Synergism between ATM and PARP1 Inhibition Involves DNA Damage and Abrogating the G₂ DNA Damage Checkpoint

DNA damage response maintains genome integrity by coordinating DNA damage with cell cycle arrest, DNA repair, and apoptosis. Central to the response is the activation of several phosphatidylinositol 3 kinase-related kinases including ATM and ATR (1). ATM phosphorylates histone H2AX (producing γH2AX), consequently generating docking sites for proteins involving in DNA repair, including BRCA1 and MRN complexes. ATM also links DNA damage to the cell cycle by controlling major DNA damage checkpoints (2). ATM-dependent phosphorylation of p53 and MDM2 stabilizes and activates p53, which then triggers G₁ arrest through the increase of the CDK inhibitor p21^CIP1/WAF1. ATM also phosphorylates and activates CHK1 and CHK2 (3), which in turn act on CDC25 and WEE1 to inactivate CDK1 and induce G₂ arrest (4).

Another early event in DNA damage response is the binding of poly(ADP-ribose) polymerase 1 (PARP1) to the damaged sites. This promotes PARylation (poly(ADP-ribosyl)ation, PAR) on PARP1 itself and surrounding proteins (5). PAR modifications serve as scaffolds for the recruitment of DNA repair and chromatin modifying complexes. PARP1 is involved in the repair of both single-strand breaks (SSB) and double-strand breaks (DSB). In the base excision repair and single-strand break repair (BER/SSBR) pathway, PARP1 is recruited to SSBs after modified bases are processed by DNA glycosylases and endonucleases. PARylation then attracts various BER/SSBR components (including XRCC1, DNA ligase III, PCNA, DNA polymerase β, APTX, and condensin I) to initiate repair (6). The most widely accepted role of PARP1 in DSB repair is in A-NHEJ (alternative nonhomologous end joining; also known as microhomology-mediated end joining). Higher eukaryotes mainly use classic NHEJ to repair DSBs, partly because the high affinity of Ku for DSBs normally limits the contribution of PARP1. Nevertheless, PARP1 is recruited to DSBs when NHEJ is compromised (7, 8). PARP1 then promotes the accumulation of MRN complexes and ATM at the DSBs (9). Unlike classic NHEJ, A-NHEJ requires XRCC1–DNA ligase III complexes, proteins otherwise function in the BER/SSBR pathway. Histone H1, a target of PARP1, facilitates DNA ligase III activation and the alignment of DNA ends before the DSB is ligated (10).

Several PARP inhibitors (PARPi) have been developed (11). As standalone agents, they induce synthetic lethality in tumors deficient in homologous recombination repair (HRR) such as those possessing BRCA1/BRCA2 mutations (12). The prevailing view is that PARPi block the repair of spontaneous single-strand lesions. These become DSBs when they are encountered by replication forks, which in turn become lethal in cells deficient in HRR. This model has been challenged by several observations including the lack of detectable SSBs in PARP1-inhibited cells (13). An alternative model involves the concept of “PARP trapping,” in which inactive PARP1 are trapped at damage sites by the inhibitors, which in turn require the HRR pathway to resolve (14).

A limitation of the clinical use of PARPi is that many cancer cells do not carry mutations in the HRR pathway. Accordingly, inhibitors targeting key enzymes of the HRR pathway such as ATM are believed to be able to increase the effectiveness of PARP inhibitors (15). In agreement with this, synthetic lethality has been demonstrated between PARP inhibitors and ATM deficiency in various cancer cells (16). Furthermore, a high response rate to the PARP inhibitor olaparib was found in patients with metastatic prostate cancer (17) or metastatic gastric cancer (18) possessing low expression or mutations of ATM. Cancers containing wild-type ATM can be sensitized to PARP inhibitors by ATM inhibitors. For example, the ATM inhibitor KU-55933 increases the sensitivity of Chinese hamster ovary cells (19), colon cancer cells (20), and mantle cell lymphoma cells (21) to PARP inhibitors.

