The resurrection of the PIDDosome – emerging roles in the DNA-damage response and centrosome surveillance

The PIDDosome was described in 2004 by Antoine Tinel and the late Jürg Tschopp, who spearheaded research in the field of cell death and inflammation for many years. Their initial findings provided evidence for a long-sought function of a highly conserved member of the caspase family caspase-2 (CASP2) as a cell death effector in the DNA-damage response (Tinel and Tschopp, 2004). Yet, after a short period of excitement, interest in the PIDDosome as a regulator of CASP2 ceased. This was mostly because of the lack of phenotypes in mice depleted of individual PIDDosome components (Berube et al., 2005; Kim et al., 2009; Manzl et al., 2009; O'Reilly et al., 2002); further, CASP2 could still become activated in the absence of p53-induced death domain protein 1 (PIDD1), e.g. in cell extracts in vitro, but also in dying neurons (Manzl et al., 2009; Ribe et al., 2012). As the biology of CASP2 has been extensively reviewed elsewhere (Bouchier-Hayes and Green, 2012; Fava et al., 2012; Puccini et al., 2013), it will not be the focus of this review. Here, we aim to give an overview on the biology of PIDD1. We discuss potential roles in DNA damage, inflammation and normal cellular physiology and aim to reconcile some of the remaining controversy that surrounds the PIDDosome, first described as a mediator of cell death in the DNA-damage response, containing PIDD1, CASP2- and RIPK1-domain-containing adaptor with death domain (CRADD, hereafter referred to as RAIDD) and CASP2 (Bock et al., 2012; Janssens and Tinel, 2012).

The official nomenclature for PIDD is now PIDD1 (p53-induced death domain protein 1). This was deemed necessary to avoid confusion with primary immune deficiency disorders, often abbreviated the same way in the literature. Please note that there are no reported PIDD1 orthologues in non-vertebrates, neither have PIDD1 paralogues been found in vertebrates. PIDD1 was originally also known as leucine-rich repeat and death-domain-containing protein (LRDD) and had been independently described by two groups in the year 2000 (Telliez et al., 2000; Lin et al., 2000). In a bioinformatics screen for proteins containing a death domain (Box 1) similar to the one found in human receptor-interacting serine/threonine kinase 1 (RIPK1, hereafter referred to as RIP1), Telliez et al. identified a protein and named it, according to its structural features, LRDD (Telliez et al., 2000). The characterisation of its sequence revealed leucine-rich repeats (LRRs) at the N-terminus, ZU5 domains (i.e. domains present in ZO-1 and Unc5-like netrin receptors) in the intermediate region, as well as a death domain (DD) at the C-terminal end (Fig. 1A-C). Another structural domain called uncharacterised protein domain in UNC5, PIDD and ankyrins (UPA) was later defined between the ZU5 and the DD (Wang et al., 2009). Moreover, the authors observed evidence for processing of PIDD1 as they detected truncated forms of the protein in mammalian cells when overexpressing LRDD cDNA. Based on structural features, they speculated that LRDD functions as an adapter for small G-proteins that had been shown to interact with LRR sequences (Telliez et al., 2000).

Lin and colleagues identified PIDD1 as a direct transcriptional target of p53 by differential display analysis in an erythroleukemia cell line (Lin et al., 2000). Consistently, the sequence of the Pidd1 gene locus contained a non-canonical p53-responsive element in the Pidd1 promoter and upon γ-irradiation of mouse embryonic fibroblasts its mRNA was induced at the transcriptional level in a p53-dependent manner to an extent similar to that of cyclin-dependent kinase inhibitor 1 (p21, officially known as CDKN1A). Similar findings were made in MCF7 breast cancer and AML-4 leukaemia cells (Lin et al., 2000). Also, overexpression of PIDD1 suppressed cell growth by inducing apoptosis in p53-deficient cells, and this effect was reversed by PIDD1 knockdown; PIDD1 was, thus, assumed to be an essential component of the apoptotic arm of p53 (Lin et al., 2000). Around the same time, alternative cell death regulators that are induced by p53, such as phorbol-12-myristate-13-acetate-induced protein 1 (PMAIP1, hereafter referred to as NOXA) (Oda et al., 2000) and Bcl-2-binding component 3 (BBC3, hereafter referred to as PUMA) (Han et al., 2001; Nakano and Vousden, 2001; Yu et al., 2001), BH3-only members of the B-cell CLL/Lymphoma 2 (BCL2) family, were also described. Notably, subsequent loss-of-function analyses in mice confirmed roles for these two BH3-only proteins in the regulation of p53-induced cell death; however, they failed to provide equally compelling evidence for a role for PIDD1 in this process (Kim et al., 2009; Manzl et al., 2009; Shibue et al., 2003; Villunger et al., 2003). Certainly, a basal expression of PIDD1 was detectable in p53-deficient HCT116 and HEK-293 cells that express the large T antigen, which supported the notion that PIDD1 also has roles outside the canonical p53 response to DNA damage (Cuenin et al., 2008; Tinel et al., 2007).

