The SARS-CoV-2 Nsp3 macrodomain reverses PARP9/DTX3L-dependent ADP-ribosylation induced by interferon signaling

SARS-CoV-2 nonstructural protein 3 (Nsp3) contains a macrodomain that is essential for coronavirus pathogenesis and is thus an attractive target for drug development. This macrodomain is thought to counteract the host interferon (IFN) response, an important antiviral signalling cascade, via the reversal of protein ADP-ribosylation, a posttranslational modification catalyzed by host poly(ADP-ribose) polymerases (PARPs). However, the main cellular targets of the coronavirus macrodomain that mediate this effect are currently unknown. Here, we use a robust immunofluorescence-based assay to show that activation of the IFN response induces ADP-ribosylation of host proteins and that ectopic expression of the SARS-CoV-2 Nsp3 macrodomain reverses this modification in human cells. We further demonstrate that this assay can be used to screen for on-target and cell-active macrodomain inhibitors. This IFN-induced ADP-ribosylation is dependent on PARP9 and its binding partner DTX3L, but surprisingly the expression of the Nsp3 macrodomain or the deletion of either PARP9 or DTX3L does not impair IFN signaling or the induction of IFN-responsive genes. Our results suggest that PARP9/DTX3L-dependent ADP-ribosylation is a downstream effector of the host IFN response and that the cellular function of the SARS-CoV-2 Nsp3 macrodomain is to hydrolyze this end product of IFN signaling, rather than to suppress the IFN response itself.

SARS-CoV-2 nonstructural protein 3 (Nsp3) contains a macrodomain that is essential for coronavirus pathogenesis and is thus an attractive target for drug development. This macrodomain is thought to counteract the host interferon (IFN) response, an important antiviral signalling cascade, via the reversal of protein ADP-ribosylation, a posttranslational modification catalyzed by host poly(ADP-ribose) polymerases (PARPs). However, the main cellular targets of the coronavirus macrodomain that mediate this effect are currently unknown. Here, we use a robust immunofluorescence-based assay to show that activation of the IFN response induces ADP-ribosylation of host proteins and that ectopic expression of the SARS-CoV-2 Nsp3 macrodomain reverses this modification in human cells. We further demonstrate that this assay can be used to screen for on-target and cell-active macrodomain inhibitors. This IFN-induced ADP-ribosylation is dependent on PARP9 and its binding partner DTX3L, but surprisingly the expression of the Nsp3 macrodomain or the deletion of either PARP9 or DTX3L does not impair IFN signaling or the induction of IFNresponsive genes. Our results suggest that PARP9/DTX3Ldependent ADP-ribosylation is a downstream effector of the host IFN response and that the cellular function of the SARS-CoV-2 Nsp3 macrodomain is to hydrolyze this end product of IFN signaling, rather than to suppress the IFN response itself.
The SARS-CoV-2 pandemic has highlighted the need for an improved understanding of coronavirus pathogenesis for the development of novel antiviral strategies. The interferon (IFN) response is a central component of innate immunity and essentially precludes viral infection, as long as it is properly activated (1,2). Therefore, successful replication of a virus within host cells requires active suppression or evasion of the host IFN response, which is often mediated by multiple viral proteins acting via separate mechanisms (3).
Type I IFN signaling is initiated upon recognition of viral nucleic acids in the cytoplasm, leading to the production and secretion of type I IFNs, such as IFNα and IFNβ, by virusinfected cells, whereas type II IFN, or IFNγ, is secreted by immune cells (1,4). Binding of these cytokines to transmembrane IFN receptors induces activation of JAK kinases such as TYK2, JAK1, and JAK2, leading to phosphorylation and nuclear translocation of transcription factors of the STAT family, mainly STAT1, and subsequent induction of several hundred interferon-stimulated genes (ISGs) (1,2). Among these ISGs are several members of the poly(ADP-ribose) polymerase (PARP) family, which catalyze the posttranslational modification of proteins with ADP-ribose units using NAD + as a substrate (5).
ADP-ribosylation has recently emerged as a critical regulator of the IFN response, modulating central steps of this signalling cascade, both upstream of Type I IFN production and downstream of Type I or Type II IFN receptor binding (6,7). Several IFN-regulated PARPs are also antiviral effectors, either by modifying host proteins involved in protein translation, stress granule formation and intracellular protein trafficking, or via modification and inhibition of viral proteins directly (6). Interestingly, some of these IFN-responsive PARPs, including PARP9, are rapidly evolving in the primate lineage, suggesting that they are engaged in an "arms race" with viral pathogens (8).
Here, we show that activation of Type I or Type II IFN signaling induces ADP-ribosylation of host proteins that can be reversed by ectopic expression of the SARS-CoV-2 Nsp3 macrodomain in human cells. This ADP-ribosylation is dependent on the IFN-responsive PARP9/DTX3L heterodimer, but does not seem to modulate IFN signaling itself, since Nsp3 macrodomain expression or PARP9/DTX3L knockout had no effect on STAT1 phosphorylation or the induction of ISGs. We propose that PARP9/DTX3Ldependent ADP-ribosylation of host proteins, which can be reversed by the SARS-CoV-2 Nsp3 macrodomain, is a downstream effector of the IFN response.

