Biphasic DNA damage and non-canonical replication stress response govern radiation-induced senescence in glioblastoma Free

Glioblastoma multiforme (GBM), a World Health Organization-classified grade 4 brain tumor, is the most common and aggressive malignant primary brain tumor in humans. The current mode of therapy includes surgery followed by radiation with concomitant temozolomide treatment. Despite multimodal therapy, the median survival of patients with GBM remains at approximately 15 months, with a 5-year survival rate of less than 5% (Stupp et al., 2005, 2009; Wen and Kesari, 2008). Furthermore, profound heterogeneity within the tumor, a hypoxic microenvironment, therapy resistance and highly recurrent properties contribute to poor patient outcomes (Friedmann-Morvinski, 2014; Huang et al., 2016; Sturm et al., 2014). Therefore, understanding the cellular and molecular mechanisms of therapy resistance is imperative for developing better therapeutic approaches.

Traditionally, senescence is defined as a cellular state of an irreversible growth-arrested condition, orchestrated through highly regulated multiple steps that include, but are not limited to, the development of senescence-associated secretory phenotypes (SASPs), change in morphology and multinucleation. Conventionally, replicative senescence (telomere attrition) and oncogene-induced senescence are the two major types of cellular senescence (Gorgoulis et al., 2019; Nardella et al., 2011). Accumulating evidence in recent years suggests another type of cellular senescence termed stress-induced premature senescence, wherein endogenous stressors such as reactive oxygen species (ROS), endoplasmic reticulum stress, exogenous stressors such as oxidative agents, and ionizing radiation act as stimuli (Hernandez-Segura et al., 2018; Sikora et al., 2021). Although cancer cells are highly proliferative, they can also undergo a senescent state after therapeutic treatment with DNA-damaging agents, and this is named therapy-induced senescence (TIS) (Fitsiou et al., 2022; Mikuła-Pietrasik et al., 2020; Prasanna et al., 2021). In recent years, TIS has gained a lot of interest because of its ability to halt uncontrolled growth of cancer cells. However, there are contrasting lines of evidence suggesting its role as a pro-tumorigenic stimulus (Wang et al., 2020). Moreover, TIS has also been shown to be reversible. Different studies including ours have shown induction of TIS and escape from TIS as one of the primary causes of therapy resistance and recurrence in various cancers including GBM (Basbous et al., 2022; Kaur et al., 2015; Saleh et al., 2019b, 2020; Wang et al., 2013). In our previous studies, we generated an in vitro radiation-induced therapy-resistant GBM model that identified clinically relevant residual resistant (RR) disease cells. These RR cells showed transient TIS with a multinucleated giant cell (MNGC) phenotype and eventually led to a relapse population (Kaur et al., 2015, 2019). Recently, this reversible TIS phenomenon has been attributed to ‘tumor dormancy’ during the therapy response in cancer (Saleh and Gewirtz, 2022; Saleh et al., 2019a). Although TIS shares some common features with the other three types of senescence, there is a lack of in-depth understanding and characterization of the cellular and molecular mechanism of TIS induction, which could be exploited for novel therapeutic interventions in GBM.

Replication stress (RS) is defined as any genetic assault that occurs during genome duplication leading to stalling of the replication fork and reduced DNA synthesis in a cell. Such DNA perturbations happen spontaneously or due to DNA-damaging agents encountered endogenously or exogenously (Saxena and Zou, 2022). In response to RS, cells initiate RS management through the protection of single-stranded DNA (ssDNA) by RPA32 (encoded by RPA2) coating and phosphorylation at threonine 21 and serines 4, 8 and 33, which in turn triggers ATR phosphorylation followed by elevated RPA32 phosphorylation downstream of ATR activation. This leads to a chain of events wherein ATR can phosphorylate a plethora of substrate proteins that play a concerted role to halt cell cycle progression for repairing stress-induced DNA damage. Unrepaired DNA damage or prolonged RS leads to double-strand breaks (DSBs) followed by either cell death or senescence (López-Contreras and Fernandez-Capetillo, 2010; Patel and Weiss, 2018). RS is shown to be a determinant factor in replicative senescence, oncogene-induced senescence and stress-induced premature senescence, which are generated during telomere attrition, oncogene-induced hyperproliferation and ROS-mediated DNA breaks in S phase, respectively (Di Micco et al., 2006; Kotsantis et al., 2018). RS-induced endogenous DNA damage-mediated senescence was reported in rat myoblasts upon H₂O₂ treatment (Venkatachalam et al., 2017). Radiation can induce senescence in cancer cells such as breast cancer, but the involvement of RS remains unclear (Lukášová et al., 2017). GBM cells harbor heightened RS and DNA damage responses with better management skills to tolerate such high RS through a proficient ATR–CHK1 pathway (Bartkova et al., 2010). Another report showed that inherent RS in glioma stem-like cells governed constitutive activation of DNA damage responses and radioresistance without any senescence induction (Carruthers et al., 2018). However, the mechanisms of RS in manifestation of TIS leading to radioresistance in GBM remains elusive.

In this study, using GBM cell lines, we demonstrate that radiation-induced ‘biphasic’ γH2AX signaling, corresponding to DNA break formation, is a hallmark of TIS. We provide evidence that the second phase of DNA damage originates in RR GBM cells in the S phase of the cell cycle, suggesting the prevalence of RS, which in turn leads to TIS. Using several cell-based assays and biochemical studies, we further report a non-canonical RS response with persistent levels of phosphorylated (p)ATR (pATR) in these cells. Finally, genetic ablation and pharmacological inhibition of ATR eradicate RR cells by inhibiting TIS and ultimately inducing cell death. Our findings provide new insights into the cellular and molecular mechanism of TIS and show that targeting such mechanisms would result in better clinical outcomes in GBM.

