the UbV to bind to the cognate DUB and compete with substrate Ub.wt without direct interaction with the catalytic site residues. For example, the Cterminal tail of UbV 8.2 for USP8 (34) binds to a cleft between the blocking loop and the fingers subdomain that is remote from the catalytic groove yet still compete with Ub.wt binding. For reference, UbV 8.2 is rotated about 40° when compared to the binding mode of Ub.wt in other USP complexes (47) . Therefore, targeting the substrate Ub.wt binding region of the JAMM domain by UbVs is novel compared to current non-specific chemical reagents targeting metalloproteases through the chelation of the catalytic zinc ion.

Furthermore, improved affinity and specificity has been achieved for several UbVs (e.g., UbV 15.1a-e for USP15 and UbV Fl11.1 for SKP1-FBL11 complex) through additional residue insertions in the b1-b2 loop, extending the interaction surface of the UbV (45, 47, 50) . The UbVs developed for STAMBP and STAMBPL1 have multiple mutations in b1, b2, and the loop they encompass; however, these mutations have little effect on UbV binding due to the orientation of distal Ub binding to the JAMM domain. The b1-b2 loop is poised to make substantial interactions with b-strands 2, 3, and 6 of STAMBPL1 with a similar peptide insertion as was employed previously for USP15 and SKP1-FBL11. Thus, specificity and affinity could potentially be improved with insertions in the b1-b2 loop and can be explored in the future.

Intriguingly, UbV SP.1 has nanomolar affinities for both STAMBP and the paralog STAMBPL1, while UbV SP.3 has a similar affinity for STAMBP but a reduced affinity for STAMBPL1. Disparity in the affinities of the two UbVs was also translated to the inhibitory effects of the respective UbVs on isopeptidase activity. UbV SP.1 was equally effective at inhibiting STAMBP and STAMBL1 activity, whereas UbV SP.3 had no apparent effect on STAMBPL1 activity. Structural and sequence similarities of STAMBP and STAMBPL1 suggest that UbV SP.1 likely binds in a similar manner to both enzymes. The noticeably stronger affinity (3-fold, Fig. 3A ) of UbV SP.1 for STAMBP is likely due to the substitution of the Val328 (STAMBPL1) in the ins-1 loop by a glutamic acid (Glu316 in STAMBP), which may introduce an additional salt bridge with Arg72 from UbV SP.1 (Fig. S3A) . Of the two hotspot mutations (R42G and L73H) identified in UbV SP.1 , UbV SP.3 also harbors the L73H mutation but preserves the Arg42 of Ub.wt (Fig. 1B) . The weaker affinity of UbV SP.3 for STAMBPL1 and the resulting lack of inhibition of STAMBPL1 activity may be explained by the difference in the interaction with these residues. The preservation of Arg42 in UbV SP.3 prevents its Arg72 from flipping towards ins-1 and participating in the hydrogen bond network as observed in the STAMBPL1:UbV SP.1 structure. In contrast, the side chain of Glu316 of STAMBP could form waterbridged hydrogen bonds with Arg42 and Arg72, thereby resulting in a reorganization of the ins-1 loop. Although no structure of human STAMBP in complex with Ub.wt has been reported, such a water-bridged interaction was observed in the complex structure of the yeast STAMBP orthologue Sst2 in complex with K63-diUb (PDB :4NQL, Fig. S3B) . The increased affinity of UbV SP.3 for STAMBP over Ub.wt is partially explained by the same L73H substitution as in UbV SP.1 (as demonstrated from the observed 7-fold increase in affinity from ITC experiments). Additional affinity could be caused by additional hydrophobic interactions of the introduced R74F mutation with residues Tyr310, Phe395, and Val347 (Tyr322, Phe407, and Val359 STAMBPL1 numbering). The five residues of UbV SP.3 tail (Phe74-Met78) differ significantly from those in UbV SP.1 (Fig. 1A) and may form additional enzyme-specific interactions divergent from those observed for the respective residues in the STAMBPL1:UbV SP.1 structure. While a relative specificity is observed for UbV SP.1 and UbV SP.3 against other DUBs, a thorough proteomics analysis may be required before using these molecules in target validation experiments. For example, there are more than 20 different Ub-binding domains (UBDs) identified to date and they are quite diverse at the structure level (51) . We noted that STAMBP UbVs do not have the key mutation at position 10 (i.e. G10A/V mutation that can lead to UbV dimerization) necessary to increase affinity for binding to Ub-interacting motifs (UIMs) (46, 52, 53) . However, UbV SP.1 and UbV SP.3 may potentially interfere with certain Ub-UBD interactions. Like the off-target effects in drug discovery, whether UbV SP.1 or UbV SP.3 may bind to other proteins, including those do not contain UBDs, should be tested in separate cellular experiments.

