PARG inhibition in ATM-deficient prostate cancer: from mechanistic discovery to therapeutic potential

Abstract

Background

ATM deficiency is frequently observed in castration-resistant prostate cancer (CRPC). However, effective therapeutic vulnerabilities associated with this genetic alteration remain poorly defined. This study aimed to identify synthetic lethal strategies that selectively target ATM-deficient prostate cancer cells.

Methods

An unbiased small-molecule compound screening was performed to identify agents exhibiting selective cytotoxicity in ATM-deficient prostate cancer cells. Candidate vulnerabilities were validated across multiple prostate cancer cell lines with genetic depletion or restoration of ATM. Mechanistic studies were conducted using molecular and biochemical assays to assess DNA damage, replication stress, and PARylation dynamics. In vivo efficacy was evaluated using ATM-deficient xenograft tumor models.

Results

ATM-deficient prostate cancer cells exhibited marked sensitivity to pharmacological inhibition of poly (ADP-ribose) glycohydrolase (PARG) using with PDD00017273, an effect that was consistently observed across multiple cell lines and partially restored by ATM re-expression. Mechanistically, PARG inhibition induced persistent PARylation in ATM-deficient cells, not via canonical DNA double-strand break signaling, but via misincorporated ribonucleotides processed by topoisomerase 1 during DNA replication. This replication-associated PARylation resulted in severe replication stress, checkpoint activation, and accumulation of DNA double-strand breaks, ultimately leading to cell death. This cytotoxic mechanism is distinct from classical PAR-dependent cell death pathways, including NAD⁺ depletion and parthanatos. In vivo, PARG inhibition significantly suppressed the growth of ATM-deficient xenograft tumors cells.

Conclusions

This study identifies PARG inhibition as a previously unrecognized synthetic lethal vulnerability in ATM-deficient prostate cancer. These findings establish a mechanistic link between ATM loss, aberrant ribonucleotide processing, and replication-associated PARylation, supporting the clinical development of PARG inhibitors as a precision therapeutic strategy for ATM-deficient prostate cancer and potentially other malignancies harboring ATM deficiency.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12967-026-08208-9.

Background

Deleterious mutations in DNA damage response (DDR) genes are common in advanced prostate cancer (PCa) and they can be inherited or somatically acquired [1]. BRCA2 and ATM are the two most commonly mutated genes [2]. Although PARP inhibitors (PARPi) such as olaparib (OLA) and rucaparib (RUC), are the standard therapy for patients with BRCA1/2 mutations, the response rate to PARPi in patients with ATM mutations is relatively low [3–5].

ATM is a Ser/Thr kinase that plays a central role in the repair of DNA double strand-breaks (DSB) [6, 7]. When activated, it phosphorylates many downstream effectors, regulating cell cycle checkpoints and activating DNA repair [8]. ATM can also be activated by oxidative stress via disulfide-dependent covalent dimerization [9]. These protective mechanisms are impaired in ATM-deficient cells, precipitating the accumulation of DNA damage, high levels of reactive oxygen species (ROS), and/or cell death [10, 11].

The regulation of intracellular ADP-ribosylation (ADPr) homeostasis is important for cell survival against genotoxic stressors [12]. PARP1 is the primary DNA damage response PARP [13, 14]. Recent research has revealed that PARP1 works with the cofactor histone PARylation factor 1 (HPF1) to modify its itself, histones, and other proteins with poly (ADPribose) (PAR), which must be dynamically removed by poly (ADP- ribose) glycohydrolase (PARG) to prevent the trapping of proteins recruited in a PAR-dependent manner [15, 16]. PARP hyperactivity has been observed cells with ATM deficiency, which may result from high levels of DNA damage [17]. This implies that, ATM-deficient cells may be particularly sensitive to PARP inhibitors, providing a rationale for using them as potential therapeutic targets.

PCa is a common cancer among newly diagnosed male cancer cases in the United States [18]. The FDA has approved two PARP-targeting drugs, OLA and RUC, for treating patients with metastatic castration-resistant prostate cancer (mCRPC). As the disease advances, resistance inevitably emerges after the initial therapeutic response, resulting in a significant decline in treatment effectiveness [19]. Several genetic mutations have been implicated in the development and progression of CRPC, with mutations in tumor suppressor genes such as PTEN, TP53, and RB1 being the most commonly observed [2, 20]. These mutations are frequently associated with tumor metastasis, disease recurrence, and therapeutic resistance. However, mutations in DNA repair genes, such as ATM, although less frequent, are emerging as significant contributors to CRPC biology [21].

In this study, we used small-scale drug screening to identify candidate compounds from a library influenced by ATM alterations that could inform novel therapies. These results indicate that PCa cells lacking ATM exhibit increased sensitivity to PARG inhibitors due to elevated PARylation at the cellular level.

Methods

Drug screening

The CRISPR-Cas9 system was used to generate ATM-KO cells [22]. To screen the drug sensitivity profile of ATM KO cells, a comprehensive oncology drug library consisting of 265 preclinical and investigational compounds was used. Cells were treated with the compound at a final concentration of 10 µM, while all control wells were supplemented with an equal concentration of DMSO to account for solvent effects. After 5 days of incubation, the inhibition rate was determined for each well.

