The deubiquitylase USP9X controls ribosomal stalling

Clancy et al. develop a specific chemical inhibitor of USP9X and characterize its effects upon the cellular proteome. This analysis reveals a central role in the regulation of ribosomal stalling through control of critical E3 ligases.


Introduction
Prompt sensing and resolution of aberrant protein translation is essential for the maintenance of protein homeostasis. Several circumstances can give rise to stalled ribosomes, such as insufficiency of a cognate acylated tRNA, defective mRNA, or faulty ribosomes (Arthur and Djuranovic, 2018;Ishimura et al., 2014). The most common cause of ribosomal stalling is thought to be the translation of poly(A), when a nascent mRNA is inappropriately polyadenylated within its coding region to generate a "non-stop" mRNA transcript lacking a stop codon (Arthur et al., 2015;Ozsolak et al., 2010). If a ribosome stalls during translation, it risks being rear-ended by a trailing ribosome. This collision generates a stable di-ribosome complex with a defined structure, which is resolved by the engagement of a dedicated machinery (Juszkiewicz et al., 2018;Collart and Weiss, 2020). In such cases, the E3-ligase ZNF598 ubiquitylates 40S complexes at specific sites on eS10 and uS10 subunits at the di-ribosome interface (Garzia et al., 2017;Juszkiewicz et al., 2018;Juszkiewicz and Hegde, 2017;Sundaramoorthy et al., 2017). This prevents further translation and initiates quality control processes (e.g., degradation of the associated mRNA) and ribosomal recycling pathways through partially understood mechanisms (Joazeiro, 2017). ZNF598 is a human Really Interesting New Gene (RING) domain protein that shares homology with the yeast protein Hel2, the deletion of which promotes read-through of polybasic sequences (Brandman et al., 2012;Juszkiewicz and Hegde, 2017). A recent report has provided evidence that the E3-ligases Makorin 1 (MKRN1) and Makorin 2 (MKRN2) may complement the activity of ZNF598 in the ribosomal quality control pathway by promoting the initial stalling of the leading ribosome as it encounters a polyA tract (Hildebrandt et al., 2019).
The function of E3-ligases can be opposed by ∼100 deubiquitylase (DUB) enzymes drawn from seven families (Clague et al., 2019). RING E3s show a tendency to auto-ubiquitylate, leading to their destabilization, which can be rescued by the activity of interacting DUBs. The best known such example is provided by the association between USP7 and MDM2, which has rendered USP7 a prominent drug target, as a means to regulate levels of p53 (Li et al., 2004). Recent work focused on this enzyme has established proof of principle that selective small-molecule inhibition among the ubiquitin-specific protease (USP) family can be achieved (Gavory et al., 2018;Kategaya et al., 2017;Lamberto et al., 2017;Turnbull et al., 2017). USP9X is one of the most abundant members of the USP family and has been linked with many processes, including centrosome function, chromosome alignment during mitosis, EGF receptor degradation, chemo-sensitization, and circadian rhythms (Harris et al., 2012;Li et al., 2017;Savio et al., 2016;Wang et al., 2017;Zhang et al., 2018). Loss-of-function mutations in females lead to congenital malformations and intellectual disability (Reijnders et al., 2016). USP9X localization is predominantly cytosolic, but it can also influence events within the nucleus such as the DNA damage response (Murtaza et al., 2015;Nathan et al., 2008;O'Dea and Santocanale, 2020;Urbé et al., 2012).
In this study, we identify the E3 ligases ZNF598 and MKRN1/ 2 as USP9X binding partners and show that USP9X governs the stability of ZNF598 and Makorins. The loss or inhibition of USP9X leads to a substantive reduction in steady-state levels of ZNF598 and MKRN1/2 and disables an effective response to the presence of stalled ribosomes.

Results and discussion
In a large-scale proteomic study of the ribosome interactome, USP9X was one of only two DUB family members to be identified along with OTUD4 (Simsek et al., 2017). Furthermore, USP9X is also apparent within the set of ZNF598 interacting proteins, previously identified in a label-free proteomic study (Garzia et al., 2017). We sought to confirm this interaction by immunoprecipitating FLAG-tagged ZNF598 transiently expressed in HEK293T cells (Fig. 1). USP9X is clearly present in the immunoprecipitate containing ZNF598-FLAG and is absent from control lanes.
