Selective small molecule PARG inhibitor causes replication fork stalling and cancer cell death

Poly(ADP-ribose)ylation (PARylation) by PAR polymerase 1 (PARP1) and PARylation removal by poly(ADP-ribose) glycohydrolase (PARG) critically regulate DNA damage responses; yet, conflicting reports obscure PARG biology and its impact on cancer cell resistance to PARP1 inhibitors. Here, we found that PARG expression is upregulated in many cancers. We employed chemical library screening to identify and optimize methylxanthine derivatives as selective bioavailable PARG inhibitors. Multiple crystal structures reveal how substituent positions on the methylxanthine core dictate binding modes and inducible-complementarity with a PARG-specific tyrosine clasp and arginine switch, supporting inhibitor specificity and a competitive inhibition mechanism. Cell-based assays show selective PARG inhibition and PARP1 hyperPARylation. Moreover, our PARG inhibitor sensitizes cells to radiation-induced DNA damage, suppresses replication fork progression and impedes cancer cell survival. In PARP inhibitor-resistant A172 glioblastoma cells, our PARG inhibitor shows comparable killing to Nedaplatin, providing further proof-of-concept that selectively inhibiting PARG can impair cancer cell survival.

P oly(ADP-ribose)ylation (PARylation) by PAR polymerase 1 (PARP1) and its reversal by PAR glycohydrolase (PARG) dictate multiple DNA damage responses [1][2][3] . PAR homeostasis regulates DNA damage responses by rapid and dynamic modulation of protein-protein and protein-DNA interactions essential for genomic integrity and cell survival 2 . Consequently, blocking PAR synthesis by PARP inhibitors (PARPi) impairs signaling and the repair of damaged DNA 4 and causes centrosome amplification 5 . Accordingly, PARP1-null mice are more susceptible to carcinogenesis induced by DNA alkylating agents 6,7 . In contrast, PARP1 hyperactivation causes excessive PAR accumulation that is cytotoxic and triggers the release of apoptosis-inducing factor (AIF) from mitochondria, leading to cell death 8,9 . Similarly, mice lacking PARG show excessive PAR accumulation, resulting in early embryonic lethality and cell death 10 ; whereas, PARG-deficient cells exhibit delayed repair of DNA single-and double-strand breaks (DSBs) 11,12 , plus aberrant cell cycle progression 13 .
PARG, a single gene alternatively spliced to generate five isoforms, is an endo-and exo-glycohydrolase that rapidly degrades PAR generated by PARP1 to coordinate DNA repair 14,15 . PARG hydrolyzes α(1′′-2′) O-glycosidic linkages in PAR chains to release ADP-ribose and oligo(ADP-ribose) chains that may signal genotoxic stress 1 . Bacterial, protozoan, and mammalian PARG structures [16][17][18][19][20][21] reveal a macrodomain fold with conserved active site residues and 'tyrosine clasp' (Tyr clasp) unique to PARG enzymes, which suggest a common mechanism for PAR hydrolysis. However, PARG does not cleave the terminal ADP-ribosyl group linked to the target residue, which is turned over by macrodomain-containing mono(ADP-ribose) (MAR) hydrolases [22][23][24] . Yet, deletion of all PARG isoforms is embryonically lethal 10 , and deletion of the nuclear isoform causes sensitivity to alkylating agents and ionizing radiation (IR) 25 . PARG is recruited to DNA damage sites through interaction with PCNA and PARP1 26 , prevents IR-induced mitotic catastrophe 13 and maintains replication fork stability in a BRCA2dependent manner 27 . Conversely, PARG depletion can prevent replication fork restart 28 . PARG genetic knockdown sensitizes various cancer cells to chemotherapeutic agents and radiation 11,13,29,30 and may cause tumor-specific killing in BRCA2-deficient cancer cells 27 . Furthermore, reduced PARG activity can decrease cytotoxicity associated with inflammation, ischemia, and stroke 3,31 and reduces liver metastases of colon carcinoma 32 . PARG furthermore regulates proliferation and differentiation of dendritic cells and T cells via PARP/NF-κB in tumor metastases of colon carcinoma 4 . Yet, PARG inhibition as a therapeutic strategy is directly questioned by reports, showing that PARG downregulation enables PARPi resistance in BRCA2 and p53-null mouse cell-line 33 , and that PARG deficiency had little impact on BRCA1-and/or PTENdeficient tumor cells 34 .
With these conflicting reports in mind, herein, we examined the TCGA database and found that in most human cancers PARG expression is upregulated, supporting the notion that PARG enzymatic function is required for tumorigenesis 35 . We sought to test PARG inhibition in cancer cells and found that although existing PARG inhibitors (PARGi) have proven valuable [29][30][31]36 , issues with potency, bioavailability, rapid clearance, selectivity, or defined mechanism of action can cloud their applications for biology. Reasoning that a bioavailable and specific PARG inhibitor would be an enabling tool, we employed chemical library screening that led to the discovery of thio-xanthine/methylxanthine derivatives as bioavailable, potent and specific PARG inhibitors. Our high-resolution crystal structures of human PARG bound to these methylxanthine inhibitors revealed detailed interactions that support a mechanism of competitive inhibition. In particular, we find that conformational switching of the Tyr clasp that distinguishes PARG from MAR and other glycohydrolases enables specific engagement of methylxanthine derivatives in the PARG active site. Crucially, our cell-based assays show effective cellular activity, including PAR accumulation in cells, inhibited PARP1 dePARylation and increased γH2AX foci formation. Indeed, H2AX phosphorylation levels in cells treated with PARGi uncover synergistic sensitization to IR. Furthermore, PARGi induces replication fork defects resembling those reported with PARG knockdown 36 , and causes cellular sensitivity in PARP1-inhibitor-resistant cells. Moreover, our compounds are PARG-selective and do not target other cellular glycohydrolases at the relevant doses, in contrast to other reported PARGi's. Our selective inhibitor reveals PARG functions that support and extend previous genetic findings, providing insights into PAR metabolism following IR damage and establishing these PARGi as enabling tools for investigating PAR biology with potential for advanced therapeutic strategies.

Results
Identification of specific PARG inhibitors. Our TCGA database analysis revealed that in most human cancers PARG expression is upregulated (Fig. 1a). Importantly, PARG enzymatic function can be required for tumorigenesis 35 . Therefore, we sought to identify a selective PARG inhibitor through our robust PARG enzyme kinetic assay 37 , which quantitatively monitors real-time PARmediated PARP1-XRCC1 interactions and disassembly in a multi-well format. A strong TR-FRET signal was generated upon assembly of the Tb 3+ -chelated XRCC1 (FRET donor) with PARylated PARP1, which incorporates fluorescein-labeled ADPribose moieties into PAR chains (FRET acceptor). The addition of PARG caused a time-dependent loss of the TR-FRET signal by reversing PARylation and disassembling the PARP1-XRCC1 complex ( Supplementary Fig. 1) 37 .
