Title: Identification of novel human USP2 inhibitor: might involve in SARS- CoV-2 papain-like (PLpro) protease deubiquitination activity

The ubiquitin-specific protease 2 (USP) belongs to the family of deubiquitinases and plays a critical role in tumors cells’ survival and therefore signifies an important therapeutic target. Previous studies have indicated promising efficacies of potent human USP2 inhibitors including, thiopurine analogues against SARS-CoV papain-like proteases (PLpro). The PLpro have significant functional implications in the innate immune response during SARS-CoV-2 infection and considered an important antiviral target. Both proteases share strikingly similar USP fold with right-handed thumb–palm–fingers structural scaffold and conserved catalytic triad Cys-His-Asp/Asn. In this urgency situation of COVID-19 outbreak, there is a lack of in-vitro facilities readily available to test SARS-CoV-2 inhibitors in whole-cell assays. Therefore, we adopted an alternate route to identify potential USP2 inhibitor through integrated structure-based virtual screening efforts. After a subsequent virtual screening protocol, the best compounds were selected and tested. The compound Z93 showed significant IC50 value against Jurkat (9.67 μM) and MOTL-4 cells (11.8 μM). The binding mode of Z93 was extensively analyzed through molecular docking, followed by MD simulations, and molecular interactions were compared with SARS-CoV-2. The relative binding poses of Z93 fitted well in the binding site of both proteases and showed consensus π-π stacking and H-bond interactions with histidine and aspartate/asparagine residues of the catalytic triad. These results led us to speculate that compound Z93 might be the first potential chemical lead against SARS-CoV-2 PLpro, which warrants in-vitro evaluations.


Introduction
Leukemia is a hematological malignancy originate in the bone marrow and results in abnormally increased production of improperly functioning white blood cells. It causes prognosis based acute or chronic illness commonly associated with lymphocyte or myeloid cells. With the growing availability of treatment options, the morbidity of acute lymphoblastic leukemia (ALL) is responsible for mortality in 20% pediatric and 50% adult cases. T-cell associated ALL (T-ALL) is aggressive from of ALL with a prevalence of 15% pediatric and 25% adult patients 1 .
Deubiquitinating peptidases (DUBs) are a broad class of proteases that dissociate ubiquitin from proteins and disrupt proteasomal and lysosomal degradation of its substrate proteins. Many members of DUB have found to play their role in the progression of different cancers 2 . Such as, USP2 is specific deubiquitinase of a proto-oncoprotein cyclin D1 and stabilizes it via restriction of ubiquitin-dependent degradation 3 . This affirms that the type of cancers which depend on cyclin D1 for their growth, e.g., T-ALL 4 , can be suppressed by using USP2 inhibitors.
The active site of human USP2 shares a similar architecture with SARS-CoV-2 PLpro, which has deubiquitinating activity and directs important implications on viral replication and pathogenesis 5 . The structural analysis revealed a similar active site of SARS-CoV-2 PLpro, including the signature catalytic triad (Cys-His-Asp/Asn) located at the interface of the thumb and Palm sub-domains 6 . Various studies have reported the efficacy of USP2 potent inhibitors including thiopurine analogs on SARS 7,8 and MERS 9 which were found to interact with the signature catalytic triad (IC50 values ranges from 5 to 21.6 μM). With the recent COVID-19 outbreak which resulted in more than 126,000 deaths worldwide 10 , there is a dearth of in-vitro facilities available to test SARS-CoV-2 inhibitors in whole-cell assays. Therefore, an alternate route might be the selective inhibition of USP2 by small molecules which may plausibly lead to the identification of SARS-CoV-2 PLpro inhibitors due to structurally conserved similar active site fold. Based on these assumptions, the present study focuses on the identification of potential USP2 inhibitor through in-depth structure-based virtual screening workflow. The computational work was carried out in sequential way followed by molecular dynamics (MD) simulation and interaction energetics using AMBER simulation package. The identified compounds were purchased and tested on Jurkat and MOLT-4 cells against T-ALL. The binding mode analysis was extensively performed to examine the molecular interactions together with SARS-CoV-2 PLpro. To the best of our knowledge, the utilized method is the first study reported in on-going COVID-19 outbreak where we propose SARS-CoV-2 PLpro inhibitor with a proof of human USP2 inhibition based on highly similar structural attributes.