Given that ATM and PARP1 are implicated to play multiple roles in DNA damage response, the precise mechanism underlying the cytotoxicity after coinhibition of ATM and PARP1 remains ambiguous. On the one hand, as ATM is involved in DNA damage repair (22), it is possible that DSBs induced by PARP1 inhibition require ATM-dependent repair. On the other hand, it is also possible that the inhibition of ATM itself promotes DNA damage that is repaired by PARP1-mediated mechanisms. Moreover, as ATM is believed to be a key component of the DNA damage checkpoints (22), inhibition of ATM is expected to promote premature exit from DNA damage–mediated arrest. Here we adopted approaches including gene deletion and small-molecule inhibitors to demonstrate that inhibition of ATM and PARP1 promotes cell death by increasing DNA damage followed by checkpoint abrogation, but does not require PARP trapping.

A construct expressing CRISPR–Cas9 targeting PARP1 (CCATCCGGAGCGAGTCCTTG) was obtained from Horizon Discovery. To generate CRISPR–Cas9 targeting ATM, oligonucleotides (5′-CACCGTTGTTTCAGGATCTCGAATC-3′ and 5′-AAACGATTCGAGATCCTGAAACAAC-3′) were ligated into BbsI-cut pX330 (Addgene). pimEJ5GFP was a gift from Jeremy Stark (Addgene plasmid #44026); pDRGFP (Addgene plasmid #26475) and pCBASceI (Addgene plasmid #26477) were gifts from Maria Jasin.

NPC cell lines C666-1 (23), CNE2 (24), HNE1 (25), and HONE1 (25) were obtained from NPC AoE Cell Line Repository (The University of Hong Kong). H1299 cells were obtained from American Type Culture Collection. HeLa cells used were as described previously (26). Cells were propagated in Dulbecco's modified Eagle's medium supplemented with 10% (v/v) calf serum (for HeLa) or fetal bovine serum (for other cells) and 50 U/mL of penicillin–streptomycin (Life Technologies). Cells were cultured in humidified incubators at 37°C with 5% CO₂. Telomerase-immortalized nasopharyngeal epithelial cell lines NP361, NP460, and NP550 (27) were propagated in keratinocyte serum-free medium supplemented (Life Technologies) with 50% v/v Epilife (Sigma-Aldrich) and 50 U/mL penicillin–streptomycin (Life Technologies). Cells were treated with the following reagents at the indicated final concentration unless specified otherwise: AZD0156 (4 μmol/L; ref. 28), AZD7762 (500 nmol/L; ref. 29), KU-60019 (5 μmol/L; ref. 30), MK-1775 (250 nmol/L; ref. 31), AZD2281 (olaparib; 1 μmol/L), VE-821 (5 μmol/L; ref. 32), ABT-888 (veliparib; 1 μmol/L; ref. 33; all from Selleck Chemicals), nocodazole (100 ng/mL; Sigma-Aldrich), and Z-VAD-FMK (20 μmol/L; Enzo Life Sciences; ref. 34).

HONE1- and HNE1-expressing histone H2B-mRFP were generated by infecting the cells with histone H2B-mRFP-expressing retroviruses in the presence of 5 μg/mL of polybrene (Sigma-Aldrich). The transduced cells were selected with 200 μg/mL of hygromycin B (Life Technologies) for ∼2 weeks before individual colonies were isolated. Cells deficient in ATM or PARP1 were generated by cotransfecting the cells with the respective CRISPR–Cas9 constructs and a plasmid carrying a puromycin-resistance gene. The transfected cells were selected with puromycin (Sigma-Aldrich, 1.5 μg/mL) for 48 hours. The cells were then grown in fresh medium without puromycin for ∼2 weeks before individual colonies were isolated. Knockout was confirmed by immunoblotting. For genomic sequencing, genomic DNA was extracted using a phenol/chloroform method (35). Genomic sequences of the ATM or PARP1 were amplified using the following PCR primers: ATM FOR (5′-ACCTCCCTTTCTTTCCAACCCTCAAACAGTCCC-3′), and ATM REV (5′-GCCTGGAGGCTTGTG- TTGAGGCTGATACATTTGG-3′), PARP1 FOR (5′-CAAGACCAG- CCTGGGCAACATGGAGAGATTCC-3′), PARP1 REV (5′-ACTCATGGACCTCTTGCCTTCTACCTACCATCCC-3′). Gene disruption of ATM and PARP1 was confirmed by sequencing of the PCR products using the primers ATM FOR (5′-CTTATCTGCTGCCGTCAACTAGAAC-3′) and PARP1 FOR (5′-AGGCATCAGCAATCTATCAGGGAAC-3′), respectively. ATM^KO HONE1 cells expressing histone H2B-GFP were generated by infecting ATM^KO cells with histone H2B-GFP–expressing retroviruses in the presence of 5 μg/mL of polybrene (Sigma-Aldrich). The transduced cells were sorted using a FACSAria IIIu flow cytometer with 488-nm laser (Becton Dickinson).