We screened the National Cancer Institute (NCI) Diversity Set II library (1990 compounds) against rat PARG. Then, human PARG was used to validate and optimize PARG inhibitors. The accuracy of the screen was determined by the calculated Z-scores of control reactions, which were between 0.6 and 0.8 throughout the plates, indicative of high reproducibility ( Supplementary  Fig. 2). Two different enzyme concentrations (12.5 nM and 25 nM) were used in subsequent analyses, with the rationale that potent inhibitors would inhibit PARG activity at both enzyme concentrations, while weaker inhibitors would inhibit PARG activity only at the lowest PARG concentration. The rate of TR-FRET signal loss in the presence of each candidate inhibitor was compared to that of control reactions, containing DMSO only (Fig. 1b). Initially, we selected 104 compounds that inhibited PARG enzymatic activity at both enzyme concentrations. Thirtynine of these compounds were discarded as they altered fluorescence intensity of donor or acceptor (F 520 and F 495 ) by >40% at the first time point. An additional twelve compounds containing metals (e.g., Pt, Hg, or As) and five compounds resembling short chain fatty acids and steroids were also discarded, leaving 51 candidate inhibitors (hit rate of 2.6%), including 12 potent hits showing~100% inhibition at both PARG concentrations (Supplementary Table 1). For the selected compounds, gel-based assays were performed to directly monitor inhibition of PARP1 dePARylation by PARG (Fig. 1c).
Most initial hits from the High Throughput Screening (HTS) contained 5-or 6-member rings and were grouped into five main chemotypes ( Supplementary Fig. 3): succinimide (JA1: 2 hits), xanthine/methylxanthine (JA2: 13 hits), hydroquinone (JA3: 13 hits), nitrobenzofuranzan (JA4: 9 hits), and benzyl-sulfonamide (JA5: 14 hits). A total of 34 compounds containing representative hits and available analogues from each chemotype were then analyzed in dose-response curves, using both the TR-FRET assay   Fig. 3). The JA2 xanthine/methylxanthine series was selected as the lead pharmacophore, based on its structural similarities to adenine, potency in vitro, and favorable drug-like characteristics. Dose-dependent inhibition of PARG activity by the JA2 series compounds was quantitatively analyzed using a gel-based PARG activity assay ( Supplementary Fig. 4 Fig. 4 and Supplementary Table 2). The TR-FRET assay is an indirect measurement of PARG activity, as it monitors PARG-mediated reversal of a PARdependent PARP1-XRCC1 interaction. Furthermore, loss of the TR-FRET signal was not detected until all PAR chains were degraded to lengths shorter than 7 ADP-ribose units 37 . Nonetheless, this assay enabled the measurement of steady-state PAR turnover rate, which was adequate for detecting enzyme inhibitors. In contrast, the gel-based assay directly monitors enzymatic conversion of PARylated PARP1 to an unmodified product by PARG, enabling a semi-quantitative measurement of enzyme inhibition (Supplementary Fig. 4). The IC 50 value of a known PARG inhibitor, adenosine diphosphate (hydroxymethyl) pyrrolidinediol (ADP-HPD), determined by the gel-based assay (1.07 μM, Supplementary Fig. 5), which was in agreement with its published values (0.1-1.4 μM) 20,38,39 . Therefore, we chose the gelbased PARG activity assay in subsequent studies.
Inspection of the five identified chemotypes of PARG inhibitors ( Supplementary Fig. 3) suggests that the JA2 series of xanthine/methylxanthine derivatives were the most attractive for several reasons. First, these compounds are structurally analogous to the adenine base of ADP-ribose (ADPR), an enzymatic product of PARG and have good potential for site-specific binding. Second, xanthine/methylxanthine-based compounds typically show good bioavailability, as evidenced by caffeine and the widely used bronchodilator, theophylline 40 . Third, the fused 5-and 6-membered rings of the xanthine core are synthetically tractable for analoging 40 . Most importantly, three thio-xanthine/ methylxanthine derivatives (JA2-3/NSC29192, JA2-4/NSC99657, and JA2-5/NSC99667) were among the most potent dosedependent inhibitors of PARG activity as assessed by HTS, with IC 50 values between 0.1 and 1.0 μM (Fig. 1d).
To identify the binding sites of JA2-3, JA2-4, and JA2-5, these compounds were soaked into crystals of human PARG (hPARG) catalytic domain (residues 448-976). Crystals of hPARG diffracted to high resolution, and the crystal packing arrangement was more amenable to binding compounds in the enzyme active site 20 , compared to crystals of rat and mouse PARG 18,19 . The hPARG active site has two deep pockets that specifically engage the adenine and diphosphate group of ADPR. Tyr795 at the tip of the Tyr clasp points towards the heart of the active site and simultaneously interacts with the adenine and diphosphate moieties.
The crystal structure of JA2-4 bound to hPARG refined at 1.7 Å resolution, (Fig. 2 and Supplementary Tables 3) identified a single JA2-4 molecule in the adenine-binding pocket. The 6´-thiocarbonyl sulfur of JA2-4 was assigned by the size and orientation of the electron density compared to that of the corresponding 6´-carbonyl oxygen of JA2120 (Figs. 2a and 3b). JA2-4 makes extensive site-specific interactions with hPARG, consistent with its potent inhibitory activity (IC 50 of 0.9 μM) (Fig. 1d). The methylxanthine head of JA2-4 makes π-stacking interactions with Phe902 side chain. The tail of JA2-4 extends from the 8´-position of the xanthine ring and is highly curved, which serves to make van der Waals contacts with Asn869. The terminal pyrrole of JA2-4 is rotated~90°and is perpendicular to the xanthine ring system. Importantly, most protein-JA2-4 interactions come from the conserved macrodomain fold. Despite the large conformational change (described below), the Tyr clasp generates limited direct contact with JA2-4; only the C β of Tyr795 interacts with the thioether group of JA2-4 (Fig. 2a). JA2-4 has a surface area of 484 Å 2 of which 89% (432 Å 2 ) is masked by direct contacts with hPARG, which can explain the relatively high potency of the JA2-4 series compounds.