In humans, PIDD1 mRNA transcript variant 1 gives rise to a full-length (FL)-PIDD1 protein that is 910 aa in length (∼100 kDa) and can be processed into three fragments (Box 2): the N-terminal 48 kDa fragment PIDD-N, and the two C-terminal fragments PIDD-C and PIDD-CC that have a molecular mass of 51 kDa and 37 kDa, respectively (Fig. 1A,B). Cleavage occurs at S446 and S588 by an autoproteolytic process that is similar to the mechanism described for self-cleaving protein segments, such as inteins, or for other proteins, such as nucleoporin Nup98 (Hodel et al., 2002). These proteins contain a conserved HSF tripeptide motif that enables a hydrophilic attack of the hydroxyl-group within the serine residue on the preceding peptide bond, converting it into an ester bond that is vulnerable to breakage through a second nucleophile (Hodel et al., 2002; Mills et al., 2014; Tinel et al., 2007). The processing of FL-PIDD1 into either PIDD-C or PIDD-CC appears to be a constitutive event and, as a result, FL-PIDD1 is barely detectable even upon induction of p53 (Tinel et al., 2011, 2007). Notably, PIDD-CC can be generated by autoproteolysis of FL-PIDD1(Tinel et al., 2011, 2007) or PIDD-C (our unpublished observations). Yet, the fact that accumulation of endogenous PIDD-C in response to DNA damage precedes the detection of PIDD-CC is consistent with PIDD-C being the most-prominent source for PIDD-CC (Tinel and Tschopp, 2004: Tinel et al., 2007). Furthermore, alternative splicing and protein autoprocessing may act redundantly to generate PIDD1 protein variants (Box 2). However, with the current tools it is difficult to scrutinize which PIDD1 isoforms are expressed in a given cell type at protein level.

PIDD1 autoprocessing requires binding of the chaperone heat shock protein 90 (Hsp90) and the co-chaperone p23 to facilitate the optimal conformation needed for self-cleavage (Table 1). Hsp90 interacts directly with FL-PIDD1 and recruits p23, a HSP90 co-chaperone (Weaver et al., 2000; Tinel et al., 2011). Yet, another heat shock protein, Hsp70, binds FL-PIDD1 as well as PIDD-N and PIDD-C; however, the role of this interaction remains to be investigated. Alongside autoproteolysis, the stability and function of PIDD1 also depends on Hsp90, which indicates an essential role for chaperones in the regulation of PIDD1 self-processing and its protein abundance (Table 1). Inhibition of Hsp90 allows rapid degradation of PIDD1 by the E3 ubiquitin-protein ligase CHIP (also known as STUB1), another co-chaperone that appears to preferentially ubiquitylate PIDD-C over PIDD-CC (Tinel et al., 2011). CHIP directly interacts with PIDD1, but also with Hsp70, which might explain the role of Hsp70 in the regulation of PIDD1. Although the PIDDosome can form upon a temperature shift in vitro and dissociation of HSP90 is needed for its formation, the initial binding appears to be required for PIDD1 function: disruption of the interaction of Hsp90 with PIDD1 impairs its autoprocessing and binding to different effector proteins (Tinel et al., 2011). Yet, it is tempting to speculate that these effects are secondary to impaired PIDD1 autoprocessing when Hsp90 is absent. Given the complex regulation of the autoprocessing mechanism and the stability of PIDD1, we anticipate that the stoichiometry and localisation of protein complexes that contain PIDD1 domains are tightly regulated in order to elicit the desired biological response.

