DNA damage and the balance between survival and death in cancer biology

The constant deleterious modification of DNA by reactive molecules, endogenously or exogenously generated, is offset by protective processes that are initiated by the DNA damage response. The interplay of the diverse signalling cascades (DNA damage response) that originate from the interference of DNA lesions with replication and the transcriptome leads to the activation of DNA repair, autophagy, senescence, apoptosis and necroptosis. Aspects of how post-translational modifications of the tumour suppressor p53 determine the switch between these end points are discussed. The crosstalk between autophagy, senescence, apoptosis and regulated necrosis is also discussed, focusing on the importance of thresholds for deciding cell fate. Throughout this Review, emphasis is placed on how DNA damage and DNA repair fit within the complex cellular context. The constant deleterious modification of DNA by reactive molecules, endogenously or exogenously generated, is offset by protective processes that are initiated by the DNA damage response. The interplay of the diverse signalling cascades (DNA damage response) that originate from the interference of DNA lesions with replication and the transcriptome leads to the activation of DNA repair, autophagy, senescence, apoptosis and necroptosis. Aspects of how post-translational modifications of the tumour suppressor p53 determine the switch between these end points are discussed. The crosstalk between autophagy, senescence, apoptosis and regulated necrosis is also discussed, focusing on the importance of thresholds for deciding cell fate. Throughout this Review, emphasis is placed on how DNA damage and DNA repair fit within the complex cellular context. Understanding how DNA damage determines cell fate — DNA repair and cell survival or death — is important for gaining insight into carcinogenesis and in promoting successful cancer therapy. This Review describes key decision-making nodes in the complex interplay between DNA damage responses and cell fate signalling. DNA is vulnerable to damage resulting from endogenous metabolites, environmental and dietary carcinogens, some anti-inflammatory drugs, and genotoxic cancer therapeutics. Cells respond to DNA damage by activating complex signalling networks that decide cell fate, promoting not only DNA repair and survival but also cell death. The decision between cell survival and death following DNA damage rests on factors that are involved in DNA damage recognition, and DNA repair and damage tolerance, as well as on factors involved in the activation of apoptosis, necrosis, autophagy and senescence. The pathways that dictate cell fate are entwined and have key roles in cancer initiation and progression. Furthermore, they determine the outcome of cancer therapy with genotoxic drugs. Understanding the molecular basis of these pathways is important not only for gaining insight into carcinogenesis, but also in promoting successful cancer therapy. In this Review, we describe key decision-making nodes in the complex interplay between cell survival and death following DNA damage.

DNA adducts themselves are able to activate cell death pathways or whether DSBs are the principal apoptotic trigger. Evidence suggests that DSBs are strong activators of apoptosis 19 , and most DNA base damage (for example, small adducts induced by reactive oxygen species (ROS) and reactive nitrogen species (RNS) and simple alkylating agents) induces necrosis 20 . The induction of necrosis is a consequence of poly(ADP-ribose) polymerase 1 (PARP1) activation during the repair of, for example, N-methylpurines and 8-oxo-guanine by BER. PARP1 activation depletes the cellular NAD + and ATP pools, which triggers necrotic death (also known as parthanatos) 21 . Incidentally, part hanatos, like apoptosis, is thought to be a protective mechanism that counter acts carcinogenesis if the need for the repair of small DNA adducts exceeds the repair capacity of the cell 22 . It should be noted that DNA repair by non-homologous end joining (NHEJ), homologous recombination (HR), MMR, BER, nucleotide excision repair (NER) and protein-linked DNA break (PDB) repair in combination with DDR signalling and damage tolerance mechanisms [23][24][25] are key cell survival pathways (FIG. 1b). However, it is not solely the lack of repair that initiates death programmes, but is rather a series of molecular interactions intertwined with the DDR that switches off repair and stimulates cell death.
In order for the cell to respond to critical DNA damage, the damage must first be detected, typically once it has manifested as a DSB or a replication fork-stalling event. Three interlinked sensor systems have been described that have the ability to detect a single DSB within minutes of its formation 26 . These immediate-early sensors in the DDR are the PI3K-related kinases (PIKKs): ataxia telangiec tasia mutated (ATM), ataxia telangiec tasia and Rad3-related (ATR) and DNA-dependent protein kinase (DNA-PK). From these sensors, through a series of kinase reactions the DDR cascade becomes activated, targeting many dual-function proteins that are key nodes in promoting Protein-DNA crosslinks are repaired by the combined action of protein-linked DNA break (PDB) repair, NER, non-homologous end joining (NHEJ) and HR, and DNA double-strand breaks (DSBs) are repaired by NHEJ or HR. Unrepaired DNA lesions can be neutral, mutagenic or toxic. b | Many sources give rise to DNA damage, which can cause mutation and DNA recombination, and can prevent DNA replication or block RNA transcription. Depending on the type of DNA lesion and the cellular background DNA damage is either tolerated or lethal. 6-4PP, 6-4 pyrimidine photoproduct; 8-oxo-G, 8-oxo-guanine; CPD, cyclobutane pyrimidine dimer; IR, ionizing radiation; Me, methyl; SSB, single-strand break; TDP, tyrosyl-DNA phosphodiesterase.

Necroptosis
A form of regulated necrosis that the cell uses when apoptosis cannot be initiated. During necroptosis the content of the cell is released into the extracellular space and causes inflammation.