Structural superposition of hPARG bound to JA2-4 and ADP-HPD revealed that the methylxanthine core of JA2-4 occupies the same binding site as the adenine of ADP-HPD (Fig. 2b), consistent with a competitive block of substrate binding activity. However, the ring system of the xanthine core is flipped horizontally and rotated~15°clockwise, relative to the adenine ring of ADP-HPD, and translated 0.8 Å towards the tip of the Tyr clasp. This distinctive ring orientation reflects a direct interaction of the 6´-thiocarbonyl group with the main chain amide of Asn869 (Fig. 2a). As predicted, a thiol substituent at the 6ṕ osition of the methylxanthine pharmacophore greatly increases potency (Fig. 3). Additionally, the 2´-carbonyl oxygen of JA2-4 hydrogen bonded with the main chain amide of Ile726. This interaction is not made by ADP-HPD, and it displaces the methylxanthine ring by 0.8 Å with respect to the adenine ring of ADP-HPD (Fig. 2a, b). The 2´-carbonyl oxygen of JA2-4 also serves as a pivot point for rotation of the methylxanthine ring. Collectively, these structures support specific and potent inhibition of PARG activity by JA2-4 and suggest a mechanism of competitive inhibition by JA2 series xanthine/methylxanthine derivatives.
Identification of key contact points of JA2-4. The most striking feature of the JA2-4 interaction with hPARG is a significant change in the Tyr clasp conformation (Fig. 2c). The JA2-4 binding site is obstructed by the side chain of Tyr795 in the unbound hPARG and in hPARG bound to ADP-HPD. To accommodate JA2-4, the Tyr795 side chain is flipped out from the adenine-binding pocket (with~90°rotation), which positions the thioether of JA2-4 in contact with C β of Tyr795. Furthermore, the interaction of JA2-4 with Tyr795 pushes on the tip of the Tyr clasp, which is displaced~0.9 Å towards the Nterminal helical bundle (Fig. 2c).
Twisting of the Tyr clasp induced by JA2-4 binding involves several new interactions with a neighboring helix α6 (Fig. 2c). Most notably, JA2-4 binding caused the Arg671 side chain to swing~9 Å towards the tip of the Tyr clasp, where it makes an electrostatic interaction with Glu797 side chain and hydrogen bonds to the main chain carbonyl of Ala750 (not shown in this view). This Arg671 pivoting motion constitutes a switch that serves as a backstop to secure the Tyr clasp in contact with the N-terminal helical bundle, which is~30 Å away from Tyr795. Taken together, this previously undetected structural switching of the Tyr clasp may explain the basis of JA2-4 binding and its inhibition of PARG activity.
Structure-based design improved PARGi potency and selectivity. We examined the structure-activity relationships (SAR) of JA2-4 analogs (Fig. 3) in competitive inhibition experiments. JA2131/NSC98003 contains a terminal morpholine in place of the pyrrole of JA2-4 and shows improved activity, which may derive from a polar interaction mediated by the 1´-oxygen of the morpholine (Fig. 3a, c). The 6´-thiocarbonyl group of JA2131 is another driver of inhibitory potency. Substitution with a carbonyl at this position decreases potency more than 50-fold (IC 50   To further test the molecular basis for these SAR data, crystal structures of hPARG bound to JA2131 and JA2120 were determined at 1.9 and 1.8 Å resolution, respectively ( Fig. 3a, b). JA2131 and JA2120 bind to the adenine-binding pocket of hPARG in the same orientation as JA2-4. As for JA2-4, the 6´-thiocarbonyl of JA2131 makes a direct van der Waals contact with the main chain amide of Asn869 (3.6 Å) from the conserved macrodomain fold (Fig. 3a). The morpholine oxygen of JA2131 appears to accept a hydrogen from the side chain amine of Asn869, explaining the modest increase in potency compared to JA2-4 ( Fig. 2a).
In contrast, the distance between 6´-carbonyl of JA2120 and the main chain amide of Asn869 is significantly longer than that of 6´-thiocarbonyl of JA2131 and is beyond the range of direct polar or van der Waals interactions (4.3 Å) (Fig. 3b). Instead, a bound water molecule, which is absent in the JA2131-and JA2-4bound structures, is positioned between JA2120 and Asn869. The bound water molecule relays an indirect hydrogen bond from 6´-carbonyl of JA2120 to the main chain amide of Asn869 and to the main chain carbonyl of Phe902 (Fig. 3b). A bound water molecule is also present at this site in structures of hPARG and hPARG bound to ADP-HPD. Notably, the 6´-thiocarbonyl group of JA2-4 series inhibitors displaces this water molecule to form improved interactions with the adenine-binding pocket of hPARG.

PARG regulates PARP1 modification in cells.
To identify suitable cell lines to test PARGi, we evaluated PARG protein expression patterns on a panel of prototypic cancer cell lines.
Cells were fractionated into cytoplasmic, nuclear and chromatin, and then PARG expression and localization were evaluated using an anti-PARG antibody (Fig. 4a). A varying level of PARG was detected in three cellular compartments. In prostate adenocarcinoma (PC3) and breast mammary epithelial (MCF-7) cells, PARG level was the highest in the cytoplasm, followed by the nucleus, with the lowest seen in the chromatin fraction. The triple-negative MDA-MB-231 breast cancer and U2OS osteosarcoma cells showed comparable expression between cytoplasm and nucleus. The PC3 cell line, with one of the highest levels of PARG expression, was selected for our initial PARGi testing. PC3 cells were treated with either DMSO or JA2131 for 2 h, then exposed to 7 Gy of IR and allowed to recover for a designated time period before harvesting and cell fractionation. Chromatin fractions of the cell lysates were analyzed by immunoblotting with PARP1 specific antibody (Fig. 4b), and the PARylation states of PARP1 were monitored as a marker for PARG inhibition. A high molecular weight (HMW) PARP1-containing band in the chromatin fractions, suggestive of hyper-PARylated PARP1, was only seen in cells treated with JA2131 PARGi. Cells treated with both JA2131 and IR displayed enhanced levels of PARylation-trapped PARP1 (Fig. 4b, 0.5 h versus 1.0 h). In these cells, PARylated PARP1 in response to PARG inhibition appeared as a discrete HMW (DHMW) proteins with limited smearing, as observed previously 41 . The DHMW was only seen when chromatin fractions were immunoblotted with the PARP1 antibody; by contrast, nuclear fractions and total cell lysates showed a contiguous smear, typically associated with PARylated PARP1. To assess whether DHMW was due to PARylation and not any other post-translational modifications, chromatin fractions of the PARGi-treated PC3 cell-lysates were treated with purified PARG, followed by anti-PARP1 western blotting. This led to the restoration of LMW PARP1, indicating PARylation was indeed causing the DHMW (Fig. 4c, d). Thus, a dynamic equilibrium between DHMW and LMW nuclear PARP1 exists in cells, and PARGi shifts this equilibrium towards DHMW. We conclude that PARGi induces PARP1 DHMW as a result of persistent PARylation, which can be reversed in vitro with exogenous purified PARG, suggesting a selective role of PARG in the dynamic regulation of PARP1 modification and signaling.
