ADP-ribosylation: from molecular mechanisms to human disease

Abstract Post-translational modification of proteins by ADP-ribosylation, catalysed by poly (ADP-ribose) polymerases (PARPs) using NAD+ as a substrate, plays central roles in DNA damage signalling and repair, modulates a range of cellular signalling cascades and initiates programmed cell death by parthanatos. Here, we present mechanistic aspects of ADP-ribose modification, PARP activation and the cellular functions of ADP-ribose signalling, and discuss how this knowledge is uncovering therapeutic avenues for the treatment of increasingly prevalent human diseases such as cancer, ischaemic damage and neurodegeneration.

PARP members, structure and activity ADP-ribosyl transferases, also known as poly(ADPribose) polymerases (PARPs), are specific enzymes that transfer the ADP-ribose moiety from b-nicotinamide adenine dinucleotide (NAD + ) to a target macromolecule, mainly proteins. This activity was identified in the 1960s (Chambon et al., 1963), and almost 20 years later, singleand double-strand DNA breaks were determined as enzyme activators in cell extracts (Benjamin and Gill, 1980a;1980b). Since then, ADP-ribosylation of proteins has been recognized as a central posttranslational modification in a range of cellular processes, such as DNA damage signalling and repair, transcription, Wnt signalling and programmed cell death (Gibson and Kraus, 2012;Virag, 2013;DaRosa et al., 2015;Kraus, 2015).
ADP-ribosylation can occur either as a single mono(ADP-ribose) unit (MAR) or as poly(ADP-ribose) (PAR) chains, which can be linear or branched. Since the ADP-ribosyl (ADPr) group contains a high density of negative charges, the addition of ADPr units can dramatically change the biophysical properties of a target protein or promote protein-protein interactions (Figure 1). For example, long PAR chains have been proposed to produce a halo of negatively charged density around the target protein, disrupting the liquid phase in which the protein is embedded (Altmeyer et al., 2015).
There are 17 known members of the PARP family in the human genome (Barkauskaite et al., 2015), and most of these possess the ability to auto-modify, often on multiple sites (Vyas et al., 2014). However, only a few are bona fide poly(ADP-ribose) polymerases, while most are in fact mono(ADP-ribosyl) transferases (Vyas et al., 2014). In human cells, the majority of PARP activity is exerted by PARP1 (85%-90%) and by PARP2 (10%-15%) (Szanto et al., 2012).
PARPs are multidomain proteins that contain a common structurally related catalytic domain that is also found in a range of pathogenic toxins from both gram-positive and gram-negative bacteria such as Bacillus sphaericus, Clostridium sp., Corynebacterium diphteriae, Salmonella enterica, Vibrio cholera and Escherichia coli, (Laing et al., 2011;Karlberg et al., 2013;Simon et al., 2014;Barkauskaite et al., 2015;Langelier et al., 2018). The catalytic domains of 5 of the 17 human members -PARP1, 2, 3, 4 and 16 -, contain an additional subdomain known as helical domain (HD), which has autoinhibitory functions by sterically hindering NAD+ binding and has to be removed for every catalytic cycle (Dawicki-McKenna et al., 2015;Langelier et al., 2018).
( Figure 1B, C). The substrate to be PARylated binds to the acceptor site on the surface of the ART domain, defined by the acceptor loop (residues 977 to 988 in PARP1) that is also thought to regulate polymer length and chain branching (Vyas et al., 2014;Chen et al., 2018). NAD + binds to the donor site in an extended conformation, such that the ADP-ribose moiety interacts with the D-loop in the ART domain (residues 875 to 894 in PARP1) (Gibson and Kraus, 2017), while the nicotinamide moiety forms three hydrogen bonds with Gly863, Ser904 and Tyr907 (PARP1 numbering) ( Figure 1B, C) (Langelier et al., 2018).