Ferroptosis
A regulated, non-apoptotic form of cell death that is dependent on iron.
survival or cell death. These proteins signal downstream checkpoint activation to prevent cell cycle progression and the recruitment of DNA repair proteins to facilitate DSB repair via the stimulation of NHEJ or HR (for example, BRCA1 and Fanconi anaemia complementation group D2 (FANCD2)) 12 , depending on the cell cycle phase. In this capacity of promoting repair and cell cycle arrest, ATM and ATR are unquestionably pro-survival, as corresponding knockdown cell lines are clearly more sensitive to DSBs following DNA damage 27 . Therefore, one would predict a low penetrance of dominant-negative ATM and ATR mutations in cancer. However, contrary to expectation, ATM is frequently mutated in tumours 28 , which is accompanied by a gain of therapeutic resistance 29 . As ATM and ATR promote cell death in instances of high DSB levels 30 , either through the activation of the p53-dependent 31 and caspase 2-dependent 32 apoptosis pathways (FIG. 2) or through the activation of E2F1, p73 and CHK1 (reviewed in REF. 33) in a cell type-specific manner 34 , loss of ATM will render tumours resistant to these cell death pathways.
Patients with ataxia telangiectasia, which is caused by inactivating mutations in ATM, have a particularly increased risk of breast cancer 35 and a moderately increased risk of colon cancer 36 , and patients with functionally compromised ATR also show an increased risk of malignancy 37 . Although ATM seems to be more frequently altered than ATR 38 in tumours, it is clear that both ATM and ATR are necessary for maintaining genome stability either by stimulating DNA repair or by signalling death 39 in compromised cells. However, as the tumours in these patients still contain either mutated ATM or mutated ATR, it suggests that the ability to survive extends beyond the capacity of these kinases to recognize and signal the downstream repair of DSBs or the stabilization of stalled replication forks. One explanation is that there are many sub-pathways for circumventing DNA polymerase stalling and DSB repair, so it is unlikely that DSBs will remain unrepaired in ATM-or ATR-mutated cells 40 , but they may not be faithfully repaired, which may contribute to genomic instability. One hypothesis states that ATM-and ATR-mediated recombination repair operates at stalled replication forks and may compete via proliferating cell nuclear antigen (PCNA) ubiquitylation with translesion synthesis (TLS; which can be error-prone) 41 to prevent fork collapse, thereby promoting cell survival through lesion bypass. ATM and ATR can switch the balance to cell death by favouring the activation of pro-death functions (discussed further below) such as Ser46 phosphorylation of p53. Therefore, the regulation of survival and death functions is not strictly separable but embedded in complex networks with regulatory feedback and feedforward loops.

Survival and death strategies
Survival strategies include the repair of pre-toxic lesions that rests on the constitutive and induced activities of several repair pathways (FIG. 1a), bypass of pre-toxic lesions by TLS, autophagy, senescence, inhibition of apoptosis by anti-apoptotic factors such as survivin, inhibitors of apoptosis (IAPs) and X-linked inhibitor of apoptosis proteins (XIAPs) (FIG. 2), and activation of transcription factors such as nuclear factor-κB (NF-κB) and AKT (also known as PKB) regulating the expression of pro-survival genes (BOX 1). Cellular death strategies include apoptosis (FIG. 2), regulated necrosis (parthanatos, necroptosis and ferroptosis) 42 , checkpoint adaptation following DNA damage 43 and mitotic death 44 . These pro-death and pro-survival pathways, and how they interact, are discussed in more detail below. p53 in the balance between survival and death following DNA damage. Different levels of damage may trigger different responses; this paradigm states that low levels of DNA damage trigger repair and survival mech anisms and high levels trigger cell death. It is reasonable to hypothesize that p53 is central in this situation given the roles of p53 target genes in survival and death 39 (FIG. 2). How can the cell quantify DNA damage, and how are the decisions made? Theoretical models predict different activation levels of p53 that are dependent on different DNA damage levels, which are controlled by the conflicting actions of positive (ATM/ATR-CHK1/CHK2-p53) and negative (MDM2) feedback loops [45][46][47] . Simply, low DNA damage levels transiently activate p53 and high DNA damage levels lead to sustained p53 activation. The different stabilization levels of p53 may contribute to the differential expression of pro-survival and pro-apoptosis genes, as the affinity of p53 for its binding sites in promoters is high for genes that are associated with cell cycle arrest (for example, CDKN1A, which encodes p21) and is low for genes that are associated with apoptosis (for example, p53-induced gene 3 (PIG3; also known as TP53I3)) 48 . Promoter selectivity by p53 is also dependent on RNA and chromatininteracting partners. Thus, by using an inducible knockout mouse model it was shown that the expression of Cdkn1a in normal cells is dependent on the RNA helicase p68 (also known as DDX5) 49 , whereas in lung and breast cancer cells the expression of PIG3 and p53 regulated apoptosis inducing protein 1 (p53AIP1) relies on cellu lar apoptosis susceptibility protein (CAS; also known as EXP2 and CSE1L) 50 . Furthermore, the direct interaction between p53 and apoptosis-stimulating of p53 protein 1 (ASPP1) and ASPP2 (REF. 51) triggers apoptosis, but its inter action with inhibitor of ASPP protein (iASPP) 52 prevents apoptosis. Crosstalk between pro-survival path ways, for example, NF-κB (BOX 1) and p53, also influences the role of p53 as a pro-survival or a pro-death factor. Clearly, the regulation of p53 and the resulting downstream effects are complex and are very likely to be cell-type and cancer-type specific.
In addition to the above-mentioned mechanisms, the change in function of p53 from 'arrestor' and 'repairer' to 'killer' is dependent upon various post-translational amino-terminal phosphorylations and carboxy-terminal acetylations of p53 (REF. 53) (FIG. 3). Phosphorylation at Ser15, Ser37, Thr18 and Ser20 by ATM, ATR, CHK1, CHK2 and DNA-PK, among others, uncouples p53 from its inhibitory MDM2 binding partner and aids in the nuclear sequestration and transcriptional activation of p53 that is necessary for its arrestor functions,

