Repair-independent functions of DNA-PKcs protect irradiated cells from mitotic slippage and accelerated senescence

When cancer cells are subjected to genotoxic stress, failure to detect or repair DNA double-strand breaks (DSBs) may result in mitotic catastrophe or lethal aneuploidy, leading to the presumed benefits of radiation and chemotherapy (Ciccia and Elledge, 2010). A common rationale for targeting the DNA damage response (DDR) in cancer treatment is to potentiate genotoxic therapy by blocking checkpoint arrest and repair (O'Connor, 2015). Of the 500-some proteins that mediate the DDR (Pearl et al., 2015), the DNA-dependent protein kinase catalytic subunit (DNA-PKcs, also known as PRKDC and XRCC7) is considered a central player in DNA damage signaling and DSB repair. An early event at many DSBs is the binding of the Ku70 and Ku80 (also known as XRCC6 and XRCC5, respectively) proteins, which can recruit DNA-PKcs and initiate repair via the conventional DNA ligase 4 (Lig4)-dependent non-homologous end joining (NHEJ) pathway (Jette and Lees-Miller, 2015). The logic that targeting NHEJ may confer or restore sensitivity to cancer therapies has led to substantial efforts to develop DNA-PKcs inhibitors as cancer drugs (Davidson et al., 2013; Salles et al., 2006). Although several clinical candidates remain under study, others have been abandoned during development and none have reached the clinic.

DNA-PKcs and ataxia-telangiectasia mutated kinase (ATM) (Yang et al., 2003) are closely related members of the phosphatidylinositol 3-kinase-related kinase (PIKK) superfamily with shared functions in the DDR, including phosphorylating Ser139 in the C-terminal tail of histone H2AX in nucleosomes adjacent to DSBs, forming γH2AX foci (Stiff et al., 2004). DNA-PKcs and ATM also phosphorylate each other (Chen et al., 2007), with DNA-PKcs serving as a negative regulator of ATM (Finzel et al., 2016; Zhou et al., 2017). This negative feedback may explain the seemingly inconsistent observations that while ATM inhibitors block γH2AX foci formation (Burma et al., 2001) and suppress checkpoint arrest and cellular senescence (Kang et al., 2017), DNA-PK inhibitors delay γH2AX foci resolution and promote checkpoint arrest and cellular senescence (Azad et al., 2011; Ciszewski et al., 2014; Sunada et al., 2016; Zhao et al., 2006).

A complementary concern is that cell lines deficient for DNA-PKcs often display reduced ATM expression (Neal and Meek, 2019). Although mechanisms have yet to be fully defined, the effect can be recapitulated by siRNA knockdown of DNA-PKcs (Peng et al., 2005) and has been linked to overexpression of the microRNA miR-100 in DNA-PKcs^−/− cells (Ng et al., 2010). Nevertheless, while the most parsimonious explanation for the DNA repair defects and radiation sensitivity in DNA-PKcs-deficient cells is their lack of DNA-PKcs activity, this fails to account for the confounding effects of ATM downregulation, which may suppress all aspects of the DNA damage response including DSB repair.

By using MCF7 breast cancer cells as a model, we observed that specifically inhibiting DNA-PKcs conferred the expected increase in sensitivity to radiation, but this was not linked to a DSB repair defect. As with chemical inhibition, partial knockdown of DNA-PKcs allowed DSBs to be repaired without delay. Despite apparently having completed repair, the γH2AX foci formed at chromosomal breaks failed to resolve, indicating a persistent DDR. When these cells progressed to mitosis, they displayed high rates of cytokinesis failure. The surviving binucleate cells adopted the characteristic senescence phenotype of flattened cell shape and expression of senescence-associated β-galactosidase (SA-βGal), ascribed to increased activity of the GLB1 lysosomal β-galactosidase. By contrast, knockdown of the Ku proteins or other core NHEJ factors was able to block DSB repair but γH2AX foci resolved on schedule, followed promptly by cell division resulting in mitotic catastrophe.

Prior studies have linked DNA-PKcs inhibition to defects in mitosis, potentially mediated by loss of interactions with polo-like kinase 1 (PLK1) and/or protein phosphatase 6 (PP6) (Douglas et al., 2014). However, we find that the persistent γH2AX, mitotic slippage and onset of accelerated senescence after irradiation of cells with decreased DNA-PKcs activity could all be suppressed by subsequent inhibition of ATM. Thus, while DNA-PKcs clearly plays a key role in regulating the DDR, our data unlink DNA-PKcs from NHEJ repair and instead define a new role in protecting cells from persistent ATM-dependent DNA damage signaling. Although blocking DNA-PKcs can inhibit proliferation after irradiation, cells remain viable after undergoing therapy-induced senescence. Given recent studies suggesting the reversibility of senescence (Chakradeo et al., 2016; Wang et al., 2013) and implicating senescent tumor cells in cancer recurrence after therapy (Demaria et al., 2016), our results suggest reevaluating the rationale for clinical development of DNA-PKcs inhibitors.