PARGi induces PAR accumulation and γH2AX foci in nuclei. As we found that inhibition of PARG increases PARP1 in a PARylated DHMW form, we tested whether our PARGi augments the level of PAR accumulation and of double-stranded DNA (dsDNA) breaks in cells treated with IR. Following IR, cells treated with JA2131 showed a significant increase in PAR accumulation in comparison to untreated cells (20-fold increase with JA2131). DMSO treated cells showed a modest post-IR PAR accumulation that was comparable to the untreated cells ( Fig. 5a, b, DMSO versus JA2131). A low magnification 3 × 3 image montage showed distinct PAR accumulation and γH2AX foci formation ( Fig. 5a and Supplementary Fig. 6). Quantification revealed that PAR accumulation coincides with γH2AX staining and foci formation, which was >10-fold higher in PARGi-treated cells than in control cells (Fig. 5c). Furthermore, at least 50% of cells showed colocalization of PAR and γH2AX foci (Fig. 5d, γH2AX/PAR foci against total cell counts Fig. 5e). Western blot analyses also indicated that JA2131-induced PAR accumulation (Fig. 5f). Inspection of images at high magnification validates the automated quantification of observed JA2131-induced PAR/γH2AX colocalization. DMSO treated cells display diffuse, low-level PAR staining in the nucleus of cells with limited γH2AX foci. However, for cells treated with the PARGi JA2131, PAR staining was localized to distinct nuclear foci that also contained the phosphorylated H2AX (γH2AX) (Fig. 5g). We saw some γH2AX foci in DMSO treated cells, but only if the contrast was independently adjusted (see Supplementary Fig. 6a, b, bottom panels). Indeed, PAR accumulation caused by the treatment of cells with IR was further augmented by the PARGi, indicating that it functions as a DNA damage sensitizer. It was previously reported that genetic knockdown of PARG results in sensitization of cancer cells to chemotherapeutic agents and radiation 11,13,29,30 , and tumor-specific killing in BRCA2-deficient cancer cells 27 . Here, we reveal an effective PAR accumulation that coincides with elevated γH2AX following treatment with a novel PARGi in PC3 cells that have intact BRCA1 and BRCA2 genes 42 .

JA2131 kills cancer cells through selective PARG inhibition.
To determine the radiation sensitization effect of PARGi, a clonogenic cell survival assay was used to measure radiation sensitization in PC3, MDA-MB-231, and MCF-7 cell lines treated with JA2131. First, we defined the radiation dose response and the optimum cell plating number for each cell-line ( Supplementary  Fig. 7). Secondly, DMSO and the PARPi Olaparib were used as a negative and positive control respectively ( Supplementary Fig. 8).
The results show that PARG inhibitor JA2131 inhibits colony formation in all three cell lines. MCF-7 cells were less sensitive to JA2131 than the PC3 cells. The triple-negative breast cancer cells MDA-MB-231 were the most sensitive among the three cell-lines treated with JA2131 (Fig. 6a). Interestingly, in MCF-7 cells with the highest level of cytoplasmic PARG showed greatest sensitivity to the commercially available PARGi PDD00017273 (PDD herein) (Supplementary Fig. 8). These data suggest that a b  Fig. 3 The 6´-thiocarbonyl group contributes strongly to the potency. a, b Crystal structures of human PARG bound to JA2131 a and JA2120 b. JA2131 and JA2120 occupy the adenine-binding pocket of hPARG in the same orientation as JA2-4. The 6´-thiocarbonyl of JA2131 makes a direct van der Waals contact with the main chain amide of Asn869 (3.6 Å) a, whereas the 6´-carbonyl of the less potent inhibitor JA2120 makes water-mediated hydrogen bonds to the main chain amide of Asn869 (4.3 Å) and the main chain carbonyl of Phe902 b. These findings indicate that the 6´-thiocarbonyl group of JA2-4 series inhibitors increases potency by introducing a direct interaction with hPARG. c The structure-activity relationships (SAR) for JA2-4 series inhibitors.
The direct interaction between the 6´-thiocarbonyl group of JA2131 and hPARG increases potency more than 50-fold in comparison to JA2120. PDB code: 6OA1 (PARG/JA2120), 6OA3 (PARG/JA2131). Source Data are provided as a Source Data file.
underlying genetic variations that dictate PARG protein expression patterns and signaling could play an important role in the effectiveness of PARGi with implications for vetting future PARGi patient groups. In addition, we tested the effect of sustained JA2131 treatment alone or in combination with IR in colony formation. Indeed, JA2131 alone was sufficient to inhibit PC3 survival, but when combined with IR was more effective in reducing the number of surviving cell-colonies ( Supplementary  Fig. 9). The triple-negative breast cancer cells MDA-MB-231 showed the largest response in our clonogenic survival assay (Fig. 6a). Thus, dose-dependent JA2131-induced cellular PARylation with and without IR was investigated. For combined IR treatment, MDA-MB-231 cells were treated with JA2131 for 2 h, followed by 7 Gy X-ray and then allowed to recover for an hour before cellular extraction and western blotting analysis. PARGi-treated cells alone showed a dose-dependent increase in cellular PARylation between 1 μM and 5 μM doses. However, PARGi combined with IR induced a massive increase in cellular PARylation at 100 nM (Fig. 6b). Notably, cells treated with radiation alone (Fig. 6b, lane 1) or inhibitor alone (Fig. 6b, lanes 2-5) did not show the level of PARylation observed when both treatments were combined, (Fig. 6b, lane 7-10) suggesting a synergistic, rather than an additive effect.