PML bodies
Heterogeneous nuclear structures that range in size from 0.2 to 1 micrometre. They are found in most mammalian cell nuclei and have been ascribed functions in tumour suppression and transcription regulation, among others.
for example, the induction of CDKN1A 54 . The phosphorylation of Ser46 is specifically linked to the killer function of p53 (REF. 55), whereby it is postulated to alter promoter selection and favour the induction of apoptotic genes such as NOXA (also known as PMAIP1) 56 , PTEN 57 and TP53AIP1 (REF. 58) causing apoptosis in response to chemical agents and physical stress [56][57][58] . There are a number of candidate kinases that have been shown to interact with p53 and to phosphorylate Ser46, such as homeodomain-interacting protein kinase 2 (HIPK2) 59 , p38 (REF. 60), protein kinase Cδ (PKCδ) 61 , p53-dependent damage-inducible nuclear protein 1 (p53DINP1; also known as TP53INP1) 62 and dual specificity tyrosine-phosphorylation-regulated kinase 2 (DYRK2) 63 . ATM and ATR are thought to signal upstream of Ser46 phosphorylation (possibly via XIAP-associated factor 1 (XAF1) 64 ) where they phosphorylate Ser19 of seven in absentia homologue 1 (SIAH1), an E3 ubiquitin ligase. This modification disrupts the HIPK2-SIAH1 complex to stabilize HIPK2 (REF. 65) and allows its associ ation with promyelocytic leukaemia protein (PML; a potent pro-apoptotic tumour suppressor), p300, CREB-binding protein (CBP; also known as CREBBP) and p53 in nuclear PML bodies, with the net effect being the upregulation of pro-apoptotic p53 targets 66 .
Incidentally, the dephosphorylation of p53 (at Ser46 and other sites) is thought to be dependent upon the p53-induced phosphatase WIP1 (also known as PPM1D) 67 . Therefore, opposing feedback loops of p53 (Ser46) phosphorylation and dephosphorylation may exist, and an imbalance favouring HIPK2 stabilization, perhaps initiated by high and/or persistent levels of activated ATM and/or ATR (in situations of high levels of damage) may switch the cell to apoptosis over repair. Nonetheless, it is clear that the regulation of p53 is highly complex and requires further investigation to understand how the decision is made between survival and death following DNA damage.
In cancer biology, the apoptotic switch may be kept inactive owing to the increased SIAH1 E3 ubiquitin ligase activity that forms part of the hypoxic response, which is coordinated by hypoxia-inducible factors (HIFs) 68 . Interestingly, there may be a link between mutation of isocitrate dehydrogenase 1 (IDH1) and increased HIF1 Figure 2 | DNA damage-dependent apoptosis. DNA damage is able to activate both the extrinsic (death receptor) and the intrinsic (mitochondria-dependent) apoptosis pathways. Pro-apoptosis components of the DNA damage response are JUN N-terminal kinase (JNK), p53, caspase 2 and possibly MYC. They activate the pro-apoptosis proteins Fas ligand (FASL), Fas receptor (FASR), BCL-2-interacting mediator of cell death (BIM; also known as BCL2L11), BAX, p53 upregulated modulator of apoptosis (PUMA; also known as BBC3), NOXA (also known as PMAIP1) and BH3 death domain interacting protein (BID) and downregulate the anti-apoptosis protein BCL-2. This causes the activation of protease activity of caspases and the DNase activity of caspase-activated DNases (CADs), cleaving the DNA. DNA damage may also inhibit the transcription of MAPK phosphatase 1 (MKP1), which leads to increased JNK phosphorylation and activating protein 1 (AP-1) activity that drives FASL and the extrinsic apoptosis pathway. APAF1, apoptotic peptidase activating factor 1; ATM, ataxia telangiectasia mutated; ATR, ataxia telangiectasia and Rad3-related; cFLIP, cellular FLICE-like inhibitory protein (also known as CFLAR); DISC, death-inducing signalling complex; FADD, FAS-associated death domain; ICAD, inhibitor of caspase-activated DNase; tBID, truncated BID; XIAP, X-linked inhibitor of apoptosis. Apoptosome activity in gliomagenesis 69 . It is enticing to postulate the involvement of SIAH1-mediated HIPK2 degradation, which results from increased HIF1 activity, as a resistance mechanism to chemotherapy-induced cell death of solid hypoxic tumours. Perhaps the effectors discussed here are also involved in cancer pathogenesis, as observed for WIP1 overexpression 70 . Of course, the transcription of pro-apoptotic p53 target genes will need to overcome the threshold set by the anti-apoptotic proteins BCL-2 (REF. 71), survivin and XIAP 72 in order to activate the cell death machinery (FIG. 2). Tumour cells with high levels of these anti-apoptosis proteins have a lower or an abrogated apoptotic response following chemo therapy 71,73 . Therefore, adjuvant cancer therapy that targets the SIAH E3 ubiquitin ligases in tumour cells may be a rewarding approach 74 .

Transcriptome damage and cell death.
It is important to consider the stability, the inducibility and the regulation of the transcriptome in triggering cell death following DNA

Box 1 | Activation of transcription factors regulating survival functions
There is a plethora of transcription factors involved in regulating survival. We discuss two of these factors below, nuclear factor-κB (NF-κB) and AKT.

Pro-survival role of NF-κB
An increased body of evidence supports the model that NF-κB can be activated by DNA double-strand breaks (DSBs) 186 . For the activation of NF-κB via the ataxia telangiectasia mutated (ATM)/ataxia telangiectasia and Rad3-related (ATR)-CHK1/ CHK2-p53 axis, it has been proposed that DSBs cause the nuclear localization of NF-κB essential modulator (NEMO) 187 via p53-induced death domain-containing protein 1 (PIDD1) in complex with receptor-interacting serine/threonine-protein kinase 1 (RIPK1; also known as RIP1) that promotes sumoylation of NEMO 188 . Thus, RIPK1-mediated NEMO sumoylation is essential for DNA damage-dependent activation of NF-κB 189 . DSBs also activate ATM, which complexes with and phosphorylates NEMO 187 . The ATM-NEMO complex is shuttled to the cytoplasm 190 where it binds to the inhibitor of NF-κB kinase (IKK) complex, which subsequently phosphorylates IκB, thereby leading to the transcriptional activation of NF-κB pro-survival target genes. Among others, NF-κB transcribes BCL2L1 (which encodes BCL-Xl), BCL2A1 (also known as BFL1) 191 , inhibitor of apoptosis protein 2 (IAP2; also known as BIRC2) 192 , TNF receptor-associated factor 1 (TRAF1) and TRAF2 (REF. 193) (see the figure). Transcriptional activation of NF-κB, therefore, suppresses the activation of both mitochondrial-and death receptor (TNFR)-mediated apoptosis. In addition, NF-κB also upregulates MAPK phosphatase 1 (MKP1) 194 to counteract JUN N-terminal kinase (JNK) phosphorylation and JNK-driven apoptosis, observed after sustained DNA damage induction (FIG. 2). Consequently, preventing the activation of NF-κB during treatment with DNA-damaging agents such as topoisomerase I inhibitors results in increased apoptosis and reduced tumour growth in vivo 195 .