Toward reexamining DNA-PKcs functions in the DDR following exposure to ionizing radiation (IR), we inhibited DNA-PKcs in MCF7 cells and then irradiated the cells to induce double strand breaks (DSBs). DNA-PKcs was targeted by transducing MCF7 cells with lentivirus expressing short hairpin RNA (shRNA) to target DNA-PKcs transcripts (shDNA-PKcs; Fig. S1A) or treating cells expressing a scrambled shRNA control (shScr) with Nu7026 {2-(morpholin-4-yl)-benzo-[h]chomen-4-one (Nutley et al., 2005)}, a potent DNA-PKcs kinase inhibitor with >100 fold selectivity over ATM. We monitored the dynamics of ionizing radiation-induced foci (IRIF) by immunofluorescence (IF) to detect foci of γH2AX immunoreactivity that result from phosphorylation of histone H2AX near DSBs. We also tracked accumulation of 53BP1, an adapter protein that colocalizes with γH2AX at DSBs where it promotes DNA damage signaling and favors NHEJ over homologous recombination (Panier and Boulton, 2014). As expected for irradiated controls (Rogakou et al., 1998; Schultz et al., 2000), IF analysis of shScr cells at 0.5 h after irradiation revealed γH2AX and 53BP1 colocalized at multiple nuclear foci that decreased in number by 2 h and largely resolved within 24 h (Fig. 1A,C), presumably reflecting the kinetics of normal DSB detection and repair. Consistent with prior studies (Bouquet et al., 2006), the few foci remaining at 24 h were larger and brighter, indicating spreading of γH2AX across chromatin near unrepaired single or clustered DSBs. Furthermore, as expected, shDNA-PKcs and shScr cells treated with Nu7026 each formed γH2AX and 53BP1 foci by 0.5 h, but the IRIF failed to resolve by 24 h, suggesting persistent DSBs (Fig. 1B,C). In agreement with the IF results, western blotting revealed similar γH2AX immunoreactivity at 0.5 and 2 h after radiation with or without Nu7026, but persistent γH2AX was found at 24 h with Nu7026 (Fig. 1D; Fig. S1B,C). The rapid activation of DNA-PKcs in control cells, as detected by western blotting analysis of Thr2609 phosphorylation and Ser2056 phosphorylation on DNA-PKcs (p-DNA-PKcs), was suppressed by Nu7026 (Fig. 1D; Fig. S1C).

Consistent with the high γH2AX levels that remained in shDNA-PKcs or Nu7026-treated shScr cells after irradiation, clonogenic assays revealed a decreased surviving fraction in shDNA-PKcs cells or Nu7026-treated shScr cells (Fig. 1E). Live-cell time-lapse imaging and automated cell proliferation analysis of shScr cells responding to 6 Gy in the presence or absence of Nu7026 showed that while control cells recovered within 1 day, Nu7026 suppressed cell division for up to 7 days (Fig. 1F). Given that inhibiting DNA-PKcs promotes cellular senescence (Azad et al., 2011), we examined the shScr cells after 5 days, finding that DNA-PKcs inhibition increased the fraction of enlarged cells with flattened morphology expressing SA-βGal (Fig. 1G).

Reflecting the critical role attributed to DNA-PKcs in conventional NHEJ throughout the cell cycle (Chang et al., 2017) (Fig. 2A), γH2AX foci persistence after DNA-PKcs inhibition is typically ascribed to unrepaired DSBs. To evaluate DSB repair after blocking conventional NHEJ, we constructed cell lines expressing shXRCC6 and shXRCC5 to knock down the Ku70–Ku80 heterodimer and DNA-PKcs recruitment as well as shXLF (XLF is also known as NHEJ1), shXRCC4, shPAXX, and shLig4 to target NHEJ factors downstream of DNA-PKcs (Fig. S1A). Neutral comet assays confirmed that shRNA targeting conventional NHEJ factors (herein denoted shNHEJ) significantly increased levels of unrepaired DSBs at 24 h (Fig. 2B). Irrespective of the different repair defects in each shNHEJ cell line, Nu7026 failed to increase residual damage. Like in shScr cells, each of the shNHEJ cell lines formed γH2AX foci by 2 h (Fig. S2) that resolved by 24 h (Fig. 2C,E). In turn, treating shScr or shNHEJ cells with Nu7026 induced foci persistence (Fig. 2D,E).