Public data

Somatic mutation profiles were curated from public repositories and visualized using cBioPortal to support integrative genomic analyses [23]. For the Cancer Cell Line Encyclopedia (CCLE) dataset, the gene expression matrix of PCa cell lines was obtained from the official CCLE website (https://sites.broadinstitute.org/ccle/)(2).

Cell culture

PC3, DU145, 22Rv1, and C4-2B cell lines were obtained and cultured according to the ATCC recommendations. These cell lines were cultured in complete RPMI 1640 medium (Gibco) containing 10% fetal bovine serum (FBS) and 1% penicillin/streptomycin antibiotic solution. For all PARGi treatments, unless stated otherwise, cells were stimulated with 10 µM PDD00017273 (Merck, SML1781), while control samples received an equivalent concentration of DMSO. Other compounds were used in this study were as follows: PARPi OLA (MCE, HY-10162), and nucleotide, nicotinamide phosphoribosyltransferase (NAMPT) inhibitor daporinad (MCE, FK866), and ATM Inhibitor KU-55,933 (Selleck, S1092).

Colony formation

For the colony formation assay, 500 PCa cells were seeded per well in six-well plates and cultured for 14 days. Afterward, the colonies were fixed with 4% paraformaldehyde (PFA) and stained with crystal violet for approximately 15 min.

Cell viability

Cell viability was evaluated using the Sulforhodamine B (SRB) assay as described in a previous study [24]. Cells were treated with PARGi for 5 days, fixed with 4% PFA, and stained with 0.4% SRB solution. After incubation for 10 min, cells were washed with 0.01 M acetic acid. The dye bound to cellular proteins was solubilized in 10 mM Tris, and the optical density was measured at 540 nm using a spectrophotometer.

Immunoblotting

Immunoblotting was performed as described in our previous study [25]. Primary antibodies against β-actin (CST, #4967), ATM (CST, #2873), γ-H2AX (Millipore, 05–636), PARP1 (Proteintech, 13371-1-AP), Lamin B1 (Proteintech, 12987-1-AP), GAPDH (Proteintech, 60004-1-Ig), Histone H3 (Proteintech, 17168-1-AP), APE1(Proteintech, 10203-1-AP), RPA32 (Proteintech, 10412-1-AP), phosphorylated RPA32 (Abcam, ab109394), Chk1 (Proteintech, 25887-1-AP), and phosphorylated Chk1 (CST, #2344) were used. PARylation proteins were detected using immunoblotting with an anti-PAR antibody (CST, #89190). After incubating with primary antibodies, membranes were incubated overnight on a shaker at 4 °C. The following day, the membranes were washed three times with PBST, incubated with an HRP-conjugated secondary antibody (1:2000, Proteintech) at room temperature for 1 h, and washed three times with PBST before signal detection. Images were captured using a Tanon-5200 Chemiluminescent Imaging System (Tanon, China).

DNA fiber assay

The DNA fiber assay was performed as previously described [26]. The cells were incubated with CldU and IdU after exposure to PARGi (10 µM) for 24 h. Cells were harvested, lysed, and single DNA molecules were spread on microscope slides. The cells were incubated with primary antibodies, followed by secondary antibodies. Imaging was performed using a Leica SP8 confocal microscope, and the data were analyzed using the LASX software.

Comet assay

Comet assays were performed using a Comet Assay Kit (ab238544; Abcam, United States) according to the manufacturer’s instructions. The cells were detached, resuspended in ice-cold PBS, and mixed with pre-warmed comet agarose. The cell-agarose mixture was applied to slides and incubated at 4 °C for 15 min in the dark. Then, slides were incubated in ice-cold lysis buffer for 60 min, followed by treatment with an alkaline solution for 30 min. After electrophoresis in ice-cold alkaline electrophoresis solution at 1 V/cm for 10–15 min, the slides were washed with chilled water and immersed in 70% ethanol for 1 min. After drying in the dark, the slides were imaged using a Leica SP8 confocal microscope with an FITC filter. DNA tail moments were quantified using ImageJ with the OpenComet plugin [27].

CRISPR-Cas9 and small interfering RNA (siRNA)

CRISPR-Cas9 gene editing was performed according to an established protocol [28]. The specific sgRNA sequences are provided in Supplementary Table S1. For siRNA transfection, commercially available siRNA products (siTop1, Santa Cruz, sc-36694; siPARG, Santa Cruz, sc-106355) were used, and transfection was conducted ccording to the manufacturer’s instructions using Lipofectamine™ RNAiMAX reagent (Invitrogen) for 72 h. Knockout or knockdown efficiency was determined using Western blot or quantitative polymerase chain reaction.