To understand GBM therapy resistance and capture the RR disease cells, we previously established an in vitro cellular model using cell lines and patient-derived primary cultures that recapitulates GBM resistance similar to in clinical settings (Fig. 1A). This model captures RR cells (<10% of the population that survives exposure to a lethal dose of radiation to parent cells) and relapse cells in the same experimental setup within a span of approximately 25–30 days (Kaur et al., 2015). To understand the cellular and molecular insights governing the properties of RR cells, here, we characterized distinctive features of these RR cells using four GBM cell lines, namely, U87MG, SF268, LN229 and LN18. To generate RR cells, the GBM cell lines U87MG, SF268, LN18 and LN229 were subjected to their respective lethal doses of radiation: 8 Gy, 6.5 Gy, 16 Gy and 10 Gy, respectively, as identified before (Ghorai et al., 2020; Kaur et al., 2015; Nair et al., 2021). First, we characterized the proliferative nature of RR cells, for which a bromodeoxyuridine (BrdU) pulse was delivered for 6 h in non-irradiated control cells, RR cells and relapse cells. BrdU-positive cells as well as BrdU intensities were measured. We observed that compared to control cells, the RR cell population was non-proliferative (day 10 showed maximum non-proliferation for U87MG and LN229, days 10–14 for LN18, and days 7–12 for SF268). However, the relapse population (generated spontaneously from RR cells) showed proliferation as parental cells (Fig. S1A–L). This non-proliferative property was accompanied by β-galactosidase (β-gal) positivity and MNGC formation as shown in Fig. S1M–O, with G2 phase arrest throughout the treatment regime (Fig. S1P).

As DNA damage is highly associated with the senescence phenotype (Mikuła-Pietrasik et al., 2020), we sought to detect γH2AX levels in RR cells using western blotting. As expected, we found increased γH2AX levels in the RR phase of the cell lines, which were reduced in the relapse population (Fig. S1Q). This type of senescence is known as TIS. Although TIS is a favorable therapy outcome, a few reports, including ours, have shown that TIS could be reversed (Kaur et al., 2015; Saleh et al., 2019b; Salunkhe et al., 2021). Therefore, it is important to understand TIS induction and reversal. As DNA damage can lead to a senescent phenotype in normal and cancer cells, we first evaluated the DNA damage response of the parent and RR cells. We meticulously studied the kinetics of γH2AX after ionizing radiation (IR) treatment in U87MG, LN229 and SF268 cells using an immunostaining technique. To our surprise, a biphasic mode of γH2AX response was observed in all the tested cell lines, as evident from Fig. 1B–D and Fig. S2A–F. A sharp γH2AX induction (first phase) was observed within 6 h post IR, which was quickly resolved within 24–72 h post IR in the U87MG, LN229 and SF268 cell lines, whereas the second phase of γH2AX induction seen around day 5 was more persistent as it stayed until ∼10 days depending upon the cell type. Biphasic induction of γH2AX was also seen by western blotting in LN18 (Fig. S2G). Similarly, we checked chromatin recruitment of p53-binding protein 1 (53BP1, encoded by TP53BP1), a well-known DSB indicator, in cells exposed to IR. Immunostaining of 53BP1 showed a biphasic mode in both U87MG and SF268 cells (Fig. 1E,F; Fig. S2H,I). Although the second wave was less intense in U87MG cells, it was significant. To confirm the physical presence of DNA breaks, we performed a neutral comet assay (Fig. 1G,H). In concordance with γH2AX and 53BP1 staining, the comet data confirmed the biphasic mode of DNA damage.

To understand the molecular basis of the reappearance of γH2AX after almost complete resolution of the first phase of DSBs, we first investigated whether the appearance of γH2AX in the second phase was dependent on the cell cycle as cell cycle phases have differential radiosensitivity (Pawlik and Keyomarsi, 2004). For this, we labeled the cells with 5-ethynyl-2′-deoxyuridine (EdU) to mark S-phase cells and performed colocalization studies with γH2AX at different time intervals (Fig. 2A). We found that the second phase of γH2AX signaling was significantly high in S-phase cells compared to that in non-S-phase cells post IR in both U87MG and LN229 cells, indicating that these cells contributed significantly to maintain the biphasic mode of DNA damage in RR populations (Fig. 2B–E; Fig. S3A,B). However, γH2AX levels were also increased in non-S-phase cells, which we speculate is due to the bystander effect of S-phase cells and accumulation of DSBs due to error-prone repair after the first phase of DNA damage. We also performed colocalization analysis between EdU and γH2AX in the RR phase. We observed a significant increase in the colocalization of γH2AX with EdU in the RR phase, indicating that the second phase of γH2AX is due to replication collapse (Fig. S3C,D). Furthermore, we checked 53BP1 levels in S-phase and non-S-phase cells, which showed that the second phase of 53BP1 recruitment was significantly higher in S-phase cells (Fig. S3E). This result provoked us to hypothesize that RS is instrumental in the induction of the second phase of DNA damage in residual cells. To test this hypothesis, we performed the DNA fiber assay in U87MG and LN229 cells in the presence of 5-iodo-2′-deoxyuridine (IdU) and 5-chloro-2′-deoxyuridine (CldU) following the experimental paradigm mentioned in Fig. 3A. We analyzed different replication parameters to detect RS, which included the percentage of ongoing forks, stalled forks, terminated forks, newly fired origins, fiber length and fork speed. Of these parameters, we observed a significant increase in stalled and terminated forks with a subsequent decrease in ongoing forks at 24 h post IR, which were maintained or enhanced at all the time points in LN229 cells (Fig. 3B), whereas U87MG cells showed a similar trend until day 5 (Fig. 3D). These differential patterns of RS endpoints seen in both the GBM cell lines might be because they are non-isogenic to each other. Moreover, a significant decrease in the fork speed and fiber length was observed after 24 h post IR, which continued to remain low in the RR phase (days 7 and 10) in both the cell lines, indicating disruptions in replication fork progression (Fig. 3C,E; Fig. S4A,B). Next, to rule out the direct contribution of broken tracts due to IR to the fiber analysis, we labeled total DNA by incubating U87MG cells with CldU for up to 24 h (unirradiated control) and 72 h (RR cells) to complete at least one cell cycle, which was followed by EdU and IdU pulse labeling as shown in the experimental scheme (Fig. S4C). Upon analysis, we found similar results with increased stalled and terminated forks in the RR phase (Fig. S4D), as shown earlier (Fig. 3B–D). Taken together, these data implied that the γH2AX response, majorly contributed by S phase cells, was associated with RS, which was detected after 24 h post IR and persisted until RR phase, thereby leading to the second phase of DNA damage.