Data collection and docking structure preparation
For the structure-based study, the co-crystalized structure of human USP2 complex with 6-thioguanine (5XU8) was utilized for docking. The structure was prepared accordingly, as described previously 11,12 . Briefly, all heteroatoms were removed, hydrogens and charges were added into USP2. The structure was minimized for the 1000 steepest descent steps at root means square gradient of 0.02 having an update interval of 10 and using the AMBER ff14SB force field. A chemical library of ~4.2 million molecules was retrieved from ZINC. The library was reduced after applying various drug-like parameter, which mainly includes the Lipinski's Rule of five (with 2 violations), oral bioavailability rule including: rotatable bonds ≤10 and TPSA ≤140 Å 2 . The resulted dataset was utilized for structure-based virtual screening workflow.

Virtual screening workflow
Before subsequent docking runs, the docking grid was defined by selecting the specific area around the catalytic triad residues (Cys276, His557, and Asn574). The filtered database was finally uploaded into the Mcule drug discovery platform and AutoDock Vina was utilized to screen the docking library 13 . Diversity filter was also applied in workflow, which reduced the docking library by removing the closest analogs and maximized the coverage of chemical space to identify active scaffolds. Autodock Vina was selected among the top-ranked scoring functions in terms of docking power and screening test according to CASF (Comparative Assessment of Scoring Functions) benchmark 2016 14 . The compounds were ranked accordingly on docking score which is based on the Vina empirical scoring function that approximates the ligand binding affinity in kcal/mol. The top hits were analysed for ADMET criteria and top hits were analysed in Chimera in terms of binding poses and interactions with the catalytic triad. The overall stability of the best complexes was analysed over a period of 20ns using AMBER 18 simulation package 15 . We used the same MD simulation protocol as described previously 12,16 . Briefly, the Antechamber was utilized to generate the general AMBER force field (GAFF) parameters for the studied ligands. After a stepwise minimization, heating and equilibration in explicit solvent environment (TIP3), a production run of 20ns was performed at standard pressure (p = 1.0 atm) and temperature (T = 300 K). The binding free energies (ΔGtol) of USP2 complexed with screened compounds were calculated using the MM-GBSA method of AMBER 18. The compounds with significant interaction profile were purchased based on availability and price. The tested compounds were later redocked using AD Vina on recently resolved crystal structure of SARS-CoV-2 PLpro (6W9C) and compared the molecular interactions. For molecular docking, docking grid was setup around the catalytic triad (Cys112, His272 and Asp286). The same MD simulation protocol was utilized for SARS-CoV-2 PLpro complexes.

MTT Assay
The cells were cultured in RPMI-1640 medium accompanied with 10% heat-inactivated FCS, 100 µg/ml streptomycin and 100 U/ml penicillin and incubated in a humidified atmosphere at 37°C and 5% CO2. The cell suspension was adjusted to a final concentration of 3×10 5 cells/ml and 100 µl cell suspension dispensed per well. The plate was incubated at 37 °C for 12 h in 5% CO2 incubator. The stock solutions (10 mM) of the test compound and cisplatin were prepared in DMSO and the test concentrations were prepared by diluting the stocks with the growth media. The test dilutions (100 µl) were added to the wells giving final concentrations of 1, 5, 10, 20 and 40 µM. MTT reagent (20 µl, 5 mg/ml in PBS) was added to each microwell after 48 h, followed by incubation for 5 h. The contents were solubilized with 100 µl 10% sodium dodecylsulfate in 10 mM HCl. The absorbance was measured after 12 h with the help of BioRad PR 4100 microplate reader at a wavelength of 570 nm. Percentage viability was calculated by using following relation, and dose-response curves were plotted. The results represent mean±SD of six readings; three determinations of two independent experiments 11,17 Where, As, Ab and Ac are absorbance of sample, blank and control, respectively.

