The same, only different – DNA damage checkpoints and their reversal throughout the cell cycle

The current dogma is that the DNA damage response (DDR) affects proliferation such that the DDR can reversibly arrest cell cycle progression to allow time for DNA repair, and, upon completion of DNA repair, the DDR is turned off to allow cell cycle resumption (Box 1). It is becoming increasingly clear that the DDR not only communicates to the cell cycle machinery, but that the cell cycle machinery is also involved in regulating the DDR, such that a similar DNA break activates different arms of the DDR depending on the cell cycle phase in which the damage occurs. In turn, silencing of the DDR is different in each cell cycle phase. These cell-cycle-dependent changes are intimately intertwined with double-strand break (DSB) end-processing and the usage of non-homologous end-joining (NHEJ) and homologous recombination (HR) repair pathways. In this Commentary, we address differences in the response to DSBs throughout the cell cycle in metazoan cells, highlighting which arms of the DDR function in the different phases. We also discuss how DNA-damage-induced checkpoints are silenced in the different cell cycle phases, how cells undergo checkpoint recovery after checkpoint silencing and how cells, when arrested, maintain reversibility of the arrest.

Central to the response to a DSB is its recognition by the MRN (Mre11, Rad50 and Nbs1) complex and the subsequent recruitment and activation of the ataxia telangiectasia mutated (ATM) kinase, which is conserved throughout the cell cycle (Giunta et al., 2010; Lee and Paull, 2005). Subsequent activation of various arms of the DDR is precisely tuned to each cell cycle phase as discussed below (Fig. 1).

ATM phosphorylates S139 on histone variant H2AX (yielding γH2AX) on the DNA adjacent to a DSB, which is recognised by mediator proteins that amplify local ATM activity and recruit repair factors (Burma et al., 2001; Lou et al., 2006; Stucki and Jackson, 2006). In addition, ATM activates the Chk2 kinase that disseminates from the site of damage and contributes to the DDR (Matsuoka et al., 1998). In response to a DSB in G1, ATM and Chk2 are required for the stabilisation of p53, which in turn results in the induction of a large variety of transcriptional targets, among which the cyclin-dependent kinase (Cdk) inhibitor protein p21 (also known as CDKN1A) (Harper et al., 1993; Harper et al., 1995). Accumulated p21 binds to and inhibits cyclin–Cdk complexes to block cell cycle progression, and loss of p53 or p21 results in complete loss of the G1 checkpoint (Deng et al., 1995; Hirao et al., 2002; Kastan et al., 1991). ATM also activates the p38 MAPK family, which contributes to the G1 checkpoint through stabilisation of p21-encoding mRNA (Lafarga et al., 2009; Reinhardt et al., 2007). This relatively slow transcriptional response is complemented by ATM-, Chk2- and p38-dependent pathways that rapidly, but transiently, prevent S-phase entry by promoting the degradation of cyclin D and the Cdc25A phosphatase that reverses inhibitory phosphorylation of Cdk2 (at T14 and Y15) (Agami and Bernards, 2000; Deckbar et al., 2010; Mailand et al., 2000; Reinhardt et al., 2007; Santra et al., 2009).

Once Cdk2 activity initiates DNA replication, DSBs can trigger entirely different signalling responses as some broken ends are not ‘simply’ religated by NHEJ (for a review, see Ochi et al., 2014; Deriano and Roth, 2013) but instead are repaired by HR, which involves the extensive 5′–3′ resection of the broken ends to produce a 3′ overhang. The process of DNA end-resection commits repair to HR and is tightly controlled such that HR only takes place when a sister template is available in S- or G2-phase (Ira et al., 2004). The single-stranded DNA (ssDNA) generated by resection activates the ataxia telangiectasia and Rad3-related (ATR) kinase (Zou and Elledge, 2003), thus promoting a switch from DNA damage signalling exclusively by ATM and Chk2 kinases in G1 to additional contributions of ATR and its effector Chk1 kinase in S-phase and G2. This activation of ATR and Chk1 kinases also occurs as a consequence of replication, when the replication fork generates single-strand DNA (ssDNA) (Paulsen and Cimprich, 2007). Cdk2 phosphorylates the key exonucleases CtIP (also known as RBBP8) and Exo1 to enable DNA end-resection and ensure that end-resection occurs only after S-phase entry (Buis et al., 2012; Tomimatsu et al., 2014; Wang et al., 2013). In addition, Cdk2-dependent, activating phosphorylations of the ATR-interacting protein ATRIP (at S224) and Chk1 itself (at S286 and S301) further restrict full activation of ATR and Chk1 to S- and G2-phase (Myers et al., 2007; Xu et al., 2012). Entry into S-phase thus drastically changes the signalling cascades that are elicited by a DSB.