Results
To study the functions of the SARS-CoV-2 Nsp3 macrodomain, we first engineered a sensitive assay to detect IFNinduced ADP-ribosylation in human cells. We transfected human A549 lung adenocarcinoma cells with the RNA mimetic poly(I:C), which induces a complete type I IFN cascade (18,19), or treated these cells with recombinant IFNγ, which induces type II IFN signalling (20,21). As expected, both treatments led to robust phosphorylation of STAT1 on tyrosine Y701 (Fig. 1A), which localized to the cell nucleus (Fig. 1B). Using the Af1521 macrodomain-derived pan-ADPribose binding reagent (Millipore, MABE1016) for immunofluorescence staining and a high-content microscopy setup for robust image analysis and signal quantification ( Fig. S1 and Experimental procedures section), we observed a pronounced increase in a punctate, cytosolic ADP-ribose signal in response to either poly(I:C) or IFNγ treatment (Fig. 1C). Therefore, both type I and type II IFN signaling induce detectable ADPribosylation in human cells.
Next, we generated A549 cell lines ectopically expressing FLAG-tagged SARS-CoV-2 Nsp3 macrodomain using lentiviral vectors and confirmed that nearly 100% of the cells express the macrodomain using anti-FLAG immunofluorescence staining ( Fig. 2A). Strikingly, constitutive macrodomain expression substantially reduced both poly(I:C) or IFNγ-induced ADPribosylation compared with empty vector controls (Fig. 2B). To confirm these results, we generated cells constitutively expressing a catalytically inactive N40A macrodomain mutant (10,13) and repeated the experiment with the inclusion of recombinant IFNα and IFNβ. Again, expression of the WT macrodomain substantially reduced ADP-ribosylation induced by all of the treatments tested (Fig. 2C). In contrast, expression of the N40A mutant had no appreciable effect on IFN-induced ADP-ribosylation (Fig. 2C), indicating that catalytic activity was required for this effect. In these experiments, we observed that the N40A mutant was expressed at lower levels than the WT macrodomain (Fig. S2A), which could, in principle, account for the higher persistence of ADP-ribose signal in the population of cells expressing the mutant protein. Therefore, the analysis in Figure 2C includes only a subset of cells that express similar levels of WT and N40A macrodomain, based on anti-FLAG immunofluorescence intensity (Fig. S2B), which did not alter the original result ( Fig. 2C and Fig. S2, C and D). To further strengthen these observations, we generated A549 cell lines with doxycycline-inducible expression of either WT or N40A mutant macrodomain (Fig. S2E). As expected, IFNγ-induced ADP-ribosylation was indistinguishable between empty vector controls and either of the uninduced cells (Fig. 2D). In agreement with the results above, doxycycline-induced expression of the WT macrodomain reduced ADP-ribosylation, which was once again restored by the N40A mutation (Fig. 2D). As before, the cells used in this analysis were gated to ensure similar expression levels between WT and N40A mutant macrodomain, with no effect on the result (Fig. S2, F and G). Collectively, these data indicate that the Nsp3 macrodomain, when ectopically expressed in human cells, hydrolyzes the ADP-ribosylation of host proteins induced by both type I or type II IFN signaling.
Given the urgent need for antiviral therapies for COVID-19 and the fact that the macrodomain is an attractive therapeutic target (9,(15)(16)(17), we attempted to repurpose compounds that already have regulatory approval as potential Nsp3 macrodomain inhibitors. For this, we performed a structure-based virtual screen of a library of 6365 compounds that have been approved for human use by any regulatory agency in the world, against the deposited crystal structures of the SARS-CoV-2 Nsp3 macrodomain bound to ADP-ribose (PDB 6W02) ( Fig. 3A and methods). Of the final list of 79 compounds of interest, 69 were sourced and tested in biochemical and cellular assays (Table S1). Using thermal shift assays, which measure thermal stabilization of a protein upon ligand binding, we observed that, in contrast to the substantial shift in thermal stability of the Nsp3 macrodomain induced by the ADP-ribose positive control, none of the 69 test compounds showed any evidence of binding to the recombinant macrodomain (Fig. 3B). Similar results were obtained in cellular assays, in which none of the compounds substantially restored IFN-induced ADP-ribosylation levels in macrodomainexpressing cells (Fig. 3C), and the slight effect of atorvastatin observed in the cellular screen was not confirmed in follow-up experiments (Fig. S3).
Interestingly, we observed that some compounds reduced IFN-induced ADP-ribosylation ( Fig. 3C) and one of these was tofacitinib, an inhibitor of the JAK kinases that phosphorylate STAT1 in response to IFN receptor activation (22). As expected, tofacitinib treatment completely blocked STAT1-Y701 phosphorylation in response to IFNγ treatment ( Fig. 4A and Fig. S4A), and this resulted in a complete loss of IFN-induced ADP-ribosylation ( Fig. 4B and Fig. S4A). This suggests that ADP-ribosylation is likely to be catalyzed by an IFN-responsive PARP that is transcriptionally induced by STAT1 complexes. This was corroborated by the fact that olaparib, an inhibitor of DNA-damage activated PARPs 1 and 2 that likely also inhibits PARP3, PARP4, and TNKS1/2 (PARP5a/b) at the relatively high concentration used here (23)(24)(25) had no effect on IFN-induced ADP-ribosylation or on STAT1 phosphorylation (Fig. 4, A and B and Fig. S4A).
One such IFN-responsive PARP is PARP9, which forms a heterodimer with DTX3L and has been shown to participate in IFN-mediated antiviral responses (26,27). We had previously generated PARP9 and DTX3L CRISPR knockout cells in an RPE1-hTERT background (Experimental procedures section and Fig. S4, B and C) and decided to test whether IFN-induced ADP-ribosylation was altered in these cells. As with A549 cells, Nsp3 macrodomain and IFN-induced ADPR  IFNγ treatment induced detectable ADP-ribosylation in RPE1 retinal pigment epithelial cells, confirming that this response is not specific to A549 cells (Fig. 4C). Surprisingly, however, PARP9 KO and particularly DTX3L KO cells displayed severely reduced IFNγ-induced ADP-ribosylation compared with controls ( Fig. 4C and Fig. S4D). To determine if this was indirectly caused by an effect of PARP9 or DTX3L knockout on IFN signaling itself, we quantified STAT1-Y701 phosphorylation levels in response to IFNγ treatment in these cells, which revealed that PARP9 KO or DTX3L KO had no appreciable effect on STAT1 phosphorylation (Fig. 4D). To ascertain if the reduced ADP-ribosylation in PARP9 KO or DTX3L KO cells may be caused by a defect in IFN signaling downstream of STAT1 phosphorylation, we performed RT-qPCR to assess the transcriptional induction of four IFNstimulated genes (ISGs)-OAS1, IRF1, ISG15, and Mx1 (28-31) after treatment with either IFNγ or poly(I:C) (Fig. 4E). As expected, strong transcriptional induction of these ISGs was observed in RPE1 cells after IFNγ or poly(I:C) treatment, but knockout of PARP9 or DTX3L did not affect the induction of any of these genes under any of the conditions tested (Fig. 4E). These data indicate that PARP9 and DTX3L are essential for IFN-induced ADP-ribosylation of host proteins, which occurs downstream of ISG induction.  (Table S1). Mean ± SEM (n = 3). Atorvastatin and tofacitinib (highlighted) are discussed in the main text.
such that macrodomain expression between WT and N40A mutant was comparable (Fig. S2, A-C). Mean ± SEM (n = 6, from three separate experiments), ****p < 0.0001, ***p < 0.001. D, quantification of ADP-ribose immunofluorescence signal intensity in A549 cells transduced either with empty vector control (e.v.) or with lentiviral constructs for doxycycline-inducible expression of either WT macrodomain or catalytically dead N40A mutant, after 24 h treatment with indicated doses of doxycycline and 100 U/ml IFNγ or vehicle control. For FLAG-macrodomain-expressing samples, cells were gated such that macrodomain expression between WT and N40A mutant was comparable (Fig. S2, E-G). Mean ± SEM (n = 7-12, from four separate experiments), ***p < 0.001 and ****p < 0.0001. In light of these results, we decided to determine if Nsp3 macrodomain expression affects host IFN signaling or if its role is also downstream of ISG induction. Consistent with the latter, macrodomain-expressing A549 cells displayed normal levels of STAT1-Y701 phosphorylation by immunofluorescence staining in response to either IFNγ or poly(I:C) treatment (Fig. 5A). Expression of the Nsp3 macrodomain also had no effect on the transcriptional induction of four ISGs in response to either IFNγ or poly(I:C) treatment (Fig. 5B). In agreement with this, macrodomain expression did not impair STAT1-Y701 phosphorylation or the induction of the IFNresponsive PARP9 or DTX3L proteins after treatment with IFNα, IFNβ, IFNγ, or poly(I:C) ( Fig. 5C and Fig. S5, A and B).