Next, to get molecular insights into how RS is maintained in the RR cells, we speculated that the accumulation of ssDNA due to inefficient RS management, disrupted replication fork reversal events or deregulated fork protection might lead to continuous RS in GBM RR cells. Accordingly, we performed native IdU assay in U87MG cells to detect nascent ssDNA by IdU incorporation in S-phase cells. We marked S-phase cells by pulse labeling with EdU (15 min), followed by IdU incorporation (30 min), in U87MG cells post IR at different time points. We detected ssDNA accumulation at 6 h post IR that gradually intensified at 24 h and was retained as cells entered RR phase (day 7) (Fig. 3F,G), which could be due to destabilized fork reversal events or faulty replication fork protection in these cells. In concordance, we performed an S1 nuclease DNA fiber assay to detect ssDNA gaps formed due to extensive ssDNA accumulation. We found that CldU fiber length shortened significantly after treatment with S1 nuclease at different time points post IR (Fig. S4E,F). Taken together, these data show that there was a significantly high RS in the S phase of RR cells, which might be responsible for the second phase of the biphasic DNA damage in GBM cells exposed to radiation.

Normal proliferating cells that experience RS have stress management machinery to cope and resolve it (Zeman and Cimprich, 2014). Cancer cells especially are well equipped to tackle RS, which is generated due to rapid proliferation and a hypoxic microenvironment (Bader et al., 2021; Macheret and Halazonetis, 2015; Ramachandran et al., 2021). Having found that RS is associated with the second phase of γH2AX response that arises post IR in S-phase cells, we sought to gain further insights into how therapy-induced RS is managed in these cells. We monitored the recruitment of the apical kinase ATR, which is the master regulator of RS response (López-Contreras and Fernandez-Capetillo, 2010), using immunofluorescence imaging. Although we observed an early induction of pATR (T1989) at 6 h post IR, it was not sustained, and no staining was seen at 24 h post IR (Fig. 4A,B). Interestingly, pATR reappeared at day 5 post IR and was sustained up to day 10 (although the intensity was reduced at day 10, discrete foci were noticed in irradiated cells compared to unirradiated cells), in agreement with the presence of RS as well as the second phase of γH2AX observed earlier. We also validated this by western blotting, which suggested an initial increase of pATR until 24 h post IR, with no further increase during the RR phase, but it remained in the phosphorylated form compared to that in unirradiated control U87MG cells (Fig. S5A,B). Next, to confirm whether the second phase of pATR is localized in S-phase cells, we checked for pATR recruitment in EdU-labeled cells post IR at different time points. Indeed, pATR showed a gradual increase in S-phase cells (Fig. S5C,D). Next, we checked the levels and recruitment of major proteins that are regulated by ATR in RS response induction. As RPA32 is the first protein recruited to ssDNA that is generated due to RS (Patel and Weiss, 2018), we stained cells with an antibody against pRPA32 (T21) in U87MG cells at different time points post IR. We observed increased number of foci at 6 h and 24 h post IR (Fig. 4C,D). This corroborated with earlier data, where at 24 h post IR, cells experienced increased stalled and terminated forks. Unexpectedly, pRPA32 foci decreased at subsequent time points in the RR phase (days 7 and 10), although RS still prevailed with ssDNA accumulation (as shown in Fig. 3F,G). Next, to check RPA32 loading, we stained the cells with an antibody against total RPA32. We observed defective RPA32 loading at all time points post IR (Fig. S5E,F). These results suggest a deregulation in RPA32 loading (Toledo et al., 2013), which might be a distinct feature for therapy-induced RS-mediated RR cell generation. We also checked for RAD17, a target of ATR, and found higher pRad17 (S656) levels at 24 h post IR, but they were comparable to those seen untreated cells in a later time point (day 10) in RR phase (Fig. 4E,F). Next, we checked RAD51, which is regulated by ATR during the RS response and forms filaments leading to the resolution through the homologous recombination repair pathway (Petermann et al., 2010), and observed discrete foci formation at 6 h and 24 h post IR. However, similar to pRPA32 and pRAD17, RAD51 foci were decreased in the RR phase (Fig. 4G,H) suggesting the loss of RAD51 activity in the RR phase. Finally, we checked pCHK1 (S345) (encoded by CHEK1), a substrate of ATR and an effector kinase of the RS-mediated ATR signaling pathway, wherein phosphorylation of CHK1 plays a crucial role in RS-mediated checkpoint control to resolve the stress by arresting the cell cycle (López-Contreras and Fernandez-Capetillo, 2010). However, pCHK1 was not consistently increased during the second phase of γH2AX signaling when RS was observed (Fig. 4I), suggesting that a deregulated RS response existed in the therapy-induced RR GBM cells. These results show that cells might resolve the RS at the onset of the second phase of the γH2AX response, but in subsequent time points, cells are unable to activate the RS response, although pATR is persistent (Fig. 4J).

As we did not find any bona fide RS response proteins downstream to ATR in residual cells during the second phase of γH2AX despite persistent RS, we asked whether these cells are capable of inducing a canonical RS response. For this, we induced RS exogenously in U87MG cells by hydroxyurea (HU), which inhibits the synthesis of nucleotides and is known to induce a canonical RS response (Saxena and Zou, 2022). We then checked the recruitment of pRPA32, RAD51 and pATR along with γH2AX at different time points in the presence of 2 mM HU, which was administered continuously up to day 5 to U87MG cells (Fig. S5G). We observed elevated levels of γH2AX, pRPA32 and pATR upon prolonged RS until day 5 in HU-treated cells compared to those in the control cells; for RAD51, the increase was observed until 24 h (Fig. S5H–K). These data show that the canonical RS response is fully functional in GBM cells.