Pro-survival role of AKT
The AKT kinase suppresses apoptosis in a PI3K-dependent manner 196 by phosphorylating, and therefore inhibiting, pro-apoptotic proteins or by activating an anti-apoptosis system. Thus, AKT phosphorylates BCL-2-associated agonist of cell death (BAD) 150 , apoptosis signal-regulating kinase 1 (ASK1; also known as MAP3K5) 151 , human caspase 9 (REF. 152) and the E3 ubiquitin ligase MDM2 (REF. 153), thereby suppressing apoptosis (see the figure). AKT also suppresses apoptosis by promoting the degradation of IκB, which leads to the activation of NF-κB and to the suppression of apoptosis via the transcription of anti-apoptotic genes (see the figure).
Depending on the DNA-damaging agent, AKT can be activated in a manner dependent on DNA-dependent protein kinase (DNA-PK) 146 , ATM 147 , ATR 148 and/or MRE11 (REF. 149). Phosphorylated AKT (Ser472) colocalizes with phosphorylated ATM (Ser1981) at DNA damage-induced γH2AX foci 149 , and this colocalization is facilitated by the action of the E3 ubiquitin ligase RNF168. AKT also influences the stabilization of p53 following activation of the DNA damage response (DDR) as AKT contributes to p53 degradation, probably through its action with MDM2 (REF. 146). Suppression of apoptosis by AKT leads to increased survival on DNA damage, which has been shown in numerous cancers 197 .
In addition to the role of AKT described above, AKT was also shown to directly stimulate DNA repair. Thus, following DNA damage, AKT binds to DNA-PK, phosphorylates it and regulates the accumulation of DNA-PK at damaged sites to improve the efficiency of DSB re-joining by non-homologous end joining (NHEJ) 198 . AKT, therefore, regulates both DSB repair and survival and, conversely, the p53-dependent and p53-independent responses linked to cell death (see the figure). damage and not just DNA damage. On the one hand, the inhibition of RNA polymerase II during transcription is a known apoptotic trigger 75 , but, on the other hand, epi genetic deregulation during cancer formation of pro-survival and anti-survival factors also affects cell death 76 .
First, considering the inhibition of RNA polymerase II, the model states that, in general, pro-survival factors are transcribed from large genes (many kilobases in size), whereas pro-death factors are transcribed from small genes (few kilobases) 77 . Consequently, DNA adducts that inhibit transcription are considered to be more likely to inhibit the transcription of pro-survival genes, as larger genes are greater targets for damage than smaller genes. In this case, unrepaired transcription-blocking adducts prevent the transcription of pro-survival factors, shifting the balance to cell death.
This mechanism can tilt the balance between survival and death, whereby transcriptional inhibition triggers phosphorylation and the accumulation of p53 either through ATM-and/or ATR-directed signalling 78 or indirectly through the inhibition of MDM2 transcription 79 (FIG. 3). The same transcription-blocking DNA damage could also prevent the transcription of pro-survival genes that are regulated by p53 such as damage-specific DNA binding protein 2 (DDB2) and xeroderma pigmentosum complementation group C (XPC) 80-82 and thereby affect the survival-death balance.
Bulky DNA adducts are particularly effective transcription inhibitors. Such adducts are repaired by transcription-coupled NER (TC-NER) 83 following RNA polymerase stalling. When TC-NER fails to remove bulky adducts, an accumulation of phospho-p53 (Ser15) is observed 84 . Whether the same is seen for phospho-p53 (Ser46) is unknown. The possibility that DNA damage-dependent transcriptional inhibition activates p53-triggered apoptosis is particularly important in cancer biology, as one would expect higher transcription rates given the increased proliferation (and metabolic) rates of cancer cells.
Another example of the interplay between transcription and DNA damage was provided by studies with cells deficient in the transcription factor FOS. Following UV radiation and treatment with agents that induce bulky DNA lesions, FOS is promptly induced in an immediate-early 85 and late 86 response. A hallmark of FOS-deficient cells is their hypersensitivity to genotoxins that cause the formation of bulky DNA lesions, such as UV radiation and benzo(a)pyrene 87,88 , which is related to severe replication and transcription blockade 89 . The repair of DNA adducts is slow in FOS-deficient cells because they fail to express the NER proteins XPF and XPG, which are regulated via FOS-activating protein 1 (AP-1) 89 . The immediate-early induction of FOS causes the stimulation of XPF and XPG transcription, which is important because it maintains the NER capacity of the cell. This early response is known as maintenance regulation and compensates for a shortage of these unstable NER gene transcripts 90 . XPF and XPG are endonucleases that cleave the DNA 5ʹ and 3ʹ from the damage 91 , and it is reasonable to suppose that cells express ataxia telangiectasia mutated (ATM) and ataxia telangiectasia and Rad3-related (ATR) phosphorylate p53, thereby leading to the expression of pro-survival genes such as cyclin-dependent kinase inhibitor 1A (CDKN1A), DNA binding protein 2 (DDB2) and the E3 ubiquitin ligase MDM2. This is very likely an early response that is required for checkpoint activation and DNA repair. Pro-death signalling: ATM and ATR cause the stabilization of homeodomain-interacting protein kinase 2 (HIPK2) that stimulates apoptosis by phosphorylating Ser46 of p53, which results in its targeting to pro-apoptosis genes. HIPK2 also causes the degradation of carboxy-terminal binding protein (CtBP) that leads to the activation of pro-apoptotic proteins NOXA and BAX. This may be a late response when DNA repair has failed. SIAH1, seven in absentia homologue 1.