We next compared the engineered USP9X −/0 HCT116 colon cancer cells that have been described previously (Harris et al., 2012) with wild-type WT cells of the same origin. As expected, the USP9X −/0 cells show reduced levels of previously identified peri-centrosomal substrates CEP55, CEP131, and PCM1 (Li et al., 2017;Wang et al., 2017;Fig. 2 A). ZNF598 levels are also greatly diminished in these cells (Fig. 2 A and Fig. S1). Two lines of argument suggest that this is not an effect on transcription: (1) endogenous ZNF598 mRNA levels are similar between the two cell lines (Fig. 2 B), and (2) levels of exogenous HA-ZNF598 expression that is driven by a non-native promoter are also diminished in transfected cells (Fig. 2 C). We treated these cells with cycloheximide and monitored the decay of the expressed HA-ZNF598. In WT cells, levels of HA-ZNF598 remained stable over the 6 h of incubation, while in the USP9X −/0 cells, the levels significantly decline to ∼60% in the same period (Fig. 2 D). The most parsimonious explanation of these combined results is that USP9X interacts with ZNF598 and regulates its steady-state levels through the control of its stability.
To demonstrate that stabilization of ZNF598 requires the catalytic activity of USP9X, we took advantage of a highly selective small-molecule inhibitor FT709 (Fig. 3). We identified USP9X inhibitors using a ubiquitin-TAMRA fluorescence polarization high-throughput screening assay, screening the inhibitory potential of a diverse collection of ∼140,000 compounds available at FORMA Therapeutics. Primary hits were further validated for direct USP9X binding by biophysical techniques such as surface plasmon resonance. Optimization of hits with respect to activity and physicochemical properties resulted in a series of compounds that includes FT709.
FT709 is potent in a biochemical assay with an half-maximal inhibitory concentration (IC 50 ) of 82 nM (Fig. 3 B). Modulation of CEP55 expression in BxPC3 pancreatic cancer cells showed an IC 50 of 131 nM (Fig. 3 C). The selectivity of FT709 was tested against >20 DUBs in a biochemical assay (Fig. S2 A) and was otherwise inactive across the panel (IC 50 >25 µM). FT709 shows vastly improved specificity over the compound WP1130, which has previously been used as a USP9X inhibitor tool compound (Peterson et al., 2015;Ritorto et al., 2014). FT709 competes with an active site probe (HA-UbC2Br) with an IC 50 of ∼0.5 µM and ∼5 µM when applied to MCF7 breast cancer cell extracts and to intact MCF7 cells, respectively (Fig. 3,and Fig. S2). Immunoprecipitation from cell lysates labeled with the active site probe HA-UbC2Br revealed that USP9X is uniquely sensitive to this compound, within a set of 21 DUBs quantified by mass spectrometry (Fig. 3,G and H;and Fig. S2).
Acute inhibition with FT709 recapitulates gene deletion of USP9X in HCT116 cells, leading to reduction of ZNF598 (Fig. 4 A). We conducted a wider survey of protein expression following USP9X inhibition through quantitative mass spectrometry ( Fig. 4 B and Table S1). Among a small number of proteins that decrease by more than twofold following inhibitor treatment, known USP9X substrates are prominent. These include the (peri)-centrosomal proteins PCM1, CEP55, and CEP131 (Li et al., 2017;Wang et al., 2017) and the mitotic kinase TTK, also known as monopolar spindle 1 kinase (Fig. 4, B and C; Chen et al., 2018). ZNF598 is also found within this cohort, in alignment with our Western blot analysis (Fig. 4, A and B). Intriguingly, a second RING E3-ligase, MKRN2, which has been linked to ribosome stalling, is similarly identified as a clear outlier (Fig. 4, B and C; Hildebrandt et al., 2019). We tested the interaction between USP9X and Makorins by expression and immunoprecipitation of MKRN2-FLAG or FLAG-MKRN1 in HEK293 cells (Fig. S3 A). Both HEK293T cells were transfected with ZNF598-FLAG or FLAG alone (pCMV-Tag2B), and cell lysates were subjected to immunoprecipitation (IP) with FLAG-antibody-coupled agarose beads. IPs were probed alongside 5% of the input as indicated, representative of two independent experiments. Arrow indicates ZNF598-FLAG; *, a nonspecific band.