Additionally, as none of the canonical RS response proteins (except ATR) were activated in RR cells, we asked whether RS prevailed in relapse cells, which emerged from RR cells (Fig. 1A). Therefore, we compared the levels of pATR along with pRAD17, pRPA32 and RAD51 in U87MG control and relapse cells and found that pATR and pRPA32 levels were significantly reduced, there was no significant difference in RAD51 levels, and pRAD17 levels were elevated in relapse cells (Fig. S5L–O). These data suggest that pATR and, consequently, the RS response were absent in relapse cells. We also checked the status of pATM in SF268 cells, wherein we observed an increase in pATM levels post IR, which decreased at 72 h and remained low until day 10 (except at day 7), suggesting that pATM is limited until 72 h post IR (Fig. S5P). Altogether, we observed a persistent phosphorylated form of ATR during the second phase of γH2AX signaling, whereas all the direct substrates of ATR (RPA32, RAD17 and RAD51) were not activated during the second phase despite the presence of RS, except for an inconsistent and subtle activation of CHK1, indicating a non-canonical role of ATR in the RS response, mainly in RR GBM cells.

Next, we wanted to understand the relevance of therapy-induced RS response on the senescence phenotype in RR cells. For this, we performed the β-gal assay in U87MG cells and followed different time points after irradiation up to the RR phase. A significant increase in β-gal-positive cells was observed from day 5 up to day 14 (in the RR phase), indicating that senescence induction might be associated with RS (Fig. 5A,B). To understand the association between therapy-induced RS and senescence at the molecular level, we focused on the ATR protein, as significant levels of pATR were found in the second phase of γH2AX induction in U87MG RR cells. We used 100 nM of the ATR inhibitor (ATRi) VE-821 to completely abolish ATR activity without affecting cell growth or morphology (Fig. S6A–C). We treated U87MG cells with ATRi after 6 h post IR and replenished the drug after every 72 h. Unexpectedly, we observed a significant decrease in β-gal-positive cells following IR and ATRi treatment compared to IR with DMSO treatment for both U87MG and LN229 cells (Fig. 5C,D; Fig. S6D,E). To strengthen this observation, we checked the expression of factors associated with the senescence phenotype, including p21 (encoded by CDKN1A), IL8 (encoded by CXCL8), macroH2A1.1 and macroH2A1.2 (isoforms encoded by MACROH2A1) at their transcript level. The transcript levels of p21, IL8 and macroH2A1.1 were significantly downregulated in the ATRi-treated group compared to the DMSO group post-IR in U87MG cells, whereas the transcript levels of macroH2A1.2 were not significantly decreased (Fig. 5E). In concordance with the inhibitor study, abrogation of the ATR protein by siRNA also showed a reduction in β-gal-positive cells compared to treatment with scrambled siRNA at different time points after radiation (Fig. 5F,G; Fig. S6F). These results show that RS precedes TIS and that ATR is important for the maintenance of TIS in RR GBM cells.

As we learned that ATR is required for TIS induction as well as maintenance, we wondered about the consequences of ATR inhibition in terms of RS management. We checked the fork dynamics through DNA fiber assay and observed an increased percentage of fork termination at 48 h and decreased ongoing forks consistently upon ATRi treatment in U87MG cells, although there was no significant difference in the percentage of stalled forks and newly fired origins (Fig. 6A). As the RAD51 protein is essential for the protection and repair of replication forks, we checked its levels upon ATRi treatment at different time points post IR in U87MG cells. We observed a significant increase in RAD51 levels in ATRi-treated U87MG cells during later time points (beyond 48 h) (Fig. 6B,C), suggesting that RAD51 might be needed in stressed RR cells for fork protection and reversal in the absence of ATR activity. This prompted us to check ssDNA accumulation. We found increased ssDNA accumulation in ATRi-treated cells after IR treatment at 72 h but not at a later time point, suggesting no additional effect of ATRi on ssDNA accumulation (Fig. 6D,E). It is possible that increased RAD51 may suffice to protect the ssDNA generated after ATRi treatment to some extent. To check whether the effects of ATRi are dependent on radiation treatment, we also checked for RAD51 and 53BP1 accumulation after prolonged ATRi without IR treatment. As expected, we observed that prolonged ATRi treatment reduced RAD51 levels, indicating a homologous recombination defect. However, 53BP1 foci largely remained unaltered, suggesting that the effects of ATRi on RAD51 and 53BP1 levels are IR dependent (Fig. S7A–D). Next, we estimated 53BP1 levels at different time points under ATRi treatment conditions after IR. Unexpectedly, we found that 53BP1 foci formation decreased significantly under ATRi treatment conditions, suggesting faulty DNA repair (Fig. S7E,F). To confirm this further, we performed a comet assay and observed a significant increase in DNA DSBs in ATRi-treated cells compared to DMSO-treated cells at the post-IR time points (Fig. 6F,G). Taken together, our results show that ATR inhibition sensitizes IR-treated GBM cells by preventing TIS formation and, at the same time, heightening RS, causing lethal DNA breaks as DSBs when RS management fails to protect the cells.

Although TIS delays the emergence of the recurrent population (positive effect of TIS), at the same time, cells utilize it to escape from the lethal effects of radiation. Thus, targeting TIS would be an effective strategy to deal with TIS-mediated resistance. As we showed that the inhibition of ATR abrogated TIS in GBM cell lines along with an increase in DNA DSBs, we wanted to understand whether cell death was triggered after ATRi treatment and what was the fate of the cells that did not undergo TIS. We checked apoptosis by analyzing the sub-G1 population using flow cytometry (Ghorai et al., 2015). ATRi treatment significantly increased the sub-G1 population at day 10 post IR compared to DMSO treatment (Fig. 7A). At this point, we were further excited to expand the utility of ATRi treatment to eradicate TIS. Recently, we reported that the PARP-1 inhibitor (PARPi) olaparib enhanced and prolonged TIS in GBM cells (Ghorai et al., 2020). We then interrogated whether ATRi would have a similar inhibitory effect on PARPi-mediated TIS. We used U87MG cells treated with PARPi along with radiation. We counted the cell numbers in control and radiation treatment groups, which included ATRi, PARPi, and ATRi in combination with PARPi treatment. We observed a remarkable decrease in cell number in the combined drug treatment arm after radiation (Fig. 7B). In parallel, we checked the status of TIS by β-gal staining (Fig. 7C). As expected, PARPi-treated cells showed up to a 95% senescence phenotype, whereas ATRi with PARPi treatment exhibited a very significant drop in senescent cells, which was more drastic at a later time point of day 20 post IR (Fig. 7D). Next, we checked the fate of these cells to better understand the effect of the lethal combination of both ATRi and PARPi treatment in GBM RR cells. For this, we performed annexin V (ANXA5) staining for apoptosis, followed by flow cytometry analysis. Here, we observed an overall increase in annexin V-positive cells at day 10 post IR, which was highly significant at day 20 post IR, when combination treatment showed more than 90% apoptosis than any of the single agents (Fig. 7E). Taken together, we showed that ATRi can eradicate TIS by inducing apoptosis and ATRi has a synergistic effect on PARPi-mediated TIS, thereby driving these cells toward apoptosis, which is a desirable outcome with clinical implications.