Ionizing radiation (IR) was delivered with a caesium¹³⁷ source from a Gammacell 1000 Elite Irradiator (Nordion).

Cells (200) were plated onto 60-mm plates and allowed to grow overnight. The medium was then supplemented with vehicle alone or the indicated drugs. After 2 weeks (neither the drugs or medium was replenished over the period), the colonies were fixed with methanol:acetic acid (v/v 2:1) followed by staining with 2% w/v crystal violet. The number of colonies was quantified using Quantity One (Bio-Rad). Data were generated from at least three independent experiments and are expressed as mean ± SEM.

The combination effects of two treatments were analyzed according to the median-effect method of Chou and Talalay (36). Each dose–response curve was used to calculate the isobologram and the linear correlation coefficient of the median-effect plot (r). The combination index (CI) at different effective doses (ED) was calculated using Calcusyn Version 2.1 (Biosoft). CI value <1 indicates synergistic effect (0.1–0.5 strong synergism; <0.1 very strong synergism); CI value of 1 indicates additive effect; and CI value >1 indicates antagonistic effect.

The pDRGFP and pimEJ5GFP reporter plasmids were used for measuring HRR- and NHEJ-mediated DNA damage repair, using a modified method according to (37) and (38), respectively. HONE1 cells were transfected with one of the reporter plasmids, a plasmid expressing I-SceI (pCBASceI), and a plasmid expressing histone H2B-mRFP. At 12 hours after transfection, cells were treated with different drugs and incubated for another 48 hours. The percentage of GFP-positive cells among RFP-positive cells were quantified using a FACSAria IIIu flow cytometer with 488 and 561-nm lasers (50,000 cells were counted per sample). Efficiencies of HRR- and NHEJ-mediated DNA damage repair were calculated with the formula: (number of cells that are GFP- and RFP-positive/number of RFP-positive cells) x 100. Data were generated from at least three independent experiments and are expressed as mean ± SEM.

Antibodies against β-actin (Sigma-Aldrich), ATR, ATM, CHK1, CHK2, phospho-CHK2^Thr68, 53BP1, PARP1, and WEE1 (Santa Cruz Biotechnology), CDK1 (a gift from Tim Hunt, Cancer Research UK), γH2AX (Bethyl Laboratories), PAR (EMD Millipore), phospho-CDK1^Tyr15 and cleaved PARP1(Asp214) (BD Biosciences) were obtained from the indicated suppliers. Immunoblotting was performed as previously described (39) except that a ChemiDoc Touch imaging system (Bio-Rad) was used to detect the signals.

For immunofluorescence microscopy, cells grown on poly-L-lysine-coated coverslips were gently washed with PBS before they were fixed with ice-cold methanol:acetic acid (v/v 1:1) for 10 minutes at −20°C. The cells were then washed three times with PBS (5 minutes each) before they were permeabilized and blocked with 0.4% Triton X-100 in PBS and with 2% BSA in PBS, respectively, for 30 minutes at room temperature each. The cells were then incubated with primary antibodies against γH2AX or 53BP1 (1 hour; 25°C), followed by incubation with Alexa Fluor-594 goat anti-rabbit IgG secondary antibodies (Life Technologies; 1 hour; 25°C). The cells were then stained with Hoechst 33258 in PBST (0.1% Triton X-100 in PBS) for 10 minutes at room temperature. The cells were washed three times with PBST for 5 minutes between each staining steps. Coverslips were mounted with 0.1 mol/L N-propylgallate in glycerol. Images were taken using a Nikon TE2000E-PFS microscope (Nikon) equipped with a EMCCD BOOST camera (Andor Technology Ltd.) under 400× magnification. The number of foci was quantified using ImageJ 1.50f software (NIH, Bethesda, MD).