Even though ATM, ATR, Chk2 and Chk1 all contribute to the stabilisation of p53 (Meek and Anderson, 2009), downstream p21 accumulation is prevented during DNA replication by the PCNA-associated CRL4^Cdt2 ubiquitin ligase (a cullin4A RING E3 ubiquitin ligase complex containing Cdt2), which is present on the replication fork (Abbas et al., 2008; Havens and Walter, 2011; You et al., 2002). Interfering with p21 degradation in S-phase compromises the control over origin licensing and results in detrimental re-replication (Kim et al., 2008), underscoring the importance of this mode of checkpoint attenuation in S-phase. With the continuous degradation of p21, the intra-S checkpoint instead relies on Wee1 kinase, which becomes expressed in S-phase and phosphorylates Cdk2 to inhibit its activity (Chow et al., 2003; Watanabe et al., 1995). Chk1 and Chk2 deposit activating phosphates on Wee1 kinase and target the counteracting phosphatase Cdc25A for degradation to impose an immediate break on further Cdk activation (Beck et al., 2010; Hughes et al., 2013; O'Connell et al., 1997).

Wee1-dependent phosphorylation of Cdks remains crucial for checkpoint control in G2. Wee1 inhibitors override the G2 checkpoint in p53 mutant cells (Hirai et al., 2009). In addition, the slower transcriptional induction of p21 that is reinstated in G2 is required for sustained inhibition of Cdk activity and cell cycle arrest (Bunz et al., 1998). In contrast to the checkpoint in response to DSBs in G1, a checkpoint arrest in G2 does not absolutely depend on ATM, p53 and p21. Although ATM and Chk2 control the establishment of an arrest after damage (Painter and Young, 1980; Rainey et al., 2008), ATR- and Chk1-dependent signals are essential for checkpoint maintenance in G2 (Brown and Baltimore, 2003; Liu et al., 2000; Shibata et al., 2010; Takai et al., 2000). Conversely, ATR and Chk1 are neither necessary nor sufficient for a G1 arrest in response to DSBs (Chen et al., 1999; Deckbar et al., 2010; Nghiem et al., 2001). This suggests that lesions engaged in HR are determinant for checkpoint maintenance in G2; this hypothesis is supported by a failure to maintain a G2 checkpoint arrest in the absence of PALB2 and BRCA2, which mediate strand invasion (Menzel et al., 2011). In addition, ATM- and ATR-dependent activation of p38 might contribute to the maintenance of a G2 checkpoint by inhibiting both Cdc25A and Cdc25B (Reinhardt et al., 2007).

The DDR is potently inhibited during mitosis, suggesting that at least some forms of mitotic DNA repair are undesirable. Even though a DSB sustained or generated in mitosis still activates ATM, downstream activation of Chk2 and recruitment of repair factors does not occur (Giunta et al., 2010; van Vugt et al., 2010). Mitotic attenuation of the DDR is enforced at various levels. The mitotic kinase Plk1 phosphorylates Chk2 on multiple sites to prevent its dimerization and recruitment to the DSB, both of which are required for its activation (van Vugt et al., 2010). Similarly, Plk1 abrogates Chk1 activation by inducing SCF^βTRCP (a SKP1–cullin-1–F-box complex that contains βTRCP as the F-box protein)-mediated degradation of the scaffold protein claspin that is required to bridge ATR and Chk1 (Mailand et al., 2006; Mamely et al., 2006; Peschiaroli et al., 2006). It is conceivable that checkpoint kinases that inhibit Cdks must be quelled to maintain the full Cdk activity that is necessary for the execution of mitosis. However, the mitotic inhibition of the DDR is not limited to the checkpoint. Cdk1 phosphorylates the ubiquitin ligase RNF8 (at S198) to prevent the formation of K63-linked polyubiquitin at damage sites (Orthwein et al., 2014). Additional phosphorylation of residues T1609 and S1618 of 53BP1 by Plk1 and an unidentified kinase interferes with its recruitment to damaged DNA (Lee et al., 2014; van Vugt et al., 2010). Phosphorylation of RNF8 and 53BP1 prevents DNA repair during mitosis (Orthwein et al., 2014). Additionally, Plk1 and Cdk1-dependent phosphorylation of XRCC4 contributes to the suppression of NHEJ during mitosis (Terasawa et al., 2014). Collectively, these mechanisms eliminate most checkpoint signals and DNA repair in mitosis.

Although the picture is far from being complete, it has become clear that a myriad of mechanisms operate to fine-tune the DDR in each phase of the cell cycle. Attenuated checkpoint responses are necessary for the orderly completion of DNA replication in S-phase (Kim et al., 2008), to prevent deleterious fusions of telomeres in mitosis (Orthwein et al., 2014), and probably more.

Activation of the DDR leads to a highly complex series of events orchestrated by a plethora of posttranslational modifications (PTMs), such as phosphorylation, acetylation, methylation, poly(ADP-ribosyl)ation (PARylation), ubiquitylation and small ubiquitin-like modifier (SUMO)ylation. As these modifications trigger checkpoint activation, cell cycle arrest and DNA repair, removal of these modifications by dedicated enzymes or degradation of modified proteins is essential for checkpoint silencing and recovery. Thus far, this emerging field has identified redundancy and complex cross-talk between the modifications as a common theme, as discussed below.