Taken together, these data indicate that γH2AX foci formation may depend on DNA damage but not DNA-PKcs, while foci resolution depends on DNA-PKcs but not DNA repair. To directly examine DSB repair in the absence of DNA-PKcs, shScr and shDNA-PKcs cells, with or without Nu7026, were irradiated and neutral comet assays performed after 24 h, revealing similar levels of residual DBSs (Fig. 2F). A time course in shScr cells showed less damage at 2 h after treatment with Nu7026 and a similar return to baseline by 24 h (Fig. 2G). A second potent and selective DNA-PKcs inhibitor Nu7441 [8-dibenzothiophen-4-yl-2-morpholin-4-yl-chromen-4-one (Leahy et al., 2004)], recapitulated the effects of Nu7026 on MCF7 cells (Fig. S3A–C). As with MCF7, Nu7026 similarly blocked foci resolution without inhibiting DSB repair in multiple cell lines, including MDA-MB-435, MDA-MB-231, B16-F10, and CT26 cells (Fig. S3G–K).

When conventional NHEJ fails, alternative end joining (altEJ) (Iliakis, 2009; Lieber, 2010) can complete DSB repair, a potential mechanism to compensate for DNA-PKcs inhibition. Thus, MCF7 cells were developed expressing shRNAs to knockdown the altEJ factors XRCC1 or Lig3 (Iliakis, 2009) (Fig. S1A). Much like in shScr or shNHEJ cells, foci formed and resolved in the altEJ shRNA cells on schedule but persisted upon treatment with Nu7026 (Fig. S4A–D). In turn, the minor DSB repair defect in shXRCC1 and shLig3 cells was not enhanced by Nu7026 (Fig. S4E), arguing against activation of altEJ upon DNA-PKcs inhibition.

ATM has long been considered to serve a unique role in the DDR (Yang et al., 2003). Upon activation by DSBs (Hartlerode et al., 2015; You et al., 2007), ATM mediates phosphorylation of H2AX in proximal nucleosomes (Burma et al., 2001), leading to 53BP1 accumulation and foci formation. ATM also recruits NHEJ factors to stimulate repair (Blackford and Jackson, 2017). Recent reports (Finzel et al., 2016; Zhou et al., 2017) of negative regulation of ATM by DNA-PKcs raise the question of whether DNA-PKcs inhibition phenotypes could be explained mechanistically by ATM deregulation. Thus, we used MCF7^GFP-IBD cells expressing GFP fused to the 53BP1-IRIF binding domain (IBD) (Efimova et al., 2010) as a live-cell reporter to examine the relative contributions of DNA-PKcs and ATM to foci formation and persistence (Fig. 3A). Consistent with the IF analysis, Nu7026 did not appreciably alter GFP-IBD foci numbers at 2 h but blocked foci resolution at 24 h (Fig. 3B). The selective ATM inhibitor Ku55933 [2-(4-morpholinyl)-6-(1-thianthrenyl)-4H-pyran-4-one (Hickson et al., 2004)], alone or combined with Nu7026, blocked foci formation. To examine a role for ATM in foci persistence, we applied the inhibitors in sequence (Fig. 3C). MCF7^GFP-IBD cells were treated with either Nu7026 or Ku55933 starting 1 h before irradiation and until 24 h after IR. Cells were then washed and incubated with either Nu7026 or Ku55933 for an additional 24 h and foci were compared to the DMSO control (Fig. 3D). Cells treated with only Nu7026 or Ku55933 displayed foci persistence and lack of foci formation, respectively. However, persistent foci induced by Nu7026 resolved upon transfer to Ku55933. To distinguish among potential mechanisms, we examined the dynamics of GFP-IBD in single foci by fluorescence recovery after photobleaching (FRAP) analysis (Fig. S5A,B). GFP-IBD remained nearly as dynamic at 24 h as 30 min, whether cells were treated with DMSO or Nu7026. Taken together, these data suggest that foci persist as long at ATM remains active, perhaps reflecting a requirement for ATM to maintain γH2AX.

To further examine the order of function between DNA-PKcs and ATM, we established MCF7 cells expressing shRNA targeting ATM (Fig. S1A). Additionally, we stabilized activated DNA-PKcs in MCF7 cells via shRNA targeting RNF144A, an E3 ubiquitin ligase purported to promote DNA-PKcs degradation (Ho et al., 2014) (Fig. 4A; Fig. S5C,D). shRNF144A cells displayed persistent p-DNA-PKcs (at Ser2056) at 24 h after IR and induced the loss of p-ATM (at Ser1981) across the time course, but had only minor, if any, impact on the total amounts of DNA-PKcs or ATM (Fig. S5D). Neutral comet assays revealed similar DSB repair defects in shRNF-144A and shATM cells (Fig. 4B). Strikingly, inhibition of DNA-PKcs with Nu7026 suppressed the DSB repair defect not only in shRNA-144A cells but in shATM cells. The latter effect may reflect reactivation of the remaining ATM protein expressed by shATM cells.