BrdU labeling and cell cycle analysis

BrdU labeling was performed as described previously [29]. BrdU labeling was performed by immunofluorescence staining using a mouse anti-BrdU primary antibody (BD Biosciences), followed by DNA denaturation with 2 M HCl for 30 min at room temperature (RT). Subsequently, nuclear staining with DAPI was performed, and images were acquired using an inverted fluorescence microscope. Cell cycle profiling was performed using a Cell Cycle Analysis Kit (Beyotime, China). Briefly, 1.0 × 10⁵ cells were seeded into six-well plates and cultured for 48 h under standard conditions. Following enzymatic dissociation and PBS washes, cells were fixed in 70% ice-cold ethanol at 4 °C for 24 h. After fixation, cells were gently rinsed with chilled PBS two times and incubated in the dark at 37 °C for 30 min in staining buffer (0.5 mL) supplemented with 10 µL RNase A and 25 µL PI. Then, the analysis of the cell cycle was analyzed using a BD Fortessa flow cytometer.

Xenograft model

All animal experiments were approved by the Ethical Review Board of Shanghai Changzheng Hospital. Thirty-six BALB/c nude mice (male, 6–8 weeks) were obtained from GemPharmatech company in Nanjing, China. Animals were housed and bred in an SPF facility. For the xenograft model, experimental procedures were performed as described in previous studie [30]. Approximately 1 × 10⁶ DU145 or 22RV1 cells were subcutaneously injected into each mouse. Then, at 5–7 days after inoculation, mice were randomly divided into two groups (n = 10 or 8 per group) and treated with daily intraperitoneal injections for 14 consecutive days with either vehicle control (30% Solutol) or PARGi (COH34, 10 mg/kg formulated in 30% Solutol containing 0.1% DMSO, MCE, HY-128760) [31, 32]. Tumor size and volume were measured and recorded every three days until the tumor reached 1000–1500 mm³.

Statistical analysis

The Mann–Whitney statistical test was performed for the DNA fiber experiments and the comet assay. For other comparisons between two groups, we performed t-tests. For comparisons involving more than two groups, we used analysis of variance to evaluate overall differences. A p-value < 0.05 was considered statistically significant. Immunoblotting experiments were performed in at least three independent replicates. DNA fiber, immunofluorescence, and comet assays were conducted with at least two independent replicates. All statistical analyses were conducted using GraphPad Prism (version 8.4.3).

Results

ATM-deficient cancer cells are exquisitely sensitive to PARG inhibition

A comparison of primary tumors from TCGA with CRPC from the SU2C dataset, along with mCRPC data from MCTP, revealed that, in addition to the commonly observed mutations in PTEN, TP53, and RB1, ATM mutations, predominantly characterized by deep deletions and missense mutations, were found at a higher frequency in CRPC (6%–8%) than in primary tumors. The ATM mutation rate in the mCRPC dataset from MCTP was 12%. Although this difference is modest, even low-frequency mutations in DNA repair genes, such as ATM, can contribute to tumor progression and therapy resistance in CRPC, underscoring this as an important area for further investigation (Fig. 1A).

Fig. 1.

ATM deficiency confers selective vulnerability to PARG inhibition in prostate cancer cells. (A) Frequency of gene alterations in PTEN, TP53, RB1, and ATM across primary prostate tumors from TCGA and CRPC samples from the SU2C dataset. (B) Schematic overview of the compound screen testing 265 small-molecule inhibitors with defined targets in DU145 ATM-wild-type (ATM-WT) and ATM-knockout (ATM-KO) prostate cancer cells. (C) Representative structures and ranking of the top hits from the inhibitor screen. PDD00017273 (PDD), a selective PARG inhibitor, emerged as a top compound that selectively suppressed ATM-KO but not ATM-WT cell growth. (D) Colony formation assays validating the selective sensitivity of ATM-KO cells to PDD at concentrations of 10 or 15 µM (n = 3). “ns” indicates no statistically significant difference; ***:p < 0.001. (E) ATM expression levels in prostate cancer cell lines. (F) ATM-deficient PCa cells are more sensitive to PDD than ATM-proficient lines, as measured by colony formation assays at concentrations of 10 or 15 µM (n = 3). “ns” indicates no statistically significant difference; *:p = 0.0272, ***:p < 0.001. (G) Rescue of PDD sensitivity in ATM-deficient C4-2B cells by stable overexpression of ATM cDNA, confirming ATM dependency of the phenotype. (H) Combined treatment with ATM inhibitor (2.5 µM) and PARG inhibitor (5 µM) synergistically suppresses colony formation in DU145 and 22Rv1 cells (n = 3). “ns” indicates no statistically significant difference; ***:p < 0.001