Phosphorylation is the most widespread and best-studied PTM of the DDR. Many serine/threonine phosphatases are involved in dephosphorylation, and these proteins act in a largely redundant manner (Fig. 2) (Lee and Chowdhury, 2011; Shimada and Nakanishi, 2013). For instance, PP2A (in complex with the B55α and B56ε subunits), PP4, PP6 and Wip1 all dephosphorylate γH2AX in vitro and control γH2AX levels in vivo (Chen et al., 2014; Chowdhury et al., 2005; Chowdhury et al., 2008; Douglas et al., 2010; Macůrek et al., 2010). This redundancy is most likely a reflection of different spatiotemporal regulation of these phosphatases during the DDR. Indeed, PP2A phosphatase activity towards γH2AX-containing nucleosomes is repressed in the first hour, but increases by twofold 5 hours after DSB induction by camptothecin (Chen et al., 2014). PP1, PP2A, Wip1 and PP6 are regulated through distinct mechanisms by DNA damage, and both PP2A and Wip1 have been shown to accumulate at DSBs (Chen et al., 2014; Chowdhury et al., 2005; Dozier et al., 2004; Guo et al., 2002; Macůrek et al., 2010; Naito et al., 2012; Shouse et al., 2011), whereas PP4 appears to be selective for chromatin-dissociated γH2AX in yeast (Keogh et al., 2006), revealing that these phosphatases modulate specific pools of γH2AX at different times. Although less thoroughly assessed for other checkpoint phosphorylation events, redundancy and spatiotemporal regulation are a recurrent theme. For example, phosphorylated ATM (S1981) and Chk2 (T68) are both dephosphorylated by PP1, PP2A and Wip1 (Carlessi et al., 2010; Dozier et al., 2004; Fujimoto et al., 2006; Goodarzi et al., 2004; Liang et al., 2006; Peng et al., 2010; Shreeram et al., 2006). Recognition of T68-phosphorylated Chk2 by the B56α subunit of PP2A can be prevented by additional ATM-dependent phosphorylation of Chk2 at S33 and/or S35, suggesting that T68 dephosphorylation is facilitated by prior loss of ATM activity and dephosphorylation of S33 and/or S35 by an additional phosphatase (Freeman et al., 2010). Formation of a Chk2–Wip1 complex, by contrast, requires Chk2 kinase activity (Yoda et al., 2006). These examples illustrate that checkpoint phosphorylation events are counteracted by multiple redundant phosphatases that are regulated themselves in time and space, and, in addition, show regulated activity towards specific substrates.

Protein phosphorylation also controls the extensive ubiquitylation at sites of damage to create a landing platform for additional signalling and repair proteins. The best-characterised events are the sequential recruitment of the E3 ubiquitin ligases RNF8 and RNF168 that produce K63-linked polyubiquitin chains (Doil et al., 2009; Stewart et al., 2009), but many other ubiquitin ligases with largely unknown substrates take part in the DDR (Jackson and Durocher, 2013). Several deubiquitylating enzymes (DUBs) have been implicated in removal or editing of the K63 ubiquitin chains formed at DSBs. USP3 and USP44 overexpression reduces the size and number of DNA damage foci and depletion of POH1, BRCC36 or USP16 results in the defective removal of ubiquitin chains at the site of damage (Butler et al., 2012; Doil et al., 2009; Mosbech et al., 2013; Shanbhag et al., 2010; Shao et al., 2009). Both DUBs POH1 and BRCC36 are recruited to K63-linked polyubiquitin chains to reverse these same RNF8- and RNF168-dependent ubiquitylation events at the chromatin, yet their impact on DNA repair processes can be antagonistic, illustrating the complexity of regulation of ubiquitin at DSBs (Butler et al., 2012; Kakarougkas et al., 2013; Shao et al., 2009). Recently, a systematic screen of DUBs identified an additional 29 DUBs that localise to DNA damage when overexpressed as GFP fusions and potentially regulate checkpoint silencing (Nishi et al., 2014).

Direct reversal of protein modifications is not the only means to dismantle the DDR. Proteasomal destruction is the end point of many of the modified proteins. The proteasome is recruited to sites of DNA damage, possibly by the associated DUBs POH1 or UCHL5 (Blickwedehl et al., 2007; Butler et al., 2012; Nishi et al., 2014). The ubiquitin-ligase RNF4 recognizes SUMOylated MDC1 and RPA1 at DNA repair foci and promotes their proteasomal degradation (Galanty et al., 2012). RNF8 not only facilitates the formation of K63-linked polyubiquitin chains by RNF168, but also targets Ku and Chk2 for proteasomal degradation via distinct K48-linked ubiquitin chains (Feng and Chen, 2012). Similarly, the E3-ligase Iduna is recruited to and activated by PAR chains, where it targets the PAR polymerase for proteasomal degradation (Kang et al., 2011). Alternatively, eviction and replacement of heavily modified nucleosomes might also aid in terminating checkpoint signals (see Box 2).