These results raised the question of how shATM and shRNA-144A might affect foci formation and resolution. Compared to what was observed in shScr cells, irradiation of shATM generated fewer γH2AX and 53BP1 foci at 0.5 or 2 h (Fig. 4C). These residual foci were lost upon treatment with Nu7026, consistent with previous observations that DNA-PKcs can initiate γH2AX foci formation in ATM-deficient cells (Stiff et al., 2004). By 24 h, the foci in shATM cells resolved (Fig. 4C,E). shRNF-144A, with or without Nu7026, displayed similar foci kinetics to shScr cells (Fig. 4D,E). Together, these results support an on-off mechanism whereby ATM, activated upon sensing damage, feeds forward to stimulate DNA-PKcs and enhance detection of DSBs but then once repair is complete, the active DNA-PKcs feeds back to downregulate ATM.

Several prior studies have observed γH2AX and/or 53BP1 foci in the absence of DSBs. Persistent foci, whether associated with damage or not, may enhance cellular senescence (Fumagalli et al., 2014; Rodier et al., 2009). However, shNHEJ cells, although lacking persistent foci, displayed enhanced senescence compared to shScr cells (Fig. 5A,B), supporting a direct role for persistent DSBs in senescence. That treating shNHEJ cells with Nu7026 further increased the percentage of senescent cells (Fig. 5A,B) suggests independent contributions of DSBs and foci to signaling, although each may require ATM to have its effects.

Along with DSB repair defects, DNA-PKcs inhibition has been associated with prolonged checkpoint arrest as well as mitotic defects leading to failed cytokinesis (Shang et al., 2010), resulting in mononucleated or binucleate tetraploid cells that progress into senescence (De Santis Puzzonia et al., 2016). To examine mitotic progression after irradiation, MCF7 cells were treated with 0 or 6 Gy, with or without Nu7026, incubated for 24 h, immunostained for γH2AX, counterstained with DAPI and analyzed by flow cytometry. In non-irradiated cells, Nu7026 appeared to suppress S phase, increasing the proportions of 2N (presumptive G1) and 4N (presumptive G2/M) cells (Fig. S6A). After irradiation, control cells displayed increased γH2AX, accumulation in G1 and decreased S and G2/M fractions (Fig. 6A). Nu7026 led to a further increase in γH2AX and an apparent shift toward 4N DNA content, consistent with a prolonged G2/M DNA damage checkpoint arrest or mitotic slippage. Thus, we applied live-cell time-lapse imaging to track cell cycle progression in MCF7 cells expressing FUCCI fluorescent cell cycle reporters (Sakaue-Sawano et al., 2008), where G1 cells express mCherry–hCdt1 (red) and S/G2 cells express mVenus–hGeminin (green). Early in S phase, cells express both reporters and between M and G1, they express neither (Fig. 6B). Nu7026 had no appreciable effects on mitosis or cytokinesis in non-irradiated cells (Fig. S6B; Movie 1). However, Nu7026-treated irradiated cells displayed mitotic slippage. They appeared to enter mitosis (green), round up and initiate cytokinesis but, due to a cytoplasmic bridge, nascent daughter cells (red) collapsed together to generate binucleate cells that adopted a flattened morphology characteristic of senescence (Fig. 6B; Movie 1). Blocking ATM with Ku55944 partially rescued cells from cytokinesis defects and cellular senescence driven by DNA-PKcs deficiency (Fig. S6C; Movie 2).

While shDNA-PKcs and shLig4 each promoted senescence after irradiation, the cells displayed distinct patterns of DSB repair and foci persistence. To examine whether they followed distinct pathways to senescence, we used live-cell time-lapse imaging to record shScr, shDNA-PKcs and shLig4 cells after irradiation in the presence or absence of Nu7026. For each condition, we tracked the trajectories of 20 cells through one or more rounds of cell division to determine whether they yielded independent daughter cells (Fig. 6C) and assembled their trajectories over time (Fig. 6D; Fig. S6C). Here, to indicate abortive cytokinesis, nascent daughter cells were counted as two cells even if they remained connected by a cytoplasmic bridge. While shDNA-PKcs cells displayed cytokinesis failure and accumulated as binucleated cells, shLig4 cells returned to cell division within a day, presumably undergoing mitotic catastrophe. When shScr, shLig4 and shDNA-PKcs cells were treated with Nu7026 and then irradiated, all three displayed cytokinetic failure. We also examined the accumulation of flat SA-βGal-positive senescent cells at 4 days after irradiation, noting the fraction that appeared mononucleate or binucleate (Fig. 6E). Although shLig4 cells displayed an increased proportion of SA-βGal-positive cells after 6 Gy, the majority of these senescent cells were mononucleated. Together with the flow cytometry analysis, this suggests that senescence in shLig4 cells may reflect an irreversible cell cycle arrest prior to onset of S phase rather than after mitotic slippage.