In this study (summarized in Fig. 7F), we show that a lethal dose of radiation induces two phases of γH2AX associated with the presence of DSBs. The second phase of γH2AX mainly originates from the S phase, resulting in huge RS in GBM cells. Notably, a cellular senescence state is detected after the onset of RS, which is abrogated when ATR is genetically perturbed or inhibited by a specific inhibitor. Further, a combinatorial approach of ATR and PARP-1 inhibitors shows a potential benefit to eliminate TIS by inducing apoptotic cell death.

Cellular senescence is a state of permanent growth arrest, induced by various endogenous and exogenous stresses. The purpose is to restrict the cells from reverting and entering the cell cycle until the stress is removed. Cancer cell senescence or TIS has been recognized as a strategy to prevent tumor growth, but the latest research advancements in this field suggest that it may be a cause of recurrence in the long run (Zhang et al., 2021). Primarily, TIS is the mechanism of cancer cells to overcome therapeutic vulnerabilities, which is now well accepted as a major cause of tumor dormancy (Saleh and Gewirtz, 2022). Therefore, escape from TIS is paramount for better therapy outcomes. Although our previous studies established TIS as the survival strategy of radioresistant residual disease cells of GBM (Ghorai et al., 2020; Kaur et al., 2015, 2020; Rajendra et al., 2018, 2021), how TIS is manifested and maintained during the treatment phase remained unexplored. Interestingly, we discovered a biphasic mode of γH2AX signaling leading to TIS in GBM cells after radiation treatment. A recent report by Venkatachalam et al. (2017) also observed two waves of DNA damage upon a sub-lethal dose of H₂O₂ treatment in normal myoblast and retinal epithelial cells. The primary mode of action of radiation-induced DNA damage is mediated by ROS generated due to IR. Therefore, the residual ROS with an inefficient ROS-scavenging toolkit could be one of the plausible underlying mechanisms of biphasic DNA damage. Another possibility is having DNA repair intermediates as a byproduct of defective homologous recombination-mediated DNA repair (Feringa et al., 2018). This could be the reason for the persistent second phase of DNA damage, as evident from the comet assay, leading to persistent pATR and senescence induction, discussed later. Herein, we, for the first time, report the biphasic mode of DNA damage in GBM RR cells as a characteristic molecular event after radiotherapy and demonstrate it as a promising therapeutic aspect.

To further understand the genesis of the second phase of γH2AX signaling in GBM cells, we postulated that RS plays a key role in inducing the second phase of DNA damage. Indeed, EdU-labeled γH2AX tracking disclosed its major connection with RS in the S phase of cells, even though there was a basal level of γH2AX in non-S-phase cells during the second phase. This was confirmed by slowed replication fork progression and an increased percentage of terminated forks. Few studies have suggested that RS-induced senescence is distinctive compared to the role of RS in oncogene-induced senescence. For example, progerin-induced RS leads to premature senescence in Hutchinson–Gilford progeria syndrome (Wheaton et al., 2017). In another report, treatment with a purine analog induced RS-activated senescence and cell death in a cell type-specific manner (Lukášová et al., 2019). Li et al. (2017) showed resveratrol-induced S-phase arrest and cellular senescence in U2OS and A549 cells. Interestingly, none of these studies explored the contribution of RS in therapy resistance. In line with these, our data suggest that RS precedes TIS in GBM cells. However, at this stage, the exact source of RS is undetermined, but its presence is evident from our data. We speculate that accumulation of IR-induced ROS, unresolved DNA breakpoints, repair intermediates or inefficient resection encountered at replication forks could be the triggers of RS, followed by amplification of DNA damage during the second phase. Nevertheless, we provide novel insights into the association of biphasic γH2AX induction with the TIS and show that RS is associated with the induction and maintenance of TIS in GBM cells, leading to radiation resistance.

RS management is the surveillance mechanism to monitor and repair stress. Mechanistically, ATR, the apical kinase, activates and orchestrates a signaling hub consisting of many downstream substrates, such as RPA32, RAD17, CHK1 and ATRIP (Saxena and Zou, 2022). However, RS management during the establishment and maintenance of TIS in cancer cells, specifically in GBM, is not known. During the onset of RS, we observed the recruitment of pRPA32 and pRAD17 and the induction of pCHK1, followed by RAD51 foci formation, which is the molecular cascade of the RS response in general or seen under HU treatment. Such recruitment was sustained for not more than 48 h, whereas persistent pATR was observed throughout the TIS phase. At this point, we do not have evidence to explain why these downstream substrates of ATR were not in activated form, but it indicates the existence of a ‘non-canonical’ RS management that governs TIS. Our previous study showed reduced MRE11 in GBM RR cells (Kaur et al., 2020). MRE11 is a nuclease known for generating long stretches of ssDNA after fork degradation during RS (Ying et al., 2012). Here, we observed increased ssDNA during the progression of the RR phase with the absence of RAD51 filament formation, indicating limited or no fork regression. It is highly intuitive that the end resection for fork reversal is compromised, although possibilities of the involvement of other nucleases and helicases such as CtIP (also known as RBBP8), EXO1, DNA2 and WRN in the absence of MRE11 cannot be ruled out (Lemaçon et al., 2017). In another way, it could be further speculated that RADX (antagonist of RAD51) might play an inhibitory role by preventing RAD51 accumulation at the stalled fork for protection during the RR phase of GBM cells (Dungrawala et al., 2017). Despite this, RR cells recover from such huge RS, which might be regulated by the high levels of pATR observed in these cells. ATR can activate fork remodelers such as SMARCAL1 to prevent fork collapse (Couch et al., 2013). Surprisingly, a higher level of RAD51 was noticed in ATR-inhibited conditions during the second phase associated with RS, indicating a compensatory role towards fork protection in the absence of ATR activity, demonstrating an attempt of RR cells for survival, although it failed in the later time points. Our findings further raise possibilities of the compensatory role of other RAD51 paralogs for the management of RS (Bhattacharya et al., 2022), which can produce more vulnerabilities of these resistant GBM cells to newer therapeutics. Earlier studies showed that GBM tissues and glioma stem-like cells inherently harbor RS, which dictates radioresistance (Bartkova et al., 2010; Carruthers et al., 2018). A recent study also reported that high RS and reduced RAD51-mediated repair in oligodendrocyte precursor cells lead to radiosensitivity (Berger et al., 2022). However, our study shows radiation therapy can induce a non-canonical RS response with a sustained phosphorylated form of ATR and is crucial for acquiring resistance properties in GBM cells.