Given the success of generating firstin-class UbV inhibitors for STAMBP and STAMBPL1, further development of UbV modulators is conceivable for the remaining 13 members of the JAMM family, including the 7 pseudo-DUBs. While the structural elements of the JAMM domain persist, many of the residues composing the canonical Ub binding site are divergent. Employing UbV selection strategies opens the opportunity of utilizing the deviation from the canonical binding site as a unique target for protein-protein interactions. Of particular interest are the BRCC36-ABRAXAS or -KIAA0157 complexes in the BRCA1-A and BRISC holocomplexes, respectively (54) . The aforementioned heterodimers with BRCC36 form the catalytic cores of their respective holocomplexes, incorporating a dimer of JAMM and pseudo-JAMM domains. Targeting these structures with UbVs would provide a more rapid and specific option than conducting small molecule inhibitor screens. Finally, the recombinant UbV modulators of the pseudo-DUBs will be ideal probes for evaluating the biological functions of these pseudo-enzymes without ablating their potential roles as molecular scaffolds in cellular processes.

All constructs for structural biology and isothermal titration calorimetry (ITC) binding measurement were cloned into a pET28-MHL vector, which encodes a TEV protease cleavable N-terminal His6-tag that will leave a non-native glycine residue after TEV treatment of the target protein, and overexpressed in E. coli strain BL21 (DE3) with 250 µM isopropyl-β-Dthiogalactopyronoside (IPTG) induction using a protocol described previously (55) . 1×PBS with 5% glycerol was used as the base buffer for the whole purification process with added supplements in different purification steps. Briefly, cell pellets from 2 L culture were used for each construct. 

X-ray diffraction data for structure determination were collected at the Canadian Light Source Inc. (CLSI) beamline 08ID-1 (CMCF-ID) (56) . The dataset was processed with the HKL-3000 suite (57) . The structures were solved by molecular replacement using PHASER v.2.6.1 using PDBs 2ZNR and 1UBQ as the search template for STAMBPL1 and UbV SP.1 , respectively. COOT was used for model building and visualization, REFMAC (v.5.8.0135) for restrained refinement. The final model was validated by MOLPROBITY. All molecular graphics were prepared with PyMOL v.2.4.0 (Schrödinger, Inc.)

For ITC measurement, proteins were dialyzed overnight with PBS buffer containing 1 mM TCEP and adjusted to a final concentration of 50 µM. All ITC measurements were performed at 25°C on a NanoITC (TA Instruments). A total of 25 injections, each of 5 µL Ub.wt, mutants, or UbVs were delivered into a sample cell of 200 µL containing one of the STAMBP JAMM or STAMBPL1 JAMM . The data were analyzed using the NanoAnalyzer software supplied by the manufacturer and fitted to a one-site binding model.