Nature Reviews | Cancer
Leads to the dephosphorylation of AKT them at a low basal level in order to avoid undesirable genomic changes. A fine-tuned (homeostatic) upregulation of these repair genes, depending on the genotoxic stress level, is obviously important in order to maintain genomic stability. Notably, priming the system with a low dose of a DNA-damaging agent prevents cell death that is triggered by a subsequent challenge dose 90 . This indicates a survival adaptation resulting from late FOS activation 86 . Therefore, the balance between survival and death might be different following acute and chronic genotoxic exposure. The induction of FOS following DNA damage is part of a general stress response that involves the activation of the JUN N-terminal kinase (JNK) and p38 kinase, which results in an increase in AP-1 activity. NER-defective mutants display sustained JNK and p38 kinase activation, indicating that DNA damage is responsible for the response 92 . The resulting long-term activation of AP-1 results in an increased expression of Fas ligand (FASL; also known as FASLG) 93 that drives the extrinsic apoptosis pathway (FIG. 2).
Collectively, unrepaired toxic DNA lesions, which are usually substrates of NER, elicit a sustained upregulation of JNK and p38 kinases 92 , a pro-death switch, by causing the downregulation of MAPK phosphatase 1 (MKP1; also known as DUSP1) -probably as a result of transcription blockade -which deactivates JNK and p38 (REF. 94) (FIG. 2). This in turn leads to late FASL upregulation that is controlled by JUN-ATF2 (REF. 95). In this case, the balance between survival and death following DNA damage rests on the proper resynthesis of the repair proteins XPF and XPG, the level of MKP1 downregulation, the duration of JNK and p38 kinase activation, the level of FASL induction and the availability of the FAS (also known as CD95 and APO1) death receptor system, which is often silenced in tumours, such as in mela nomas 96 . In this context, for DNA damage that induces the upregulation of FOS and XPF-XPG-mediated NER repair, the levels of FOS in cancer cells may have a deciding role. It is conceivable that the manipulation of this system bears a potential for cancer therapy.
Changes in the transcriptome that are caused by epigenetic alterations should also be taken into account when considering the response of cells to DNA-damaging agents as epigenetics has a role in regulating DNA repair and the activation and execution of the death pathways. For example, in tumours, notably gliomas, the DNA repair gene O 6 -methylguanine-DNA methyltransferase (MGMT) is frequently silenced by the hypermethylation of CpG islands in its promoter region 97 . MGMT removes alkyl adducts from the O 6 -position of guanine, thereby preventing the cascade of events leading to DDR activation and cell death. Therefore, gliomas in which MGMT is silenced respond to alkylating chemotherapeutics 98 . Silencing of the factors that are required for apoptosis has also been reported in other cancers. As discussed above, the sustained stabilization of p53 caused by unrepaired DNA damage triggers apoptosis. A critical target gene of p53 is hypermethylated in cancer 1 (HIC1), which encodes a transcriptional repressor of sirtuin 1 (SIRT1). SIRT1 deacetylates p53, thereby negatively regulating its transcriptional activity 99 . Therefore, the often observed epigenetic silencing of HIC1 in cancer cells removes the transcriptional suppression of SIRT1 and leads to pre mature p53 inactivation, which counteracts p53-triggered apoptosis 100 . In addition, epigenetic silencing of the factors that are directly involved in apoptosis execution is also found in cancers. For example, melanomas contain silenced apoptotic peptidase activating factor 1 (APAF1) 101 and caspase 8 promoters 96 , rendering them resistant to genotoxic therapies.
DNA damage-triggered senescence. DNA damage is an inducer of cellular senescence 102 . Current knowledge of senescence originates from the initial observation that human diploid fibroblasts (HDFs) can only undergo a limited number of cell divisions (replicative senescence) and that HDFs exposed to high doses of DNA-damaging agents such as ionizing radiation or mitomycin C (MMC) stop proliferating while remaining metabolically active 103 . This phenomenon of DNA damage-induced senescence has been harnessed for generating cell feeder layers 103 .
There is evidence that the signal for DNA damageinduced senescence originates from DSBs in telomeres, which activate continuous DDR signalling 102 and thus activate a cell cycle arrest in a p53-dependent 104 or p16 (also known as INK4A)-dependent 105 manner (FIG. 4). A fundamental question is why DSBs in genomic DNA only transiently activate the DDR but DSBs in telomeric DNA activate continuous DDR signalling. The answer lies in the observation that DSBs in or next to telomeres are not repaired. Telomeric repeat-binding factor 2 (TRF2), a component of shelterin that binds to double-stranded TTAGGG repeats in telomeres in vitro, localizes to all telomeres in vivo 106 and inhibits the

End-to-end fusion
The ends of linear DNA are protected by telomeres. When double-strand breaks (DSBs) occur in telomeres and telomeres are partially lost, DSB repair can fuse two broken telomeric ends. This can be seen in mitosis as two chromosomes connected at their ends.
completion of DSB repair by NHEJ by preventing the recruitment of ligase IV to DSBs that are located in or next to telomeric DNA 107 (FIG. 4). Therefore, owing to persistent DSBs in telomeres, the DDR is continually active, the cell cycle arrest is prolonged and the cell enters a quiescent state, but in genomic DNA DSBs are repaired and DDR activation and cell cycle arrest are transient. Thus, TRF2 protects human telomeres from end-to-end fusion 108 , which may result in malignant transformation 109 . This explains why high doses of DNA-damaging agents (for example, >70 Gy of ionizing radiation) are required for generating viable, nonproliferating HDFs, as the probability for DSBs forming in telomeres is a fraction of the probability of their formation in the rest of the genome. It is important to note that p53 is required for activating this pathway, which explains why HDFs do not die following high-dose treatment, whereas human transformed cells that harbour activated RAS or mutated p53 undergo apoptotic death rather than senescence 110 . Notably, already moderately toxic doses of chemical genotoxins are able to trigger senescence. Thus, temozolomide, a methylating agent that is used in glioma therapy, was shown to induce ATM and ATR activation and senescence in a sub-fraction of the cell population already at low clinical concentrations 111 . It is therefore unlikely that unrepaired DSBs in telomeres are the only mechanism by which senescence is induced following the exposure of cells to DNA-damaging agents. Thus, intriguing unanswered questions that still need to be addressed include: how do low-dose genotoxins induce senescence in a telomere-independent manner, and are these silenced cells able to re-enter a proliferative state? If cells re-enter proliferation, the implication would be far-reaching, as tumour cells would escape therapy by entering a non-proliferating state that is reversed after the therapy cycle has been completed.