To confirm a mechanism linking mitotic slippage to senescence via formation of binucleated cells, even without prior DNA damage (Panopoulos et al., 2014), we treated MCF7 FUCCI cells with the small molecule kinase inhibitors AZD1152-HQPA (barasertib) and GSK461364 to target Aurora B or PLK1, respectively, and thereby disrupt mitosis and/or cytokinesis (Fededa and Gerlich, 2012). Time-lapse imaging confirmed each inhibitor conferred a cytokinesis defect in non-irradiated cells much like that observed with DNA-PKcs inhibition in irradiated cells, leading to binucleate cells that flattened out to adopt a senescent cell phenotype (Fig. S6E, Movie 3). In turn, staining analysis after 4 days treatment with the kinase inhibitors revealed increased SA-βGal, with a high proportion of binucleate senescent cells (Fig. S6E–G). A caveat is that most binucleate cells formed after Aurora B or PLK1 inhibition died rather than entering senescence (Fig. S6H). The ATM/p53 target p21^CIP1 (also known as CDKN1A) promotes cell cycle arrest, blocks apoptosis and can induce senescence on its own (Kagawa et al., 1999). When combined with the Aurora B or PLK1 inhibitors, p21^CIP1 overexpression rescued the binucleate cells from death, yielding nearly homogeneous senescence (Fig. S6E,H; Movie 4).

DNA-PKcs is a PIKK activated by binding of its Ku70 and Ku80 partners to DNA ends (Suwa et al., 1994). The search for endogenous substrates initially led to transcription factors such as p53, Sp1, c-Myc and c-Jun but eventually yielded a wide range of other proteins (Goodwin and Knudsen, 2014). Mutant cells display DNA repair defects, a prolonged proliferative arrest and decreased survival after exposure to radiation or other genotoxic agents (Lees-Miller et al., 1995). While attention initially focused on p53 activation, DNA-PKcs was also found to phosphorylate the Ku proteins along with other NHEJ factors, including XRCC4, Lig4 and XLF, implicating DNA-PKcs in NHEJ, the primary mode of DSB repair throughout the cell cycle (Chang et al., 2017). Remarkably, our current understanding that DNA-PKcs serves an essential role in NHEJ has remained largely untested, built on studies in a small number of DNA-PKcs-deficient cell lines [e.g. DNA-PKcs^−/− MO59J human glioma (Lees-Miller et al., 1995)] complemented with constructs derived from a reassembled DNA-PKcs cDNA (Kurimasa et al., 1999). A common feature of DNA-PKcs^−/− models is the coinciding downregulation of ATM (Neal and Meek, 2019). Strikingly, Peng et al. (2005) found that even a transient knockdown of DNA-PKcs with siRNA recapitulated this effect, resulting in loss of both ATM transcript and protein within days. Then, as the siRNA effects were lost, transcript and protein levels were restored not only for DNA-PKcs but also ATM. Perhaps, lacking negative regulation by DNA-PKcs (Finzel et al., 2016; Zhou et al., 2017), constitutively active ATM may induce autoregulatory factors [e.g. microRNA miR-100 (Ng et al., 2010; Zhang et al., 2011)] that mediate its downregulation. Thus, a caveat in assigning DNA-PKcs an essential role in DSB repair is that this ignores our long-standing knowledge of co-regulation with ATM, which itself plays multiple roles in the DDR. In short, while a direct role for DNA-PKcs in NHEJ is well accepted, it remains to be rigorously established. Nonetheless, only limited evidence has been published to date (e.g. Sunada et al., 2016) that appears to challenge the prevailing model.

Here, we showed that MCF7 human breast cancer cells either treated with the specific DNA-PKcs inhibitor Nu7026 and/or expressing an shRNA targeting DNA-PKcs displayed unperturbed DSB repair in neutral comet assays. By contrast, shRNA knockdown of Ku70, Ku80, Lig4 and other NHEJ factors (shNHEJ) conferred a dramatic repair defect after irradiation, whether DNA-PKcs was active or not. Arguing against alternative end joining (alt-EJ) compensating for the NHEJ defect (Iliakis, 2009), shRNA knockdowns of Lig3 and XRCC1 failed to block DNA-PKcs-independent DSB repair.

While we obtained no evidence for a role in NHEJ, DNA-PKcs activity was critical in mediating recovery from the DDR. Cells lacking DNA-PKcs displayed persistent γH2AX foci and cell proliferation arrest even after apparently completing DSB repair. In turn, despite a significant DSB repair defect in shNHEJ cells, γH2AX foci resolved on schedule and cells returned to proliferation. Moreover, inhibiting DNA-PKcs in shNHEJ cells led to γH2AX foci persistence and cell cycle arrest, confirming uncoupling between DNA damage and the DDR. Inhibiting ATM both blocked DSB repair and overcame the phenotypes of DNA-PKcs inhibition. A mechanism consistent with our data is for ATM to serve as the key driver that initiates and maintains the DDR and activates NHEJ repair while DNA-PKcs functions downstream, initially working in concert with ATM but then opposing ATM signaling to terminate the DDR. Remarkably, even when DSBs persist, the negative feedback from DNA-PKcs can still terminate the DDR. Indeed, inhibiting DNA-PKcs partly suppressed the DSB repair defect in shATM cells and several shNHEJ lines. Here, DNA-PKcs appears to antagonize DSB repair and promote aneuploidy rather than preserve genomic integrity.