Two reaction mechanisms have been proposed, with detailed structural evidence supporting the second mechanism (Tsurumura et al., 2013). One is an S N 2 displacement mechanism, with the formation of a penta-coordinated transition state (Marsischky et al., 1995), while the other is an S N 1 strain-alleviation mechanism that involves the formation of a stable furanosyl oxocarbenium ion (van Rijssel et al., 2014), and a rotation around the phosphodiester bond (Simon et al., 2014;Cohen and Chang, 2018). In either case, the nucleophilic attack is performed by an oxygen or a nitrogen atom from the side chain of the target amino acid, which can be glutamic acid, aspartic acid, serine, cysteine, arginine, lysine or tyrosine (Ogata et al., 1980;Laing et al., 2011;Rosenthal and Hottiger, 2014;Bonfiglio et al., 2017;Leslie Pedrioli et al., 2018). Subsequently, the product can have chemical reorganisations: glutamate and aspartate modifications undergo a C1'-C2' transfer, and lysine linkages suffer an Amadori rearrangement to form a stable ketoamine Morgan and Cohen, 2015;Cohen and Chang, 2018) ( Figure  1A). Ultimately, nicotinamide is released as a by-product. Linear PAR chains are formed using the hydroxyl group in C2 of the ADP-ribose moiety, and branching involves the oxygen in C2' for the nucleophilic attack (Juarez-Salinas et al., 1982;Chen et al., 2018).
Recently, an important modifier of PARP catalytic activity, termed histone PARylation factor (HPF1), was described (Gibbs-Seymour et al., 2016). HPF1 is responsible for switching specificity of PARP1/2 towards serine and tyrosine residues and from auto-PARylation to PARylation of chromatin components and remodellers Leslie Pedrioli et al., 2018). HPF1 also seems to modulate the length of ADPr polymers and can itself be mono(ADP)ribosylated by PARP1 (Leslie Pedrioli et al., 2018). Recent studies revealed that serine could be the predominant PARylation site at chromatin after DNA damage (Leidecker et al., 2016;Palazzo et al., 2018). Strikingly, some results indicate that despite the presence of a hydroxyl group and the resemblance with serine, threonine is not modified by PARP in mammalian cells (Leslie Pedrioli et al., 2018).
In PARP1, the WGR domain is not involved in the initial recognition and binding of DNA-breaks (Eustermann et al., 2015). Instead, three zinc fingers (ZnFs) make the primary contact with the DNA. The first two ZnFs at the PARP1 N-terminus are necessary and sufficient for protein recruitment to DNA-damage sites in vivo, using structurally equivalent residues (Ali et al., 2012). Recent NMR studies suggest that ZnF2 is the leading domain that binds to the 3' end of the break, followed by ZnF1, which recognises the 5' end. This complex promotes ZnF3 recruitment, which leads to WGR domain binding to a surface formed by ZnF1, ZnF3 and DNA (Eustermann et al., 2015). Interestingly, PARP1 makes much more extensive contacts with the DNA surrounding the break than at the break site per se, allowing for the recognition of DNA breaks from a variety of sources. In contrast, the WGR domains of both PARP2 and PARP3 (which do not have ZnFs) play a key role in DNA binding and discriminate between different DNA ends by recognising the presence of a 5'phosphate group at the DNA break site (Langelier et al., 2014;Grundy et al., 2016;Obaji et al., 2018).
In addition to the domains involved in DNA-break recognition and catalytic activation, PARP1 contains a BRCT-like (BRCA1 C-terminus) domain where most of the auto-modification sites have been identified Tao et al., 2009) and which is implicated in mediating protein-protein interactions (Liu et al., 2011;Noren Hooten et al., 2011;Hsu et al., 2019) ADP-ribosylation of DNA ADP-ribosylation was long considered a protein modification exclusively. However, recent reports have independently shown that DNA-dependent PARPs can add ADPr covalently to DNA ends, at least in vitro (Talhaoui et al., 2016;Munnur and Ahel, 2017;Zarkovic et al., 2018). PARP1, PARP2 and PARP3 all modify both 3' or 5' terminal phosphate groups via a phosphodiester bond, and PARP1 and PARP2 can also modify free 3'hydroxyl groups to generate a ribose-ribose bond (Talhaoui et al., 2016;Munnur and Ahel, 2017;Zarkovic et al., 2018). Modification of 5' phosphorylated ends may protect them from phosphatase activity, offering a possible function for this modification in vivo. Surprisingly, ADP-ribosylation of single-stranded DNA gaps promoted their ligation by DNA ligases even in the absence of ATP, suggesting that DNA modification "activates" these ends for ligation . However, it is currently unclear if and how this promotes DNA repair in vivo.

Cellular Functions of ADP-ribosylation DNA damage signalling and repair
Perhaps the best-studied cellular role of ADP-ribosylation is the crucial function of PARP1 and PARP2 in promoting the repair of DNA strand breaks (Ray Chaudhuri and Nussenzweig, 2017). PARP1 is a sensor of DNA breaks with high affinity for DNA and a lesion recognition mechanism that allows it to be activated by DNA breaks induced by a broad range of sources (Eustermann et al., 2015).