Nsp3 macrodomain and IFN-induced ADPR
To investigate the relative kinetics of IFN-induced JAK/ STAT signaling, PARP9/DTX3L protein induction, and ADPribosylation, we performed a series of time-course experiments in response to IFNγ, IFNβ, or poly(I:C) (Fig. 5, D and E and Fig. S5, C-L). As expected, STAT1-Y701 phosphorylation peaked at 30 min in response to recombinant IFNs and slowly faded over time, whereas poly(I:C) induced robust STAT1 phosphorylation only at 24 h due to the delayed production of type I IFN by the transfected cells themselves (Fig. 5E and Fig. S5, C, F, G, J and K). Consistent with our previous observations, macrodomain expression did not affect the magnitude or the kinetics of JAK-dependent STAT1 phosphorylation under any of the conditions tested ( Fig. 5E and Fig. S5, C, F, G, J and K). Crucially, the kinetics of DTX3L protein induction coincided with the formation of macrodomain-sensitive ADP-ribosylation, both of which started increasing from 4 to 6 h after recombinant IFN treatment, much later than the peak of STAT1 phosphorylation (Fig. 5, D and E and Fig. S5, C-L). Collectively, these data indicate that the Nsp3 macrodomain does not affect the host IFN signaling cascade or the induction of interferonstimulated genes, but is instead involved in suppressing downstream biological processes triggered by IFN-induced ADP-ribosylation (Fig. 6).