To investigate cellular vulnerabilities conferred by ATM deficiency, we examined the differential sensitivities of parental ATM-wild-type (ATM-WT) and ATM-knockout (ATM-KO) DU145 PCa cells (Supplementary Fig. 1A) to 265 inhibitors whose targets have been elucidated (Fig. 1B). Among the hits were compounds targeting previously described dependencies conferred by ATM mutations, including histone deacetylase, DNA synthesis, and topoisomerase inhibitors. Unexpectedly, PDD00017273 (PDD) (Fig. 1C, Supplementary Fig. 1B), which selectively inhibits the PARG, was among the top hits that reproducibly suppressed the growth of ATM-KO cells but not ATM-WT cells. The selective sensitivity of ATM-KO cells to PDD was validated by measuring cell survival in colony formation assays (Fig. 1D, Supplementary Fig. 1F). In the presence of PARGi, ATM-WT cells exhibited a slight inhibition, although the effect was significant in ATM-KO cells. Given that different PCa cell lines exhibited differential ATM expression, we assessed ATM expression in PC3, 22RV1, and C42-B cells (Fig. 1E). Next, we tested the sensitivity of these three PCa cell lines to PDD. This analysis included one ATM-deficient and two ATM-proficient cell lines. ATM-deficient cancer cells were more sensitive to PDD than ATM-proficient cancer cells, as validated by measuring cell survival in colony formation assays (Fig. 1F). The preferential lethality of PDD in C42B ATM-deficient cancer cells was partially rescued by stable over expression of the ATM cDNA (Fig. 1G), confirming that ATM deficiency was responsible for sensitivity to PDD. Next, the combination of the ATM inhibitor and PARGi significantly inhibited the colony formation in DU-145 and 22RV1 cells (Fig. 1H, Supplementary Fig. 1E).

PARG inhibition leads to a trend of PARylation accumulation in ATM-deficient cells

PDD is a potent and selective inhibitor of PARG that catalyzes the hydrolysis of O-glycosidic linkages in ADP-ribose polymers. Inhibition of PARG has been demonstrated to cause the intercellular PARylation accumulation. Inhibition of PARG has been demonstrated to cause the intercellular PARylation accumulation. Therefore, we examined whether the lethal effects of PDD on ATM-deficient cells were due to PARylation accumulation. PDD treatment increased PARylation levels in ATM-KO cells in a dose- and time-dependent manner, with a significant increase compared to ATM-WT cells (Figs. 2A-B). Small punctate PARylation foci were observed in the nuclei of ATM-KO cells upon treatment with PDD, as indicated by immunofluorescence (Fig. 2D). In contrast, the same PDD concentrations did not affect MARylation activity (Supplementary Fig. 3A). The increase in PARylation observed in ATM-KO cells was reproduced by siRNA-mediated depletion of PARG, excluding potential off-target effects of PDD (Fig. 2C, Supplementary Fig. 1D). Similarly, PARG inhibition combined with ATM loss in 22Rv1 cells yielded similar results, underscoring the reproducibility and general relevance of these observations (Supplementary Fig. 1D).

Fig. 2.

PARG inhibition induces PARylation accumulation specifically in ATM-deficient cells. (A–B) Immunoblot analysis showing dose- and time-dependent PARylation accumulation in ATM-KO but not ATM-WT DU145 cells following PDD treatment(dose: 0, 1, 5, 10, 15 µM; time:0, 0.5, 2, 4, 6, 8, 12 h). (C) SiRNA-mediated PARG knockdown recapitulates PARylation accumulation in ATM-KO cells, excluding off-target effects of PDD. (D) Immunofluorescence analysis of PAR expression in ATM WT and ATM KO cells treated with PARGi and/or PARPi. Cells were treated with DMSO (Ctrl), PARGi 10 µM, and PARGi 10 µM + PARPi 1 µM for 24 h. (Left) Representative images of ATM WT cells showing DAPI (nuclear staining), PAR, and merged images. (Right) Representative images of ATM KO cells with the same treatments. PAR expression is observed in the cytoplasm and nucleus, with reduced PAR accumulation in the presence of PARPi, especially in ATM WT cells. Bar graph showing the percentage of cells with greater than 5 PAR foci. “ns” indicates no statistically significant difference; **:p = 0.0019. (E) Western blot analysis of PAR levels in ATM WT and ATM KO cells treated with PARGi (10 µM) and/or PARPi (1 µM). Cells were treated for 24 h with PARGi and/or PARPi, and PAR expression was detected by Western blotting. The blot shows increased PAR accumulation in ATM WT cells treated with PARGi, which is reduced in the presence of PARPi. (F) Colony formation assays in DU145 and 22RV1 cells treated with PARGi and/or PARPi. ATM WT and ATM KO cells were treated with increasing concentrations of PARGi (0, 10, 15 µM) and/or PARPi (1 µM) for 10 days. The colonies were stained with crystal violet, and the images show the effects of treatment on colony formation. The presence of PARPi significantly reduces colony formation in both DU145 and 22RV1 cells, with a stronger effect observed in ATM WT cells

Next, we examined PARylation signals by co-treating cells with PARG and PARP inhibitors to determine whether the PARylation accumulation in ATM-KO cells was PARP-dependent. The addition of low concentrations of the PARP1/2 inhibitors OLA (PARPi) suppressed the increased PARylation level in both ATM-KO and ATM-WT cells (Fig. 2E, Supplementary Figs. 2E and 3B-C). PARP inhibition by OLA rescued the phenotype of PARG inhibition (Fig. 2F).