PARP1 activation leads to extensive HPF1-assisted PARylation of chromatin components surrounding DNA damage sites (Boulikas, 1988;Gibbs-Seymour et al., 2016). PARylation of histone H1 and all four nucleosomal histones, as well as HMG proteins, occurs on a number of modification sites, predominantly serines Palazzo et al., 2018), but whether these have differing functions or are simply a chromatin attachment site for PAR chains is currently unclear. In addition to changes in the chromatin environment (discussed below), PARylation leads to the recruitment of a myriad of DNA repair factors, which often contain dedicated PAR-binding domains such as BRCT, PBZ, WWE and macrodomain, or a short positively charged peptide sequence termed the PAR-binding motif (PBM) (Beck et al., 2014a). Crucially, PARP1 automodification reduces its affinity for DNA, allowing the repair machinery to access the damage site (Satoh and Lindahl, 1992).
In the case of DNA single-strand break repair (SSBR), PARP1 and PARP2-dependent ADP-ribosylation leads to the recruitment of the central scaffolding protein XRCC1, which contains a PAR-binding BRCT domain (Caldecott, 2008;Breslin et al., 2015;Polo et al., 2019). Lesions repaired by this pathway arise predominantly from oxidative damage to the DNA, but are also formed as intermediates of the base excision repair pathway or by the abortive activity of topoisomerases and DNA ligases (Caldecott, 2014). XRCC1 interacts with DNA and a range of DNA modifying enzymes that process these lesions to restore canonical 3'OH and 5'P termini required for subsequent re-ligation of the damaged strand by DNA ligase III (Caldecott, 2008;Polo et al., 2019) (Figure 2A).
In cycling cells, this pathway prevents the collision of unrepaired single-strand breaks (SSBs) with the DNA replication machinery, which would convert SSBs into much more deleterious DNA double-strand breaks (DSBs) (Figure 2B). As these DNA replication-induced DSBs are oneended, their accurate repair requires homologous recombination using the sister chromatid (Saleh-Gohari et al., 2005;Cortes-Ledesma and Aguilera, 2006). This leads to a distinctive requirement for functional homologous recombination in cells with defective single-strand break repair, as discussed in the context of PARP inhibitors below. Interestingly, SSBR was recently shown to serve as a backup mechanism for the "repair" of unligated Okazaki fragments during DNA replication (Hanzlikova et al., 2018) and is also thought to play a role in a sub-pathway of DNA double-strand break repair termed microhomology-mediated end-joining (Sfeir and Symington, 2015).
PARP1 also plays a crucial role in promoting the reversal of dysfunctional DNA replication forks (Ray Chaudhuri et al., 2012;Ray Chaudhuri and Nussenzweig, 2017). Fork reversal is an active process that occurs when DNA replication stalls due to impediments to the progression of the replisome (Zellweger et al., 2015) and involves the formation of a "chicken foot" structure in which the newly synthesised daughter strands anneal to each other (Quinet et al., 2017) ( Figure 2C). The molecular mechanisms of this process are currently under intense investigation, but PARP1 seems to stabilise reversed forks by preventing the helicase RECQ1 from dismantling the reversed DNA arm (Berti et al., 2013).
PARP1 engagement of DNA breaks, particularly DSBs, has to be carefully coordinated with other endbinding proteins to ensure genomic stability. The Ku70/Ku80 heterodimer is a sensor of DSBs for repair by the non-homologous end-joining pathway (NHEJ) (Shibata et al., 2018). PARP1 is thought to compete with Ku for DSB binding so that PARP1 loss allows Ku to engage DNA ends aberrantly and vice-versa, leading to damage hypersensitivity and genomic instability (Hochegger et al., 2006;Cheng et al., 2011). This is highlighted by a recent report suggesting that PARP1 may participate in the eviction of Ku from breaks that are destined for repair by NHEJindependent pathways (Yang et al., 2018). Conversely, the Mre11/Rad50/Nbs1 (MRN) complex, which also plays a very early role in the signalling and repair of DSBs, has been suggested to require PARP1 for its recruitment to break sites (Haince et al., 2008;Bryant et al., 2009). However, there are also instances in which PARP1 activation must be actively suppressed, such as at telomeres, where the shelterin complex, and in particular TRF2, prevents PARP1 binding to avoid attempts of "repairing" telomeric DNA ends (Schmutz et al., 2017).