As PTMs are central to both the coordination of DNA repair and checkpoint signalling, disruption of enzymes that remove such PTMs can affect cell survival after DNA damage in multiple ways. Disruption of PP2A activity impairs DSB repair, and the observed persistence of γH2AX in PP2A-depleted or -inhibited cells might therefore be caused by persistence of DSBs, lack of checkpoint silencing or both (Chen et al., 2014; Chowdhury et al., 2005). Chemical inhibitors of checkpoint kinases, such as caffeine, present a useful tool to separate defects in DNA repair from defects in terminating the checkpoint downstream of these kinases (Davies et al., 2004; van Vugt et al., 2005). In contrast to PP2A, loss of Wip1 or PP4 have limited effects on DSB repair, but strongly affect checkpoint silencing, cell cycle restart and survival after DNA damage, even in the presence of checkpoint kinase inhibitors (Lindqvist et al., 2009a; Moon et al., 2010; Nakada et al., 2008; Shaltiel et al., 2014). Crucial functions of these phosphatases in checkpoint reversal are limited to a select number of substrates, possibly owing to the extensive redundancy between phosphatases. Wip1 dephosphorylates γH2AX, ATM, p53, Mdm2, Chk1 and Chk2 to repress the DDR (Fujimoto et al., 2006; Lu et al., 2007; Lu et al., 2005; Macůrek et al., 2010; Shreeram et al., 2006), yet its role in checkpoint recovery is restricted to G2 where it is required to antagonise p53 (Fig. 3) (Lindqvist et al., 2009a). Wip1-depletion in p53-deficient cells does not affect recovery from DNA damage in G2, underscoring that Wip1 activity towards these other substrates is dispensable (Lindqvist et al., 2009a). Similarly, PP4 dephosphorylates γH2AX, RPA2, 53BP1, CHD4 and KAP1, but recovery in its absence is mostly affected in G1 cells (Fig. 3), and can be effectively overcome by expression of a non-phosphorylatable mutant of KAP1 (S473A) (Lee et al., 2012; Shaltiel et al., 2014).

Although several DUBs are required for both DNA repair and cell viability after irradiation (Delgado-Díaz et al., 2014; Ismail et al., 2014; Lancini et al., 2014; Nishi et al., 2014), the role of DUBs in checkpoint silencing and cell cycle restart after DNA damage remains to be addressed. Of the 32 DUBs found to localise to sites of DNA damage, 13 appear to be dispensable for overall DNA repair, as judged by single-cell gel electrophoresis, with depletion of USP4, USP37 or POH1 affecting checkpoint kinase signalling. Their contribution to checkpoint silencing needs to be assessed and mapped to relevant substrates.

Checkpoint recovery follows largely the same activation pathway as unperturbed mitotic entry (Lindqvist et al., 2009b), but there are clear differences: the DDR inhibits the activities of the pro-mitotic kinases Cdk1, Aurora A and Plk1 (Krystyniak et al., 2005; Lock and Ross, 1990; Smits et al., 2000), and causes the degradation of several proteins that promote mitotic entry, such as Cdc25A and Bora (Falck et al., 2001; Qin et al., 2013). The resultant rewiring of the cell cycle machinery is most clearly demonstrated by the acquired dependence on Plk1 activity for G2 checkpoint recovery (Fig. 3). Plk1 is dispensable for normal mitotic entry and its inhibition in unperturbed cells merely leads to a delay in mitotic entry. However, after a DNA-damage-induced G2 arrest, Plk1 and its upstream activators Aurora A and Bora become essential for re-entry into the cell cycle, by contributing to the activation of cyclin-B1–Cdk1 (Lindqvist et al., 2009b; Macůrek et al., 2008; van Vugt et al., 2004). In parallel, the degradation of Cdc25A reduces the redundancy of Cdc25 phosphatases normally observed during mitotic entry, and recovering cells become dependent on a specific isoform of Cdc25B, which is stabilised after damage (Jullien et al., 2011; van Vugt et al., 2004).

Aside from its roles in re-activation of cyclin-B1–Cdk1, Plk1 additionally controls silencing of DDR signals. The same mechanisms that enable Plk1 to prevent checkpoint signalling in mitosis operate during checkpoint recovery in G2 (Figs 1 and 3). Plk1 activity starts to rise in G2, several hours before mitosis, and at early stages of the DNA damage response Plk1 facilitates strand invasion for HR (Yata et al., 2014). During checkpoint recovery, Plk1 targets claspin for degradation, disabling Chk1 activation (Mailand et al., 2006; Mamely et al., 2006; Peschiaroli et al., 2006). Failure to degrade claspin leads to enhanced and continued Chk1 activity and inability to recover from a G2 arrest. Similarly, Plk1 negatively regulates 53BP1 and Chk2 to promote recovery (van Vugt et al., 2010; Vidanes et al., 2010). Plk1 might also contribute to the inactivation of p53 through direct and indirect mechanisms (Ando et al., 2004; Dias et al., 2009; Liu et al., 2010; Yang et al., 2009), and proteomic approaches have identified many potential new binding partners and substrates of Plk1, including DNA damage checkpoint and repair proteins (Halim et al., 2013; Lowery et al., 2007). Plk1 thus drives transitions from arrest to cell cycle re-entry and from checkpoint signalling to checkpoint silencing and recovery.