Autophagy in the balance between survival and death
following DNA damage. The regulation of autophagy by DNA damage occurs on three levels. DNA damage prevents the suppression of autophagy by mTOR complex 1 (mTORC1) while activating the autophagy-related protein 1 homologue (ATG1; also known as ULK1) and the Beclin 1 complexes. This is accomplished via several pathways involving PARP1, JNK, ATM and p53 (FIG. 5).
In autophagy, p53 has multiple roles that are mediated through the transcriptional activation of PTEN 112 and the lysosomal protein DNA damage-regulated autophagy modulator protein 1 (DRAM1) 113 . The transcriptional activation of PTEN by p53 counteracts the suppression of autophagy by preventing the activation of AKT by receptor tyrosine kinases (RTKs). There are, therefore, opposing effects at play between the activation of AKT by growth factors 114 and the DNA damage-dependent inactivation of AKT by PTEN 115 . DRAM1 does not seem to have a role in autophagy initiation. The three splice variants (SV1 (REF. 113), SV4 and SV5 (REF. 116)) of DRAM1 localize to the lysosomes or endosomes (SV1) or to the peroxisomes and autophagosomes (SV4 and SV5). As none of these organelles has a role in the initiation of autophagy, DRAM1 expression by p53 probably prepares the cells for dealing with the increased load of autophagy that is induced by DNA damage.
PARP1 is a chromatin-associated enzyme that is involved in BER of small adducts such as those induced by alkylating agents and ROS 117 . It modifies several proteins by poly(ADP-ribosyl)ation and in the process consumes large amounts of NAD + . Treating PARP1 wild-type cells with doxorubicin, a topoisomerase II inhibitor that also induces oxidative stress 118 , causes cells to indirectly activate autophagy, by the depletion of ATP and NAD + . When PARP1 is knocked out, doxo rubicin no longer triggers a strong autophagy response 119 . Interestingly, PARP1 also regulates starvation-triggered autophagy. When PARP1 is inhibited or deleted, the onset of starvation-triggered autophagy is delayed 120 . Starvation increases the intracellular ROS levels 121 and the DNA damage caused by ROS leads to the depletion of energy by PARP1 activation. Therefore, the repair of DNA damage during which PARP1 becomes activated is a pre-requisite for the optimal induction of autophagy 120 . All DNAdamaging agents, such as UV-A, that cause the formation of ROS, trigger autophagy 122 . Therefore, it seems that DNA damage and autophagy are tightly connected through PARP1.
DDR signalling can directly activate autophagy. In this context, ATM activates AMPK, which subsequently leads to the activation of autophagy 123 . Autophagy has been observed following the induction of DNA damage that does not require PARP1 for its repair. An example is DSBs that arise from the processing of O 6 MeG 124 and that lead to the activation of ATR and ATM 125 . O 6 MeG-triggered autophagy is dependent on ATM 111 . The use of ionizing radiation has expanded our understanding of the role of ATM in autophagy. DSBs that are induced by ionizing radiation activate ATM, which in turn activates the tumour suppressor tuberin (TSC2) via the serine/threonine-protein kinase 11 (STK11) and AMPK metabolic pathway to repress the autophagy suppressor mTORC1 (REFS 126,127) and activate the ATG1 complex. Interestingly, chaperonemediated autophagy has been shown to have a role in the negative feedback regulation of the DDR as it causes the degradation of activated CHK1 (REF. 128).
The Beclin 1 complex can be inhibited by BCL-2 (REF. 129). BCL-2 is a downstream target of phosphorylation by JNK 130 , thus BCL-2 is no longer able to inhibit the Beclin 1 complex and pre-autophagosomal membrane structures can form. As discussed above, MKP1 inactivates JNK and, therefore, the downregulation of MKP1 resulting from transcription-blocking DNA lesions will lead to sustained JNK activation. Therefore, it can be hypothesized that all transcription-blocking DNA lesions (above a particular threshold) bear the potential to activate the Beclin 1 complex (FIG. 5).
At low DNA damage levels, autophagy is a survival mechanism. With high damage levels, however, autophagy becomes a process that is out of control, and the balance is shifted to death. In molecular terms, persistent DNA damage leads to the long-lasting activation of ATM that causes the repression of mTORC1 and the induction of autophagy 127 that is not switched off 131 . Under these conditions, excessive autophagy will cause the cell to cannibalize itself and die. Within this context, ATM can be activated by forkhead box protein O3 (FOXO3) 132 , which is normally bound to DNA. Upon DNA damage, it dissociates from the DNA, interacts with ATM and contributes to the sustained activation of the ATM/ATR-CHK1/CHK2-p53 axis. In addition, FOXO3 also regulates the transcription of LC3 (also known as MAP1LC3A) and BCL-2/adenovirus E1B 19 kDa protein-interacting protein 3 (BNIP3), both of which are required for autophagy 132 . As FOXO3 activates both the p53-dependent cell death and autophagy pathway, it may have a role at the interface between survival and death. Studies have shown that death by autophagy in malignant cells following treatment with arsenic trioxide (As 2 O 3 ) 133 . Arsenic trioxide affects the function of PML bodies 59 , which are important in the phosphorylation of p53 at Ser46 by HIPK2. Therefore, a link may exist, centring on PML bodies 134 , between autophagy cell death and phospho-p53 (Ser46). Although survival through autophagy may be a tumour-promoting mechanism 135 and a target for therapy 136 , autophagy-mediated cell death may be an important tumour suppressor 137 that also contributes to the outcome of cancer therapy.