Discussion
Successful viral evasion or suppression of the host interferon response is key for the establishment of a viral infection and is often mediated by several viral factors acting on multiple host targets (3). The coronavirus Nsp3 macrodomain is thought to represent an important mediator of coronavirus pathogenesis by reversing host antiviral ADP-ribosylation and is thus an attractive drug target (7,9,(15)(16)(17). However, little is known about the molecular targets of IFN-responsive PARPs and how reversal of these modifications by the Nsp3 macrodomain may exert its proviral effects.
We have established a cellular immunofluorescence-based assay that can detect the physiological levels of ADPribosylation induced by both type I and type II IFN signaling in response to the viral RNA mimetic poly(I:C) or recombinant interferon treatments (Figs. 1 and 2), which is consistent with previous studies showing that IFNγ treatment results in detectable ADP-ribosylation during the process of macrophage activation (32). We demonstrated that ectopically expressed SARS-CoV-2 Nsp3 macrodomain can hydrolyze these modifications in human cells (Fig. 2), which, to our knowledge, had so far only been shown in vitro using recombinant coronavirus Nsp3 macrodomains (10,13).
We have employed this assay in a repurposing screen for Nsp3 macrodomain inhibitors (Fig. 3), which turned out unsuccessful, but highlighted the reasonable throughput of our setup, which allowed two people to perform triplicate screens for 69 compounds in less than a week. Given the shortage of BSL-3 facilities required to test anticoronavirus therapeutics on the SARS-CoV-2 virus directly, and the inherent uncertainty in these tests regarding on-target effects in case of successful reduction in viral loads (33), we believe that a cellular assay such as the one presented here may be a critical intermediate step in the development of Nsp3 macrodomain inhibitors, since a positive result in this assay ensures that candidate compounds are cell-permeable and have on-target effects in human cells.
This microscopy-based assay also allowed us to investigate the origins and functions of the IFN-induced ADP-ribosylation. Interestingly, we observed that Nsp3 macrodomain expression was associated with a slight, but not statistically significant, reduction in ADP-ribosylation even in control cells (Fig. 2, B-D). As this effect was also observed upon tofacitinib treatment (Fig. 4B), this basal ADP-ribosylation signal may stem from low-level endogenous IFN signaling in A549 cells. Crucially, the signal we detected is predominantly cytosolic and has a punctate pattern (Fig. 1C) but, despite extensive efforts, we have so far been unsuccessful in colocalizing this signal with known cytosolic structures and organelles (LCR and NH, unpublished observations). Naturally, it will be critical to identify not only this structure but also the actual protein targets of the Nsp3 macrodomain-sensitive ADPribosylation induced by the IFN response, in order to elucidate the putative antiviral function of this modification.
Given that inhibition of DNA damage-responsive PARPs with olaparib had no effect on IFN-induced ADP-ribosylation, but inhibition of JAK kinases completely prevented its formation (Fig. 4, A and B), our data indicate that the ADPribosylation detected here is catalyzed by a PARP whose expression is induced by IFN signaling. Importantly, we identified that the PARP9/DTX3L heterodimer, whose expression is IFN-stimulated (Fig. 5), is absolutely necessary for this ADP-ribosylation, as the signal was lost upon knockout of either of these genes (Fig. 4). Whether PARP9 or DTX3L is individually required for this activity is difficult to determine, as KO of either gene severely affected the stability of the binding partner (Fig. S4C), in agreement with previous studies  showing codependence of these proteins (26,27). Interestingly, PARP9 and DTX3L expression is also induced in cells infected with SARS-CoV-2, as long as viral load is high enough to induce an IFN response (34). In agreement with our results, PARP9/DTX3L induction under these conditions is completely prevented by treatment with a JAK kinase inhibitor (34). While it is tempting to speculate that PARP9/DTX3L directly catalyze the formation of the IFN-induced ADP-ribose signal we detected, we cannot rule out a role of these proteins in a putative signaling cascade that culminates in ADPribosylation by another PARP.