PARylation accumulation in ATM-deficient cells is replication-coordinated

These results demonstrate that ATM deficiency contributes to endogenous PARylation; therefore, identifying the predominant sources of PARylation in ATM-deficient cells is important. With respect to DNA damage signaling, PARP1 binds to and is activated by both DNA double-strand breaks. However, despite the ATM deficiency alone increasing the γH2AX level, a marker of DNA double-strand breaks, no additional γH2AX induction was observed following short-period suppression of PARG (Fig. 2B). Moreover, methyl methanesulfonate (MMS), an alkylating agent, significantly increased PARylation levels in ATM-WT cells (Fig. 3A). Conversely, only a slight increase in PARylation was observed in ATM-KO cells, despite MMS causing more DNA damage (Fig. 3A). These data suggest that the majority of detectable PARylation in ATM-KO cells is not caused by DNA damage.

Fig. 3.

PARylation in ATM-deficient cells originates from replication-associated ribonucleotide processing. (A) Immunoblot showing that methyl methanesulfonate (MMS; durations: 0, 5, 10, 20, 40, 60 min) strongly induces PARylation in ATM-WT but only marginally in ATM-KO cells. (B) Co-immunostaining of PAR and PCNA showing PARylation foci colocalized with replication sites during S phase in ATM-KO cells. (C) Short incubation with the DNA polymerase inhibitor Aphidicolin (APH) abolishes PARGi (10µM) -induced PARylation. (D) BrdU labeling showing no ssDNA foci in ATM-WT or ATM-KO cells under basal or PARGi (10µM) -treated conditions, excluding replication-gap-induced PARP activation. (E) Western blot analysis of PAR, APE1, TOP1, γH2AX, and ACTIN levels in DU145 ATM KO cells and ATM WT/KO cells treated with PARGi (10 µM) and siRNA targeting APE1 or TOP1. (Left) DU145 ATM KO cells were transfected with empty vector (EV) or APE1 knockdown (APE1 KD) and treated with PARGi for 24 h. (Right) ATM WT and ATM KO cells were transfected with siCTRL or siTOP1 and treated with PARGi for 24 h. The blots show PAR accumulation, APE1 or TOP1 expression, and γH2AX as a marker for DNA damage response. (F) Quantification of genome-embedded ribonucleotides showing higher levels in ATM-KO vs. WT cells. Exogenous nucleotide supplementation restores resistance to PARGi. Colony formation assays for ATM-WT, ATM-KO, and ATM-KO + Nuc (200µM) treated with varying concentrations of PARGi (0, 10, 15 µM). (G) Immunofluorescence showing reduced PARylation signals after nucleotide supplementation in ATM-KO cells. (H) Immunoblot confirming decreased PARylation protein levels upon nucleotide supplementation (PARGi: 10µM)

We considered whether PARylation occurs during DNA replication. Most of the PARylation detected was present in S phase and localized to or at sites of DNA replication, as indicated by co-immunostaining with anti-proliferating cell nuclear antigen (PCNA) antibody (Fig. 3B). To further confirm that PARylation was accumulated during the S phase, we treated cells with the DNA replication inhibitor aphidicolin (APH). APH is an inhibitor of DNA replication by targeting α, ε, and δ DNA polymerases. Studies have demonstrated that APH specifically binds to α DNA polymerase, forming a pol α-DNA-APH ternary complex that blocks DNA replication. Strikingly, short incubation with APH almost completely blocked the appearance of PARylation, confirming that DNA replication was the source of PARylation accumulation in ATM deficient cells (Fig. 3C).

Processing embedded ribonucleotide in ATM-deficient cells causes PARylation accumulation

To explain the above result, we examined the possibility that PARylation was triggered during replication at at one or more canonical DNA replication intermediates. For example, in BRCA1/2 deficient cells, PARP1 is activated during S phase by replication gaps due to Okazaki fragment processing (OFP) defects. However, we cannot detected ssDNA foci labeled with BrdU in either ATM WT or KO cells under unperturbed conditions or under PARG inhibition (Fig. 3D, Supplementary Fig. 2B). In contrast, small punctate BrdU foci were observed following short- term incubation of PARPi. Elevated PARylation in ATM KO cells was also does not occur during their excision by DNA base excision repair (BER), because depletion of the APE1 endonuclease that excises abasic sites during BER failed to reduce PARylation levels.

Next, we considered the possibility that misincorporated ribonucleotides contributed to the PARylation accumulation observed in ATM-deficient cells. Ribonucleotides are incorporated into DNA by Polymerases α, ε, and δ during replication. The presence of ribonucleotides in DNA can cause mutations and may lead to the generation of single-and double-strand breaks. We agree with this idea; we observed the higher levels of genome-embedded ribonucleotides in ATM-KO cells than in wild-type cells (Fig. 3F). The resistance of ATM KO cells to PARGi was restored after exogenous nucleotide supplementation (Fig. 3F). Furthermore, we used immunofluorescence to demonstrate that after exogenous supplementation of nucleic acids, the PARylation levels in the KO cells also decreased after exogenous nucleic acid supplementation (Fig. 3G). A similar change was observed in the PARylation expression at the protein level (Fig. 3H). To process the embedded ribonucleotides in the genome, cells relay on RNase H2 and, alternatively, topoisomerase 1 (TOP1)-mediated ribonucleotide excision pathway. Short-term TOP1 depletion using short interfering RNAs (siRNAs) reduced the levels of PARylation accumulation in ATM-deficient cells (Fig. 3E). Furthermore, TOP1 depletion reduced the level of γH2AX in ATM-deficient cells to levels similar to those in WT cells (Fig. 3E). These results suggest that TOP1 processing of genome-embedded ribonucleotides during replication leads to PARylation accumulation in ATM-deficient cells.