Interplay between survival and death
The crosstalk between the major end points auto phagy, necrosis and apoptosis following genotoxic insult is complex and not fully understood. Some aspects have already been discussed above and so we focus here on examples that underline this complexity. On a basic level, DNA damage activates the cell cycle checkpoints, stops cell cycle progression and activates DNA repair via the ATM/ATR-CHK1/CHK2-p53 axis. Upon damage resolution, the ATM/ATR-CHK1/CHK2-p53 cascade is switched off by dephosphorylation 67 , and cell death by necrosis or apoptosis does not occur. The complexity arises from the observation that the DDR simultaneously activates pro-death mechanisms such as apoptosis 138 and necroptosis 21 and pro-survival mech anisms including DNA repair 90 and autophagy 111 , often via the same signalling proteins and pathways. Within this context, thresholds have an important role. Under optimal growth conditions, autophagy is suppressed owing to signalling from receptor tyrosine kinases (RTKs) through the PI3K-AKT pathway to mTOR complex 1 (mTORC1). mTORC1 comprises mTOR, mammalian lethal with SEC13 protein 8 (mLST8), PRAS40 (also known as AKT1S1) and regulatoryassociated protein of mTOR (RAPTOR). AKT suppresses autophagy by phosphorylating PRAS40 directly or indirectly through the phosphorylation of tuberin (TSC2) (at Ser939, Ser1086/Ser1088 and Thr1422). The ATG1 complex, comprising autophagy-related protein 1 homologue (ATG1; also known as ULK1) and ULK2, ATG13 and FIP200 (also known as RB1CC1), is a principal initiator of autophagy. mTORC1 inhibits the ATG1 complex by direct interaction of RAPTOR with ATG1-ULK2 and mTOR phosphorylation of ATG13 and ATG1-ULK2. These phosphorylations suppress ATG1-ULK2 kinase activity and, as ATG1-ULK2-AGT13-FIP200 kinase activity is required for autophagy, this prevents autophagy initiation. Contrary to the suppression of autophagy by AKT and mTORC1, autophagy can be activated by the energy-sensitive kinase AMP-dependent protein kinase (AMPK). As shown, AMPK is activated by high cellular levels of AMP and deactivated by high ATP levels. Once activated, AMPK phosphorylates RAPTOR and TSC2 (at Thr1227 and Ser1345), thereby removing the suppression of autophagy by mTORC1. AMPK furthermore phosphorylates ATG1, thereby stimulating the activity of the ATG1 complex and promoting autophagy. Activated ATG1-ULK2 phosphorylates both FIP200 and ATG13, which then lead to the translocation of the entire complex to the pre-autophagosomal membrane and to autophagy induction. The canonical initiation of the pre-autophagosomal membrane structures is mediated by Beclin 1 in complex with VPS34 (also known as PIK3C3), a class III PI3K. DNA damage leads to the activation of autophagy by inactivating mTORC1 and activating the ATG1 and the Beclin 1 complexes. ATM, ataxia telangiectasia mutated; ATR, ataxia telangiectasia and Rad3-related; DRAM, damage-regulated autophagy modulator protein; JNK, JUN N-terminal kinase; MAPK1, MAPK phosphatase 1; PARP1, poly(ADP-ribose) polymerase 1. Thresholds in cell death are established by downstream mechanisms that inhibit pro-death pathways such as apoptosis. Important death modulators are survivin, IAPs and XIAP. Survivin, together with IAP and XIAP, inhibits caspases 72,139 , thus preventing cell death by apoptosis. Survivin is highly expressed in replicating tissues during embryonic and fetal development 140 and is also found to be overexpressed in tumours 141 . Knockout of survivin sensitizes cells to DNA damage-triggered apoptosis as the threshold of activation is reduced 142 . During embryonic development, cells in tissues undergo continuous cycles of cell division. Replication stress that is caused by endogenous DNA damage 143 could potentially drive these cells into apoptosis. Survivin prevents this by suppressing the premature activation of apoptosis by inhibiting caspase activation. In tumours, the same mechanism leads to resistance to anticancer drugs 144 . Therefore, in order for apoptosis to occur, sufficient caspases must be activated to overcome the threshold preventing apop tosis that is established by the anti-apoptosis proteins survivin and XIAPs. Interestingly, as no threshold for the protective function of autophagy has been described, it is reasonable to posit that autophagy protects the cell from apoptosis at low DNA damage levels. In addition to overexpressing survivin, cancers have been shown to modulate the threshold for cell death activation by other means. For example, melanomas suppress the expression of the pro-apoptosis proteins APAF1 (REF. 101) and caspase 8 (REF. 96), and lymphomas express high levels of anti-apoptosis proteins such as BCL-2 (REF. 145). The creation of this threshold can be strengthened by active signalling. Thus, depending on the DNA-damaging agent, AKT can be activated in a manner dependent on DNA-PK 146 , ATM 147 , ATR 148 and/or MRE11 (REF. 149). AKT strengthens the threshold via its activating or inhibitory action on BCL-2-associated agonist of cell death (BAD) 150 , apoptosis signal-regulating kinase 1 (ASK1; also known as MAP3K5) 151 , caspase 9 (REF. 152) and MDM2 (REF. 153) (BOX 1). Upon persistent DNA damage and sustained DDR and apoptosis signalling, sufficient caspases become activated to overcome the inhibitory threshold, and these caspases cleave Beclin 1 (REF. 154), thus preventing further autophagy initiation and the protection gained from it. Importantly, the cleavage fragment of Beclin 1 interacts with mitochondria and further stimulates apoptosis. The threshold has thus been exceeded and the cell is able to enter the apoptotic pathway.
Upon DNA damage, if the cell is unable to activate the required amount of caspases to switch from autophagy to apoptosis, and the DNA damage is not resolved, the cell can die by regulated necrosis. It is important to note that most studies on regulated necrotic cell death have been carried out in systems in which apoptosis was inhibited, either chemically using small-molecule inhibitors 155 or by knockout 156 or knockdown 157 of death receptors and associ ated domains. Under these conditions, cells preferentially undergo regulated necrosis following DNA damage. This is in agreement with the biological role of regulated necrosis in nature. Cells infected with viruses that contain proteins that inhibit apoptosis trigger regulated necrosis as an alternative cell death mechanism 158 . In a similar manner, cancer cells that show apoptosis resistance are susceptible to the induction of necrosis 159 . Notably, apoptosis is an energy-dependent process whereas necrosis requires much less energy 160 . Excessive PARP1 activation that is caused by high amounts of base damage or DNA single-strand breaks will therefore switch the cell from apoptosis to regulated necrosis 161 . Consequently, two crucial factors for the initiation of regulated necrosis over apoptosis are, first, the effectiveness of the apoptosis machinery within a cell and, second, the available energy required for apoptosis initiation and completion.
In this scenario, two forms of regulated necrosis, namely, necroptosis and parthanatos, are briefly discussed. Necroptosis depends on the activation of receptor-interacting serine/threonine-protein kinase 3 (RIPK3) 162 by tumour necrosis factor receptor signalling via RIPK1 (REF. 163), cytosolic DNA via DNA-dependent activator of IFN-regulatory factors (DAI; also known as ZBP1 and DLM1) 164 and by lipopolysaccharides or double-stranded RNA via the Toll-like receptors and the action of TIR domain-containing adaptor mol ecule 1 (TICAM1) 165 . Caspase 8 and cellular FLICE-like inhibitory protein (cFLIP; also known as CFLAR) 166,167 in activate RIPK3; therefore, inhibitors of caspase 8, such as IAP1 and IAP2 (REF. 168), will stimulate necroptosis while inhibiting apoptosis. RIPK3 in conjunction with RIPK1, DAI or TICAM1 causes the trimerization and plasma membrane localization of mixed lineage kinase domainlike protein (MLKL) 169 and, consequently, the loss of the osmotic homeostasis of the cell.
The term parthanatos was coined to describe the form of necrosis that is triggered owing to the hyperactivation of PARP1 (REF. 170). PARP1 hyperactivation leads to the depletion of cellular NAD + (REF. 171) that causes a catastrophic failure in ATP production via oxidative phosphorylation 172 and glycolysis 173 . Parthanatos is the most common form of regulated necrosis that is triggered by DNA damage. Whether necroptosis is triggered by DNA damage is largely unknown. Parthanatos is induced by genotoxins such as ionizing radiation 174 , chemical ROS producers, and methylating and ethylating agents 21 . Massive amounts of DNA damage will lead to excessive PARP1 activation and consequently to the loss of cellular NAD + and ATP. Given the central role that ATP has in both apoptosis and necrosis, it is reasonable to state that the cellular energy status is an important factor involved in the decision of which pathway will be activated following DNA damage 175 . In support of this, autophagy and the energy sensor AMPK have a central role in preventing DNA damage-triggered parthanatos 176 . Following the induction of DNA damage, multiple energy-dependent processes (for example, BER) become activated that deplete cellular NAD + and ATP levels. The increase in AMP will activate AMPK to facilitate the restoration of ATP levels via autophagy and other mechanisms. When DNA damage overwhelms the capability of the cell to repair it, or when the energy-dependent processes that are activated by the DNA damage deplete ATP and NAD + Jasplakinolide A drug originally isolated from a sponge used for the stabilization and polymerization of the actin filament. It triggers apoptosis in a DNA damage-independent manner.
(PARP1-mediated repair) beyond the ability of AMPK to restore sufficient energy to sustain life, the cell will die. In this scenario, the activation of AMPK protects against DNA damage-triggered cell death by activating autophagy or stimulating apoptosis instead of parthanatos by providing sufficient ATP for apoptosis execution.
Shifting the balance from death to survival Apoptosis is characterized by several features (mitochondrial membrane dysfunction, cell shrinkage, membrane changes and activation of caspases) that are indicative of early and late apoptosis 177 . Can the process that triggers apoptosis be reverted if DNA damage signalling is interrupted? A point of no return was postulated, although this precise point has not been clearly defined 178 . This model of cell death interruption was verified by the demonstration that transient treatment with jasplakinolide induces mitochondrial damage and caspase activation, and that this could be reverted by interrupting the treatment. However, the process of apoptosis was no longer reversible when nuclear fragmentation occurred 179 , supporting the idea that there is a point of no return in the pathway shifting cells from survival to death. This point is probably determined by a sufficiently high level of executing caspases and caspase-activated DNases (CADs) needed to cleave the genomic DNA completely. This may be influenced by anti-caspases such as survivin or XIAPs, which are overexpressed in cancers 73 . Rescue from apoptosis (known as anastasis) might be a tumour-relevant process, as it has been shown that incomplete apoptotic DNA cleavage may result in chromosomal aberrations and genomic instability 179 . Whether the process of DNA damage-induced apoptosis can be abrogated by the attenuation of the damage-triggered signalling following damage repair is an intriguing open question.