PIDD-CC nucleates a complex with the dual adapter RAIDD that is needed for the recruitment and activation of CASP2 and, potentially, cell death initiation. This complex is commonly referred to as the PIDDosome. We propose the term ‘Caspase-2−PIDDosome’ for historic reasons, complex-I, to discriminate it from other PIDD1-containing complexes discussed below (Fig. 1C, Table 1). In contrast, signalling that involves PIDD-C has been associated with NF-κB activation and cell survival (Fig. 2B). DNA damage can cause translocation of PIDD-C to the nucleus, where it forms a complex with RIP1 and inhibitor of NF-kappa-B kinase subunit gamma (NEMO, officially known as IKBKG), the ‘NEMO-PIDDosome’ (Janssens et al., 2005; Tinel et al., 2007). Subsequent sumoylation, phosphorylation and ubiquitylation of NEMO leads to its export and to activation of NF-κB signalling in the cytoplasm, which is considered an anti-apoptotic signal (Janssens et al., 2005; Tinel et al., 2007). Although no protein complex that contains PIDD-N has been described, it might be a negative regulator of PIDD-C, as its overexpression appears to dampen NF-κB activation (Janssens et al., 2005; Tinel et al., 2011).

Of note, these studies nicely document a role for PIDD1 in NF-κB activation that is induced by DNA damage; the evidence that PIDD1 activation, indeed, promotes cell survival under these conditions is sparse (Bock et al., 2012). Whereas we were able to confirm a deficit in NF-κB activation in response to DNA damage in the absence of PIDD1, comparable to that caused by loss of RIP1 or PARP1, we were unable to document a role for PIDD1 in cell survival under the conditions tested (Bock et al., 2013). Primary cells derived from Pidd1^−/− mice showed normal cell death responses, similar to those we noticed upon lack of Raidd or Casp2 (Manzl et al., 2009). Furthermore, the DNA-damage repair potential of mouse hematopoietic stem cells exposed to γ-irradiation, as read out by colony formation, appeared unaffected by loss of PIDD. Yet, we observed a clear deficit in cytokine production in response to DNA damage in bone marrow-derived macrophages and mouse embryonic fibroblasts (MEFs) of these mice (Bock et al., 2013). Also, production of tumour necrosis factor (TNF) in the gastrointestinal tract after γ-irradiation was found to be reduced in PIDD-deficient animals. Moreover, survival and tumour transformation potential of these mice were unaffected, which suggested that the role of the ‘NEMO-PIDDosome’ is to promote inflammation, rather than to control cell survival or DNA repair (Bock et al., 2013).

Formation of the Caspase–2–PIDDosome (Fig. 1C) involves interaction with the adapter molecule RAIDD, which contains a DD as well as a caspase-recruitment domain (CARD) (Duan and Dixit, 1997). RAIDD and PIDD-CC associate through their DD to form a high molecular weight complex, whereas the N-terminal CARD in RAIDD acts as a docking site for the pro-form of CASP2. Interaction with RAIDD is restricted to PIDD-CC, which is generated from the human PIDD1 transcript variant 1, because the minor deletion in transcript variant 3 appears to be sufficient to abrogate RAIDD binding (Tinel et al., 2007; Tinel and Tschopp, 2004) (Fig. 1A,B). Whereas RIP1 and NEMO are immediately recruited to PIDD1, the interaction with RAIDD occurs later. This is in line with the observation that PIDD-C and PIDD-CC are generated in a sequential manner (Janssens et al., 2005; Tinel et al., 2007). Yet, given the overlapping kinetics with which the different complexes are formed, we speculate that both types, the Caspase–2–and NEMO-PIDDosome, can be present in the cell at a given time in different locations and amounts. It is also not entirely clear whether autoprocessing of PIDD-CC is prevented once PIDD-C enters the nucleus or upon its binding to RIP1.