Virtual screening workflow validation using set of known USP2 binders
To validate the docking protocol, a total of 12 reported USP2 inhibitors were subjected to an established virtual screening pipeline. The IC50 values of inhibitors against USP2 were converted to pIC50 values (-log of IC50) and plotted against the AD Vina docking scores and MMGBSA total binding free energy scores calculated by Amber-mmgbsa method. The corresponding values are listed in Table S1 and correlation plot between pIC50 values and docking scores is displayed in Figure 1. The AD Vina docking scores showed weak correlation with the experimental pIC50 values (R 2 = 0.21), whereas MMGBSA scores showed better correlation (R 2 = 0.53). Here, we speculated that the weak correlation between AD Vina docking scores and experimental pIC50 values might be indorsed by side chains conformational flexibility of binding site residues, which were adjusted during MD simulations in explicit solvent environment. Hence, the MMGBSA total binding free energy showed better correlation with experimental values, as it incorporates all electrostatic and van der Waals (vdW) interactions. Overall, combining AD Vina followed by MMGBSA calculations suggested a reliable approach to select the compounds against USP2 (together with expected false positives and negatives).

Virtual screening
To identify potential inhibitors against UPS2, structure-based virtual screening was employed. A substantial number of compounds were eliminated through several ADMET filters, which include drug-like and toxicity assessment. A total of 52 compounds were selected based on the AD Vina score from a known USP2 binder (CHEMBL3392809, AD Vina score: -7.3 kcal/mol) as cut-off value. Among these hits, each compound was investigated for molecular interactions with signature catalytic triad (Cys276, His557, and Asn574), which further reduced the list to 14 compounds. The final selection of compounds was made after extensive MD simulation analysis of these 14 complexes which included, 1) constant interaction with any of the catalytic triad residues, 2) lowest MMGBSA total binding free energy values and 3) stable conformation of ligand over a period of 20 ns ( Figure S1). A total of 3 compounds were finally selected and only one compound (ZINC9325709 denoted as Z93) was purchased for biological evaluations based on the specific scaffold, its availability and price (Figure 2 and Table S2). The detailed ADMET properties are tabulated in (Table S3).

Figure 2:
Chemical structure of selected compounds from in silico screen. CHEMBL3392809, potent USP2 inhibitor is included as reference.

Biological evaluation
Jurkat and MOLT-4 cell lines were used to examine cytotoxicity of selected USP2 inhibitor (Z93) against T-ALL while cisplatin was used as a control. Both selected cell lines display versatile characteristics to study cytotoxicity of drugs against T-ALL. The dose-response curve is presented in Figure 3. The

Binding mode of Z93 with USP2 and SARS-CoV-2 PLpro
Recent studies have determined the crystal structure of MERS, SARS-CoV and CoV-2 PLpro, which corroborated the similar palm-thumb-fingers structural scaffold and conserved catalytic triad Cys-His-Asn/Asp like in human USP2, albeit the low sequence homology 6,18,19 . After superimposition, the corresponding domain architecture of both PLpro displayed strikingly similar representation of conserved USP fold and orientation of catalytic triad residues (Figure 4A-4D). These findings led us to speculate that identified Z93, a potential USP2 inhibitor might be able to inhibit the SARS-CoV-2 PLpro because both proteases share a similar catalytic triad. Overall, the integrated virtual screening identified 14 hits which were reduced to 3 after analysing the binding free energies ( Figure 4E). To gain further insight into the binding mode, we performed molecular docking followed by MD simulations against both PLpro and most favourable binding conformation of active compounds, Z93 was obtained from the largest cluster after 20ns production run through clustering analysis. The RMSD plot indicated the stable backbone conformation of binding site residues with docked Z93 ( Figure 4F). Overall, both complexes fluctuated in the start and remained stable for last 20 ns indicated that Z93 adopted more favourable conformation (Figure 4G and 4H). Despite the similar docked sites on both PLpro, the binding poses were found different. More insights revealed the consensus binding interactions of Z93 with the catalytic triad in both PLpro. In UPS2, the major interaction was established due to stable stacking interaction between terminal benzene of Z93 and His557 of catalytic triad. Moreover, the benzothiazole moiety of Z93, apart from establishing a H-bond with the side chain nitrogen atom of Asn574 also formed extensive hydrophobic interactions with Thr572, Thr559, Ser550 and Tyr561 ( Figure 4I). In relation to SARS-CoV-2 PLpro, the Z93 also found interacted with the catalytic triad where the benzothiazole moiety was positioned to make direct stacking interaction with His272, and terminal benzene was positioned to form stacking interaction with Trp106 ( Figure 4J). Additionally, the nitrogen atom next to benzothiazole established H-bond with the sidechain oxygen atom of conserved Asp286 of catalytic triad. The predicted binding site was close to that found for SARS-CoV PLpro 7 .
Overall, the stable binding mode of Z93 in both proteases was mainly governed due to stacking interactions and complexes were remained converged with an RMSD value of less than 0.5 Å. Moreover, the estimated total binding energies after 20 ns MD simulation for Z93 binding to USP2 and SARS-CoV-2 PL pro were -38.07 and -33.05 kcal/mol respectively. Z93 participated well through vdW interactions as evident from strong stacking interactions with relatively less electrostatic contributions (Table S4).