Conclusions
DNA damage triggers a plethora of cellular responses. Thus, it activates cell cycle checkpoints that provide time for the cell to repair the damage before it interferes with the replication machinery. If repair fails or DNA repair is saturated, the remaining DNA damage impedes replication and transcription, and thereby the activated DDR signals downstream cell death pathways. The competence of a cell to survive DNA damage is therefore proportional to the amount of critical DNA damage, the repair capacity of the cell, the proliferation level, the status of p53 and key DDR proteins including ATM, ATR and DNA-PK, the effectiveness of activating DNA repair genes (which is dependent on epigenetic silencing and cellular transcription factors), and the execution of downstream cell death pathways. The DNA repair capacity differs greatly between cell types; for example, human embryonic stem cells repair most types of DNA lesion more effectively than differentiated cell types 180 , whereas monocytes and muscle cells are defective in BER 181,182 , and some cancers show upregulation of repair, for example, metastatic melanoma 183 , and highly variable MGMT repair activity, such as in gliomas 97,98 . The current paradigm states, in simple terms, that a low level of DNA damage activates DNA repair (hallmarked by the upregulation of repair genes such as XPF, XPG, DDB2, XPC, XRCC1 and others 90 ), whereas at high DNA damage levels DNA repair is saturated, and unrepaired DNA damage activates one of the death programmes, including apop tosis, regulated necrosis and extensive self-cannibalism by autophagy. It is not well understood how the cell switches between these pathways, but it seems that the phosphory lation status of p53 and anti-apoptosis thresholds are key nodes in determining whether a cell lives or dies. The functions of ATM and ATR extend into each possible outcome following DNA damage. For this reason, ATM and ATR seem to be the main decision makers, informing effectors such as p53 how to function. The increased resistance of tumours carrying mutations in ATM 29 demonstrates the importance of ATM in initiating cell death pathways. It is important to note that inactivation of p53 in cancer cells can lead to either drug sensitization or resistance, depending on the genotoxic agent used. Thus, bulky adducts induce a p53-triggered upregulation of DNA damage repair and tolerance genes (XPC, DDB2 and DNA polymerase η (POLH)), which is a dominant survival trait 81, 184 , but upon small adducts, such as O 6 MeG, upregulated apoptosis genes play a key role 138 . Although much work has focused on how different modifications of p53 dictate differential promoter activation of pro-survival and pro-death genes, it is not entirely clear how p53-deficient cells undergo cell death following DNA damage. Thus, the p53-independent pathways that are activated by DNA damage through MKP1, JNK, caspase 2 and possibly other factors need further exploration. The importance of DDR-triggered pro-survival factors (for example, DNA repair and damage tolerance), as well as thresholds that need to be overcome before cell death pathways can be initiated, have been recognized. Inhibitors that specifically target components in these pathways are in active development and in clinical trials (for examples see Supplementary Information S1 (Table)). Thus, inhibitors targeting the ATM/ATR-CHK1/CHK2 axis, cell cycle effectors and DNA repair show promise in combination with genotoxin-based chemotherapeutics, but might also have benefits per se in cancers that exhibit replication stress and extensive endogenous ROS production (for example, inhibitors of PARP1 and MTH1). By targeting anti-apoptosis proteins and pathways it is conceivable that the threshold for death is lowered for genotoxic-and biological-based therapies. In the future, it needs to be explored in more detail how specific DNA lesions activate and coordinate the complex interplay between survival and death. A deep understanding of the repair and processing of specific DNA lesions during replication and transcription and how they activate survival and death pathways is of fundamental importance for cancer therapy. The ultimate goal is to apply this knowledge to protect normal tissue during therapy with classic genotoxic anticancer drugs while simultaneously sensitizing cancer cells to die. The protection of normal tissue has far-reaching implications for stem cells, as they have been shown to activate DNA damage-triggered apop tosis easily 185 , and for the elimination of genomically compromised cells from the healthy population as a cancer prevention strategy.