Furthermore, we showed that TIS is dependent on ATR activity. ATR inhibition abrogated senescence induction followed by its maintenance, as detected by β-gal positivity and the expression of SASP genes. This hints towards an interesting hypothesis of ATR-mediated chromatin regulation of the expression of these genes, which might be independent of its typical substrate interactions, as shown in this study. In addition, we noticed a decrease in MNGC formation upon ATR inhibition, which collectively enhanced the clearance of residual disease cells in GBM.

Finally, we show the high clinical implication of targeting GBM RR cells via dual inhibition of ATR and PARP-1. Our data demonstrate that ATRi has a profound benefit when combined with the PARPi olaparib. PARP-1 inhibitors, including olaparib, are in clinical trials (phase II) for newly diagnosed GBM as well as recurrent GBM (Fulton et al., 2018; Lesueur et al., 2019). We recently showed the clinical importance of olaparib in GBM therapy by accelerating and extending the TIS phase and at the same time, cautioned for inhibitor resistance (appearance of regrowth), even reported by others (Fleury et al., 2019; Ghorai et al., 2020). On the other hand, ATR inhibitors are in clinical trials for other cancers but not for GBM (Ferri et al., 2020). Intriguingly, our data strongly support the use of ATR inhibitors to kill the notorious TIS population as well as olaparib-induced TIS. One way to explain this synergy is the dependency of TIS on the ATR-mediated RS response. One of the major mechanisms of the PARP-1 inhibitor is trapping PARP-1 to the sites of DNA damage, resulting in RS induction inside the cancer cells. Hence, ATR inhibition might become the deciding factor to eliminate such TIS having huge RS, facilitating cell death. Thus, there is the possibility of using ATRi in olaparib-resistant or other PARPi-resistant cancers to bring out the maximum lethality.

Overall, our study elucidates for the first time the mechanism of radiation-induced senescence as a strategy of tumor dormancy or resistance in GBM after therapy and identifies persistent pATR with non-canonical RS management as the characteristic feature of TIS. Furthermore, targeting ATR or in combination with PARP-1 would serve as TIS-specific therapeutic targets with the potential to improve the outcomes of patients with GBM.

Human glioblastoma cell lines U87MG, SF268, LN229, and LN18 were grown in Dulbecco's modified Eagle's medium (DMEM, Gibco, USA) supplemented with 10% fetal bovine serum (Gibco) and antibiotics at 37°C in a humidified incubator containing 5% CO₂. Cell lines were authenticated by short tandem repeat (STR) profiling and routinely checked for mycoplasma contamination by using the commercially available kit (HiMedia, India). For irradiation, cells were treated with the corresponding lethal doses of IR (γ-radiation and X-ray) to generate RR disease cells (Ghorai et al., 2020; Nair et al., 2021). In general, cells were passaged at 48–72 h intervals after IR treatment to eliminate the dead cells and given medium change whenever there were some floating dead cells. Cells were seeded at different time points and collected as needed for experimentation. 100 nM of ATRi (VE-821, Selleck Chemicals) was used 6 h post IR treatment in the culture medium and replenished every 72 h until the RR cell state was reached. 5 μM of PARPi (Olaparib, Selleck Chemicals) was used and replenished until the RR state was reached, as mentioned in our earlier study (Ghorai et al., 2020). 2 mM HU (Sigma-Aldrich) was prepared in DMSO and added to the culture medium as indicated to induce DNA replication arrest. The thymidine analogs CldU (Sigma-Aldrich) and IdU (Sigma-Aldrich) were dissolved in ultrapure distilled water (Invitrogen) and used as indicated. EdU (Sigma-Aldrich) was prepared in DMSO as a 10 mM stock solution.

Anti-BrdU antibody (347580; against IdU) was from BD Biosciences. Anti-BrdU (ab6326; against CldU), anti-53BP1 (ab21083), anti-RAD51 (ab213), anti-pRPA32 (ab109394), anti-RPA32 (ab2175), anti-pRAD17 (ab76012), anti-pATR (ab227851), anti-actin (ab8224), goat anti-rabbit IgG (ab205722) and goat anti-mouse IgG (ab205719) were purchased from Abcam. Anti-γH2AX (9718), anti-ATR (2790), anti-CHK1 (2360) and anti-pCHK1 (#348) were purchased from Cell Signaling Technology. Anti-rabbit Alexa Fluor 594 (A-11037), anti-mouse Alexa Fluor 488 (A-11001), anti-rat Alexa Fluor 488 (A-11006) and anti-rabbit Alexa Fluor 488 (A-11034) were purchased from Thermo Fisher Scientific. Cy3-conjugated anti-mouse secondary antibody (C2181) was purchased from Sigma-Aldrich. BrdU, Trypan Blue, DAPI, sodium deoxycholate, Complete Mini EDTA-free Protease Inhibitor Cocktail Tablets and PhosSTOP Phosphatase Inhibitor Cocktail Tablets were purchased from Sigma-Aldrich. Bovine serum albumin (BSA), paraformaldehyde, Triton X-100 and glycine were purchased from HiMedia. Phenylmethylsulfonyl fluoride (PMSF) and the bicinchoninic acid (BCA) protein quantification kit were purchased from Thermo Fisher Scientific. VECTASHIELD mounting media (H1000) was purchased from Vector Laboratories.

siRNA was used to transiently suppress the expression of ATR (predesigned and validated; SASI_Hs01_00176271, Sigma-Aldrich) in U87MG cells. Non-specific siRNAs (FAM labeled) (Sigma-Aldrich) were used as a control. Cells were transfected with 90 pmol of siRNAs using Lipofectamine RNAi MAX (Invitrogen) as per the manufacturer's instructions. All the experiments were performed within 72 h post transfection. For long-duration experiments, cells were re-transfected every 72 h and viable cell counts were recorded for cell proliferation assay.