PARP1 and PARP2 are partially redundant, as illustrated by the early embryonic lethality of the double knockout mouse (Menissier de Murcia et al., 2003). Phenotypically, PARP2 can replace PARP1 for many of the roles described above, but is restricted in part by a more limited specificity for DNA breaks with 5'P ends (Langelier et al., 2014). Both enzymes are redundant for XRCC1 recruitment to oxidative lesions  and the repair of DNA base damage (Ronson et al., 2018), but only PARP1 seems to generate ADP-ribose in response to topoi- A singlestrand break activates PARP1/2, leading to HPF-1 assisted PARylation of chromatin. PARP auto-modification causes its release and PAR chains surrounding the break site recruit XRCC1 complex. APTX and PNKP process break termini, Polb fills the gap by DNA synthesis and DNA Ligase IIIa seals the remaining nick. PARG removes PAR chains and XRCC1 complex is released, completing the repair. B) Defective single-strand break repair causes a reliance on homologous recombination. An unrepaired single-strand break is encountered by an ongoing replication fork, which converts it into a one-ended double-strand break. This lesion is repaired by BRCA1 and BRCA2-dependent homologous recombination. C) Role of PARP1/2 in fork reversal. A replication fork encounters an obstacle to its progression and reverts. PARP is activated either by the obstacle/lesion itself or by the DNA end at the regressed fork. PAR chains prevent RecQ1 binding/activity. Upon resolution of the block, PARP release (and presumably PAR chain degradation by PARG) allow RecQ1 helicase to access the reversed fork and remodel it back into a canonical replication fork. somerase poisons (Hoch et al., 2017). Surprisingly, PARP2 seems unable to modify the same target sites as PARP1, suggesting that this redundancy is indirect (Leslie Pedrioli et al., 2018). Intriguingly, a recent study suggested that PARP2 extends PARP1-generated PAR chains, introducing branching points that are recognised by branchingspecific factors .
The other DNA-dependent ADP-ribosyl transferase PARP3, although activated by DNA breaks in vitro, has less clear roles in DNA repair, and has been implicated in double-strand break (DSB) repair by non-homologous end-joining (Rulten et al., 2011), particularly during IgG class switching (Robert et al., 2015), regulation of DSB repair pathway choice (Beck et al., 2014b) and most recently the repair of G4-quadruplex containing DNA lesions (Day et al., 2017;Layer et al., 2018).
ADP-ribosylation also controls telomere length. TRF1, a telomere-binding protein, is PARylated by the tankyrase PARP5a (or TNKS1), which reduces its affinity for the telomere and allows telomerase to access the DNA end for elongation (Smith et al., 1998). Similarly, PARP2 has been shown to contribute to telomere homeostasis by modifying TRF2 (Dantzer et al., 2004). Other PARPs, such as PARP9, PARP10 and PARP15 also play roles in DNA repair (Yan et al., 2013;Nicolae et al., 2014;Nicolae et al., 2015), suggesting that the interplay between ADP-ribosylation and genomic stability may be even more extensive than currently known.

DNA-dependent PARPs and chromatin
PARP1 can be thought of as an integral component of chromatin that modifies chromatin structure directly (Clark et al., 2012). For example, PARP1 was shown to compete with histone H1 for binding to linker DNA (Poirier et al., 1982;Kim et al., 2004) and is reported to have intrinsic histone chaperone activity in vitro, mediated in part by the highly negatively charged nature of the PAR polymer (Muthurajan et al., 2014). This is further illustrated by the extensive PARP1-dependent modification of core and linker histones (Boulikas, 1988), as well as the existence of H2A variants with PAR-binding domains that may well mediate long-range PARP-dependent chromatin interactions (Timinszky et al., 2009). PARP1 also regulates chromatin accessibility indirectly by recruiting chromatin remodellers, such as ALC1, SMARCA5 and CHD2 (Ahel et al., 2009;Smeenk et al., 2013;Luijsterburg et al., 2016). Interestingly, processing of PARP1-generated PAR chains by PARG and NUDIX5 hydrolases has been suggested to provide a localised pool of ATP in the nucleus for ATP-consuming chromatin remodelling complexes (Wright et al., 2016). PARP1 activation at gene promoters also controls the induction of transcription, such as chromatin "puffing" of heat shock-inducible genes in Drosophila polytene chromosomes (Tulin and Spradling, 2003), and at gene promoters responsive to transcription factors such as NFkB, PPARg and hormone receptors such as ER, AR and RAR (Kraus and Hottiger, 2013). Recently, PARP1 has also been linked with the regulation of RNA polymerase II pausing via the negative elongation factor NELF-E (Gibson et al., 2016). However, a more precise understanding of the molecular mechanisms involved in most of these processes and reconciliation with the fact that PARP1 KO mice have very mild phenotypes are still lacking.