Cells were seeded onto 96-well plates and placed onto a TE2000E-PFS microscope (Nikon) equipped with a SPOT BOOST EMCCD camera (SPOT Imaging Solutions) and an INU-NI-F1 temperature, humidity and CO₂ control chamber (Tokai Hit). Alternatively, cells were seeded onto 24-well plates and imaged using a Zeiss CellDiscoverer 7 (Zeiss). Images were captured every 5 minutes for 24 hours. Data analysis was carried out manually using ImageJ 1.52a software (NIH, Bethesda, MD).

Flow cytometry analysis after propidium iodide staining was performed as described previously (40) except that a Coulter Epics XL flow cytometer (Beckman Coulter) or a FACSAria III sorter was used (BD Biosciences) was used.

Statistical significance of data was analyzed using the Student t test (unpaired).

Nasopharyngeal carcinoma (NPC) is one of the most aggressive head and neck cancers. Cells from NPC were adopted as models because they expressed relatively high level of PARP1 in comparison with normal epithelial cells from the same tissue (41). We found that most NPC cells also expressed ATM and other components of the G₂ DNA damage checkpoint signaling pathway, including ATR, CHK1, CHK2, and WEE1 (Fig. 1A). Most of our studies were also repeated with HeLa cells (cervical carcinoma).

To eliminate the contribution of genetic differences other than ATM, we first generated isogenic HONE1 cell lines containing either wild-type ATM or CRISPR–Cas9-disrupted ATM (Fig. 1B). An inhibitor of ATM (KU-60019, ATMi herein) induced a G₂–M delay in the parental cells but not the ATM-knockout (ATM^KO) cells, verifying the specificity of ATMi (Fig. 1C). Notably, lower concentrations of an inhibitor of PARP1/2 (olaparib) were sufficient to induce G₂–M arrest in ATM^KO compared with wild-type (WT) cells (Fig. 1D). The increase in CHK2^Thr68 and CDK1^Tyr15 phosphorylation was consistent with the activation of the G₂ DNA damage checkpoint by PARPi (Fig. 1E). The sensitivity of ATM^KO cells to PARP inhibitor was not limited to olaparib, as similar results were obtained using another PARP inhibitor (ABT-888; Supplementary Fig. S1A). Similarly, we generated ATM^KO in HeLa cells (Supplementary Fig. S1B) and showed that they were more sensitive than WT cells to PARPi (Supplementary Fig. S1C). These experiments confirmed that the G₂ DNA damage checkpoint induced by PARP inhibitors can be promoted by a loss of ATM function.

Given that disruption of ATM increases the sensitivity to PARP inhibitors, we next investigated if ATM-containing cells could be sensitized using ATM inhibitors. Figure 2A shows that at concentrations of ATMi or PARPi that individually did not affect the cell cycle, a potent G₂–M arrest was induced by combining the two inhibitors. This effect was not limited to HONE1 cells, as similar results were also obtained with a different NPC cell line (HNE1; Supplementary Fig. S2A) or HeLa cells (Supplementary Fig. S2B), albeit the concentrations of the drugs needed to achieve synergism were cell line–specific. Potent activation of the G₂ DNA damage checkpoint was also induced with PARPi and another ATM inhibitor (AZD0156; Supplementary Fig. S2C), and, conversely, ATMi with another PARP inhibitor (ABT-888; see Fig. 7C).

Consistent with the cell-cycle arrest, clonogenic survival assays revealed that long-term cell survival was reduced more significantly by the combination of ATMi and PARPi than by the individual drugs alone in both HONE1 (Fig. 2B) and HeLa cells (Supplementary Fig. S2D). As previous studies have not formally demonstrated that coinhibition of ATM and PARP1 represents synergism, we incubated cells with different concentrations of ATMi and PARPi and performed isobologram analysis according to the median-effect method of Chou and Talalay (36). Figure 2C shows that the CI of ATMi and PARPi was less than 1 (indicating synergism) over a range of ED. Collectively, these results indicated that ATMi and PARPi act synergistically in activating the G₂ DNA damage checkpoint and inhibiting cell growth.