Dose-dependent (0–1 µM) growth inhibition assay for ATRi was performed in U87MG, SF268 and LN229 cells for 72 h, as described earlier (Ghorai et al., 2020). In brief, 0.5×10⁵ cells were seeded in six-well plates overnight, followed by inhibitor treatment. Viable cells were counted based on the Trypan Blue dye exclusion principle using a hemocytometer. Cell morphology was imaged to observe any morphological changes upon inhibitor treatment.

To score actively proliferating cells, cells were grown on glass coverslips, incubated with 10 µM BrdU for 6 h and taken at different time intervals as indicated. Cells were fixed and permeabilized with 4% formaldehyde and 0.5% Triton X-100 for 15 min each. Following PBS washes, 2 M HCl treatment was performed, followed by immunostaining with anti-BrdU (1:200). The slides were then labeled with DAPI and mounted using VECTASHIELD mounting media. Slides were visualized under a confocal microscope (LSM 780, Zeiss) with a 40× objective and images were captured.

Treated cells were seeded at an appropriate density in sterile culture plates and washed with 1× PBS, followed by fixation with 0.5 ml of fixative solution provided in the Senescence Detection Kit (9860, Cell Signaling Technology) for 15 min at room temperature. Cells were washed twice with 1× PBS and incubated for 4–6 h with 0.5 ml of staining solution containing 20 mg/ml of X-gal as per the manufacturer's protocol. Plates were analyzed for positive staining and images were acquired using an Olympus IX-71 microscope.

MNGCs were counted as described in our earlier reports (Ghorai et al., 2020; Kaur et al., 2015). In brief, cells with more than two nuclei with cell size >50 μm were classified as multinucleated and cells containing only one nucleus >50 µm were classified as giant cells. 100 cells were counted for each data point and the final data were generated from three independent experiments.

Cells were grown on glass coverslips and processed at the indicated time points after treatments. To analyze protein localization specifically in the S phase of cell cycle, cells were incubated with 100 µM EdU for 15 min along with treatments as indicated. Cells were fixed and permeabilized with 4% paraformaldehyde for 15 min and 0.5% Triton X-100 for 15 min, respectively. Click reaction mixture [100 mM sodium ascorbate, 2 mM CuSO₄, 1 µM biotin azide (Invitrogen), 10 µM Alexa Fluor 488 azide (Invitrogen)] was prepared and cells were incubated for 1 h in a dark humid chamber. Following PBS washes, blocking was performed using 1% BSA for 1 h at room temperature. Overnight incubation with the following primary antibodies was performed: anti-γH2AX (1:300), anti-pRPA32 (1:300), anti-pRAD17 (1:300), anti-RAD51 (1:300), anti-pATR (1:200), anti-RPA32 (1:200) and anti-53BP1 (1:300). After PBS washes, coverslips were incubated with specific Alexa Fluor-conjugated secondary antibodies (1:400). Cell nuclei were then counterstained with DAPI and mounted using VECTASHIELD mounting medium. Slides were visualized under a confocal imaging system (Zeiss LSM 780) and images were captured accordingly. For immunofluorescence, cells were processed similarly as mentioned above without the EdU incorporation and click reaction step.

For total protein extraction, cells were collected and lysed in ice-cold RIPA lysis buffer (50 mM Tris-Cl, pH 7.4, 150 mM NaCl, 1 mM EDTA, 1 mM sodium fluoride, 0.5% sodium deoxycholate, 1% SDS, 1% NP-40, 0.1% sodium azide, 1 mM sodium orthovanadate) with freshly added protease inhibitor cocktail (1×) and PMSF (1 mM) for 1 h, and sonicated using Bioruptor Plus (Diagenode). Protein lysates were centrifuged at 16,000 g and supernatants collected in separate tubes. 30–40 µg protein sample was used for immunoblotting as per standard methods. Blots were developed using ECL reagent (Bio-Rad) and images were captured using the Chemidoc imaging system (Bio-Rad). Uncropped blots are shown in Fig. S8.

To check DNA DSBs, neutral comet assay was performed as mentioned in our earlier work (Nair et al., 2021). Briefly, pre-cleaned microscope slides were thinly coated with 1% agarose and allowed to solidify at room temperature. After harvesting, cells were mixed with molten 1% low melting point agarose and the mixture was spread evenly on the slide in a uniform layer with the help of a coverslip. Slides were kept on ice for gel solidification. Cells were lysed by submerging slides in lysis buffer (10 mM Tris, 2.5 M NaCl, 100 mM EDTA, proteinase K, 1% Triton X-100, 10% DMSO, pH 8.5) and incubated for 1.5–2 h at room temperature in the dark. Slides were rinsed briefly in 1× TAE buffer (40 mM Tris-acetate, 1 mM EDTA) and electrophoresis was performed for 15 min in 1× TAE at 15 V. Slides were washed in PBS and incubated in absolute ethanol for 5 min, followed by air drying. Nuclei were stained with 2.5 µg/ml propidium iodide solution and visualized under a fluorescence microscope (Zeiss) with 40× magnification. Images were analyzed using CometScore software (http://rexhoover.com/index.php?id=cometscore) and graphs were plotted for tail moments using GraphPad Prism software.

To analyze DNA replication dynamics at the single-molecule level, DNA fiber assay was carried out as described by Nieminuszczy et al. (2016). Cells were pulse labeled with 50 µM IdU and 250 µM CldU following treatments as indicated in the experimental schemes. Cells were harvested and lysed, and fibers were spread out on frosted microscope slides. Immunodetection of labeled tracks was carried out using the following primary and secondary antibodies: mouse anti-IdU/BrdU (1:100), rat anti-CldU/BrdU (1:100), Cy3 anti-mouse (1:200) and Alexa Fluor 488 anti-rat IgG (1:200).