Toward resolving the apparent paradox, our results are most consistent with recent studies establishing DNA-PKcs as a negative regulator of ATM after DNA damage (Finzel et al., 2016; Zhou et al., 2017). A conservative model is that the Ku70–Ku80 heterodimer and ATM bind to DSBs, initiating the DDR. Once recruited by Ku, DNA-PKcs initially functions in concert with ATM to activate the DDR. Together, they phosphorylate substrates in chromatin surrounding the DSB. Phosphorylation of H2AX at Ser139 in adjacent nucleosomes leads to γH2AX foci that promote DNA repair and induce cell cycle arrest (Fig. 6F). While ATM makes the major contribution to DDR activation, DNA-PKcs activity is critical for normal recovery, mediated by downregulating ATM as DNA repair progresses. A model compatible with much of the literature is that Ku proteins bound at DSBs indeed recruit and activate DNA-PKcs kinase activity but also tether it in place. Our study suggests that subsequent release of active DNA-PKcs might serve as a mechanism to signal successful NHEJ repair, permitting downregulation of ATM and local foci resolution. Thus, in the absence of DNA-PKcs, while NHEJ can proceed, ATM remains active, resulting in γH2AX foci persistence, mitotic slippage and accelerated senescence. One inference is that the repair defects classically observed in DNA-PKcs-deficient cell lines could reflect compensatory changes to accommodate ATM deregulation rather than loss of DNA-PKcs activity per se.

Nevertheless, our results do not rule out a specific role for DNA-PKcs in promoting DSB repair. One report found DNA-PKcs dispensable for rapid rejoining of compatible ends but suggested a role in Lig4-mediated repair of complex damage (Reynolds et al., 2012). Complex DSBs may require end processing such as that mediated by the DNA-PKcs substrate Artemis (also known as DCLRE1C), an endonuclease that accelerates NHEJ by clipping 5′ overhangs (Ma et al., 2002). Nonetheless, the ⁶⁰Co γ-radiation (1.25 MeV) used here to induce DSBs produces more clustered damage than simple ligatable breaks (Gulston et al., 2002), suggesting that DNA-PKcs-independent repair can also rejoin complex DSBs (Magnander and Elmroth, 2012). Furthermore, despite lacking DNA-PKcs, yeast can perform rapid and efficient NHEJ of a wide range of lesions (Aylon and Kupiec, 2004), although other PIKKs may provide the critical functions. Even then, insofar as diverse prokaryotes (Weller et al., 2004) can perform end-joining repair and express homologs of Ku proteins and Lig4 (Weller et al., 2002) but appear to lack PIKKs altogether, DNA-PKcs was apparently not required for NHEJ for the first few billion years of evolution (Gu and Lieber, 2008).

Along with persistent γH2AX, we observed a profound cytokinesis defect, consistent with known roles for DNA-PKcs in cell division after DNA damage. Cells lacking DNA-PKcs performed mitosis and initiated cell division but displayed persistent cytoplasmic bridges that caused nascent daughter cells to collapse together, resulting in binucleate tetraploid cells that progressed to senescence. Consistent with prior work linking mitotic slippage to senescence (Panopoulos et al., 2014), we confirmed that inhibition of Aurora B or PLK1 was sufficient to produce cytokinetic failure and enhanced senescence in MCF7 cells, even without prior irradiation. Taken together, the senescent phenotype after irradiation of cells lacking DNA-PKcs activity may be linked primarily to mitotic defects rather than any impacts on DNA repair.

The underlying rationale for development of DNA-PKcs inhibitors has long been to target NHEJ repair and thereby augment genotoxic cancer therapy. However, preclinical studies have begun to implicate alternative mechanisms of action. While Nu7441 induces chemo- and radio-sensitization in xenograft models (Ciszewski et al., 2014; Zhao et al., 2006), these effects may not depend on DSB repair (Sunada et al., 2016). The activity of the orally available DNA-PKcs inhibitor M3814 (nedisertib), an effective radiosensitizer in xenograft tumors (Damstrup et al., 2016) currently under investigation in clinical trials, has recently been ascribed to deregulation of ATM and p53 (Guo et al., 2018). Our data add to the emerging picture that the major role of DNA-PKcs in radiation tolerance may be in recovery after repair.