One possible explanation of the G₂–M arrest induced by ATMi and PARPi is that DNA damage caused by PARPi could accumulate in the absence of ATM-dependent repair. As ATM is an integral component of the G₂ DNA damage checkpoint, another possible mechanism is that ATMi could disrupt the checkpoint, allowing damaged cells to enter mitosis prematurely. The resulting mitotic catastrophe would also result in a similar G₂–M DNA contents. To distinguish these possibilities, HONE1 cells expressing histone H2B-mRFP were generated, and the fate of individual cells was monitored using live-cell imaging (Fig. 3A). In contrast to the inhibition of either ATM or PARP1, coinhibition of ATM and PARP1 markedly delayed mitotic entry—about half of the cells failed to enter mitosis over the 24-hour imaging period (Fig. 3B). Similar results were obtained with another NPC cell line (HNE1; Supplementary Fig. S3B) or HeLa (Supplementary Fig. S3C). These results indicated that coinhibition of ATM and PARP1 triggers G₂ delay and that inhibition of ATM does not immediately overcome the G₂ DNA damage checkpoint.

Although, in general, cells exhibited a delay in interphase in the presence of ATMi and PARPi, the cells that were able to enter mitosis underwent unusually prolonged mitosis (Supplementary Fig. S3A). An increase in mitotic and post-mitotic cell death was observed at a later time window (24–48 hours after treatment), suggesting that after transiently arrested in G₂, the ATMi- and PARPi-treated cells were able to overcome the G₂ DNA damage checkpoint (Supplementary Fig. S3D). This was more prominent in HeLa cells compared with HONE1 cells. Mitotic entry was confirmed by the decrease in CDK1^Tyr15 phosphorylation and accumulation of phosphorylated histone H3^Ser10 (Supplementary Fig. S3E). In agreement with subsequent increase in cell death, flow cytometry analysis revealed substantial sub-G₁ population at 48 hours after treatment with ATMi and PARPi, which could be reversed with the caspase inhibitor Z-VAD-FMK (Supplementary Fig. S3F). Collectively, these data suggest that coinhibition of ATM and PARP1 first triggers G₂ delay before promoting checkpoint bypass, resulting in mitotic and post-mitotic cell death.

The above results suggested that the effect of ATMi differs from that of other inhibitors of the G₂ DNA damage checkpoint. In contrast to ATMi, an inhibitor of ATR (VE-821, ATRi herein) did not delay G₂–M in the presence of PARPi (Supplementary Fig. S4A). The concentration of ATRi used was expected to be sufficient to inhibit ATR because it was able to overcome IR-mediated G₂ arrest and force cells into mitosis (Supplementary Fig. S5A). Similarly, no G₂–M delay was observed when PARPi was applied together with and an inhibitor of CHK1/CHK2 (AZD7762, CHK1i herein) (Supplementary Fig. S4B) or WEE1 (MK-1775, WEE1i herein; Supplementary Fig. S4C). Similar to ATRi, both CHK1i and WEE1i could overcome the IR-mediated G₂ arrest (Supplementary Fig. S5A).

Interestingly, unlike inhibitors of ATR, CHK1, and WEE1, ATM inhibitors were unable to bypass the IR-induced G₂ DNA damage checkpoint (Supplementary Fig. S5A and S5B). To determine whether the complete loss of ATM affects the checkpoint, ATM^KO cells were irradiated and tracked using live-cell imaging. Supplementary Fig. S5B shows that mitotic entry was abolished by IR in both WT and ATM^KO cells. The expression of markers of the G₂ DNA damage checkpoint also verified that in contrast to ATRi, ATMi was neither able to disrupt the G₂ arrest (CDK1^Tyr15 phosphorylation) nor restore mitotic entry (histone H3^Ser10 phosphorylation; Supplementary Fig. S5C).

Consistent with the idea that ATMi and PARPi initially induced DNA damage without abrogating the G₂ DNA damage checkpoint, further incubation with ATRi, CHK1i, or WEE1i triggered mitotic entry in ATMi- and PARPi-treated cells (Fig. 3C). Collectively, these data indicated that inhibitors of the ATR–CHK1/2–WEE1 pathway can bypass the G₂ delay induced by coinhibition of ATM and PARP1.