To prevent the analysis of broken tracts, U87MG cells were labeled with 50 µM CldU for 24–72 h, which was enough time to finish one round of replication and cell cycle. Cells were then pulse labeled with 100 µM EdU and 250 µM IdU as indicated in the experimental schemes. Next, cells were spotted and lysed, and fibers were spread, fixed and denatured on the frosted slide, followed by immunodetection using the respective antibodies. To detect EdU labeling, a click reaction mixture was prepared as mentioned above, and slides were incubated for 3 h in a dark humid chamber. Slides were washed with PBS and mounted with VECTASHIELD mounting medium. Fibers were visualized and imaged with the Zeiss LSM 780 confocal imaging system.

To perform DNA fiber assay using S1 nuclease, cells were pulse labeled with IdU and CldU, followed by cell permeabilization using CSK buffer [100 mM NaCl, 10 mM MOPS (pH 7), 3 mM MgCl₂, 300 mM sucrose, 0.5% Triton X-100] for 10 min and PBS washes. Cells were then washed once with S1 nuclease buffer (30 mM sodium acetate, 10 mM zinc acetate, 5% glycerol, 50 mM NaCl, pH 4.6) and treated with S1 nuclease (Invitrogen, 18001016) (20 U/ml) for 30 min at 37°C. Next, the S1 buffer was removed, and the cells were scraped and collected in PBS containing 0.1% BSA. Cells were spotted, lysed, fixed and denatured on the slide, followed by immunodetection using the respective antibodies. Fibers were visualized and imaged with the Zeiss LSM 780 confocal imaging system. Images were processed and analyzed using ImageJ software.

To detect nascent DNA in actively replicating cells, cells were pulsed with 100 µM EdU for 15 min, followed by a quick PBS wash, and a pulse of IdU was given for 30 min. For immunofluorescence-based detection of ssDNA, cells were fixed with 4% paraformaldehyde and permeabilized for 15 min with 0.5% Triton X-100 at 4°C. To detect EdU labeling, click reaction mixture (100 mM sodium ascorbate, 2 mM CuSO₄, 1 µM biotin azide, 10 µM Alexa Flour 488 azide) was prepared and cells were incubated for 1 h in a dark humid chamber. Following PBS washes, cells were blocked in 1% BSA and incubated with mouse anti-BrdU antibody (1:150) overnight at 4°C in 1% BSA. For IdU detection, cells were incubated with Cy3-conjugated secondary antibody (1:200) for 1 h. Nuclei were labeled by DAPI and mounted with VECTASHIELD mounting medium. Slides were visualized under a confocal microscope (Zeiss LSM 780), and images were acquired.

Total RNA was extracted from approximately 1 million cells by TRIzol reagent (Invitrogen). 1 μg RNA was subsequently treated with RQ1 DNase (Promega), followed by cDNA synthesis using the Improm-II reverse transcription system (Promega). Expression levels of target genes were quantified using the QuantStudio 5 Real-Time PCR cycler system with SYBR green (Applied Biosystems). Each assay was performed in triplicate and the final data were generated from three independent experiments. The expression level of each gene was analyzed and calculated relative to GAPDH expression using the ΔΔCt method with formula: 2^{[−ΔCt(experimental) – ΔCt(control)]}. The details of primers used in this study are: p21 F, 5′-GACACCACTGGAGGGTGACT-3′; p21 R, 5′-ACAGGTCCACATGGTCTTCC-3′; IL8 F, 5′-CAGTTTTGCCAAGGAGTGCT-3′; IL8 R, 5′-TTGGGGTGGAAAGGTTTGGA-3′; macroH2A1.1 F, 5′-GCCTCTTCCTTGGCCAGAA-3′; macroH2A1.1 R, 5′-CACTGTCGATCGAGGCAATG-3′; macroH2A1.2 F, 5′-CTTTGAGGTGGAGGCCATAA-3′, macroH2A1.2 R, 5′-TCTTCTCCAGCGTGTTTCCT-3′; GAPDH F, 5′-AATCCCATCACCATCTTCCAG-3′; and GAPDH R, 5′-AAATGAGCCCCAGCCTTC-3′.

Cells were seeded in six-well plates, harvested and washed with PBS after treatment, followed by fixation in 70% ethanol for 30 min at 4°C. Ethanol was removed by centrifugation and PBS washes were given to remove residual ethanol. Cells were further incubated in propidium iodide cocktail containing RNase A (40 µg/ml) for 15–30 min on ice. Flow cytometry data acquisition was performed using Attune NxT flow cytometer (Thermo Fisher Scientific) and analyzed using the ModFit software (Verity Software) for sub-G1 and cell cycle phase (G1, S and G2) analysis. Annexin V staining was performed according to our earlier report (Nair et al., 2021) and apoptotic cells (annexin V positive) were evaluated.

For immunofluorescence experiments, foci and intensity were calculated per nucleus; for MNGCs, they were calculated per cell. Image analysis was performed using ImageJ software. The numbers of samples and biological replicates are indicated in the figure legends. Furthermore, the generation of RR cells leads to a low number of actively proliferating cells (10–20% of the total population). Therefore, to get meaningful data, we considered a higher number of cells (technical repeats for each biological data point) during days 7 and 10 post IR, particularly for the DNA fiber assay, ssDNA assay, S1 nuclease fiber assay and EdU+γH2AX colocalization studies. For statistical analysis, two-tailed unpaired Student's t-test and one-way ANOVA with Tukey's multiple comparison test were performed. Graphs were plotted using GraphPad Prism 8.0 software and Microsoft Excel. Final figure panels were generated using Adobe Illustrator.

We acknowledge the Advanced Centre for Treatment, Research & Education in Cancer (ACTREC) Imaging Facility for their generous help during confocal microscopy imaging. We express our sincere gratitude to all the members of Shilpee laboratory and Dr Amit Dutt laboratory (ACTREC) for their constructive intellectual inputs during this study.