Nsp3 macrodomain and IFN-induced ADPR
We propose that the PARP9/DTX3L heterodimer promotes this ADP-ribosylation as a downstream effector of the IFN response, as we did not observe an effect of PARP9/DTX3L deletion on STAT1 phosphorylation or ISG induction (Fig. 4). These data are surprising, considering previous reports showing that PARP9 and DTX3L overexpression increases ISG induction in response to IFN signaling in human fibrosarcoma cells (27) and that PARP9 silencing reduces IFNγ-induced STAT1 phosphorylation and ISG induction in macrophages (35). While further studies are clearly required to clarify the role of PARP9 and DTX3L in the IFN response, we speculate that differences in genetic manipulation of PARP9/DTX3L levels (overexpression, RNA interference, or CRISPR/Cas9 KO) or cell types (fibrosarcoma, macrophage, or retinal pigment epithelia) could account for these differences.
In agreement with our model for an effector function of this PARP9/DTX3L-dependent ADP-ribosylation, the ectopic expression of the Nsp3 macrodomain also had no effect on STAT1 phosphorylation or ISG induction (Fig. 5), despite reversing PARP9/DTX3L-mediated ADP-ribosylation (Fig. 2). This was again surprising, since macrodomain 1000 U/ml IFNα, 1000 U/ml IFNβ, 100 U/ml IFNγ, or transfected with 0.1 μg/ml poly(I:C). Same experiment as Fig. S2A, including parts of the same FLAG panel, which came from the same membrane as phospho-STAT1. Actin loading control from same membrane as PARP9. D, quantification of ADP-ribose immunofluorescence signal intensity in A549 cells transduced with empty vector control (e.v) or with a lentiviral construct for constitutive expression of FLAG-tagged macrodomain, at indicated timepoints after treatment with 100 U/ml IFNγ. Mean ± SEM (n = 3). E, representative image of immunoblot analyses for STAT1 phospho-Y701, DTX3L, FLAG, and actin loading control in A549 cells transduced either with empty vector control (e.v.) or with a lentiviral construct for constitutive expression of FLAG-tagged macrodomain, at indicated timepoints after treatment with 100 U/ml IFNγ.
Nsp3 macrodomain and IFN-induced ADPR mutation in a mouse coronavirus model (mouse hepatitis virus-MHV) was previously shown to cause an increase in IFN production by virus-infected murine macrophages (14). In this MHV model, the macrodomain was proposed to mainly counteract PARP14 activity (14), which is thought to act upstream of the production of type I IFN by virusinfected cells (36). These results are not necessarily conflicting, as our methodology may not detect the full complement of IFN-induced ADP-ribosylation or the MHV macrodomain may hydrolyze different PARP targets that play a more prominent role in the mouse IFN response relative to humans (37). Alternatively, macrodomain mutation in the context of a full-length Nsp3 protein may have secondary effects on the activity of other domains, such as the papain-like protease known to suppress IFN signaling (38,39).
In conclusion, we show here that the SARS-CoV-2 Nsp3 macrodomain hydrolyzes PARP9/DTX3L-dependent ADPribosylation induced by IFN signaling and uncover a role for this modification as a putative effector, rather than modulator, of the IFN response (Fig. 6). As part of this study, we developed a cellular assay with the potential to substantially impact drug discovery efforts currently underway to target the Nsp3 macrodomain as a novel anticoronavirus therapy.