We next addressed the hypothesis that inhibition of ATM promotes DNA damage, which is then repaired by PARP1-mediated mechanisms. As indicated by the accumulation of γH2AX, ATMi alone was sufficient to induce DNA damage (Fig. 4A). Consistent with this, more PAR signals were detected in ATM^KO than in WT cells (Fig. 4B). Although the increase in PARylation after ATMi treatment was less obvious in HONE1 cells (due to the relatively high PAR background in these cells; Fig. 4A and E), it was readily observable in HeLa cells (Supplementary Fig. S6).

PARylation induced by ATMi was abolished with PARPi (Fig. 4A; Supplementary Fig. S6). This explains the higher level of DNA damage (as indicated by γH2AX) after cotreatment of ATMi and PARPi than the individual drugs alone (Fig. 4A and C). As ATM itself is one of the kinases that phosphorylate histone H2AX (42, 43), 53BP1 foci formation was also used as an additional DNA damage response marker (Fig. 4C).

To determine if coinhibition of ATM and PARP1 affects DNA repair, reporter assays for HRR or NHEJ were performed in the presence of ATMi and/or PARPi. Figure 4D shows that HRR (but not NHEJ) was abolished in the presence of ATMi.

Major events of the G₂ DNA damage checkpoint include phosphorylation of CHK2^Thr68 and CDK1^Tyr15. Phosphorylation of both proteins increased over time following exposure to ATMi and PARPi, indicating the accumulation of DNA damage after coinhibition of ATM and PARP1 (Fig. 4E). Consistent with this, although cells treated with either ATMi or PARPi displayed transient G₂–M delays before resuming normal cell-cycle progression after 24 hours, coinhibition of ATM and PARP1 promoted a more sustained G₂–M arrest followed by apoptosis (Fig. 4F). Collectively, these data suggest that although inhibition of ATM or PARP1 individually induces transient DNA damage, inhibition of both proteins prevents DNA repair.

The above results showed that low level of DNA damage could be induced by PARPi alone, which was then enhanced by ATMi. To explore if “PARP trapping” at damage sites (14) is required for synergism with ATMi, PARP1-deficient (PARP1^KO) HONE1 cells were generated using CRISPR-Cas9 (Fig. 5A; Supplementary Fig. S7), with the rationale that the absence of PARP1 would not enable PARPi to trap PARP1 on DNA. As expected, PARP1^KO cells were insensitive to PARPi in assays for G₂–M delay (Fig. 5B) and clonogenic survival (Fig. 5C). Significantly, the absence of PARP1 increased sensitivity to ATMi (Fig. 5D). Similarly, we generated PARP1^KO with CRISPR-Cas9 in HeLa cells (Supplementary Fig. S8A) and confirmed that they were insensitive to PARPi (Supplementary Fig. S8B–S8C) but displayed an increased sensitivity to ATMi (Supplementary Fig. S8D). These results were also consistent with the higher expression of γH2AX induced by ATMi in PARP1^KO than in WT cells (Fig. 6A, lanes 2 and 6). As expected, further increase in γH2AX was induced by ATMi and PARPi together in WT but not in PARP1^KO cells (lanes 4 and 8). Consistent with the γH2AX results, reduction of survival was induced by increasing concentrations of PARPi in ATMi-treated WT but not in PARP1^KO cells (Fig. 6B). Both PARPi-treated WT and untreated PARP1^KO cells showed similar sensitivity to increasing concentrations of ATMi (Fig. 6C), suggesting that either inhibition or loss of PARP1 could act synergistically with ATMi.