Although the partial redundancy between PARP1 and PARP2 is clear for DNA damage-related functions, whether this extends to chromatin remodelling and transcriptional regulation is unclear. A screen for PARP2 targets revealed an enrichment of proteins associated with transcriptional regulation and RNA splicing, suggesting this might be the case. Similarly, targets of PARP3 were enriched in RNA processing, transcription and chromatin organization (Bartolomei et al., 2016), suggesting that all three DNA-dependent PARPs may well be involved in important DNA-damage independent aspects of chromatin biology.

ADP-ribose in cellular signalling
ADP-ribosylation is also involved in controlling several signalling cascades, such as Wnt/b-catenin, NFkB and the unfolded protein response. The two tankyrases PARP5a and PARP5b (TNKS1 and TNKS2) PARylate axin, a central component in the b-catenin destruction complex, leading to its proteasomal degradation via RNF146, a PARdependent E3 ubiquitin ligase (Huang et al., 2009). Wnt signalling is further promoted by PARP10-dependent mono-ADP-ribosylation of GSK3b, which inhibits its kinase activity and also stabilises b-catenin . PARP10 additionally suppresses NFkB signalling via MARylation and inactivation of NEMO , and PARP16 was shown to MARylate and activate PERK and IRE1a, central signalling hubs in the unfolded protein response in the endoplasmic reticulum (Jwa and Chang, 2012).
Many PARPs are involved in cellular antiviral mechanisms, with PARP7, PARP9, PARP12 and PARP14 all implicated in the interferon response, and PARP13 is involved in direct degradation of viral transcripts (Atasheva et al., 2014;Welsby et al., 2014;Zhang, Y. et al., 2015;Iwata et al., 2016). Intriguingly, many of these enzymes, as well as PARP4 and PARP15, are under diversifying selective pressure in primates, suggesting an ADP-ribose "arms-race" between hosts and viral pathogens (Daugherty et al., 2014).
With the recent development of better tools to detect ADP-ribose modification of proteins (Chang, 2018), many additional roles of ADP-ribosylation in a variety of cellular signalling pathways are likely to emerge in coming years.

PARP1 and cell death
Active PARP1 produces large amounts of PAR and at high levels of DNA damage up to 80% of the cellular NAD + pool can be depleted within 5-15 min (D'Amours et al., 1999). Since NAD + is necessary for glyceraldehyde 3phosphate dehydrogenase activity during glycolysis (Tan et al., 2013), a reduction in NAD + leads to lower pyruvate production, reducing carbon flow into the mitochondrial TCA cycle, and hence ATP production. Conversely, ATP is required for NAD + synthesis, and therefore the uncontrolled use of NAD + by PARP1 can lead to a bioenergetic collapse (Figure 3). PARP1 hyperactivation initiates a programmed cell death pathway termed parthanatos, which is independent of canonical apoptosis, necrosis or autophagy (Yu et al., 2002;Galluzzi et al., 2018) and is mediated by the apoptosisinducing factor (AIF) (Yu et al., 2002;Andrabi et al., 2006) ( Figure 3). AIF is a mitochondrial membrane-anchored protein that is required for the assembly of the mitochondrial electron transport chain and exists in an equilibrium between monomeric and dimeric forms, with NAD(H) binding favouring dimer formation (Brosey et al., 2016). Upon PARP1 hyperactivation, AIF is released from the mitochondria and translocates to the nucleus to drive parthanatos (Yu et al., 2002;Otera et al., 2005). How this occurs is currently unclear, but surprisingly, the AIF transmembrane fragment does not need to be cleaved (Wang, Y. et al., 2009). One possibility is that NAD + depletion itself promotes AIF release either by mitochondrial dysfunction (Alano et al., 2010;Baxter et al., 2014) or by inducing conformational changes in AIF (Sevrioukova, 2009;Brosey et al., 2016). Alternatively, a direct interaction between AIF and protein-free PAR polymers has been suggested to mediate AIF release (Andrabi et al., 2006;Wang, Y. et al., 2011). Free AIF then promotes translocation of the nuclease MIF to the nucleus, which cleaves genomic DNA inducing cell death (Wang, Y. et al., 2016a) (Figure 3). Many of the molecular mechanisms of parthanatos remain to be clarified, but an improved understanding of this pathway is critical for the development of novel treatment avenues for a potentially large number of diseases (see below).