Using shRNA, next, we generated a stable MDA-MB-231 PARG knockdown (PARG-KD) cell-line (Supplementary Fig. 10). We compared PARGi JA2131-induced PARylation in PARG-KD and control cells. We reasoned that PARG-KD cells would show a reduced response to a selective PARGi as the stable PARG-KD cells would have to adopt a PARG-independent survival mechanism. MDA-MB-231 control cells and PARG-KD cells were treated with JA2131, then total cell lysates were subjected to immunoblotting with anti-PAR antibody, followed by immunodetection for anti-PARG, PARP1 and Actin antibody (Fig. 6c). As expected, untreated PARG-KD cells showed a higher baseline level of PARylation than the control cells. Treatment with JA2131 did not affect the level of PAR accumulation in PARG-KD cells; whereas, a measurable increase in PAR was seen in the MDA-MB-231 cells. Treatment of the cells under the same inhibitor regime with IR showed equivalent results (see Supplementary  Fig. 11). Dose-response curves comparing control and PARG-KD cells further validate the selective action of JA2131 on control cells expressing PARG, but not on the PARG-KD cells ( Supplementary Fig. 12). As a comparison, we also tested the inhibitor PDD, which induced PARylation in both control and PARG-KD cells. However, both cell-lines responded with increased PARylation from the PDD inhibitor, suggesting potential non-selectivity towards other cellular glycohydrolase ( Supplementary Fig. 13).
To further evaluate the selectivity of JA2131, we performed Sulforhodamine B (SRB) 43 cytotoxicity tests in MDA-MB-231 control and PARG-KD cells (Fig. 6d). The results show approximately a 10-fold higher IC 50 value for PARG-depleted cells, compared to control cells with a normal level of PARG (see Fig. 6d, IC 50 table). Notably, only 50% of the PARG-KD cells responded to the PARGi to produce the indicated IC 50 . In assay (CETSA) 44 on PC3 cells treated with either DMSO or JA2131. The results show JA2131 treatment induced a 9°C stabilization of PARG proteins in cells ( Fig. 6f and Supplementary  Fig. 14); whereas, the PCNA control showed no significant change in thermal stability in response to JA2131 treatment (see Fig. 6f).
We also evaluated JA2131's specificity in vitro as an inhibitor of purified hPARG in the presence and absence of detergent (Triton-X 100, 0.1%) or a nonspecific target protein (bovine serum albumin BSA, 0.1 mg/ml). This concentration of Triton-X 100 is~10-fold higher than its critical micelle concentration (CMC, 0.0155%), so it can effectively distinguish promiscuous and aggregating compounds 45 . Neither the addition of Triton-X 100 nor BSA significantly altered the inhibitory potency of JA2131 ( Supplementary Fig. 15a). In addition, we also examined whether or not JA2131 could nonspecifically interfere with PARP1 activation. PARP1 is an abundant nuclear enzyme and PARylated PARP1 is an abundant substrate for PARG, following DNA damage. JA2131 did not inhibit PARP1 activation up to 100 μM; whereas, a potent PARP inhibitor Olaparib completely shuts down PAR synthesis by PARP1 (Supplementary Fig. 15b). These results, together with the observed extensive physical interactions between JA2131 and its binding site on hPARG (Fig. 3), support the conclusion that JA2131 selectively inhibits PARG activity.

JA2131 causes replication fork defects.
In the BRCA2-negative background, the depletion of PARG inhibits cell growth 27 . Yet, we find that PC3 cells with intact BRCA2 show exquisite DNA damage induction by JA2131 as seen by γH2AX foci formation (Fig. 5) and poor long-term survival (Fig. 6a). Therefore, we investigated if this JA2131-induced massive γH2AX foci formation leads to growth arrest and cell death. JA2131-induced cellular cytotoxicity in PC3 cells with or without IR, along with commonly used DNA damaging agents and the chemotherapeutic agents Doxorubicin and Nedaplatin (Fig. 7a, inset table). Surprisingly, there was no added benefit for combining PARGi with IR, in fact, cells actually survived slightly better in this 72 h assay period (see Fig. 7a, IC 50 table). This is contrary to long-term clonogenic survival (Fig. 6a) and massive γH2AX foci formation (Fig. 5c) observed with combined PARGi and IR treatment. Indeed, we also observed a similar phenomenon with Olaparibtreated PC3 cells, where IR induced higher IC 50 values compared to the non-IR treated samples (see Supplementary Fig. 16). Therefore, we reasoned that IR exposure activates homologydirected repair (HDR), which counteracts PARGi effects in the short term. However, PARGi-treated dividing cells accumulate damage from replication stress over time that decreases long-term survival as observed in clonogenic assays. Furthermore, we tested our inhibitors in A172 glioblastoma cells that are resistant to PARPi and platinum-based chemotherapeutic agents. Cellular cytotoxicity assay showed our PARGi was more effective than Veliparib, but comparable in potency to Nedaplatin (Fig. 7b,  Recently, PARG depletion was shown to cause recruitment of double-strand-break-repair factors to chromatin in the absence damage and to slow replication fork progression, 36 consistent with PARG function in replication stress protection through the resolution of dysfunctional replication-structures. Because our PARGi effectively killed cancer cells in the absence of DNA damage (Fig. 6d, Fig. 7a, b), we reasoned that JA2131 could be acting at the replication fork. Therefore, we tested whether our PARGi could induce replication stress defects. HeLa cells pretreated with JA2131 or DMSO for 2 h were used to evaluate fork progression in DNA fiber assay (Fig. 7c, d). Indeed, PARGi alone was sufficient to induce a significant reduction in replication progression in the absence of genotoxic stress. Under these conditions, PARGi induced a statistically significant reduction in the average IdU tract from 7.5 to 6.0 μm. Thus, we tested replication progression in the presence of mild genotoxic stress delivered by a low-dose of hydroxyurea (Fig. 7e, f). Under this mild genotoxic stress-condition, we observed a reduction in the average IdU tract length to 5.0 μm. These findings show that our PARGi recapitulates reported replication defects caused by PARG depletion 36 , a finding that further validates JA2131 selective action against PARG.