PARG inhibition interferes with DNA replication and causes DNA damage in ATM-deficient cells

Next, we investigated the mechanisms underlying the cytotoxicity of PARG inhibition. Parthanatos, a caspase-independent programmed cell death, is triggered by the PAR accumulation. Excess PAR induces mitochondrial release of apoptosis-inducing factor (AIF), which binds to macrophage migration inhibitory factor (MIF) and transports MIF into the nucleus, where it cleaves genomic DNA. However, we failed to detected AIF nuclear translocation following PARG inhibition (Supplementary Fig. 2A). PARylation is a major cellular sink for NAD+, and PARG inhibition has been shown to reduces cellular NAD+ levels in the presence of DNA damage inducers. Thus, we examined whether PARGi-induced accumulation of PARylation in ATM-KO cells was accompanied by NAD+ depletion. Although there was a slight trend toward a reduction in NAD+ levels in ATM-KO cells, no significant reduction in NAD+ levels was detected in either ATM-KO or WT cells after PARGi treatment (Supplementary Fig. 2D). In contrast, treatment with FK866, a highly specific inhibitor of NAMPT, an essential enzyme in NAD+ biosynthesis, significantly reduced NAD+ levels in both WT and ATM-KO cells, thereby confirming the sensitivity of the employed NAD+ assay (Supplementary Fig. 2D). Furthermore, FK866 treatment reduced PARylation levels and rescued the phenotype caused by PARG inhibition (Figs. 4A-B). These results suggest that NAD+ depletion is unlikely to be the major cause of cell death following PARG inhibition in ATM-KO cells. Then, we determined whether persistently elevated PARylation in PARGi-treated ATM-deficient cells disrupts cell-cycle progression. PARGi treatment significantly increased the proportion of S-phase cells in ATM-KO (Supplementary Figs. 2B-C). To better understand the phenotype associated with the prolonged S phase, we evaluated replication dynamics using DNA combing to visualize single DNA molecules and measure replication fork progression. DNA fiber imaging and quantification revealed that ATM-KO cells exhibited significantly shorter nascent DNA tracks than WT cells after PARGi treatment, demonstrating impaired replication progression (Fig. 4C).

Fig. 4.

PARG inhibition disrupts DNA replication and induces DNA damage in ATM-deficient cells. (A) Colony formation assay of ATM WT and ATM KO cells treated with PARGi and/or NAMPTi. ATM WT and ATM KO cells were treated with increasing concentrations of PARGi (0, 10, 15 µM) and 10 µM NAMPTi. (B) Western blot analysis of PAR levels in ATM WT and ATM KO cells treated with PARGi (10 µM) and/or NAMPTi (10 µM). Cells were treated for 24 h with PARGi and NAMPTi, and PAR expression was detected by Western blotting. The blot shows significant PAR accumulation in ATM WT cells treated with PARGi, which is reduced upon co-treatment with NAMPTi. (C) DNA fiber assays showing shortened nascent DNA tracks in ATM-KO cells upon PARG inhibition, indicating replication-fork slowing. “ns” indicates no statistically significant difference, in ATM WT, p = 0.0741; in ATM KO, **:p = 0.0019. (D) Immunofluorescence of γH2AX showing marked DNA double-strand-break accumulation in ATM-KO but not WT cells after prolonged PARG inhibition (10 µM, 24 h). (E) Immunoblot showing phosphorylation of CHK1 and RPA32, consistent with replication-stress signaling activation in ATM-KO cells following PARG inhibition (10 µM). (F) Western blot analysis of DNA damage response markers in ATM WT and ATM KO cells treated with PARGi (10 µM) and/or PARPi (1 µM). Cells were treated for 24 h, and the levels of phosphorylated RPA (p-RPA), total RPA, phosphorylated CHK1 (p-CHK1, S137), and total CHK1. (G) Alkaline comet assay revealing DNA strand breaks in ATM WT and ATM KO cells after treatment with PARG inhibitor (PARGi, 10 µM). Cells were treated with DMSO (control) or PARGi for 24–48 h, with and without the addition of PARP inhibitor (PARPi) at 48 h. The images show significant DNA damage in ATM KO cells upon PARG inhibition, which is alleviated by the addition of PARPi. (H) Quantification of DNA damage from the alkaline comet assay. The average tail moment was measured in ATM WT and ATM KO cells treated with DMSO, PARGi (24 and 48 h), and PARGi + PARPi (48 h). Data points represent individual cells, with the bars indicating the mean ± SEM. Statistical significance is indicated by *** (p < 0.001). (I) In vivo tumor growth analysis of DU145 cells with and without PARG inhibition. (Top) Representative images of tumors from control and PARGi-treated mice. (Bottom) Tumor volume (mm³) measured over time (days) in control and PARGi-treated groups (n = 10). Statistical significance is indicated by *** (p < 0.001). Tumor growth is significantly inhibited in the PARGi-treated group compared to the control group. (Middle) Representative H&E staining and immunohistochemical analysis of Ki67 and PAR expression in DU145 tumors treated with control or PARGi, showing reduced proliferation and PAR levels upon PARG inhibition. (J) In vivo tumor growth analysis of 22RV1 cells with and without PARG inhibition. (Top) Representative images of tumors from control and PARGi-treated mice. (Bottom) Tumor volume (mm³) measured over time (days) in control and PARGi-treated groups (n = 8). Statistical significance is indicated by *** (p < 0.001). PARGi treatment significantly inhibits tumor growth compared to the control group. (Middle) Representative H&E staining and immunohistochemical analysis of Ki67 and PAR expression in 22RV1 tumors from ATM WT and ATM KO mice, showing the effects of PARG inhibition on proliferation and PAR levels