Although the above results indicated that loss of PARP1 catalytic activity was not necessary for synergism with ATMi, PAR signals were nevertheless still detected in PARP1^KO cells, which could be further abolished with PARPi (Fig. 4B). This suggested that other PARP activities (for example, PARP2) were present in PARP1^KO cells. Hence, it is possible that trapping of other PARP isoforms could account for the marginal drop in survival induced by PARPi even in the absence of PARP1 (Fig. 6B). To explore this further, a different PARP inhibitor called ABT-888 (veliparib), which is less effective than PARPi in PARP trapping at concentrations that inhibit the catalytic activity of PARP1 (44), was used. Similar to PARPi, ABT-888 induced a G₂–M delay in WT but not in PARP1^KO cells (Fig. 7A). ABT-888 promoted DNA damage synergistically with ATMi, as indicated by the increase of γH2AX and phosphorylation of CDK1^Tyr15 and CHK2^Thr68 (Fig. 7B). In agreement with this, applying ATMi and ABT-888 together promoted G₂–M arrest in WT but not PARP1^KO cells (Fig. 7C). Furthermore, ATM^KO cells were more sensitive to ABT-888 than WT cells (as a control, ATM^KO cells were not further sensitized by ATMi). Clonogenic survival assays also confirmed that ATMi and ABT-888 were more cytotoxic than the individual drugs alone (Fig. 7D). Finally, isobologram analysis confirmed that the two drugs acted synergistically (Fig. 7E). Taken together, these results suggest that PARP1 trapping may not be a requirement for the synergism between the coinhibition of ATM and PARP1. The relative insensitivity of PARP1^KO cells to PARPi suggested that other PARP isoforms played a relatively minor role in comparison with PARP1 in synergism with ATMi.

Although it is established that PARP inhibitors induce cytotoxicity in HRR-deficient cancers, drugs are currently unavailable for most components of the HRR pathway. Examples of attempts to enhance the sensitivity of PARP inhibitors include reducing the expression of BRCA1/BRCA2 using PI3K inhibition (45) or mild hyperthermia (46). The potential for ATM as a target is illustrated by the synthetic lethality between PARP inhibitors and ATM deficiency in cancers such as mantle cell lymphoma (47, 48) and colorectal cancer (20). Indeed, the FDA granted a “breakthrough therapy designation” for PARPi for treating prostate cancer carrying ATM mutations. Nevertheless, ATM mutations are uncommon in other types of cancers (16). Moreover, the functional consequences of many ATM mutations found in cancer are unknown. There is increasing evidence showing that the inhibition of ATM using small-molecule inhibitors may have advantages over ATM deficiency. For example, HRR is dysregulated when ATM activity is inhibited but not when ATM expression is disrupted (49, 50). Moreover, although ATM-knockout mice are viable, expression of a kinase-inactive mutant of ATM results in embryonic lethality (51).

In this study, we mainly focused on HeLa and NPC cell lines. NPC cell lines expressed both ATM and PARP1 (Fig. 1A). Moreover, no recurrent genomic aberration of ATM or other components of the DNA damage repair pathways was found in NPC by whole-genome sequencing (52, 53). We found that ATM^KO cells from NPC (Fig. 1D; Supplementary Fig. S1A) or HeLa (Supplementary Fig. S1C) were more sensitive to PARP inhibitors than the parental cells. Conversely, cells became hypersensitive to ATMi after PARP1 was deleted (Fig. 5D; Supplementary Fig. S8D). In agreement with studies in cell lines from other cancers (19–21), our studies showed that both HeLa and NPC cell lines could be killed by a combination of ATMi and PARPi (Fig. 2; Supplementary Fig. S2). Significantly, isobologram analysis indicated that the cytotoxic effects of ATMi and PARPi were synergistic rather than simply additive (Figs. 2C and 7E).

Mechanistically, loss of ATM function (either with ATM inhibitors or in a ATM^KO background) resulted in an increase in DNA damage and PARylation (Fig. 4). PARPi (1 μmol/L) was sufficient to abolish the majority of cellular PARylation (Fig. 4A). This concentration of PARPi alone did not have a deleterious effect on cell proliferation including cell-cycle distribution (Fig. 2A), mitotic entry (Fig. 3A), or long-term survival (Fig. 5C). This is consistent with the nonessential nature of PARP1 for normal cell growth (PARP1^KO cells are viable). At higher concentrations, however, PARPi reduced cell survival (Fig. 5C), possibly because the effect of PARP1 trapping became more pronounced at higher concentrations. The finding that PARP1^KO cells were more sensitive to ATMi than WT cells also suggested that PARP1 trapping is not essential for the synergism in the coinhibition of ATM and PARP1 (Fig. 5D). Further supporting evidence includes studies using ABT-888 (Fig. 7B–D) and ATM^KO cells (Supplementary Fig. S1A). A caveat is that although ABT-888 has less PARP1 trapping activity compared with many other PARP inhibitors, it is possible that some activity was present at the concentration used (54).