Nuclear localisation of PIDD-C is needed for NF-κB activation; but PIDD-CC has also been detected in the nucleolus (Fig. 2C), an organelle with various functions that range from ribosome biogenesis to DNA repair (Ogawa and Baserga, 2017). A yeast-two-hybrid screen using a mouse thymoma cDNA library and PIDD1 isoform 2 as a bait identified nucleolin (NCL) as a potential binding partner for PIDD1 (Table 1). The subsequent analysis documented that GFP-tagged PIDD-C is constitutively targeted to the nucleolus (Pick et al., 2006). Interestingly, whereas the cytoplasmic pool of this fusion protein was rapidly degraded after UV-treatment, potentially through CHIP-mediated ubiquitylation, the nucleolar fraction of the protein remained stable, which indicates that nucleolar PIDD-C is protected from degradation. As PIDD1 protein isoform 2 cannot be autoprocessed into PIDD-CC, it remains unclear whether the latter can also localize to or might even be generated inside the nucleoli (Pick et al., 2006). Recent work, however, documents the localisation of PIDD-CC in the nucleolus (Ando et al., 2017). Here, mass spectrometry (MS) revealed the association of PIDD1 with nucleophosmin 1 (NPM1), a key nucleolar protein and established tumour suppressor that is most frequently mutated in acute myeloid leukaemia (Heath et al., 2017). Upon DNA damage, and most prominently in response to inhibition of topoisomerase I, NPM1 binds to the LRRs in PIDD-N as well as to PIDD-CC; the latter directs activation of caspase-2 in the nucleolus in a PIDD1-dependent manner, thereby promoting cell death (Ando et al., 2017). Currently, it remains unclear what drives the interaction of PIDD1 and NPM1 in response to DNA damage and whether or how, PIDD-N is involved in PIDDosome assembly, as it has never been reported to shuttle to the nucleus. Remarkably, NPM1 can also localise to the cytoplasm (Maggi et al., 2008), which raises the possibility that it there interacts with FL-PIDD1 and shuttles it to the nucleolus where autoprocessing into PIDD-CC is completed. The study by Maggi and colleagues also raises the question whether the mutated form of NPM1 (NPM1c+), which is preferentially found in the cytoplasm (Heath et al., 2017), promotes disease, in part, by preventing PIDD1 from entering the nucleolus, thereby reducing proficiency of CASP2 activation.

Independently of the signalling pathways described above, PIDD1 participates in translesion DNA synthesis (TLS), i.e. DNA extension across a lesion, in response to UV-radiation (Fig. 2D). In the nucleus, PIDD1 was reported to form yet another complex with key components of the replication machinery, proliferating cell nuclear antigen (PCNA), replication factor C subunit 5 (RFC5) and RFC4 referred to as PCNA-PIDDosome (Table 1). These proteins were identified by MS-analysis as binding partners of overexpressed PIDD1 (Logette et al., 2011). PCNA acts as a DNA-sliding clamp that requires the RFC for its positioning onto DNA and is itself needed to load DNA polymerases for replication (Logette et al., 2011). Moreover, it exerts critical roles in DNA repair. Importantly, the interactions between PIDD1, PCNA and RFC5 are mediated by their ZU5 domains; hence, the complex can only be formed by FL-PIDD1, or − given the low abundance of FL-PIDD1 and the preferred nuclear localisation of its first autoprocessing product − by PIDD-C (Logette et al., 2011). A role for PIDD1 isoform 4 or isoform 5 (only the latter encodes the two ZU5 domains) (Fig. 1A) remains unexplored. Yet, when PIDD1 binds the C-terminal part of PCNA, p21 dissociates from PCNA and is degraded, which allows PCNA mono-ubiquitylation of lysine 164 (Bruning and Shamoo, 2004). This modification, in turn, facilitates chromatin association of the low-fidelity DNA polymerase eta (Pol η), which then performs TLS (Fig. 2D). In doing so, the cell overcomes unrepaired DNA lesions in S-phase that are not resolved by nucleotide excision repair and allows to replicate DNA in the presence of cyclobutan-pyrimidine-dimers (Logette et al., 2011). In this context, PIDD1 appears to act as a pro-survival factor in the ‘PCNA-PIDDosome’ (Fig. 1C). Consistently, cells of the human keratinocyte cell line HaCaT that were depleted of PIDD1 through RNA interference (RNAi) showed increased cell death upon UV-C treatment. Moreover, skin from Pidd1-deficient mice that were exposed to UV-radiation showed higher numbers of apoptotic keratinocytes in situ (Logette et al., 2011). Whether this increase in cell death was owing to an increased mutational load or the lack of survival signals that depend on NF-κB remains unclear. Intriguingly, it has been shown that NPM1c+ can delocalize Pol η to enhance its cytoplasmic degradation (Ziv et al., 2014). Hence, it is tempting to speculate that interactions between PIDD1 and NPM1 are not only needed to facilitate the activation of nucleolar CASP2, but also do affect TSL efficiency by fine-tuning Pol η localisation and abundance. In summary, by forming different protein complexes, PIDD1 appears critical to fine-tune a number of cellular responses that deal with the consequences of DNA damage, including the maintenance of replication capacity and inflammatory cytokine production. Yet, its role in another key response to DNA-damage, i.e. the induction of cell death, remains controversial.