Human diseases and therapeutic opportunities PARP1/2 inhibition and HR-defective cancer DNA-activated PARPs, particularly PARP1, became attractive drug target candidates for cancer therapy in 2005 when PARP inhibition (PARPi) was shown to induce synthetic lethality in cells lacking BRCA1/2 Farmer et al., 2005). As discussed above, in the absence of PARP1-dependent SSBR, unrepaired singlestrand breaks are converted into DSBs by the passage of a replication fork, leading to a distinct requirement for BRCA1/2-dependent homologous recombination (HR) ( Figure 2B). BRCA genes are tumour suppressors that are frequently mutated in breast and ovarian cancers, and four compounds (rucaparib, niraparib, olaparib and talazoparib) are currently licenced by the U.S. Food and Drug Administration (FDA) for treatment of BRCA-defective cancers (O'Connor, 2015;Bitler et al., 2017). These inhibitors bind to the nicotinamide binding site in the catalytic domain, mimicking the three H-bonds established by the nicotinamide group from NAD + . By blocking PARP catalytic activity, these compounds slow single-strand break repair in two ways: a) the lack of PARylation surrounding break sites delays the recruitment of DNA repair factors such as XRCC1; and b) by preventing PARP1 auto-modification that is required for release of the protein from the DNA break (D'Amours et al., 1999). Thus, these inhibitors lock or "trap" the enzyme bound to the DNA, preventing the access of other enzymes to the break Pommier et al., 2016;Lord and Ashworth, 2017). Novel inhibitors that induce more stable trapping of PARP seem to be better inducers of synthetic lethality in BRCA-mutated cells, suggesting that this trapping effect is crucial for PARP inhibitor efficacy (Murai et al., 2014).
The clinical success of PARP inhibitors in BRCA1/2-mutated breast and ovarian cancers has ignited a push for more widespread use of these compounds in cancers with a molecular signature of defective HR, irrespec-ADP-ribose mechanisms and disease 7 tive of which HR gene is mutated and in which tissue the tumour originated (Pilie et al., 2019). Similarly, novel inhibitors that selectively target different PARPs, including PARP3, PARP5a/5b, PARP7, PARP10, PARP11 and PARP14 are under investigation for the targeted treatment of cancers with alterations in particular pathways (Ishida et al., 2006;Lindgren et al., 2013;Iwata et al., 2016;Wang, Y. Q. et al., 2016b;Ferri et al., 2017;Yoneyama-Hirozane et al., 2017;Kirby et al., 2018;Moustakim et al., 2018;Murthy et al., 2018).
Remarkably, PARP1 inhibitors may also be of significant therapeutic value for non-oncological use both in rare neurological disorders in which excessive PARP signalling seems to be detrimental, as well as in more prevalent degenerative diseases in which parthanatos seems to play a central pathological role (Berger et al., 2018) (discussed below).

ADP-ribosylation in genetic neurodegenerative disorders
Mutations in single-strand break repair genes, such as PNKP, APTX, TDP1 and XRCC1, cause genetic neurodegenerative disorders characterised by severe cerebellar atrophy and ataxia (Moreira et al., 2001;El-Khamisy et al., 2005;Bras et al., 2015;Hoch et al., 2017). Treatment of cells from these patients with DNA damaging agents leads to excessive PARP1 activation, suggesting that defective single-strand break repair leads to overt signalling of these lesions (Hoch et al., 2017). As deletion of PARP1 partially rescued many of the cerebellar defects observed in XRCC1-deficient mice, it was suggested that PARP1-induced parthanatos and/or NAD + depletion contributes to disease pathology (Hoch et al., 2017). Although PARP1 inhibition should in principle be beneficial in this scenario, the currently available PARP1 inhibitors are unlikely to be of therapeutic value, as the PARP1 trapping effect (discussed above) further compounds the DNA repair defect in these cells (Hoch et al., 2017). In this context, inhibitors that better mimic genetic deletion of PARP1 would be desirable.