Makorins coimmunoprecipitate with USP9X, ZNF598, and the ribosomal subunit eS10. Accordingly, MKRN2 is reduced in USP9X −/0 cells with no corresponding reduction in mRNA levels (Fig. 4,D and E), reflecting decreased protein stability (Fig. S3 B). Note that ZNF598 and MKRN2 are the only RING E3-ligases contained within the proteomic dataset (>6,000 proteins) that show this magnitude of response to USP9X inhibition (Table S1). However, we could detect MKRN1 by Western blotting, and this is similarly destabilized by loss of USP9X in HCT116 cells (Fig. S3, B-D).
We next used HEK293 cells to ask if the effects of USP9X ablation upon ZNF598 and MKRN2 could be extended to another cell type. We used a set of four individual gRNAs designed to target USP9X. They were inserted in an expression plasmid that also codes for Cas9 and a puromycin resistance cassette (px459-pSpCas9[BB]-2A-Puro-v2). Plasmids were transfected either individually or as a pool, and cell populations were harvested after 168 h of selection with puromycin. USP9X protein expression was effectively ablated in cells transfected with the pool of gRNAs, leading to a correspondingly stark reduction in ZNF598 and MKRN1/2 levels. ZNF598 loss, achieved through the same transient CRISPR/Cas9-based approach, showed no reciprocal effect on USP9X levels ( Fig. 5 A), consistent with mRNA data presented in Fig. 2 The HEK293 cells used in this study (HEK293-Flp-In TREX GFP-P2A-[K AAA ] 21 -P2A-RFP) have been engineered to express a , and USP9Y (normalized to Actin) were performed with cDNA derived from the indicated cell lines. The mean of three independent biological replicates is shown, and error bars indicate the standard deviation. (C) HCT116 or HCT116 USP9X −/0 cells were transfected with 0, 0.2, 0.4, 0.8, or 1.6 µg HA-ZNF598 and 0.2 µg GFP as a transfection control, and lysates analyzed by immunoblot with the indicated antibodies. Graph shows HA-ZNF598 relative to cotransfected GFP. Panel is representative of three experiments. (D) HCT116 or HCT116 USP9X −/0 cells were transfected with 0.2 µg HA-ZNF598 and treated for the indicated times with 100 µg/ml cycloheximide (Chx). Lysates (8 µg for HCT116 and 20 µg HCT116-USP9X −/0 ) were probed with the indicated antibodies. Graph represents the results from four independent experiments. Error bars represent the standard deviation. TPS, total protein stain.  (D) or intact MCF7 cells (E) were incubated with FT709 (30 min at 25°C for cell extracts, 3 h at 37°C for cells) at the indicated concentrations. Cells were lysed, and extracts were incubated with 0.1 µg HA-UbC2Br probe for 5 min at 37°C, followed by SDS-PAGE analysis. Samples were immunoblotted with USP9X and HA antibodies as indicated. Arrow indicates HA-probe labeled band corresponding to the USP9X~Ub probe adduct. Modification of USP9X with a ubiquitin probe (USP9X~Ub) was lost with increasing concentrations of inhibitor. (F) Quantitation of Western blots shown in D and E. (G and H) HA-based immunoprecipitation of HA-UbC2Br probe-labeled DUBs from cell lysates incubated first with DMSO, 1 or 10 µM FT709, for 1 h at 37°C. Immunoprecipitated proteins were eluted and either analyzed side by side with total lysate samples by immunoblotting (TL, total lysate; EL, eluate) or subjected to mass spectrometry-based quantification in three technical replicates. Differences in DUB-probe binding were quantified for 21 identified DUBs and normalized relative to DMSO control (error bars represent standard deviation of three technical replicates). See Fig. S2 for uncropped immunoblots. conc, concentration.