Motivated by the early observations made by Tinel and Tschopp, many follow-up studies on PIDD1 aimed to position it as an initiator of CASP2-mediated cell death (Bock et al., 2012; Janssens and Tinel, 2012). Interestingly, zebrafish embryos that lack p53 are radio-resistant, similar to several human cancer cell lines that lack functional p53. However, radiation sensitivity was restored when p53-deficient zebrafish embryos were treated with a morpholino that targets checkpoint kinase 1 (CHK1) or a chemical CHK1 inhibitor (CHK1i) (Sidi et al., 2008). Remarkably, this type of DNA-damage-induced cell death appears to be independent of CASP3, CASP8 or CASP9, is not blocked by BCL2 overexpression − yet requires CASP2. This cell death pathway has, therefore, been dubbed ‘CHK1-suppressed’ (CS) pathway of cell death and was further shown to depend on an ATM−ATR−CASP2 signalling axis (Sidi et al., 2008). However, the CS pathway does not depend on canonical regulators of cell death, such as BCL2, PUMA, NOXA, p63, p73 or cytochrome c release. These findings suggest that CHK1 prevents a cell death response that becomes activated upon mitotic entry in the presence of DNA damage or, equally possible, after exit from erroneous mitosis (Sidi et al., 2008) (Fig. 2C). Importantly, this cell death pathway, which was nicely genetically delineated in zebrafish embryos appeared to be conserved in human cancer cells. However, it was impossible to recapitulate all features of the CS pathway in mice (Manzl et al., 2013). Furthermore, it remained unclear how ATM/ATR-signalling could activate CASP2. In a follow-up study, the group of S. Sidi showed that ATM but, somewhat surprisingly, not the CHK1 activator ATR binds to PIDD1 (Ando et al., 2012). It does so by engaging its LRR domains and by phosphorylating PIDD1 at Thr788 in response to irradiation, most prominently when CHK1 function was blocked simultaneously (Ando et al., 2012). This binding of PIDD1 to ATM resulted in CASP2-dependent cell death of HeLa cells. ATM-mediated phosphorylation of Thr788, which is located in the DD, was shown to be necessary and sufficient for PIDD1 to bind to RAIDD and to activate CASP2. At the same time, this prevents the binding of PIDD1 to RIP1, which would otherwise activate NEMO-sumoylation and NF-κB signalling (Ando et al., 2012). This suggests that ATM can convert PIDD1 from a survival factor to a protein that promotes cell death. How CASP2 drives cell death in this setting remains to be defined.

It remains uncertain how selective this ATM-mediated phosphorylation event is for the CS pathway. ATM appears to bind to PIDD1 also upon IR damage alone and the phosphorylation of Thr788 clearly affects the ability of cells to activate NF-κB in response to the topoisomerase IIa inhibitor etoposide, which induces DNA damage (Ando et al., 2012). This occurs in the absence of CHK1 inhibition, which suggests a role for PIDD1 in the canonical DNA-damage response. Further, the need to force cells to override the G2/M checkpoint complicates a clean epistatic analysis of events, as multiple additional signalling cascades are activated. Moreover, only HeLa cells and SV40-immortalised MEFs were tested rigorously and both lack a functional p53 response. Regardless of this, IR − in combination with CHK1i treatment − was clearly more effective in driving phosphorylation of PIDD1 and Caspase-2−PIDDosome assembly than the use of IR alone (Ando et al., 2012), which suggests that entry into mitosis, or even mitotic traverse, is needed for efficient Caspase-2−PIDDosome activation and cell death. In line with this is the observation that the mitotic pseudo-kinase BUBR1 (also known as BUB1B) can prevent PIDDosome formation in mitotic cells (Thompson et al., 2015) (Fig. 2C). BUBR1 is a component of the mitotic checkpoint complex (MCC) that executes the spindle assembly checkpoint (SAC). The checkpoint is responsible for the inhibition of the anaphase-promoting complex (APC/C), thereby preventing mitotic progression until all kinetochores have achieved proper microtubule attachment (Musacchio, 2015). Since small interfering RNA (siRNA) that targets BUBR1 triggers Caspase-2−PIDDosome formation in response to IR alone, BUBR1 has been suggested to function either downstream or in parallel to CHK1 (Thompson et al., 2015). Of note, knockdown of BUBR1 phenocopied the effects that are mediated by siRNA targeting CHK1, which leads to PIDDosome-dependent apoptosis in response to IR. As siRNA against MAD2, another key-component of the MCC, had no such effect on PIDDosome assembly, a novel non-canonical function of BUBR1 outside the SAC has been proposed (Thompson et al., 2015).