It is typically thought that cell cycle resumption upon a DSB-induced arrest occurs only following completion of DNA repair. However, cells can terminate the checkpoint and enter mitosis in the presence of DNA damage, a process well characterised in unicellular yeast and known as checkpoint adaptation (reviewed in Paulovich et al., 1997). The first hints of checkpoint adaptation in human cells came from the discovery that DNA damage in HCT116 cells resulted in aberrant mitoses (Andreassen et al., 2001), which was later strengthened by the finding that, as in yeast, checkpoint adaptation in human cells requires Plk1 (Syljuåsen et al., 2006; Toczyski et al., 1997). Plk1 must be degraded following DNA damage in order to achieve a proper G2 arrest (Bassermann et al., 2008) and the inability to silence Plk1 activity following DNA damage increases the fraction of cells that is able to enter mitosis (Smits et al., 2000). It was subsequently shown that irradiation of cells in G2 results in increased amounts of γH2AX and MDC1 foci in mitotic cells, both markers for DNA damage (Deckbar et al., 2007; Kato et al., 2009; Liang et al., 2014), strengthening the notion that maintaining a G2 cell cycle arrest requires a threshold of persistent DNA damage. Using a fluorescence-based probe for Plk1 activity and DNA damage in single cells, Liang and colleagues have recently reported that DNA damage permits a gradual increase of Plk1 activity throughout the arrest (Liang et al., 2014). Once Plk1 activity increases beyond a certain level, the cell progresses to mitosis, regardless of whether DNA damage persists. G2 checkpoint recovery might thus represent checkpoint adaptation, where DNA damage triggers an arrest whose duration is not necessarily conditioned by DNA repair. In support of this notion, the kinetics of recovery from DSBs in G2 in non-transformed cell lines and the mouse intestinal epithelium in vivo indeed show relatively synchronous mitotic entry after a delay of several hours (Chwalinski and Potten, 1986; Shaltiel et al., 2014). However, it is important to note that the inducible expression of a constitutively active mutant of Plk1 (T210D) fails to override an established DNA damage checkpoint (R.H.M., unpublished results), indicating that Plk1 activity is insufficient to drive mitotic entry of cells with a fully activated DDR. Therefore, other factors that contribute to the silencing of the DNA damage response operate in conjunction with Plk1 to drive checkpoint recovery or adaptation.

Checkpoint adaptation is a potentially dangerous process, as the propagation of damaged DNA can contribute to tumorigenesis. Mitotic entry in the presence of severe DNA damage results in cell death through mitotic catastrophe (Bunz et al., 1998; Chan et al., 1999; Chen et al., 2003). Alternatively, pre-marked breaks present in G2 cells are partially disassembled during mitosis and carried over to daughter cells where the DDR is fully (re)activated in G1 (Giunta et al., 2010; Lee et al., 2014). Activation of the DDR in G2 can even prime adapting cells for a cell cycle arrest or delay in G1. It has been recently shown that the amount of residual Cdk2 activity at mitotic exit dictates the quiescence or proliferation decision in daughter cells (Spencer et al., 2013). Although cells normally retain some cyclin-E–Cdk2 activity as they complete mitosis and swiftly progress through G1 without the requirement for mitogenic stimulation, the induction of p21 in late G2 or mitosis gives rise to quiescent daughters with hypophosphorylated retinoblastoma protein (pRb) (Spencer et al., 2013). Similarly, PP2A-B56γ-dependent dephosphorylation of pRb in G2 results in quiescent daughter cells that lack residual Cdk2 activity (Naetar et al., 2014). Inhibition of PP2A-B56γ in G2 results in increased levels of cyclin E1 in mitotic cells and accelerates Cdk2-dependent S-phase entry in daughter cells. As ATM contributes to the activation of PP2A-B56γ (Shouse et al., 2011), ATM might thus synergise with p21 to inhibit Cdk2 activity in checkpoint-adapting cells, which are then programmed to enter a state of quiescence after cell division to safeguard genomic stability.

Expression of G2-specific genes is driven by the transcription factors FoxM1, B-Myb (also known as MYBL2) and nuclear factor Y (NF-Y), and is essential for timely mitotic entry and proper progression through mitosis (Chae et al., 2004; Down et al., 2012; Laoukili et al., 2005). FoxM1, NF-Y and B-Myb are all activated by Cdk2, ensuring that they become active only when cells progress through S- and G2-phase (Chae et al., 2004; Laoukili et al., 2008; Major et al., 2004; Ziebold et al., 1997). Given that Cdk-activity is required for the transcription of G2-specific genes, decreased Cdk-activity following DNA damage can compromise their expression (Badie et al., 2000; Crawford and Piwnica-Worms, 2001; Maity et al., 1996). Some residual Cdk-activity, possibly supported by specific Cdc25B isoforms (Jullien et al., 2011), must therefore be retained. Temporary elimination of this residual Cdk activity with inhibitor compounds abolishes G2-specific gene expression and compromises the ability of cells to recover after checkpoint inhibition (Alvarez-Fernández et al., 2010). Similarly, small interfering RNA (siRNA)-mediated depletion of FoxM1 is detrimental to G2 recovery and can only be rescued by inducible siRNA-resistant FoxM1 when it is present throughout the arrest (Alvarez-Fernández et al., 2010). In conjunction with FoxM1, B-Myb contributes to the expression of G2-specific genes during an arrest in G2 and is required for checkpoint recovery in p53-deficient cells (Mannefeld et al., 2009). Damaged G2 cells must thus carefully suppress the Cdk-driven G2 transcriptional programme during the arrest to prevent premature mitotic entry, but at the same time retain sufficient expression to maintain recovery competence (Fig. 4A; for a review on fine-tuning of transcriptional regulation during a DNA damage-induced G2 arrest see Alvarez-Fernández and Medema, 2010).