Given that ATM is implicated in multiple functions in checkpoints and DNA repair, the mechanisms responsible for the cytotoxicity associated with ATMi are potentially more complex. We found that ATMi (and another ATM inhibitor AZD0156) and PARPi together triggered G₂ delay, suggesting that inhibition of ATM resulted in DNA damage without immediately overcoming the G₂ DNA damage checkpoint (Fig. 2A; Supplementary Fig. S2). This was also confirmed using live-cell imaging analysis (Fig. 3; Supplementary Fi. S3) and flow cytometry (Fig. 4F). A schematic model of the relationship between ATM and PARP1 in the G₂ DNA damage checkpoint is summarized in Fig. 7F. Following cell-cycle arrest in G₂, cells treated with ATMi and PARPi were able to enter mitosis, resulting in mitotic and post-mitotic cell death. The extent of cell death is cell line–specific (Supplementary Fig. S3D). We believe as ATM is also involved in the G₂ DNA damage checkpoint, cells treated with ATMi and PARPi were subsequently able to overcome the checkpoint and enter mitosis aberrantly, resulting in loss of viability in clonogenic assays (Fig. 2B; Supplementary Fig. S2D).

Clinical trials of ATM inhibitors are generally focusing on their roles as radiosensitizers. In this connection, we showed that neither ATMi nor AZD0156 could bypass the IR-induced DNA damage checkpoint (Supplementary Fig. S5). The checkpoint was functional even after the ATM gene was disrupted (Supplementary Fig. S5B). In contrast, IR-mediated arrest was readily overcome with inhibitors of other components of the checkpoint (ATR, CHK1/CHK2, or WEE1; Supplementary Fig. S5A). Moreover, these inhibitors could bypass the delay in G₂ induced by ATMi and PARPi (Fig. 3C). These data suggest that ATM inhibitors may function differently compared with checkpoint inhibitors. This fits into the idea that although ATM and ATR have overlapping functions, ATR may be the principal mediator of the G₂ DNA damage checkpoint. It is interesting that PARPi acts synergistically with ATMi, but not with inhibitors of ATR and CHK1. It is likely because ATM is involved in both DNA damage repair and checkpoint regulation, whereas ATR and CHK1 are mainly checkpoint regulators. The DNA damage induced by PARPi alone (in the presence of ATM) is probably not sufficient to trigger the activation of the G₂ DNA damage checkpoint.

Interestingly, phosphorylation of CDK1^Tyr15 was reduced in some experiments after cells were incubated with ATMi alone (e.g., Figs. 4E and 7B), suggesting that inhibition of ATM could dysregulate the CDK1-regulatory network in an unperturbed cell cycle. This was consistent with the notably long mitosis in a subset of ATMi-treated cells (Supplementary Fig. S3A). In contrast, ATMi did not affect the CDK1^Tyr15 phosphorylation following DNA damage induced with IR (Supplementary Fig. S5C) or PARPi (Figs. 4E and 7B). Overall, our data suggest that ATM inhibitors function as PARPi- or radiosensitizers by suppressing DNA damage repair rather than triggering checkpoint abrogation.

No potential conflicts of interest were disclosed.

Conception and design: J.P.Y. Mak, R.Y.C. Poon

Development of methodology: J.P.Y. Mak, H.T. Ma

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): J.P.Y. Mak

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): J.P.Y. Mak, R.Y.C. Poon

Writing, review, and/or revision of the manuscript: J.P.Y. Mak, H.T. Ma, R.Y.C. Poon

Study supervision: R.Y.C. Poon

We thank Charles Kam, Cylene Yang, Judy Yu, and Chaoyu Zhang for technical assistance. This work was supported in part by the Research Grants Council grants 662213 and AOE-MG/M-08/06 (to R.Y.C. Poon).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.