Experimental procedures
Cell culture and treatment conditions A549 lung adenocarcinoma cells were grown in DMEM/ high glucose media (Thermo) and RPE1-hTERT retinal pigment epithelia cells were grown in DMEM/F-12, both supplemented with 10% Fetal Bovine Serum (Thermo). HEK293 FT cells were grown in DMEM/high glucose supplemented with 10% FBS (Thermo), 50 mg/ml gentamycin (Sigma), 1 mM sodium pyruvate (Thermo), nonessential amino acids (Thermo), and 2 mM L-glutamine (Thermo). All cell lines were maintained at 37 C in a humidified atmosphere containing 5% CO 2 . Recombinant interferons α, β, and γ (Sigma SRP4594, I9032, and SRP3058) were added directly to the media at the indicated doses and poly(I:C) (Sigma) was transfected using PEI and the cells collected 24 h later. Tofacitinib (Selleckchem), olaparib (Selleckchem), or atorvastatin (Sigma) was added at the indicated doses at the same time as the induction of IFN signaling.

Lentiviral construct generation and viral transduction
SARS-CoV2-Nsp3 macrodomain (UniProt identifier P0DTC1, residues 1023-1157) was amplified from a pET30abased vector (a kind gift by A.Fehr -Univ. of Kansas) and cloned via BamHI and XhoI sites into a pCDNA3.1-based vector previously engineered to contain an N-terminal FLAG-FLAG-Strep-Strep tag. This ORF was then subcloned into a pCDH-puro lentiviral vector (System Biosciences) via NheI and XhoI sites or amplified and cloned via EcoRI and AgeI sites into the pLVX-TetOne-puro lentiviral vector (System Biosciences). The N40A mutation was generated by standard site-directed mutagenesis and all vectors were confirmed by Sanger sequencing of the inserts. These vectors were cotransfected with psPAX2 and pMD2.G packaging vectors (Systems Biosciences) into HEK-293FT cells using standard PEI transfection, the supernatant collected 48 h and 72 h after transfection, filtered through 0.45 μm filters, and used to transduce A549 cells in the presence of 8 μg/ml polybrene (Sigma). After 48 h, 5 μg/ml puromycin was added to the media for selection of transduced cells for 7 days.

Virtual screening
The crystal structure of the Nsp3 macrodomain from SARS-CoV-2 cocrystallized with ADP-ribose (Protein Data Bank code 6W02) was used for structure-based studies (40). The protonation states of residues were revised and the variable conformation of the Asp27 residue was adjusted. Hydrogen atoms were added and water molecules and ligands were removed. Molecular docking simulations were performed with Gold 5.4 and AutoDock 4.2.3 softwares. The protein was set as rigid and the ligands as flexible. For the docking studies using Gold 5.4, the radius of simulation was set to 6 Å (considering the large size of ADP-ribose). CHEMPLP score function was selected and the efficiency parameter set to very flexible, while other parameters were left at default values. For the docking with AutoDock, nonpolar hydrogen atoms were merged with respective carbons and Gasteiger charges were added. The number of points of each dimension was 25 × 40 × 50 Å and grid box was centralized in the coordinates 10.649 × 7.101 × 20.841. For each ligand, 100 simulations of molecular docking were performed. For optimization, Lamarckian genetic algorithms were employed and the molecular docking protocols were validated by redocking. For this, cocrystallized ADP-ribose was removed and redocked in PDB 6W02. RMSD distances between experimental and simulated atom ligands were calculated by VMD (41). For virtual screening, the subset of 6365 drugs approved in the world available in Zinc15 (http:// zinc15.docking.org/) was selected as target library.