p53 counteracts the transcription of many G2-specific genes to contribute to the maintenance of a G2 arrest, and upregulation of p53 is sufficient to permanently arrest cells in G2, even in the absence of DNA damage (Agarwal et al., 1995; Bunz et al., 1998; Krenning et al., 2014; Mannefeld et al., 2009; Riley et al., 2008). Consequently, cells must also balance p53 activity to maintain the reversibility of a G2 arrest (Fig. 4A). This requires negative regulation of p53 by Mdm2 and Wip1, which are both transcriptional targets of p53 (Barak et al., 1993; Fiscella et al., 1997). Mdm2 is the main E3 ubiquitin ligase that targets p53 for proteasomal degradation and inhibits its function as a transcription factor (for a review on Mdm2 and p53, see Wade et al., 2013), whereas Wip1 dephosphorylates both p53 and Mdm2 to promote p53 degradation (Lu et al., 2005; Lu et al., 2007). Recently, TRIM24 was identified as an additional p53-induced ubiquitin ligase that preferentially targets phosphorylated variants of p53 for degradation (Jain et al., 2014). As a result, p53 levels oscillate in response to DNA damage and altering the oscillatory pattern of p53 results in distinct cell fate outcomes (Batchelor et al., 2008; Lahav et al., 2004; Purvis et al., 2012). Wip1 depletion results in excessive p53 signalling after DNA damage and renders cells unable to recover even when the checkpoint is silenced (Batchelor et al., 2008; Lindqvist et al., 2009a). Reintroducing siRNA-resistant Wip1 restores recovery competence when it is continuously present during the arrest, but not when it is only induced from the moment of checkpoint silencing, underscoring the importance of negative regulation of p53 in the time frame preceding checkpoint reversal for controlling the outcome of the DNA damage response (Lindqvist et al., 2009a).

The anaphase-promoting complex/cyclosome (APC/C) is a multi-protein complex with E3 ubiquitin ligase activity that is normally active from the completion of chromosome alignment in mitosis until the end of G1; it is necessary to direct the degradation of cyclins, securin and several mitotic kinases for the orderly completion of mitosis (Primorac and Musacchio, 2013). Two coactivators, Cdc20 and Cdh1, sequentially activate the APC/C. Cdc20 activates the APC/C following chromosome alignment and is inhibited by the spindle assembly checkpoint. This initiates the destruction of cyclin B1 and a decrease in Cdk activity, which is required to allow Cdh1-dependent activation of the APC/C during the late stages of mitosis and throughout G1. DNA damage in G2 can activate APC/C^Cdh1, which is essential for a persistent G2 arrest and requires p53- and p21-dependent Cdk inhibition (Bassermann et al., 2008; Lee et al., 2009; Sudo et al., 2001; Wiebusch and Hagemeier, 2010). Premature activation of the APC/C^Cdh1 in G2 by DNA damage targets many proteins required for mitotic entry for destruction, including cyclin A, cyclin B1 and Plk1, thus preventing recovery from a G2 arrest (Fig. 4A) (Johmura et al., 2014; Krenning et al., 2014; Lee et al., 2009; Lindqvist et al., 2009b; Wiebusch and Hagemeier, 2010). The resulting cells that have a G2 DNA content but lack G2-specific protein expression are committed to senescence (Johmura et al., 2014; Ye et al., 2013). In order for a G2 arrest to remain reversible, Cdk activity must thus be maintained at sufficient levels to preserve pro-mitotic gene transcription, but also to prevent premature activation of the APC/C.

Throughout G2, cyclin B1 continuously shuttles between the cytoplasm and the nucleus, but the rate of export exceeds the rate of import, resulting in a mostly cytoplasmic localisation until the onset of mitosis (Hagting et al., 1998). In response to DNA damage, cyclin B1 is bound by p21 and relocates to the nucleus within 5 hours in normal human fibroblasts (Charrier-Savournin et al., 2004; Krenning et al., 2014). The binding of p21 to cyclin-B1–Cdk1 complexes in the nucleus most likely prevents their nuclear export. Whereas Cdc25A is degraded upon DNA damage, Cdc25B and Cdc25C are exported out of the nucleus (reviewed by Boutros et al., 2006), separating them from cyclin-B1–Cdk1 complexes that have translocated to the nucleus in response to damage. These nuclear p21-bound cyclin-B1–Cdk1 complexes are inhibited by nuclear Wee1 (Charrier-Savournin et al., 2004; Krenning et al., 2014; Müllers et al., 2014). Moreover, once cyclin-B1–Cdk1 complexes have translocated to the nucleus, they become refractory to reactivation by addition of a Wee1 inhibitor, most likely due to the cytoplasmic sequestration of the remaining Cdc25 isoforms. This causes the G2 arrest to be maintained in cells with nuclear cyclin B1, even if the checkpoint is silenced prior to APC/C activation (Fig. 4B) (Charrier-Savournin et al., 2004; Krenning et al., 2014; Müllers et al., 2014). As p53 plays a central role in the nuclear translocation of cyclin B1, APC/C activation and the regulation of repression of pro-mitotic gene transcription, these mechanisms are often attenuated in cancer cells.