reporter system for terminal ribosomal stalling (Juszkiewicz and Hegde, 2017). The reporter cassette contains GFP (N-terminal) and RFP (C-terminal) separated by a FLAG-tagged stalling reporter (SR) incorporating a polyA stretch of twenty-one codons (K AAA ) 21 (Fig. 5 B). This is flanked by viral P2A sequences, at which ribosomes skip formation of a peptide bond, without interrupting translation elongation. Consequently, unimpaired translation generates 3 proteins (GFP, FLAG-SR, RFP) in equal  amounts. Stalling at the FLAG-SR aborts translation before RFP synthesis, leading to a substoichiometric RFP:GFP ratio. Failure to effectively respond to stalled ribosomes allows eventual readthrough and a consequent rise in the RFP:GFP ratio that can be assessed by fluorescence-activated cell sorting (Fig. 5 B, schematic diagram; Juszkiewicz and Hegde, 2017). As reported previously, an isogenic reporter cell line, in which the ZNF598 gene has been deleted, shows an enhanced RFP:GFP ratio when compared with parental cells consistent with read-through, due to failure of the ribosomal stalling response (Juszkiewicz and Hegde, 2017). Using the pooled USP9X gRNA cells that show highly reduced levels of both USP9X and ZNF598, we could recapitulate this phenomenon (Fig. 5 C). We also analyzed three of the cell populations treated with individual gRNAs as represented in Fig. 5 A that show varying effectiveness for USP9X depletion. Guide 1 serves as a control because it was ineffective in editing the USP9X gene and accordingly shows no change in the RFP:GFP ratio. Guide 4 generates two distinct populations with ∼50% showing an enhanced RFP:GFP ratio, while Guide 7 shows a uniform enhancement of this ratio reminiscent of the effect of ZNF598 deletion and comparable to the pooled gRNA transfected sample (Fig. 5 C). Importantly, we were able to recapitulate polyA read-through with FT709, implicating USP9X enzymatic activity, and providing a pharmaceutical approach to counteract ribosomal stalling (Fig. 5 D). Chemical inhibition leads to a smaller reduction in ZNF598 levels but was again accompanied by parallel reductions in MKRN1 and 2, consequent to decreased protein stability (Fig. 5,E and F;and Fig. S3 E). We propose that USP9X inhibition may promote read-through by a combined effect on each of these RING E3 ligases linked to the ribosomal quality control pathway. Other cell types show a similar pattern of reduction in centrosomal and ribosomal stalling-associated proteins following application of FT709 (Fig.  S3 F). In A549 and U2OS cells, FT709-dependent loss of ZNF598 is minor compared with MKRN2.

Conclusions
USP9X is for the most part a nonessential DUB family member that is nevertheless expressed at relatively high levels (Behan et al., 2019;Clague et al., 2015). Multiple biological functions have been ascribed to USP9X that include roles in apoptosis, Wnt signaling, and mitotic checkpoint control (Nielsen et al., 2019;Schwickart et al., 2010;Skowyra et al., 2018). Our proteomics data most strongly support previously established links to centrosome biology (Han et al., 2019;Li et al., 2017;Wang et al., 2017). Here, we reveal a new biological role for USP9X in the resolution of stalled ribosomes, which is supported by unbiased proteomics. We propose that this is principally related to its governance of ZNF598 and Makorin ubiquitin E3 ligase levels (Hildebrandt et al., 2019). It is possible that USP9X could also play a more direct role in the deubiquitylation of ribosomal subunits themselves during resolution of stalling. However, this is difficult to unravel from effects upon their ubiquitin conjugation by the E3 ligases described here. Moreover, other DUBs (USP21, OTUD3) have recently been linked to this function (Garshott et al., 2020). The nonuniform dynamics of ribosomal processing, duration, and resolution of stalling may have important implications for protein folding, mRNA turnover, and the integrated stress response (Harding et al., 2019;Collart and Weiss, 2020). Recent studies have also shown that ribosomal collisions can result in +1 frame-shifting when the no-go RNA decay pathway is compromised (Simms et al., 2019). Our introduction of a highly specific USP9X tool compound inhibitor will enable further enquiry into pathways previously linked to USP9X, which should now include global profiling of protein translation.

Materials and methods
Chemical compound FT709 used in this study was prepared by the procedures described in detail previously (Follows et al., 2020).

Transfection
For transient transfection, 2 µg total DNA (per well of a six-well plate) was transfected using Genejuice (Novagen) according to the manufacturer's instructions. Cells were harvested 24 or 168 h after transfection.