Discussion
Reversible PARylation at DNA breaks and replication forks enforce control of modular assemblies of dynamic complexes, enabling damage detection, signaling, compartmental localization, and repair 46,47 . The ability to control PARylation reversal by PAR glycohydrolases 48 particularly PARG, which is the predominant enzyme that removes PAR chains at sites of DNA damage 12,49 offers the potential to control uniquely susceptible damage responses in cancer cells. Yet, there is no PARGi in the clinic, and existing inhibitors show various limitations. A classic PARG inhibitor is Gallotannin (GT), a large naturally occurring polyphenol that exhibits inhibition of PARG with an IC 50 below 5 μM 50 . GT retards DNA repair 51 , protects against H 2 O 2 -mediated cell death 50 , sensitizes cancer cells to Cisplatin 52 , and can specifically kill BRCA-deficient breast cancer cells 27 . However, concerns exist about GT's possible nonspecific mode of action and membrane permeability, as well as its antioxidant activity 53 . Salicylanilide derivatives appear to be nonspecific PARG inhibitors 45 . Rhodanine-based PARG inhibitors (RBPIs) are selective PARG inhibitors in vitro with low micromolar potency, but their bioavailability and cellular activity remain questionable 45 . Quinazolinedione sulfonamide derivatives target the PARG active site; however, rapid clearance in vivo renders them unsuitable as therapeutic agents 54 . The recently reported small-molecule PARGi PDD 54 has strong effects on cellular PAR metabolism. Yet, our tests with PARG-depleted cells suggest a nonspecific mode of PDD action in mammalian cells. Such concerns with existing inhibitors motivated the efforts presented here for the development of a pharmacologically viable and selective PARGi via chemical library screening and structure-based drug design to improve potency and selectivity. These efforts led to the identification of thioxanthine/methylxanthine derivatives as chemically tractable and reasonably potent pharmacophores for specific and cell-permeable chemical knockdown of PARG (Fig. 1), for testing PARG biological functions and for therapeutic potential. Notably, methylxanthine derivatives occur in many plants and are among Cells were treated with either DMSO or 10 µM JA2131 for 2 h, irradiated with 3 Gy (MDA-MB-231 and MCF-7), 7 Gy (PC3) IR and grown for~2 weeks, then colonies were fixed with methanol and stained with crystal violet. The results of three independent experiments are shown. Error bars = the standard error of the mean (SEM). b dose-response of JA2131 in MDA-MB-231 cells with and without IR. Cells were treated with designated concentrations of JA2131 for 2 h and then either left untreated or exposed to 3 Gy IR and allowed to recover for 1 h before lysis. Total protein was immunoblotted with anti-PAR antibody (upper panel), stripped and reprobed with anti-PCNA as a loading control. c MDA-MB-231 cells with stable scrambled shRNA (WT) or stable PARG shRNA, knockdown (KD), were treated with 5 μM JA2131 for 2 h, lysed and then total lysates were immunoblotted for anti-PAR (upper panel) followed by anti-PARG (upper middle panel), anti-PARP1 (lower middle panel) and anti-actin (bottom panel) antibody. Numbers on the blot shows relative band intensities determined by ImageJ. d The SRB cytotoxicity assay to evaluate JA2131 in MDA-MB-231 cells. Stable cells with scramble shRNA (control) and shPARG (PARG-KD) cells were treated with an increasing concentration of JA2131 for growth inhibitory IC 50 determination. The data was normalized against 100% survival at the lowest inhibitor concentration. Graphs for IC 50 were fitted to the four-parameter logistic equation using Prism8 and shown in the table (see methods). Error bars show the percent coefficient of variation. Equivalent concentrations of DMSO were used as a negative control. e The cytotoxicity of JA2131 along with PDD, Veliparib and Doxorubicin against MRC-5 cells. The IC 50 curves were generated as described above and shown in the table. Error bars show the percent coefficient of variation. f CETSA for in-cell JA2131 target engagement. Normalized anti-PARG western blot intensities of three independent experiments were fitted as above and the estimated aggregation temperature (T agg ) is shown in the tabulated insert. Source Data are provided as a Source Data file.
the most broadly consumed of all pharmacologically active compounds 40 .
The active JA2-4 series compounds (JA2-4 and JA2131) identified here display multiple complementary features and specific interactions with the hPARG active site, consistent with the observed sub-micromolar IC 50 . These PARGi are small in comparison to most drugs, so they provide suitable prototypes that suggest sites for modifications to improve potency and pharmacokinetic properties. Our current structural data furthermore reveals the molecular bases of the specific and competitive PARG inhibition by thio-methylxanthine derivatives (Figs. 2, 3). From the protein-ligand interactions identified here, the 6´-thiomethylxanthine derivatives JA2-4 and JA2131 compete with the adenine moiety of PAR substrates for active site binding.
Moreover, our structures unveil how the hPARG Tyr clasp switches between two conformational states to specifically engage methylxanthine derivatives containing different functional groups. This structural plasticity of the Tyr clasp is a key feature accompanying the binding of the potent JA2-4 series inhibitors.
The Tyr clasp is a unique feature of PARG that is absent in other PAR glycohydrolase macrodomains, such as ARH3 55 , suggesting why PARG-KD cells show specific PARG inhibition by JA2131 (Fig. 3). A similar strategy of anchored plasticity to induce changes by inhibitor binding, allowed the design of potent selective inhibitors of nitric oxide synthases 56 , supporting the value of this inhibitor type for PARG.
Importantly, identification and characterization of JA2131 as a selective PARGi enabled direct examination of PARG signaling and the nature of PARP1-PARG relationship in DNA damage responses in cells. Notably, PARP1 is both the main polymerase and acceptor of PAR in response to DNA damage. Our data reveal that PARGi trapped PARP1 in a PARylated form (Fig. 4b). DePARylated PARP1 showed a single step change in molecular weight that suggests a potent glycohydrolase with a fast PAR turnover rate for alleviating PARP inhibition imposed by autoPARylation at the DNA damage site. Indeed, this idea is consistent with the observed rapid PARG activity post DNA damage. ARTICLE PARG inhibition as a therapeutic option is directly questioned by a report showing PARG downregulation as a mechanism for acquiring PARPi resistance in BRCA2 and p53-null mouse cellline 33 . Therefore, we generated BRCA2 and p53 proficient PARPi-resistant SUM149 human triple-negative breast cancer cell lines and investigated PARG protein expression. Our results detected no observable change in PARG protein expression in these cells (see Supplementary Fig. 17). However, JA2131-induced PAR accumulation was comparable to those observed for PC3 cells (Supplementary Fig. 16 and Fig. 5a). These observations suggest that selective loss of PARG in PARPi resistance may strictly depend on the underlying genetic context. Indeed, the general importance of PARG in cancer progression is underscored by finding that catalytically active PARG leads to poor prognosis and survival, while expression of enzymaticallydefective protein played no part in tumor growth 35 . This is consistent with our TCGA database analysis, where most tumors show higher PARG expression compared to matched, normal tissues (Fig. 1a) and supports the possible value of a PARGi for ongoing investigations. Indeed, recently another small molecule PARGi was shown to delay DNA repair, sensitize cancer cells to DNA damaging agents, and be effective against PARPi-resistant tumors 57 . These data are in concert with the results presented here, underscoring the potential importance of PARGi cancer therapeutic.
Overall, our results identify unique targetable structural features of PARG that allowed us to identify and optimize a pharmacologically useful specific inhibitor. This PARGi is selective, cell-permeable, prevents PAR removal by PARG, and kills cancer cells at a level similar to Olaparib. As such, it is expected to contribute to understanding the roles of PARG and PAR turnover in multiple aspects of DNA damage responses. PARG is a unique mammalian protein without paralogs, in contrast to the human PARP family consisting of 17 related enzymes 58 . Thus, PARGi merits testing as potentially producing fewer off-target effects than PARP inhibitors currently employed as cancer therapeutics. With these points in mind, we hope that JA2131 will enable many studies evaluating the potential value of specifically inhibiting PARG in various cancer therapeutic strategies.