Cloning of expression vectors and purification of recombinant protein
The SARS-CoV-2 Nsp3 macrodomain (Uniprot identifier P0DTC1, residues 1024-1192) was cloned into pNH-TrxT vector (Addgene plasmid #26106) using SLIC restriction-free cloning method (42). Briefly, the pNH-TrxT plasmid was linearized and 100 ng of linearized plasmid was mixed in a 1:4 M ratio with SARS-CoV-2 Nsp3 macrodomain (gBlock gene fragment, Integrated DNA Technologies) and incubated with T4 DNA polymerase for 2.5 min at room temperature and for 10 min on ice. The mixture was used to transform NEB5α competent E. coli cells (New England BioLabs) according to manufacturer's instructions. Colonies were grown on LB agar containing 5% sucrose using the SacB-based negative selection marker (43). The construct was verified by sequencing of the insert regions and was then used to transform E. coli BL21(DE3) cells. In total, 500 ml Terrific Broth (TB) autoinduction media including trace elements (Formedium, Hunstanton, Norfolk, England), supplemented with 8 g/l glycerol and 50 μg/ml kanamycin, was inoculated with 5 ml of overnight preculture and incubated at 37 C until an OD600 of 1 was reached. After an overnight incubation at 16 C, the cells were collected by centrifugation at 4200g for 30 min at 4 C. The pellets were resuspended in lysis buffer (50 mM HEPES pH 7.5, 500 mM NaCl, 10 mM imidazole, 10% glycerol, 0.5 mM TCEP) and stored at −20 C. For protein purification, the cells were thawed and lysed by sonication. The lysate was centrifuged (16,000g, 4 C, 30 min), filtered and loaded onto a 5 ml HiTrap HP column equilibrated with lysis buffer, and charged with Ni 2+ . The column was washed with 30 column volumes of lysis buffer and four column volumes of wash buffer (30 mM HEPES pH 7.5, 30 mM imidazole, 500 mM NaCl, 10% glycerol, 0.5 mM TCEP). The protein was eluted using wash buffer containing 300 mM imidazole. Imidazole was removed by dialysis and TEV-protease was added (1:30 M ratio, 16 h, 4 C) to cleave the His 6 -TrxT-tag followed by a reverse IMAC step to remove impurities. Sizeexclusion chromatography was carried out on a HiLoad 16/ 600 Superdex 75 pg 120 ml column in 30 mM HEPES pH 7.5, 300 mM NaCl, 10% glycerol and 0.5 mM TCEP. Pure fractions were pooled and stored at −70 C.

Thermal shift assay
The purified SARS-CoV-2 Nsp3 macrodomain was diluted to 0.3 mg/ml in assay buffer (25 mM HEPES pH 7.5, 100 mM NaCl) and mixed with 100 μM of the compounds. The sample containing Flubendazole was measured at 20 μM compound concentration due to limited solubility. Samples in the presence or absence of 100 μM ADP-ribose were used as controls. All samples and controls contained a final concentration of 1%(v/v) DMSO. Samples were loaded to glass capillaries and analysis was performed in Prometheus NT.48 (NanoTemper). Data points were recorded from 35 to 65 C, with the temperature increasing by 1 C/min. The onset of scattering and melting temperatures based on the change of intrinsic protein fluorescence (ratio 350 nm/330 nm) was calculated in PR.ThermControl software (NanoTemper).