The phage-displayed UbV library used for selection was re-amplified from Library 2 as previously described (34) . Protein immobilization and the following phage selections were done according to standard protocols (58) . Briefly, purified STAMBP proteins (full-length or JAMM domain) were coated on 96-well MaxiSorp plates by adding 100 µL of 1 µM proteins and incubating overnight at 4°C. Five rounds of phage display selections were performed as follows: a) Preparation of the phage library pool, within which each phage particle displays a unique UbV and encapsulates the UbV encoding DNA in a phagemid; b) The pool of phage-displayed UbV library are applied to immobilized STAMBP; c) STAMBP-binding phage are captured, and nonbinding phage are washed away; d) Bound phage are amplified by infection of bacterial host E. coli (OmniMAX); e) Individual phage with improved binding properties obtained from round 4 and round 5 are identified by phage ELISA (see below) and subjected to DNA sequencing of the phagemids to obtain UbV sequences (44) . High-affinity UbVs identified from phage ELISA were cloned into expression vectors, including a FLAG-tag epitope, expressed in E. coli, and purified. For the ELISA experiments, proteins were immobilized on 384-well MaxiSorp plates (Thermo Scientific 12665347) by adding 30 µL of 1 µM protein solution for overnight incubation at 4ºC. Phage and protein ELISAs with immobilized proteins were performed as described before. Binding of phage or FLAG-tagged UbVs was detected using anti-M13-HRP (GE Healthcare 27942101) or anti-FLAG-HRP antibody (Sigma-Aldrich A8592), respectively.

To measure the half-maximal effective binding concentration (EC50) of the UbVs' binding to the immobilized DUB, the concentration of UbVs in solution was varied from 0 to 4 µM by two-fold serial dilutions. EC50 values were calculated using the GraphPad Prism (version 5.0a) software with the built-in equation (non-linear regression curve).

Inhibition concentrations were determined using a FRET-based diUb cleavage reaction to measure isopeptide hydrolysis by the STAMBP JAMM domain. Fluorescence of hydrolyzed K63linked diUb (TAMRA/QXL position 3 labeled, Boston Biochem #UF-330) was measured on a BioTek synergy LX plate reader equipped with a red filter cube assembly (ex. 530 nm, em. 590 nm) for 60 minutes at ambient temperature (22°C). Enzyme (2 nM) and UbVs were preincubated for 10 minutes at ambient temperature before addition of the diUb substrate (100 nM) in 50 µL reaction buffer (50 mM HEPES pH7.5, 100 mM NaCl, 0.01% v/v Tween 20, 0.01 mg/mL BSA, 5 mM DTT). Reactions were replicated in triplicate (N=3) for varying concentrations of UbV (1 µM to 0.1 nM). The rate of reaction was determined from the slope of the initial linear phase of the reaction. Reaction rates were averaged and normalized to the negative control (no inhibitor) then plotted against UbV concentrations in GraphPad Prism 8. 50% inhibition concentrations were determined by non-linear regression using the formula for inhibitor concentration vs. normalized response (variable slope).

To 

The atomic coordinates and structure factors (Code:7L97) have been deposited in the Protein Data Bank (https://wwpdb.org). PyMOL script for generating the sausage view is uploaded to GitHub at https://github.com/tongalumina/rmsdca. All other data are included in this manuscript. K63-linked poly-Ub 

Ubiquitin signaling in immune responses

Regulation of DNA repair by ubiquitylation

Ubiquitin ligases: Cell-cycle control and cancer

Ubiquitin modifications

Nonproteolytic Functions of Ubiquitin in Cell Signaling

Deubiquitylating enzymes and drug discovery: Emerging opportunities

Toward understanding ubiquitin-modifying enzymes: From pharmacological targeting to proteomics

MINDY-1 Is a Member of an Evolutionarily Conserved and Structurally Distinct New Family of Deubiquitinating Enzymes

a Class of Its Own: A New Family of Deubiquitinases Promotes Genome Stability

Recent Advances in the Discovery of Deubiquitinating Enzyme Inhibitors

The promise and peril of chemical probes

Drug development: Allosteric inhibitors hit USP7 hard

Small-molecule inhibitor of USP7/HAUSP ubiquitin protease stabilizes and activates p53 in cells

Discovery and characterization of highly potent and selective allosteric USP7 inhibitors

Molecular basis of USP7 inhibition by selective small-molecule inhibitors

Endocytosis: the DUB version

Insights into the mechanism of deubiquitination by jamm deubiquitinases from cocrystal structures of the enzyme with the substrate and product

A genomic and functional inventory of deubiquitinating enzymes

A Histone H2A Deubiquitinase Complex Coordinating Histone Acetylation and H1 Dissociation in Transcriptional Regulation

Higher-Order Assembly of BRCC36-KIAA0157 Is Required for DUB Activity and Biological Function