Intriguingly, phosphorylated PIDD1 colocalises with histone H2AX phosphorylated at Ser139 and ATM protein kinase phosphorylated at Ser1981 at DNA lesions in interphase, which positions it as a putative substrate of ATM, but was found at kinetochores in early prophase and this localisation required BUBR1 (Thompson et al., 2015). BUBR1 itself is recruited to kinetochores to initiate the SAC; there, it competes with RAIDD to bind to PIDD1, which has been suggested to avoid the induction of apoptosis during mitosis. The formation of the Caspase-2−PIDDosome is, thus, only possible once the APC/C degrades BUBR1 at the end of mitosis, an event that triggers formation of the Caspase-2−PIDDosome for cell death in the following interphase (Thompson et al., 2015). However, CASP2 is also phosphorylated in the linker region at Ser340 through CDK1 in order to prevent its activation in mitosis (Andersen et al., 2009). Thus, BUBR1-mediated control of PIDDosome formation during mitosis appears to serve here as an additional fail-safe mechanism. Hence, in the absence of DNA damage, CASP2 might act as a sensor of the metabolic state in cells stuck in M-phase for extended periods of time (Salazar-Roa and Malumbres, 2017). This idea finds support in observations that have been made in Xenopus oocytes that were arrested in meiosis II and experienced nutrient deprivation (Nutt et al., 2009). Such a role would be consistent with reports that claim that CASP2 can contribute to mitotic cell death of cells treated with anti-mitotic drugs (Ho et al., 2008). We propose that such a metabolic sensor function of CASP2, if it exists, does not require PIDD1 (Fig. 2) because its formation is actively prevented in M-phase (Thompson et al., 2015), and that p53-activation triggered in response to extended mitosis does not involve CASP2 (Fava et al., 2017).

It remains unclear why the death of cells entering mitosis in the presence of DNA damage should be postponed until the next interphase, unless the cell manages efficient DNA repair during mitotic traverse. However, DNA repair activity in mitosis appears to be minimal and double-strand breaks are usually only marked for repair in the next interphase, whereas extensive DNA damage activates the SAC (Heijink et al., 2013). Moreover, given the fact that canonical apoptosis can also be engaged by alternative routes in cells that lack p53, e.g. through the activation of PUMA or NOXA by p73 (Flinterman et al., 2005; Melino et al., 2004), it seems unlikely to us that evolution has developed the PIDDosome solely to promote the death of p53-deficient cells in an unforeseeable artificial setting. The controversy was perpetuated by the fact that mouse genetics did not support a critical role for either Caspase-2−PIDDosome component in DNA-damage-induced cell death; further, the CS pathway is poorly conserved (Kim et al., 2009; Manzl et al., 2013, 2009). Evolutionary conservation was carefully evaluated in a range of primary haematopoietic cells, including stem cells, thymocytes, resting and mitogen-stimulated T and B cells, gastrointestinal epithelial cells, and primary as well as immortalised MEFs (Manzl et al., 2013). The latter were shown to have a clear deficit in activating the Caspase-2−PIDDosome (Ando et al., 2012; Thompson et al., 2015), but died similar to wild-type cells upon DNA damage (Manzl et al., 2013). Equally puzzling was the phenomenon that all that seems to be required for PIDDosome assembly is exit from mitosis in the presence of DNA damage. Hence, we wondered: what might all these treatments have in common such that, eventually, the Caspase-2−PIDDosome is activated?