Discussion
Ubiquitin carboxyl-terminal hydrolase 2 (USP2) is a highly conserved protein present in most eukaryotes. It is found to be involved in intracellular proteasomal degradation, cell-cycle regulation and stress response. The overexpression of USP2 has been found involved in progression and metastasis several cancers, e.g., triple negative breast cancer, colorectal cancer and non-Hodgkin's lymphoma 20 . Most of the proteasome inhibitors usually pose extensive off-target effects and an alternative approach of USP2 inhibition can selectively target proteasomal degradation of its multiple oncogenic substrates with least effect on noncancerous cells 21,22 .
It antagonizes ubiquitin-dependent degradation of cyclin D1 resulting in its stabilization. The cells involving cyclin D1 independent cell-cycle progress, for example, normal human fibroblasts or cancer cells that don't show cyclin D1 expression are not affected by the inactivation of USP2. These evidences suggest that inhibition of USP2 in cancer cells which express cyclin D1 for their growth could induce growth suppression 3 . ML364, an already reported inhibitor of USP2, has close structural similarity with Z93, with both sharing 2-((4-methylphenyl)sulfonamido)benzamide residue. ML364 induces cyclin D1 degradation resulting in cell-cycle arrest, indicating key role in progression of non-Hodgkin's lymphoma and colorectal cancer 22 . Cyclin D1 is necessarily involved in cell-cycle progression from G1 to S phase and its overexpression is evident in many human cancers 23 . It is found to be upregulated in most of the ALL patient while its expression is not affected in AML patients as compared to the normal controls 4 . Reduced survival rates have been observed in ALL patients with higher expression of cyclin D1, suggesting its involvement in the progression of aggressive forms of ALL 23 .
6-Mercaptopurine (6MP) and 6-thioguanine (6TG) has been reported effective against the treatment of various cancer including leukemia. These both compounds are hypoxanthine-guanine phosphoribosyltransferase inhibitors and possess purine residue 24 . Another potential effect of these compounds have been reported against MERS-CoV and SARS-CoV PLpro 9 . The benzothiazole moiety of Z93 resembles 6MP and 6TG purine, having [4.3.0] bicyclic system. In another study, indole based 3chymotrypsin-like protease (3CLpro) inhibitor of SARS-CoV have been reported with low micromolar range 25 . These analogies strongly potentiate the hypothesis that Z93 can inhibit SARS-CoV-2 by antagonizing its PLpro.
The modelling studies indicated the consensus molecular interactions governed by Z93 with both proteases. There was no major structural deviation observed over a period of 20 ns in both docked complexes. With both proteases, Z93 interacted with the His and Asp/Asn of catalytic triad in a similar way, albeit with different binding mode and revealed favourable total binding free energies. The Z93 bound conformation was found more stable with SARS-CoV-2 PLpro due to an additional staking interaction with Trp106, therefore showed more favourable vdW energy contributions, as compared to Z93/USP2. Conclusively, based on the in-vitro and repurposed modelling analysis, it is entirely possible to develop selective, noncovalent inhibitors against SARS-CoV-2 PLpro by using structure-based approach against USPs involved in disease processes.