Western blotting
Adherent cells were washed in PBS and lysed directly in preheated Laemmli buffer devoid of bromophenol blue and beta-mercaptoethanol. Lysates were transferred to tubes, boiled for 15 min, and the protein concentration was determined using BCA protein quantification kit (Pierce). After normalization of protein concentrations and addition of bromophenol blue and beta-mercaptoethanol, samples were boiled again for 10 min, and 15 to 50 μg of protein was loaded per sample in standard SDS-PAGE gels. Proteins were transferred to nitrocellulose membranes (Bio-Rad), visualized with Ponceau Red (Sigma), and the membranes were cut horizontally such that different portions of the same membrane could be incubated with the appropriate antibodies. Membranes were blocked with 5% BSA for 30 min and incubated with primary antibody overnight at 4 C. After extensive washing in TBST buffer, 1 h incubation in appropriate HRP-conjugated secondary antibodies (Sigma), and another round of washing, membranes were incubated with ECL Prime (Amersham), and the signal was detected using a Chemidoc MP Imaging System (Bio-Rad). Signals were quantified using ImageJ software.

Immunofluorescence staining
Cells were seeded either on 1.5H glass coverslips (Thorlabs) or in microscopy-compatible plastic 96-well plates (Corning), treated as required, washed with PBS, and fixed with 2% (for pSTAT1) or 4% EM-grade PFA (EMS) prepared in PBS, which was subsequently quenched with 0.1 M glycine. After permeabilization in 100% methanol (for pSTAT1) or 0.2% TritonX-100 in PBS, samples were blocked in 1% BSA/5% goat serum in PBS and incubated with primary antibody for 1 h at room temperature (for ADP-ribose) or overnight at 4 C. Samples were extensively washed in PBS, incubated with appropriate fluorescently labeled secondary antibodies (Thermo), washed again, stained with DAPI (Thermo), and the coverslips were mounted in Vectashield (Vector Labs) or the plates were maintained in 90% glycerol until image acquisition.

Image acquisition and analysis
Fluorescence microscopy images were acquired on a customized TissueFAXS i-Fluo system (TissueGnostics) mounted on a Zeiss AxioObserver 7 microscope (Zeiss), using 20× Plan-Neofluar (NA 0.5) or 40× Plan-Apochromat (NA 0.95) objectives and an ORCA Flash 4.0 v3 camera (Hamamatsu). For most experiments, 6 × 6 adjacent fields of view were acquired per sample using automated autofocus and image acquisition settings. Images were analyzed using StrataQuest software (TissueGnostics). For phospho-STAT1 quantification, individual nuclei were detected in the DAPI channel and the mean intensity of phospho-STAT1 signal per nucleus quantified for thousands of cells per sample. For ADP-ribose quantification, individual nuclei were detected in the DAPI channel Nsp3 macrodomain and IFN-induced ADPR and the approximate cell boundaries determined by growing outward from the nuclear area for a fixed distance, until neighboring areas touched each other or until the ADP ribose signal reached background levels. Peaks of ADP-ribose signal intensity within this cellular mask (excluding the nucleus) were detected as "dots" and the total fluorescence signal contained in all dots per cell was quantified for thousands of cells per sample. In both cases, the fluorescence intensities of all cells within a sample were averaged, and this value normalized to the average intensity of the IFNγ-treated control sample for each biological replicate experiment (Fig. S1).

RT-qPCR
Total RNA was extracted from 10 6 cells using an RNAeasy kit (Qiagen), treated with DNAseI (Ambion) and reversetranscribed using SuperScriptII (Thermo), according to manufacturer's instructions, using both oligodT and random hexamer primers (Thermo). This cDNA was used for qPCR (5 ng/ reaction) using Power SYBR green Master mix (Thermo) with three technical replicates per biological replicate, using 200 nM of the primer sets indicated in Table 1, which were chosen from PrimerBank (46). Reactions were performed on an Applied Biosystems 7500 Real-time PCR system, using  default settings. RPL19 was used as housekeeping control and standard 2 −ΔΔCt analysis performed relative to the untreated control sample.

Statistical analyses
All experiments were repeated on at least three separate occasions, often with multiple parallel replicates processed on the same day, but treated as independently as possible. All graphs and statistical analyses were generated using GraphPad Prism software and display both the distributions of individual datapoints and their mean ± SEM of the normalized values relative to the IFNγ-treated control for each replicate, considered as 100%. Statistical comparisons between samples were performed using ANOVA, with p < 0.0001 indicated by ****, p < 0.001 indicated by *** and p < 0.01 indicated by **.

Primers and antibodies
All oligonucleotides and antibodies used in this study are presented in Tables 1 and 2 Data availability The datasets generated during the current study are available from the corresponding author on reasonable request.
Supporting information-This article contains supporting information.