Protein expression and purification. The catalytic domain of rat PARG (residues 385-972) was purified as described previously 18 . The human PARG (hPARG) catalytic domain (residues 389-976) was cloned in pET28a (Novagen) with an Nterminal His-tag and expressed in E. coli Rosetta cells. hPARG 389-976 was purified by Ni-NTA (Qiagen) affinity chromatography, loaded onto a heparin column (GE Healthcare), and then eluted with a salt gradient (0-1 M NaCl). hPARG 389-976 was further purified, using a Superdex 200 size-exclusion chromatography. For the preparation of PARylated PARP1, the DNA binding domain (DBD; residues 1-374) of human PARP1 and the PARP1C catalytic domain (residues 375-1014) were purified as described previously 18 . For crystallization, human PARG catalytic domain (residues 448-976) containing the six surface entropy reduction mutations 20 was cloned in pET28a vector with an N-terminal His-tag and expressed in E. coli HMS174 cells expressing GroESL chaperon. The GST-tagged central BRCT domain of human XRCC1 (BRCT1, residues 294-417; cloned in pGEX-6p1) was expressed in E. coli Rosetta cells and purified by glutathione affinity chromatography. Following cleavage of the GST-tag with PreScission protease (GE Healthcare), the BRCT1 domain was purified using a Sephacryl 100 (GE Healthcare) size-exclusion column. For the biotinylation of BRCT1, the BRCT1 domain of XRCC1 was cloned in pGEX-6p1 with a C-terminal biotin acceptor peptide (BAP) tag and co-expressed with the BirA biotin ligase (pACYC184-BirA plasmid; Avidity) in E. coli BL21 (DE3) cells. The biotinylated BRCT1 was purified using the same protocol as the GST-BRCT1 protein.
TR-FRET PAR turnover assay. For a high-throughput screening (HTS) of PARG inhibitors, we developed a high-throughput PARG activity assay using the TR-FRET system (a manuscript describing the details of this assay will be published elsewhere). Briefly, the fluorescein molecule (FITC) was enzymatically incorporated into PARP1 in a reaction containing PARP1C (2 μM), the PARP1 DBD (2 μM), a 24 mer nicked DNA oligo (2 μM), and a mixture of unlabeled NAD + (Sigma) and FITC-NAD + (Trevigen) substrates (total NAD + concentration of 100 μM). After incubation for 1 h at 37°C, PARylated PARP1 was desalted using a PD-10 desalting column (GE healthcare) in a buffer containing 25 mM Tris-HCl (pH 7.5), 50 mM NaCl, and 0.01% NP-40 and used as the FRET acceptor. The biotin-tagged BRCT1 domain of XRCC1 (BRCT1, 5 nM) was conjugated with an equimolar concentration of Tb 3+ -chelated streptavidin (Life Technologies) in TR-FRET assay buffer (FRET donor). To monitor dose-response inhibition of PARG activity (supplementary Table 2), rat PARG 385-972 (12.5 nM) was pre-incubated with compounds for 1 h at room temperature. The reaction was started by the addition of the FRET pair of FITC-labeled PARP1 (42 nM) and Tb 3+ -BRCT1 (5 nM). The TR-FRET signal was subsequently monitored for 60 min.
High-throughput screening of PARG inhibitors. Small molecule inhibitors of PARG were identified by high-throughput screening (HTS) using the TR-FRET PAR turnover assay. Compounds (final concentration of 6.7 μM) from the NCI Diversity Set II library (1990 compounds) were pre dispensed into black polystyrene 384-well plates (Corning) using a Hummingbird liquid handler (Digilab). Then, 10 μl of rat PARG 385-972 (12.5 nM and 25 nM) were subsequently dispensed into plates, using a multi-drop combi nL reagent dispenser (Thermo Scientific) and incubated with compounds for 1 h at 4°C. The reaction was initiated by the addition of 20 μl of the premixed FRET pair of FITC-labeled PARP1 (42 nM) and Tb 3+ -BRCT1 (5 nM), and the TR-FRET signal was monitored every 4 min for 40 min (11 data points). Hits were selected by visually comparing the rate of the TR-FRET signal decrease in each well to that of negative control reactions containing DMSO (Fig. 1b and see Text for details). The HTS was performed at the High-Throughput Screening Core (HTSC) at Washington University School of Medicine.
Gel-based PARG activity assay. PARP1C (2 μM) is enzymatically auto-modified in a reaction containing the PARP1 DBD (2 μM), a 24 mer nicked DNA oligo (2 μM), and NAD + (300 μM) for 1 h at 37°C, as described previously 18 . Human PARG 389-976 (1 nM) was pre-incubated with compounds for 1 h at room temperature. The reaction was initiated by the addition of PARylated PARP1 (final concentration of 500 nM) and quenched after 30 min incubation at room temperature by adding SDS-PAGE sample buffer. The gel was visualized by sypro-ruby staining 18 and the retained PARylated PARP1C species were quantified using ImageJ 59 . PARylated PARP1C that migrates slower than unmodified PARP1C was quantified. Control reactions in the absence of PARG (100%) or compound (0%) were used for normalization of the degree of PARG inhibition by compounds (% inhibition). The IC 50 values were determined by fitting the dose-response data for each compound to a four-parameter logistic equation using SigmaPlot (Systat Software Inc.).
To test the specificity of JA2131, human PARG 389-976 (1 nM) was preincubated with JA2131 (5, 10, and 20 μM) in the presence and absence of 0.1% Triton-X 100 or 0.1 mg/ml BSA for 1 h at room temperature. The reaction was initiated, quenched, visualized, and quantified as described above.