ATR plays a key role in the response to replication stress (RS) and functions as an S-phase checkpoint kinase. Once activated, ATR phosphorylates downstream effectors such as the CHK1 effector kinase and RPA itself. Consistent with RS response activation, we observed a significant increase in CHK1 and RPA32 phosphorylation in ATM-KO cells. In contrast, this signal was barely detectable in WT cells (Fig. 4E) and was reversed upon PARPi treatment (Fig. 4F).

Prolonged inhibition of DNA replication induces double-stranded DNA breaks and cell death

Then, we investigated whether PARGi caused DNA damage in ATM-KO cells. Staining for γH2AX was performed on ATM-KO and WT cells. γH2AX was barely detectable in WT cells, whereas high levels were detected in ATM-KO cells following long-term PARGi (Fig. 4D). To further explore the type of accumulated DNA damage, we performed alkaline comet assays. These assays detect DNA strand breaks, by measuring nuclear DNA tails after electrophoresis (Fig. 4G). A slight increase in the number of nuclear tails was observed in WT cells. In contrast, ATM KO cells nuclei exhibited extensive DNA strand breaks, which were more pronounced at 48 h. PARP inhibition by OLA alleviated the strand breaks (Fig. 4H). These results demonstrate that DNA damage accounts for the preferential cytotoxicity of PARG inhibitors in ATM deficient cancer cells. Finally, we extended our findings in vivo. The xenograft tumor model of nude mice, using ATM-KO cell lines (DU145 and 22RV1), demonstrated that ATM KO enhanced the sensitivity to PARGi and increased PARylation (Fig. 4I, J).

Discussion

PCa is characterized by heterogeneity in therapeutic responses and clinical outcomes [33]. This heterogeneity is closely associated with the genomic landscape of tumors [34]. ATM, a DNA damage–sensing gene, is frequently altered in mCRPC [35–37]. ATM knockdown or inhibition sensitizes cancer cells to radiotherapy [38, 39]. A recent prospective study demonstrated that in patients with mCRPC, the presence of germline BRCA/ATM mutations was associated with improved efficacy of first-line novel hormonal therapy [40]. In this study, we integrated in vivo genomics with high-throughput in vitro drug screening to identify potential therapeutic strategies for ATM-deficient CRPC.

This study identified that ATM-deficient cancer cells are highly susceptible to PDD00017273, a selective PARG inhibitor, with the underlying mechanism being attributable to the aberrant PARylation buildup during DNA replication. In contrast to earlier studies that mainly examined the association between ATM deficiency and responsiveness to PARP inhibitors, this study identified PARG inhibition as a novel synthetic lethal target for therapeutic intervention in ATM-deficient CRPC.

We discovered that PARylation accumulates in ATM-deficient cells, largely independent of canonical DNA damage signaling, yet it is intrinsically linked to DNA replication. Furthermore, these findings indicate that misincorporated ribonucleotides are a primary source of this PARylation, which is then processed by TOP1. This mechanism stands in contrasts with BRCA1/2-deficient contexts, where defective Okazaki fragment maturation results in replication gaps and dependency on PARP activity [41, 42]. Although both ATM and BRCA1/2 deficiencies converge on elevated dependence on ADP-ribose metabolism, the upstream determinants of PARP activation are fundamentally distinct. These findings reveal a previously unrecognized role of ATM in regulating ribonucleotide-driven PARylation during replication, thereby establishing a new conceptual link between defects in the DNA repair pathway defects and ADP-ribose metabolism. Furthermore, incorporating ribonucleotide burden or TOP1 dependence holds promise for enhancing patient selection for PARP inhibitor-based therapies. These biomarkers may facilitate the identification of patients whose tumors exhibit heightened susceptibility to PARP inhibition, driven by specific RS or DNA repair defects. Although this remains an area of ongoing investigation, integrating these biomarkers with ATM status could provide a more nuanced and precise approach to select patients who are most likely to benefit from such therapeutic strategies.