Mutations in enzymes involved in removing ADPribose modifications also leads to neurodegenerative disease, as illustrated by the identification of patients with mutations in the hydrolases ARH3 and TARG. ARH3 has specificity for both poly-ADP-ribose chains as well as mono-ADP-ribose moieties attached to serines (Abplanalp et al., 2017;Fontana et al., 2017), whereas TARG hydrolyses the ester linkage between mono-ADP-ribose and aspartate or glutamate side chains (Sharifi et al., 2013). ARH3 mutations are associated with neurodegenerative defects such as ataxia and febrile seizures, while TARG1 loss causes severe developmental delay, epilepsy and quadriplegia (Sharifi et al., 2013;Danhauser et al., 2018;Ghosh et al., 2018). Whereas TARG deficient cells shown signs of DNA-repair defects, a role for ARH3 in DNA damage re-sponses is speculative at this point, although serine has been recently established as the primary acceptor of DNA damage-induced ADP-ribosylation (Palazzo et al., 2018). If excessive PAR formation, NAD + depletion and/or parthanatos are also involved in promoting the neurological defects seen in these patients, currently available catalytic PARP1 inhibitors may well be a viable therapeutic option (Danhauser et al., 2018;Ghosh et al., 2018).
Mutations in PARP10 lead to a neurodegenerative disorder associated with developmental delay and cortical atrophy, as well as delayed myelination (Shahrour et al., 2016). Although a defect in PARP10-dependent Wnt or NFkB signalling was not determined, patient cells had a DNA repair defect in response to hydroxyurea (HU) and ultraviolet light (UV), and the pathology is reminiscent of other DNA repair disorders (Shahrour et al., 2016). A more detailed understanding of the cellular consequences of PARP10 loss and which of its many functions is most important to prevent disease onset and progression will be critical to suggest possible therapeutic avenues for this disease.

Parthanatos inhibition
PARP1-dependent cell death via parthanatos has been implicated in several critical pathological processes, such as ischemia-reperfusion injury in myocardial infarction and stroke, septic shock, brain trauma and neurodegenerative diseases such as Parkinsons disease and Alzheimers disease (Pacher and Szabo, 2007;Moroni, 2008;Lee et al., 2013;Dawson and Dawson, 2017;Berger et al., 2018;Henning et al., 2018;Kam et al., 2018;Zhang, J. et al., 2018). A common theme among these disorders seems to be PARP1 hyperactivation in response to oxidative DNA damage, either as part of the reperfusion of oxygen-deprived tissues or caused by pathophysiological changes that induce the production of reactive oxygen species or nitric oxide.
Mounting pre-clinical evidence suggests that PARP1 knockout or PARP inhibitor treatment have profound beneficial effects in mouse models of parthanatos-induced pathologies, preventing cell death and tissue dysfunction (Pacher and Szabo, 2007;Dawson and Dawson, 2017;Berger et al., 2018;Henning et al., 2018). These results have prompted calls for clinical trials to repurpose PARP inhibitors for the treatment of these disorders, particularly when no other viable treatment option exists (Berger et al., 8 Hoch and Polo 2018). Conceptually, targeting other steps in parthanatos, such as preventing mitochondrial AIF release or inhibiting nuclear import or activation of the MIF nuclease may also be of therapeutic value (Figure 3). Although these compounds would have to undergo extensive pre-clinical and clinical efficacy and safety trials, their development may be warranted by a reduced potential for DNA repair-associated side-effects of systemic PARP inhibition during chronic treatment.

Concluding remarks
Detailed knowledge of the processes and pathways regulated by post-translational modifications such as phosphorylation and ubiquitination led to the development of a myriad of kinase inhibitors and molecules targeting the ubiquitin system, either already in clinical use or in clinical trials (Ferguson and Gray, 2018;Wertz and Wang, 2019). In recent years, novel tools to study ADP-ribosylation have allowed a rapid development in this field, characterising many of the "writers", "erasers" and "readers" of this modification. Taken together with the fact that the first PARP inhibitor only entered the clinic in 2014, this raises the exciting prospect that a more detailed understanding of ADP-ribose metabolism, particularly of less well-studied PARPs and hydrolases, may well yield novel therapeutic strategies in coming years.