Poly(ADP-ribosyl)ation is involved in the regulation of a variety of cellular pathways, including, but not limited to, transcription, chromatin, DNA damage and other stress signalling. Similar to other tightly regulated post-translational modifications, poly(ADP-ribosyl)ation employs ‘writers’, ‘readers’ and ‘erasers’ to confer regulatory functions. The generation of poly(ADP-ribose) is catalyzed by poly(ADP-ribose) polymerase enzymes, which use NAD+ as a cofactor to sequentially transfer ADP-ribose units generating long polymers, which, in turn, can affect protein function or serve as a recruitment platform for additional factors. Historically, research has focused on poly(ADP-ribose) generation pathways, with knowledge about PAR recognition and degradation lagging behind. Over recent years, several discoveries have significantly furthered our understanding of poly(ADP-ribose) recognition and, even more so, of poly(ADP-ribose) degradation. In this review, we summarize current knowledge about the protein modules recognizing poly(ADP-ribose) and discuss the newest developments on the complete reversibility of poly(ADP-ribosyl)ation.
Poly(ADP-ribosyl)ation (PARylation) is a highly abundant protein post-translational modification involved in a variety of essential cellular functions: DNA repair, transcription, chromatin structure alterations, cell cycle progression and division, replicative ageing, apoptosis and necrosis [1-6]. Subsequent to the discovery of poly(ADP-ribose) (PAR) five decades ago, PARylation has been most extensively studied in the context of DNA damage, where three DNA damage inducible PAR polymerases (PARPs), namely PARP1 (ARTD1), PARP2 (ARTD2) and PARP3 (ARTD3), play a role [7-10].
PARPs belong to a large family of proteins (17 recognized members in humans), which, in turn, is part of an ADP-ribosyl transferase (ART) superfamily [5, 11]. Although PARPs were assumed to be exclusively eukaryotic (with the exception of yeast), diverged orthologues were recently discovered in a number of bacterial species . Currently, at least 14 bacterial genomes appear to possess those diverged orthologues of PARP, which are considered to have been acquired through horizontal gene transfer [11, 12]. Human PARPs can be divided into three main groups: (a) the H-Y-E catalytic triad containing PARPs 1–5 capable of PARylation; (b) PARPs 6–8, 10–12 and 14–16 only capable of mono-ADP-ribosylation; and (c) PARP9 and PARP13, which are inactive as a result of the lack of NAD+ binding residues [1, 5].
PARP1 is the main producer of PAR within the cell and, despite its high abundance (0.2 to 1 × 106 copies per cell), the levels of PAR in the absence of DNA damage are low as a result of its rapid turnover and the tight regulation of PARP1 catalytic activity [9, 13-16]. PARP1 has a modular structure, comprising six domains, which can in turn be grouped into three main domains: the N-terminal DNA-binding domain, a central automodification domain and a C-terminal catalytic domain [3, 4, 17]. Recently, structure–function analyses of PARP1 shed light on its DNA-binding mode and subsequent catalytic activation [4, 17-20]. The automodification domain is the major acceptor of PAR chains and it also bears a non-essential BRCT-domain, playing a role in protein–protein interactions [3, 21]. Upon sensing single- and double-strand DNA breaks, PAR levels are rapidly elevated. The major acceptor of PAR is PARP1 itself, although a number of other proteins are also modified (e.g. DNA ligases, p53, histones), allowing their rapid recruitment to an appropriate site or altering their function accordingly in response to the signal [3, 22-25].
Protein PARylation starts with the transfer of an ADP-ribose (ADPr) moiety from NAD+ onto a protein acceptor, with subsequent release of nicotinamide and a proton (Fig. 1). The initial ADPr group is transferred primarily onto glutamate and aspartate residues via an ester bond [21, 26-29], although there are reports showing that lysines can act as acceptors as well [30, 31]. After the production of the initial ester bond, repeating units of ADPr are attached via a unique 2′,1″-O-glycosydic ribose–ribose bond to produce long (up to 200 ADPr units) linear chains of PAR in vitro, with branching every 20–50 residues (Fig. 1) [3, 32-35]. There is no definitive insight, however, into the length and branching of PAR in vivo.
A key feature of many covalent post-translational modifications is their recognition by ‘reader’ modules: specific proteins and their (usually globular) protein domains. Protein PARylation is no exception. A large number of proteins, rather than being PARylated, bind PAR, allowing their recruitment to an appropriate site or the alteration of their function in response to the signal [3, 22]. To date, the known PAR-binding proteins can be classified into four distinct classes of PAR recognition modules: PAR-binding linear motifs (PBMs), macrodomains, PAR-binding zinc-fingers (PBZs) and the WWE domains [1, 22].
PARylation regulates many important biological processes, which implies that the signal must be efficiently recognized and removed in a timely manner. Thus, PAR is a highly dynamic post-translational modification, with a half-life in the range 1–6 min . Macrodomain proteins are the only PAR-binding proteins also capable of processing ADPr. Poly(ADP-ribose) glycohydrolase (PARG) is one example of such a PAR-processing macrodomain . Along with ADP-ribosylhydrolase (ARH)3, a significantly less active PAR-degrading enzyme, PARG, accounts for the rapid removal and homeostasis of PAR, leading to mono-ADP-ribosylated protein products (Fig. 1) [37-39]. The identity of the enzyme that removes the terminal ADP-ribosyl-hydrolase has, until recently, remained elusive.