PIDD1, together with p21, has been shown to be upregulated in an Adeno-Cre inducible, KRAS^G12D-driven lung cancer mouse model after prolonged treatment with cisplatin that caused drug-resistance (Oliver et al., 2010). Interestingly, repeated long-term drug-treatment of these mice failed to increase overall survival rates, despite intermittent tumour regression. This was because the growth of more aggressive clones increased when mice did not receive any drugs. As a result these tumours presented with huge variations in chromosome copy numbers, indicative for high-grade aneuploidy (Oliver et al., 2010). Remarkably, when drug-resistant tumours were treated again with cisplatin, these aneuploid tumours showed a stronger induction of PIDD1, along with p21, compared to mice with euploid tumours that received drug-treatment for the very first time. Therefore, an enhanced up-regulation of PIDD1 is linked to increased aneuploidy in KRAS^G12D-driven lung cancer (Oliver et al., 2010). Similarly, cisplatin treatment induced up-regulation of PIDD1 in RAS-mutant but p53-proficient human non-small-cell lung carcinoma and HCT116 colon cancer cells. Remarkably, exogenous expression of PIDD1 resulted in induction of p21 and reduced growth rates, which was associated with an emerging drug resistance (Oliver et al., 2010, 2011). The reduced growth and drug responses caused by PIDD1 overexpression were, again, found to be abrogated in p53-null cells; this positions PIDD1 upstream of p53 in this setting (Oliver et al., 2010, 2011). Together, this suggests that PIDD1 has an anti-proliferative and hence, potentially, also a pro-survival function that is independent of NF-κB signalling, at least in the presence of functional p53.

Oliver and colleagues also investigated the link between the fast recurrence after initial chemo therapy (‘platinum-resistance’) and PIDD1, p21 and their common inducer of gene expression, p53. They proposed a feed-forward loop in which, upon DNA damage, p53 induces PIDD1, in turn, contributing to p53 activation and continued p21 expression that explains the reduced growth of cells overexpressing PIDD1. Importantly, this loop requires activation of caspase-2 that leads to selective cleavage of MDM2 at Asp367. The N-terminal fragments of MDM2 that are generated by CASP2-mediated proteolysis lack the RING domain that is required for E3 ligase activity. Hence, these fragments accumulate, bind to p53, and increase its stability and target gene transcription, which leads to p21-mediated cell cycle arrest (Oliver et al., 2011).

These findings, together with genetic data from mice that lack PIDD1 (Kim et al., 2009; Manzl et al., 2013, 2009; Oliver et al., 2010) argue against a clear-cut pro-death role of PIDD1. How can these discrepancies be reconciled? Our own recent findings suggest that CASP2 activation that depends on PIDD1 does neither contribute to cell death, nor to p53 activation induced by DNA damage (e.g. upon treatment with doxorubicin) or prolonged mitotic arrest (e.g. upon treatment with taxol). Yet, p53 activation that is triggered by centrosome amplification – a phenomenon that can occur as a consequence of cytokinesis failure – does clearly depend on the Caspase-2−PIDDosome (Fava et al., 2017). Remarkably, it is the number of mature mother centrioles that appears to be counted by PIDD1, thereby employing a mechanism that still needs to be discerned. Intriguingly, PIDD1 localises to the distal end of mature centrosomes in healthy cells (Fava et al., 2017). More than one mature centriole with appendages can initiate Caspase-2−PIDDosome assembly, which promotes CASP2-mediated MDM2 cleavage, p53 accumulation and p21-mediated cell cycle arrest (Fig. 2A). All this occurs in the absence of appreciable cell death. Consistently, depletion of centrosomes abrogated the proficiency of the pathway upon cytokinesis failure, whereas the generation of extra centrosomes induced by overexpression of polo-like kinase 4 (PLK4) sufficed to activate the pathway in the absence of cytokinesis failure (Fava et al., 2017). Our findings, therefore, established the Caspase-2−PIDDosome as the missing link that connects supernumerary centrosomes to activation of p53. Intriguingly, anti-mitotic drugs, DNA-damaging agents (Dodson et al., 2004; Sato et al., 2000), impaired CHK1 function (Peddibhotla et al., 2009), inhibition of BubR1 (Oikawa et al., 2005) or inhibition of Aurora B kinase (Ditchfield et al., 2003) all have the strong potential to cause cytokinesis failure or direct centrosome amplification. Hence, it seems highly likely that at least one component feeding into Caspase-2−PIDDosome activation under all these different conditions is the increase in centrosome number. Such an increase happens in a substantial fraction of cells exposed to these treatments, either in response to cytokinesis failure or centrosome amplification.