DUB biochemical assay
The assay was performed in a final volume of 6 µl in assay buffer containing 20 mM Tris buffer, pH 8, 0.03% bovine γ-globulin, 0.01% Triton X-100, and 1 mM glutathione. Nanoliter quantities of 10-point, threefold serial dilutions in DMSO were predispensed into 1,536 assay plates for a final top concentration between 25 and 33.3 mM and subsequent half-log dilutions. 2× DUB (0.025 nM final concentration) was added and preincubated for 30 min at room temperature. 2× ubiquitin (Ub)rhodamine (25 nM final concentration) was added to initiate the reaction. Fluorescence readings (excitation: 485 nm, emission: 535 nm) were acquired over 12 min (Envision reader). The slope of the data from each point was used to determine IC 50 .
For all assays, data were reported as percent inhibition compared with control wells. IC 50 values were determined by curve fitting of the standard four-parameter logistic fitting algorithm included in the Activity Base software package IDBS XE Designer Model205. Data were fitted using the Levenburg-Marquardt algorithm.
Meso Scale Discovery (MSD) assay for CEP55 BxPC3 cells were seeded in 96-well plates and exposed to FT709 (20 µM top concentration, 1:2 serial dilutions) for 6 h. Cell lysates were prepared in RIPA buffer and stored at −80°C until analysis. Samples were analyzed by an MSD ELISA assay (Pacific Biolabs) using a CEP55 antibody (Novux; 1:500 dilution in PBS) captured overnight at 4°C, 30 µl lysates per well, 30 µl of CEP55 antibody (CST; 81693) diluted 1:2,000 in 1% blocker A/PBS and 30 µl per well of a 1:4,000 diluted goat anti-rabbit sulfo-tag, 1% blocker A/PBS. Plates were read on a MSD Sector Imager 2400. Results were transformed as percentage DMSO controls and curves fitted using a nonlinear regression to determine the IC 50 . Results presented are based on duplicate values.
DUB profiling assays using Ub-based active site-directed probes Molecular probes based on the ubiquitin scaffold were generated and used essentially as described (Altun et al., 2011;McGouran et al., 2013). In brief, HA-tagged ubiquitin bromoethyl (HA-UbC2Br) was synthesized by expressing the fusion protein HA-Ub 75 -intein-chitin-binding domain in Escherichia coli BL21 strains (Borodovsky et al., 2002). Bacterial lysates were prepared and the fusion protein purified over a chitin-binding column (NEB). HA-Ub 75 -thioester was obtained by incubating the column material with mercaptosulfonate sodium salt overnight at 37°C. HA-Ub 75 -thioester was concentrated to ∼1 mg/ml using 3,000 molecular weight filters (Sartorius) and then desalted against PBS using a PD10 column (GE Healthcare). 500 µl of 1-2 mg/ml of HA-Ub 75 -thioester was incubated with 0.2 mmol of bromo-ethylamine at pH 8-9 for 20 min at room temperature, followed by a desalting step against phosphate buffer, pH 8.0, as described above. Ub probe material was concentrated to ∼1 mg/ ml, using 3,000 molecular weight filters (Sartorius), and kept as aliquots at −80°C until use.

DUB competition assays in cell extracts and in cells (in situ)
Crude MCF7 cell extracts were prepared as described previously using glass-bead lysis in 50 mM Tris, pH 7.4, 5 mM MgCl 2 , 0.5 mM EDTA, 250 mM sucrose, and 1 mM DTT (Borodovsky et al., 2002;McGouran et al., 2013). For experiments with crude cell extracts, 50 µg of MCF7 cell lysate was incubated with different concentrations of FT709 for 1 h at 37°C, followed by addition of 1 µg HA-UbC2Br and incubation for 5 min at 37°C. Incubation with Ub-probe was optimized to minimize replacement of noncovalent inhibitor FT709 by the covalent probe. Samples were then subsequently boiled in reducing SDS sample buffer, separated by SDS-PAGE, and analyzed by Western blotting using anti-HA (1:2,000), anti-USP9X (Cell Signaling Technology; D4Y7W 14898 at 1:1,000), or β-actin (1:2,000) antibodies. 5 × 10 6 intact MCF7 cells were incubated with different concentrations of inhibitors in cultured medium for 4 h at 37°C, followed by glass bead lysis, labeling with HA-UbC2Br probe, and analysis by SDS-PAGE and Western blotting as described above.