The G1 checkpoint is important in cells that encounter DSBs during G1, as well as those that have adapted from the G2 checkpoint or sustained breaks during mitosis (Giunta et al., 2010). In addition, G2-arrested cells that activate APC/C^Cdh1 are likely to be subjected to a G1-type DDR, owing to the lack of Cdk1/2 activity. Reversal of the G1 DNA damage checkpoint appears to be distinct from G2 as, firstly, the crucial players that promote checkpoint recovery in G2 are absent (Plk1 and Cdc25B) or do not affect recovery in G1 (Wip1; Shaltiel et al., 2014). Secondly, unlike the G2 checkpoint, the G1 checkpoint remains reversible for several days in untransformed cells and the mechanisms that cause an irreversible arrest in G2 (Fig. 4) are not relevant in G1 (Deckbar et al., 2010; Krenning et al., 2014; Shaltiel et al., 2014).

Given the exquisite dependence of the G1 checkpoint on p53 and p21, the oscillatory behaviour of p53 in response to DSBs provides an attractive model of ‘recurrent initiation’, in which waves of ATM activity, p53 and p21 come and go until triggers (DSBs) are eliminated (Batchelor et al., 2009). The simple take on this model is that activation of the DDR in G1 drives the cell into a quiescent-like state and, as time progresses, the DNA damage can be repaired in order for cells to re-enter G1. In the arrested state, either the DNA damage is repaired in time to drive the cell back to ‘early G1’, or the intensity and number of pulses commit the cell to an irreversible arrest. Reality, however, appears to be more complex.

First, late signalling events uncouple G1 checkpoint maintenance from ATM activity. This is illustrated by the fact that in the absence of Wip1, ATM activity, gauged by its autophosphorylation, no longer follows a pulsatile pattern, yet p53 oscillations are maintained and G1 checkpoint recovery is unaffected (Batchelor et al., 2008; Shaltiel et al., 2014). This suggests that recurrent initiation at the level of the DSB-sensing kinase ATM is not required for the reversibility of the G1 checkpoint. Strikingly, although ATM inhibitors can prevent the initiation of a G1 checkpoint arrest, ATM inhibition is insufficient to reverse the checkpoint arrest (Shaltiel et al., 2014). Instead, G1 checkpoint maintenance relies strongly on the ATM downstream targets Chk2 and p38 MAPK (Deckbar et al., 2010; Lafarga et al., 2009; Raman et al., 2007; Shaltiel et al., 2014). Both kinases can stabilise p53, and p38 further contributes to a G1 arrest by stabilising p21-encoding mRNA and p27 (also known as CDKN1B) protein (Cuadrado et al., 2009; Lafarga et al., 2009). It is currently unknown how p38 is controlled after its initial ATM-dependent activation, although the p53 target Gadd45α has been implicated in maintaining its activity during a persistent arrest (Reinhardt et al., 2010). ATM thus triggers the arrest in response to DSBs in G1, but Chk2 and p38 take over to sustain the arrest independently from ATM. Chk2 and p38 must therefore be silenced to promote a cell cycle restart. Despite the ability of Wip1 to remove the activating phosphorylations on both Chk2 and p38, spontaneous recovery from a G1 checkpoint arrest excludes a major role for Wip1 in terminating Chk2 and p38 signalling (Fujimoto et al., 2006; Shaltiel et al., 2014; Takekawa et al., 2000).

Second, p53 levels do not correlate directly with its transcriptional activity or with the subset of target genes it activates. Target genes are individually regulated by damage signals and other pathways on the levels of DNA, RNA and protein, thereby allowing control beyond p53 dynamics. This is clearly illustrated by p21, whose transcription is quickly terminated by binding of the co-repressor KAP1 to its promoter, allowing p21 mRNA levels to closely follow the activity of p53 (Gomes and Espinosa, 2010; Lee et al., 2007; Purvis et al., 2012). KAP1 itself is inhibited by ATM- and Chk1- or Chk2-mediated phosphorylation, which is counteracted by the PP4 phosphatase (Lee et al., 2012; Li et al., 2007; White et al., 2012). Tampering with the dynamic repression of p21 transcription by depleting either KAP1 itself or PP4 compromises G1 checkpoint reversal, making KAP1 and PP4 essential for G1 checkpoint recovery (Shaltiel et al., 2014). At the mRNA level, p21 transcripts must be stabilised by the RNA-binding protein HuR, which requires HuR to be phosphorylated by Chk2 and/or p38 (Abdelmohsen et al., 2007; Lafarga et al., 2009). Thus, whereas p53 remains a prerequisite for p21 induction in response to DNA damage, several pathways converge on its targets to define the reversibility of the arrest.