In conclusion, our study assigns DNA-PKcs a new primary role in the radiation response as the critical factor protecting cells against the deleterious effects of constitutive ATM activity. In the absence of DNA-PKcs, despite substantially completing DSB repair, unopposed ATM signaling leads to mitotic slippage and senescent arrest. Given concerns that therapy-induced senescence may be reversible (Chakradeo et al., 2016; Wang et al., 2013) and recent studies implicating inflammatory mediators released by senescent cells not only in tissue aging and carcinogenesis (Hoare and Narita, 2018), but also in driving adverse effects and resistance to cancer therapies (Demaria et al., 2016), these findings suggest a reevaluation of DNA-PKcs as a cancer target.

The human mammary carcinoma cell line MCF7^Tet-On and Lenti-XTM293T cell line were obtained from Takara. Cell lines MDA-MB-435 (human melanoma), MDA-MB-231 (human breast cancer), B16-F10 (mouse melanoma), and CT26 (mouse colon cancer) were obtained from the ATCC. The MCF7^GFP-IBD cell line with GFP fused to the 53BP1 IRIF binding domain (IBD) under tetracycline-inducible control has been reported previously (Efimova et al., 2010). A previously described MCF7 cell line with FUCCI cell cycle reporter constructs (Sakaue-Sawano et al., 2008) was reconstructed here by transduction with lentivirus expressing mVenus-hGeminin (1/110)/pCSII-EF-MCS and mCherry-hCdt1 (30/120)/pCSII-EF-MCS (a kind gift of Atsushi Miyawaki, RIKEN Center for Brain Science, Japan). Cells with positive expression were selected by fluorescence-activated cell sorting (FACS).

Human cells were maintained in DMEM containing 4.5 g/l glucose (Thermo Fisher Scientific) supplemented with 10% Tet-approved FBS (Atlanta Biologicals) and 1% penicillin/streptomycin (Thermo Fisher Scientific). Mouse cells were maintained in RPMI 1640 medium (Thermo Fisher Scientific) supplemented with 10% Tet-approved FBS (Atlanta Biologicals) and 1% penicillin-streptomycin (Thermo Fisher Scientific). The cells were tested for mycoplasma contamination and authenticated by short tandem repeat profile (IDEXX BioResearch) prior to performing experiments. All experiments were performed within 3 to 10 passages after thawing cells. Cells were treated with small molecules or DMSO vehicle 1 h before irradiation. All chemical probes used in this study are listed in Table S1.

Pairs of Sigma MISSION shRNAs targeting expression of PRKDC (DNA-PKcs), ATM, RNF144A, XRCC6 (Ku70), XRCC5 (Ku80), Lig4, 53BP1, PAXX, XRCC4, XLF, Lig3, XRCC1 and a non-targeting scrambled (Scr) negative control were obtained as pLKO.1-puro vectors and used according to manufacturer's instructions. Lentivirus-containing supernatant was produced by transfection of the 293T Lenti-X cell line with corresponding plasmids and packaged plasmid mix, and applied to MCF7^Tet-On (Takara) cells. Following selection in the presence of puromycin, pairs of stable MCF7^Tet-On cell lines with decreased levels of PRKDC, ATM, RNF144A, XRCC6, XRCC5, Lig4, 53BP1, PAXX, XRCC4, XLF, Lig3 and XRCC1 protein expression were established. Cells from the third passage post-selection were frozen in liquid nitrogen as a stock and most experiments were performed within 4–10 passages. At least two shRNA constructs targeting different sequences of the corresponding mRNA were evaluated for each gene. Cells were collected from passage 2 or passage 3 after selection with puromycin and whole-cell lysates were extracted. Knockdown of targeted genes was validated by western blotting (Fig. S1A) and the shRNA with greatest apparent knockdown based on protein expression was used for experiments. Phenotypes were validated by shRNAs conferring consistent effects on formation and resolution of foci compared to scrambled control. shRNAs and antibodies used in this study are described in Tables S2 and S3.

Cells were plated at 100 cells per well in six-well plates in triplicate and irradiated with doses of 1, 2, 3 and 4 Gy using a GammaCell ⁶⁰Co source (MDS Nordion) with the dose rate ranging from 10.5 to 9.4 cGy/s depending on the date of the experiment. Cells remained in culture for 3 weeks before staining with Crystal Violet (0.5%), and colonies of at least 50 cells were counted.

For neutral comet assays, cells were seeded at 10⁵ cells per well in six-well plates prior to irradiation. After 24 h, cells were mixed with Comet LM agarose and single-cell electrophoresis was performed on CometSlides (Trevigen). Slides were fixed, dried, stained with SYBR green (Thermo Fisher Scientific) and imaged on a Zeiss Axiovert 40CFL with a 10× Plan-NeoFluar objective and AxioCam digital camera controlled by AxioVision 4.8 software. Two or more replicates were performed. Images were analyzed using an ImageJ comet assay macro (https://www.med.unc.edu/microscopy/resources/imagej-plugins-and-macros/comet-assay/) and plugin OpenComet (Gyori et al., 2014; http://www.cometbio.org/index.html).