Mass spectrometry-based DUB inhibitor profiling assays Ub-probe pulldown experiments in the presence of different concentrations of the inhibitor FT709 were performed essentially as described (Altun et al., 2011;McGouran et al., 2013) with some modifications. In brief, immunoprecipitated material from 500 µg to 1mg of MCF-7 cell crude extract was subjected to insolution trypsin digestion and desalted using C18 SepPak cartidges (Waters) based on the manufacturer's instructions. Digested samples were analyzed by nano-ultra performance liquid chromatography (UPLC)-mass spectrometry (MS)/MS using a Dionex Ultimate 3000 nano UPLC with EASY spray column (75 µm × 500 mm, 2 µm particle size; Thermo Fisher Scientific) with a 60-min gradient of 0.1% formic acid/5% DMSO to 0.1% formic acid/35% acetonitrile/5% DMSO at a flow rate of ∼250 nl/min (∼600 bar/40°C column temperature). MS data were acquired with an Orbitrap Q Exactive High Field (HF) instrument in which survey scans were acquired at a resolution of 60,000 at 400 m/z, and the 20 most abundant precursors were selected for collision-induced dissociation fragmentation. From raw MS files, peak list files were generated with MSConvert (Proteowizard V3.0.5211) using the 200 most abundant peaks/spectrum. The Mascot (V2.3, Matrix Science) search engine was used for protein identification at a false discovery rate of 1%, mass deviation of 10 ppm for MS1 and 0.06 D (Q Exactive HF) for MS2 spectra, Cys carbamidylation as fixed modification, Met oxidation, and Gln deamidation as variable modification. Searches were performed against the UniProtKB human sequence database (retrieved October 15, 2014). Label-free quantitation was performed using MaxQuant Software (version 1.5.3.8), and data were further analyzed using GraphPad Prism software (v7) and Microsoft Excel.

Deep proteome workflow
Protein extracts (1.2-1.5 mg) containing SDS were reduced with 5 mM dithiothreitol, alkylated with 20 mM iodoacetamide, and then subjected to methanol/chloroform extraction. Protein pellets were resuspended in 6 M urea by vortexing and sonication, then diluted to a final concentration of 1 M before in-solution digestion with 0.2 µg/µl trypsin (sequencing grade, Promega) overnight at 37°C. Off-line high-pH reverse-phase prefractionation was performed on the digested material as previously described (Davis et al., 2017), with the exception that eluted peptides were concatenated down to 10 fractions. Peptide fractions were analyzed in technical replicates by nano-UPLC-MS/ MS using a Dionex Ultimate 3000 nano UPLC with EASY spray column (75 µm × 500 mm, 2 µm particle size; Thermo Fisher Scientific) with a 60-min gradient of 2% acetonitrile/0.1% formic acid/5% DMSO to 35% acetonitrile/0.1% formic acid/5% DMSO at a flow rate of ∼250 nl/min. MS data were acquired with an Orbitrap Q Exactive HF instrument in which survey scans were acquired at a resolution of 60,000 at 200 m/z, and the 20 most abundant precursors were selected for higher-energy collisional dissociation fragmentation with a normalized collision energy of 28.

Data analysis
All raw MS files from the biological replicates of the SILACproteome experiments were processed with the MaxQuant software suite, version 1.5.3.8, using the Uniprot database (uniprotHumanUP000005640.fasta, retrieved July 2015) and the default settings (Tyanova et al., 2016). The minimum required peptide length was set to six amino acids, and two missed cleavages were allowed. Cysteine carbamidomethylation was set as a fixed modification, whereas oxidation and N-terminal acetylation were considered as variable modifications. Pro-teinGroup text files were further processed using Excel (see Table S1), and the log 2 of the normalized ratios was plotted using JMP software (version 13.0.0). The mass spectrometry proteomics data have been deposited to the ProteomeXchange Consortium via the PRIDE partner repository with the dataset identifier PXD018662.
Online supplemental material Fig. S1 shows additional Western blot validation data for ZNF598 antibodies, supplementing Fig. 2. Fig. S2 shows IC 50 data for FT709 across a panel of DUBs and the uncropped version of HA immunoblots shown in Fig. 3, D, E, and G. Fig. S3 shows supplementary data supporting the interaction between USP9X, and MKRN1 and 2; and cycloheximide chase and quantitative RT-PCR data supporting a role for USP9X in the stabilization of MKRN1 and 2. Table S1 shows proteomic data for Fig. 4 B.