Although the regulation of recovery competence is quite well understood, little is known about factors that determine whether a cell will remain recovery competent or exit the cell cycle upon DNA damage. As many of the processes regulating recovery competence are affected by p53, cell fate decisions will most likely be made at the level of p53. Indeed, a permanent cell cycle exit in G2 requires p53 (Bunz et al., 1998; Johmura et al., 2014; Krenning et al., 2014), and cells depleted of p53 are unable to maintain a stable G2 arrest and eventually will progress through mitosis in the presence of DNA damage (Bunz et al., 1998).

Recovery competence is much more stringently regulated in G2 compared to G1 phase. Whereas an 8-hour pulse of p53 activation readily induces senescence in G2 cells, it fails to do so in G1 (Krenning et al., 2014; R.H.M., unpublished results), indicating that the cell cycle position of a DNA-damaged cell influences recovery competence. In addition, the mode of p53 activation greatly impacts upon cell fate decisions (Purvis et al., 2012). One factor that affects p53 activation and cellular outcome is the amount of DNA damage, where increasing doses of DSBs result in a stronger activation of p53 and cause more cells to become senescent (Loewer et al., 2013; Purvis et al., 2012). Furthermore, sustained DSBs cause the continuous oscillation of p53, resulting in an increased induction of p21 and a greater propensity to enter senescence (Lahav et al., 2004; Loewer et al., 2013; Loewer et al., 2010). However, recent data show that cell fate decisions in DNA-damaged G2 cells are established within 5 hours, which would correspond to the first wave of p53 (Krenning et al., 2014). This suggests that the strength of the first p53 oscillation, rather than number of oscillations, establishes cellular fate in G2. These observations, albeit contradictory, might very well operate within the same cell. Although DNA-damaged G1 cells retain recovery competence longer than their G2 counterparts, they eventually become incapable of G1 recovery (R.H.M., unpublished results). It is likely that G1 recovery competence is dictated by the number of p53 oscillations, whereas G2 recovery competence is determined by the strength of the first pulse. Another explanation for these contradictory observations might be found in the cell types used for investigation. Much of the knowledge obtained about p53 oscillations comes from transformed cell lines, in which the mechanisms for cell cycle exit are often repressed (Bulavin et al., 2002; Kleiblova et al., 2013; Müllers et al., 2014). Clearly, p53 activity is a major determinant in cellular fate decisions; however, how regulation upstream or downstream of p53 affects cellular fate remains largely unresolved.

In addition, the location of DNA damage might greatly influence the outcome of a DNA damaging insult. Up to half of the persistent DNA damage foci in cells undergoing senescence in response to ionising radiation are associated with telomeres (Hewitt et al., 2012). This telomeric DNA damage is refractory to repair and as a result causes persistent DNA damage signalling (Fumagalli et al., 2012; Noda et al., 2012). Indeed, p53-oscillations might be different in cells with similar amount of DNA damage foci (Loewer et al., 2013), indicating that the location of DNA damage, or its type or repair state, might influence p53 dynamics. As not all persistent foci in irradiation-induced senescence are found at telomeres, this raises the question as to whether there are other genomic loci that are refractory to DNA repair, but instead will activate persistent DNA damage signalling and the onset of senescence.

DNA damage checkpoint arrests are useful as long as they can be reversed when appropriate. The notion that recovery from a DNA-damage-induced arrest is merely a consequence of DNA repair is out-dated: checkpoint recovery involves the timely and spatially controlled action of numerous enzymes that disassemble DNA damage foci as the DNA is repaired and an extensive interplay between the cell cycle machinery and DDR that we are only starting to comprehend. In addition, little is known about the determinants of cell fate following a DNA damaging insult. As the first hints of processes underlying cell fate decisions, such as the mode of p53 activation or DNA damage severity (Krenning et al., 2014; Purvis et al., 2012), are being reported, many questions are left unanswered. For instance, it remains to be determined whether the location of a DSB, or the repair pathway used for its repair affect cellular fate.

Understanding these processes is vital to our understanding of normal cell biology, tumorigenesis and cancer resistance to genotoxic therapy. The accumulation of mutations during tumorigenesis likely requires several events of checkpoint recovery. Consistent with this is the finding that positive regulators of checkpoint recovery in G2 are commonly overexpressed in cancer and associated with poor prognosis (Bulavin et al., 2002; Cholewa et al., 2013; Li et al., 2002; Wang et al., 2008; Wierstra, 2013). Conversely, tumorigenesis is impaired in their absence (Bulavin et al., 2004; Wang et al., 2009), and several regulators of G2 checkpoint recovery are under (pre)clinical evaluation (Gilmartin et al., 2014; Gumireddy et al., 2005; Stadler et al., 2014). Understanding the mechanisms underlying cellular fate will allow us to devise new therapeutic strategies to sensitise cancer cells to DNA-damaging agents currently used in the clinic.