To image IRIF, MCF7^GFP-IBD cells were seeded on cover glass at 2.5×10⁴ per well in 24-well plates. Expression of the GFP-IBD reporter was induced with 1 μg/ml doxycycline treatment for 48 h. After treatment with DNA-PKcs and/or ATM inhibitors, cells were fixed with 4% PFA at the indicated time point, stained with 5 μg/ml Hoechst 33342 (Sigma-Aldrich), and mounted using ProLong Gold (Invitrogen). For IF staining, cells were fixed with 4% PFA and permeabilized with 10% Triton-X 100 for 10 min. After blocking with 5% BSA, primary antibodies for γH2AX (Millipore, 05-636, 1:1000) or 53BP1 (Novus, NB100-304, 1:1000) were then incubated on cell slides overnight at 4°C. Following PBS washes, fluorescent secondary antibodies (Jackson ImmunoResearch) were applied for 1 h at room temperature. Cell slides were mounted with ProLong Gold after PBS washes. Foci images were captured on an Zeiss Axiovert 40CFL with a 40× Plan-NeoFluar objective and pseudocolored using ImageJ. Two or more replicates were performed.

Cells were seeded at 3×10⁴ per well in six-well plates and treated with inhibitors for 1 h prior to irradiation. Cells were fixed after 4 or 5 days and assayed for SA-βGal activity as described previously (Efimova et al., 2010). Images were captured on a Zeiss Axiovert 200M microscope with a 20× Plan-NeoFluar objective and Axiocam digital camera controlled by OpenLab software. SA-βGal-positive and -negative cells were counted in multiple fields, yielding an average percentage indicated on each SA-βGal image as mean±s.e.m. Two or more replicates were performed.

5×10⁵ cells were plated in P-100 Petri dishes for at least 24 h and harvested by scraping in 1 ml lysis buffer. Whole-cell lysates were prepared using M-PER lysis reagent (Thermo Fisher Scientific) in the presence of protease and phosphatase inhibitors (Thermo Fisher Scientific). 20 μg of protein was loaded per well, separated on a NuPage 3-8% Tris-Acetate precast gels (Invitrogen), and transferred onto a PVDF membrane (Millipore). After dividing blots into strips by apparent molecular mass, immunoblotting was performed using primary antibodies as described in Table S3 and detected with peroxidase-conjugated secondary antibodies (Thermo Fisher Scientific, NA934vs or NA931). This was followed by detection with ECL peroxidase substrate (Thermo Fisher Scientific).

For sample preparation, MCF7 cells were collected 24 h after irradiation, then fixed with 2% PFA for 10 min on ice and permeabilized with 90% ice-cold methanol. Following blocking with 1% BSA, cells were incubated with an Alexa Flour 647-conjugated anti-γH2AX antibody (anti-H2AX phosphoserine 139, Cell Signaling Technology, CST9720, 1:50) for 2 h and then washed using 1% BSA. 3 µg/ml DAPI was added to samples for DNA staining 15 min before flow cytometry. Flow cytometric data were acquired using a BD Fortessa X20 using FACSDiva software. 50,000 viable cells were acquired per sample. Flow cytometric data were analyzed using FlowJo software.

ShScr, shDNA-PKcs or MCF7-FUCCI cells were seeded in a six-well plate with 30,000 cells per well. After 24 h in culture, cells were treated with DMSO or Nu7026 and/or irradiated with 6 Gy. The plates were then analyzed by time-lapse imaging in an IncuCyteS3 live-cell imaging system. Phase contrast, green and red channel images were acquired at 20× magnification with scanning every 2 h for 7 days. More than 25 non-overlapping fields were captured for each well. Quantitative analysis of cell confluency was performed using IncuCyteS3 2018 software.

Photobleaching was carried out on MCF7^GFP-IBD cells after doxycycline induction. A 40× oil immersion objective of Leica SP8 laser scanning confocal microscope was used for all FRAP experiments. Photobleaching was achieved with 405 nm laser excitation for 10 s at full intensity. Data acquisition was performed with an excitation at 488 nm with 40% intensity for image scanning. At least ten independent experiments were performed for each condition. Cells were imaged every second. Regions of interest (ROIs) of the bleached area were acquired with ImageJ and normalized with easyFRAP (Rapsomaniki et al., 2012). Fluorescence recovery plots were fitted to a one-phase association exponential curve.

Statistical significance for anti-γH2AX, anti-53BP1 and GFP-IBD foci counting and comet assays was determined using the non-paired Student's t-test. Calculations were performed using Prism software (GraphPad) or Excel. P≤0.05 were considered statistically significant.

We thank our colleagues for helpful support including Julian Lutze for comments on the manuscript and Liuyun Wu and Jordan Cooper for data analysis.