Crystallization of human PARG catalytic domain. Crystals of the unliganded hPARG 448-976 were grown by hanging drop vapor diffusion as described previously 20 . hPARG 448-976 (7.5 mg/ml) was mixed with an equal volume of well solution, containing 16-24% (w/v) PEG 3350, 0.1 M PCTP (0.04 M sodium propionate, 0.02 M sodium cacodylate, 0.04 M Bis-Tris propane) pH 7.5 and 0.15 M MgCl 2 and incubated at 22°C. To soak inhibitors, crystals were harvested in a cryoprotectant solution containing 26% PEG 3350, 0.05 M PCTP pH 7.5, 0.1 M NaCl, 0.15 M MgCl 2 , and 2.5% glycerol. Crystals were soaked with inhibitors at 5 or 10 mM concentrations for 15 h, and then were flash-cooled in liquid nitrogen. X-ray diffraction data for five structures were collected in-house at 100 K using a Rigaku MicroMax007 rotating anode X-ray generator equipped with Xenocs Mirrors and a MAR image plate, and processed using HKL2000 60 and SCALEPACK 60,61 . All crystals have one hPARG molecule per asymmetric unit and diffracted to 1.7-2.0 Å resolution. X-ray diffraction data statistics are shown in Supplementary Table 3 of the online supplementary information. The structures of hPARG bound to the methylxanthine derivatives were determined by molecular replacement using Molrep of the CCP4i suite 62 with the unliganded hPARG 448-976 structure (PDB ID: 4B1G) 20 as a search model. The crystallographic models for the hPARG-inhibitor complexes were constructed using COOT 63 and refined using REFMAC 64 . Crystallographic data statistics are shown in Supplementary Table 3. All structural figures were prepared using PyMOL (www.pymol.org) or Chimera 65 .
To show reproducibility and confirm the in-house PARG-inhibitor structures, hPARG crystals were also prepared for synchrotron data collection. hPARG crystals were grown as described, soaked with 10 mM JA2131 or JA2120 compound for 1-2 h, then flash-cooled in liquid nitrogen. X-ray diffraction data were acquired at Stanford Synchrotron Radiation Lightsource (SSRL) beamline 9-2 66,67 . Data were processed with XDS to 1.6-1.7 Å 66,68 , solved by molecular replacement as described, and refined in Phenix 69 . Diffraction and crystallographic data statistics are shown in Supplementary Table 4 of the online supplementary information. PARG-inhibitor structures determined from synchrotron data were similar in resolution and conformation to complexes determined from in-house data despite having a different space group.
Cellular thermal shift assay (CETSA). CETSA was performed as described by Molina et al. 44 . Briefly, PC3 cells were treated with 10 μM JA2131 or an equivalent volume of DMSO for 2 h, washed three times with PBS and extracted for total cell lysate. Ten percent glycerol was added to the samples before subjecting to heat. Gradient thermal cycler (C1000 Touch, Bio-Rad) was used to heat samples at 42, 57, 66, 72, 77, and 85°C for 3 min and chilled on ice for 5 min. Then, samples were centrifuged at 14,000 rpm for 20 min at 4°C. Supernatants were carefully removed and analyzed by western blotting analysis.
Antibodies and western blots. Cells were grown in 6-well plates and 10 cm dishes, treated with PARGi for 2 h, irradiated and recovered for 1 h before harvesting. Cells were irradiated, ranging from 1 to 10 Gy using RAD SOURCE, RS-2000, Biological System, RadSource.com. Cells were washed with phosphate buffer and saline (PBS) and lysed with buffer containing 50 mM Hepes, pH 7.5, 1% (vol/ vol) igepal-C630, 1 mg/ml bacitracin, 1 mM EDTA, 10 mM NaF, 1 mM sodium orthovanadate, 10% (vol/vol) glycerol, 50 mM NaCl, 1 mM PMSF, and Protease Inhibitor Cocktail Set III (EMD Millipore). The detergent-soluble fraction was used for western blotting. Cell fractionation was carried out using the Subcellular Protein Fractionation Kit, (#78840, ThermoFisher) according to the manufacturer's instruction. Primary antibodies were purchased from the following sources; Actin, GAPDH, γH2AX and Histone-H3 was from Cell Signaling Technology. H2AX and LAMB1 were from Invitrogen/ThermoFisher. PCNA and p53 were from Santa Cruz Biotechnology. Anti-PAR (10 H) was from EMD Millipore and anti-PARG was from or Cell Signaling Technology. Anti-PARP1 from Abcam. Secondary antibodies were conjugated HRP (Cell Signaling) and ECL Clarity was purchased from Bio-Rad. All primary and secondary antibodies were diluted 1:1000 with 5% non-fat Blotting Grade Blocking Reagent in 1x TBST, except for anti-PAR 1:500. Uncropped western data are presented in the Supplementary Fig. 18.
Immuno-staining and imaging. PC3 cells were grown to 75% confluence in 24well 25 μm film bottom Eppendorf black cell Imaging plates (0030741005), treated with either DMSO or designated inhibitors for 2 h, irradiated 7 Gy X-ray and allowed to recover for an hour before fixing with 4% (wt/vol) paraformaldehyde, pH 8.0. All subsequent steps were performed at pH8.0. Cells were washed 6-7 times with PBS, permeabilized 0.5% Triton X-100 in PBS on ice for 20 min, blocked overnight with PBS/5% Goat serum/3%BSA/0.5% Triton X-100. Cells were incubated with primary antibody 1:250 dilution in PBS/5%BSA/0.5% Triton X-100 overnight at 4°C. Following 6× wash with PBS pH 8.0, cells were exposed to fluorescence-labeled secondary antibody diluted 1:500 in PBS/5%BSA/0.5% Triton X-100 for 1 h at room temperature. Following 6-7× PBS washes, cells were fixed again with 4% PFA at room temperature for 20 min. After a further 6× PBS washes cells were imaged in PBS buffer using ImageXpress high-content imager (Molecular Devices Inc). Statistical analysis was performed with MetaXpress Software.
Sulforhodamine B (SRB) cytotoxicity assay. Cells were plated in 96-well plates, treated as designated in reduced 5% FBS and fixed with trichloroacetic acid after 72 h, as previously described by 43 . PARG inhibitor JA2131 was titrated typically from 300 to 0.15 μM. Colorimetric or fluorescence analysis was performed in a FlexStation 3, using SoftMax Pro 7.0, Molecular Devices. Normalized graphs were generated with Prism 8 software, non-linear 4-parameter data fitting was performed for IC 50 calculation.
TCGA analysis. We used the TCGA-Assembler suite 70 with the Assay Platform option set to gene to obtain the normalized Rsem RNA-seq gene expression data from TCGA (https://cancergenome.nih.gov) project. Data were processed with in-house scripts (C++ and Bash) and plotted with the R package "ggpllot2" and "ggpubr".
Reporting summary. Further information on research design is available in the Nature Research Reporting Summary linked to this article.

Data availability
The data supporting the findings of this manuscript are available from the corresponding author upon reasonable request. Coordinates and structure-factor files for human PARG complexed with JA2-4, JA2-8, JA2-9, JA2120, and JA2131 have been deposited in the Protein Data Bank, with accession codes; 6O9X, 6O9Y, 6OA0, 6OA1, and 6OA3.