An Adversarial DNA N6-Methyladenine-Sensor Network Preserves Polycomb Silencing

K63-specific deubiquitination by two JAMM/MPN+ complexes: BRISCassociated Brcc36 and proteasomal Poh1

Thiolutin is a zinc chelator that inhibits the Rpn11 and other JAMM metalloproteases

Capzimin is a potent and specific inhibitor of proteasome isopeptidase Rpn11

Targeted inhibition of the COP9 signalosome for treatment of cancer

The ESCRT machinery in endosomal sorting of ubiquitylated membrane proteins

Mutations in STAMBP, encoding a deubiquitinating enzyme, cause microcephaly-capillary malformation syndrome

STAM-AMSH interaction facilitates the deubiquitination activity in the C-terminal AMSH

Identification of AMSH-LP containing a Jab1/MPN domain metalloenzyme motif

Clathrin anchors deubiquitinating enzymes, AMSH and AMSH-like protein, on early endosomes

Structural basis for specific cleavage of Lys 63-linked polyubiquitin chains

Targeting the deubiquitinase STAMBP inhibits NALP7 inflammasome activity

A strategy for modulation of enzymes in the ubiquitin system

The ubiquitin code

Generation and Validation of Intracellular Ubiquitin Variant Inhibitors for USP7 and USP10

Inference of Macromolecular Assemblies from Crystalline State

The Vps27/Hrs/STAM (VHS) domain of the signaltransducing adaptor molecule (STAM) directs associated molecule with the SH3 domain of STAM (AMSH) specificity to longer ubiquitin chains and dictates the position of cleavage

NMR Reveals the Interplay among the AMSH SH3 Binding Motif, STAM2, and Lys63-Linked Diubiquitin

Mechanism of recruitment and activation of the endosome-associated deubiquitinase AMSH

Structural and Functional Analysis of Ubiquitin-based Inhibitors That Target the Backsides of E2 Enzymes

A General Strategy for Discovery of Inhibitors and Activators of RING and U-box E3 Ligases with Ubiquitin Variants

System-Wide Modulation of HECT E3 Ligases with Selective Ubiquitin Variant Probes

Inhibition of SCF ubiquitin ligases by engineered ubiquitin variants that target the Cul1 binding site on the Skp1-Fbox interface

The ubiquitin interacting motifs of USP37 act on the proximal Ub of a di-Ub chain to enhance catalytic efficiency

Structural and Functional Characterization of Ubiquitin Variant Inhibitors of USP15

Potent and selective inhibition of pathogenic viruses by engineered ubiquitin variants

Peptide and small molecule inhibitors of HECT-type ubiquitin ligases

A Structure-Based Strategy for Engineering Selective Ubiquitin Variant Inhibitors of Skp1-Cul1-F-Box Ubiquitin Ligases

Ubiquitin-binding domains from structures to functions

Structural and functional characterization of a ubiquitin variant engineered for tight and specific binding to an alpha-helical ubiquitin interacting motif

Dimerization of a ubiquitin variant leads to high affinity interactions with a ubiquitin interacting motif

Metabolic control of BRISC-SHMT2 assembly regulates immune signalling

Crystal structure and activity-based labeling reveal the mechanisms for linkage-specific substrate recognition by deubiquitinase USP9X

Beamline 08ID-1, the prime beamline of the Canadian macromolecular crystallography facility

HKL-3000: The integration of data reduction and structure solution -From diffraction images to an initial model in minutes

Identifying specificity profiles for peptide recognition modules from phage-displayed peptide libraries

We greatly appreciate help from Dr. Sachdev S. Sidhu (University of Toronto) who provided the UbV library. We sincerely thank Dr. Elton Zeqiraj (University of Leeds) for providing BRISC complex proteins. Q.L. is supported by an OGS scholarship. E.M. was a MITACS Industrial Postdoctoral Fellow with funding provided by MITACS and ProteinQure Inc. WZ is currently a CIFAR Azrieli Global Scholar in the Humans & The Microbiome Program. W.Z. is also a recipient of the Cancer Research Society/ BMO Bank of Montreal Scholarship for the Next Generation of Scientists. 

We declare no conflict of interests.