ATM-deficient cells exhibited pronounced sensitivity to PARGi, we also observed a modest effect in ATM-competent cells, particularly with PDD00017273 treatment. Although ATM-competent cells were less sensitive than ATM-deficient cells, PARG inhibition induced a significant increase in PARylation. This suggests that, although the absence of ATM enhances sensitivity, PARG inhibition can still impact ATM-WT cells, highlighting its broader applicability as a therapeutic strategy. These observations are consistent with previous studies reporting PARylation accumulation and sensitivity to PARG inhibition in wild-type cancer models [43, 44]. This underscores the potential of targeting PARP/PARG-mediated pathways for therapeutic intervention in a wide range of PCa contexts.

Inhibition of PARG in ATM-deficient cells led to disruption of replication homeostasis and genome stability. Blocking PARG activity led to pronounced RS, characterized by S-phase arrest and shortened replication tracks, as well as robust activation of RS signaling, including CHK1 and RPA32 phosphorylation. This distinct vulnerability underscores the potential to exploit RS through the accumulation of DNA double-strand breaks, as evidenced by γH2AX staining and comet assays. These cytotoxic effects were mechanistically distinct from previously described PAR-dependent cell death pathways such as NAD⁺ depletion or parthanatos, neither of which was significantly activated under our experimental conditions [45]. Instead, these results indicated that replication-associated DNA damage as the primary determinant of lethality. This distinct vulnerability highlights an unexplored opportunity to exploit RS in ATM-deficient CRPC via PARG inhibition.

In summary, these findings suggest that ATM deficiency could serve as a predictive biomarker for therapeutic response to PARG inhibitors. Given that ATM alterations are highly enriched in mCRPC, this vulnerability may have direct clinical relevance for a subsets of patients with limited treatment options. Importantly, ATM-deficient tumors often display attenuated responses to PARP inhibitors, underscoring the need for alternative synthetic lethal strategies. Although this mechanistic understanding highlights the effectiveness of PARP inhibition in certain DNA repair-deficient cancers, it does not necessarily imply that PARP inhibitors should be universally combined or sequenced with all treatments. The effectiveness of these strategies may depend on additional molecular contexts, such as the presence of other genetic alterations or a specific tumor microenvironment.

Nevertheless, this study has several limitations. This study primarily relied on in vitro models and xenograft systems, and validation in more physiologically relevant settings, such as patient-derived organoids or genetically engineered mouse models, will be critical.

Conclusions

This study identifies PARG inhibition as a novel and selective synthetic lethal strategy for ATM-deficient PCa. We demonstrate that ATM loss leads to aberrant accumulation of replication-associated PARylation driven by misincorporated ribonucleotides and TOP1-mediated processing, rather than canonical DNA damage signaling. In this context, inhibition of PARG in this context disrupts replication homeostasis, induces severe RS and DNA double-strand breaks, and ultimately results in cancer cell death through mechanisms distinct from classical PAR-dependent cell death pathways such as NAD⁺ depletion or parthanatos. ATM deficiency confers heightened sensitivity to PARG inhibition in vitro and in vivo, highlighting ATM loss as a potential predictive biomarker for therapeutic response. These findings uncover a previously unrecognized vulnerability in ATM-deficient cancers and provide a strong rationale for the clinical development of PARG inhibitors as a precision therapy for patients with ATM-deficient castration-resistant PCa and potentially other malignancies harboring ATM alterations.

Electronic Supplementary Material

Below is the link to the electronic supplementary material.

Abbreviations

PCa

Prostate cancer

CRPC

Castration-resistant prostate cancer

PARG

Poly (ADP-ribose) glycohydrolase

DSB

Double strand breaks

DDR

DNA damage response

ADPr

ADP-ribosylation

Author contributions

Shancheng Ren, Zhixiang Xin, and Weidong Xu designed this work. Xuan Zhou wrote the manuscript. Xuan Zhou and Chunyu Guo conducted all in vitro and in vivo experiments and contributed to data acquisition. Zhiguo Fan and Yihaoyun Lou were responsible for performing the statistical analyses and data interpretation. Zhiying Yue performed the data review.

Funding

This article was funded by the National Natural Science Foundation of China (8233009), Shanghai Rising-Star Program (23QA1408100), Shanghai Municipal Health Commission (2022YQ065).

Data availability

The public data utilized in this article are freely and openly accessible via cBioPortal and CCLE [2, 23]. Data and materials for all experiments are available upon request.

Declarations

Ethical approval and consent to participate

All animal experimentation proceeds according to the Standard of IACUC (Institutional Animal Care and Use Committee) and performs in according to an established protocol approved by the Ethic Committee of Changzheng Hospital. Ethics approval was obtained by the Institutional Ethics Committee of the Shanghai Changzheng Hospital Hospital.

Consent for publication

All authors have read and approved the manuscript.

Competing interests

The authors have no conflicts of interest to declare.