Given what has been discussed above, we propose that lagging chromosomes in aneuploid KRAS-driven lung tumours frequently result in failure of cytokinesis. This triggers centrosome accumulation, Caspase-2−PIDDosome-dependent p53 stabilisation and p21 induction and, in turn, may increase PIDD1 expression levels over time because it is also a p53 target. The rise in PIDD1 expression is possibly owing to subsequent DNA damage in cells that fail to arrest properly in the presence of extra centrosomes and, again, activate p53 − either along the canonical DNA-damage response route or upon delayed M-phase progression in response to problems of chromosome alignment (Lambrus and Holland, 2017). This would secure p21-checkpoint proficiency and the survival of these aneuploid cells. Consistently, CASP2 deficiency facilitates tumour recurrence in this cancer model after cisplatin treatment (Terry et al., 2015). In addition, reduced CASP2 expression that is linked to impaired function of the transcriptional regulator BCL9L contributes to aneuploidy tolerance in colon cancer (López-García et al., 2017). This phenomenon, potentially, also explains why p53-deficient KRAS tumours, which are aneuploid early on and cannot trigger this response, show a relative survival benefit after cisplatin treatment that is not observed in mice bearing p53-proficient tumours (Oliver et al., 2010). It remains to be seen whether such p53-deficient tumours are more prone to cell death when exposed to cisplatin or other drugs that promote G2/M failures, but it would fit nicely with observations made in p53-mutant zebrafish or HeLa cells exposed to DNA damaging agents in combination with RNAi of CHK1i or BUBR1 (Thompson et al., 2015). Here, Caspase-2−PIDDosome assembly is most likely also triggered by extra centrosomes in G1 cells that were forced to override the G2 and M checkpoints. However, in this situation the activation of CASP2 preferentially causes cell death as these cells cannot arrest (Thompson et al., 2015). This type of cell death does not seem to depend on components of the canonical mitochondrial apoptosis pathway (Sidi et al., 2008). However, it also largely excludes a role for the BH3-interacting domain death agonist (BID), a bona fide CASP2 substrate and potent apoptosis effector (Guo et al., 2002); thus, it will be very interesting to find out how CASP2 actually kills these cells. Given its high homology with C. elegans Ced3−the one and only caspase in the worm−one is tempted to speculate that it acts as an initiator and effector at the same time. If this were the case, it would raise the question regarding potential CASP2 substrates that cause cell death−possibly in a fashion similar to that of gasdermin D, which is activated by caspase-1 (Shi et al., 2017).

All these observations led us to assign a potentially oncogenic function to PIDD1, as it might increase the selection pressure in aneuploid, p53-proficient cells to overcome this anti-proliferative barrier. This possibility is supported by the delayed outgrowth of MYC-driven lymphomas in Pidd1^−/− mice (Manzl et al., 2012), a phenomenon that for unclear reasons is not seen in Raidd^−/− mice, (Peintner et al., 2015). Moreover, given the role of PIDD1 in TLS (Logette et al., 2011), it would be highly interesting to see whether mice lacking the PIDDosome show a delay in UV-driven skin cancer.

For many years, PIDD1 and its binding partners were considered to only have a role in the DNA-damage response, which severely hampered the search for potential pleiotropic, physiological roles of PIDD1. With the newly established role for PIDD1 as a sensor of aberrant centrosome numbers, one can now also envision a potential role of the PIDDosome in cellular differentiation processes and during organogenesis. Our own findings clearly show that the PIDDosome restricts ploidy in the liver by activating p53 in a CASP2-dependent manner. Accordingly, the DNA content of hepatocytes that are deficient of the PIDDosome is substantially increased and, thereby, copies the p53-null mutant phenotype (Fava et al., 2017; Kurinna et al., 2013). Clearly, PIDDosome-deficient animals constitute an interesting model to study the impact of ploidy on liver function and regeneration, without facing the adverse effects of global p53 deficiency. Of note, a number of cell types become polyploid during organ development or in response to infection: cardiomyocytes, for example, increase their ploidy during terminal differentiation, whereas viral infection can cause cell−cell fusion, and bacterial infection may trigger the formation of macrophage giant cells (Aguilar et al., 2013). Under all these conditions, extra centrosomes arise in cells and it is tempting to speculate that PIDD1 is called into action in some settings, marking PIDD1 also as a target for the pharmacological modulation of these processes.

We are thankful to all members of the Division of Developmental Immunology at MUI for helpful discussion.