The half-life of protein mono-ADP-ribosylation is significantly longer compared to PARylation and is proposed to be the rate-limiting step in PAR hydrolysis . The failure to hydrolyse the ester bond between the terminal ADPr and the glutamate residue has been implicated in a human hereditary neurodegenerative disease with renal failure . Although previously purified from rat liver, the protein responsible for the removal of the proximal ADPr, attached via an ester bond, had not been identified . However, macrodomains were recently identified as the missing enzymes capable of cleaving the last, glutamate-linked ADPr, further broadening our understanding of PAR metabolism [29, 42, 43]. Finally, once released, ADPr, can cause protein glycation and glycoxidation, although, normally, it is catabolized to AMP and ribose 5-phosphate .
In the present minireview, we focus on current knowledge about the various PAR-binding protein modules and the enzymes that hydrolyse the PAR signal.
PAR recognition protein domains
PBMs were discovered first and are the most abundant group of PAR-binding motifs, with more and more cases of in vivo functions being documented. The similarity of PAR to other nucleic acids and the fact that basic amino acids frequently play an important role in the recognition and binding of DNA and RNA, combined with the observations that some proteins (e.g. histones and p53)  strongly and noncovalently associate with PAR, led to the discovery of a 22–25 amino acid long motif that can interact with PAR. The observation that many proteins involved in DNA damage response and DNA repair carry such a motif led to the suggestion that PAR serves as a general recruitment platform for other proteins . The first consensus motif where positively-charged residues are followed by an HxBxHHBBHHB sequence (where ‘B’ is basic, ‘H’ is hydrophobic, and ‘x’ is any amino acid) has been suggested  and several peptides containing such a motif were shown to bind PAR in vitro.
The consensus sequence of PBMs was later refined by Gagne et al. , leading to a better-defined motif: [HKR][X][X][AIQVY][KR][KR][AILV][FILPV]. The in silico prediction of PAR-binding proteins using the newly-defined PBM again identified many DNA repair proteins, amongst others. However, not all newly-identified peptides interact with PAR, nor do all peptides interacting with PAR fit the proposed motif, reducing the direct usability of such a motif for PAR-binding prediction. Despite the difficulties with in silico prediction, a MS-based screen uncovered PAR-binding proteins with functions in DNA and RNA metabolism . Although the structural basis of PAR recognition by PBMs remains unknown, owing to the difficulty of determining the structures of complexes formed by two intrinsically flexible entities (PAR and the PBM), the interaction between PAR and PBMs can be quite strong, with an equilibrium dissociation constant (Kd) often in the nanomolar range . PBMs are considered to primarily function as recruitment motifs bringing proteins to sites of PARylation, especially to DNA lesions, where PARP1 is activated. In addition, the PBMs of p53 appear to modulate its DNA-binding function . Arguably, the most important PBM function was demonstrated for apoptosis-inducing factor, where the identified PBM was necessary for PAR-mediated cell death .
In addition to the usual PBMs motifs, some alternative ones were identified. Among those, particularly interesting are the RNA recognition motifs. The specificity for PAR over RNA, however, has been questioned as a result of the similarities between both molecules. Yet the ability of the domain to bind both RNA and PAR provides a way of controlling protein function, where PAR could compete with RNA and thus prevent protein functions such as RNA splicing . Furthermore, some chromatin remodellers (such as chromodomain-helicase-DNA-binding protein 4 in humans and Drosophila Mi-2) were found to carry alternative PAR-binding motifs and these motifs were shown to be required for the correct localization of these ATP-dependent nucleosome remodelling enzymes [51, 52].
An excellent recent review by Krietsch et al.  covers current knowledge about PBMs in great detail.
Macrodomains: globular ‘readers’ and ‘erasers’ of PARylation
The evolutionary ancient macrodomain was named after the histone variant macroH2A, where, in addition to the histone fold, an additional 140–190 amino acid conserved domain was detected and shown to play a role in X chromosome inactivation [54, 55]. Macrodomains are found in all kingdoms of life, including some viruses . In humans, macrodomains appear in proteins together with other domains, such as the PARP catalytic domains, the histone fold domain, sucrose nonfermenting 2 helicase ATPase domain, cellular retinaldehyde-binding and triple functional domain, or as stand-alone domains. Many human macrodomain containing proteins have been linked to ADP-ribosylation. The first biochemical characterization and evidence for macrodomain modules binding ADPr came from the structure of Af1521 from a thermophilic archaea [56-58]. Together with subsequent structural data, this revealed that the macrodomain is a globular domain comprising a six-stranded mixed β-sheet and five α-helices, which form a cleft for the nucleotide ligand, where binding occurs through stacking interaction with the adenine ring, strengthened via the interactions with the pyrophosphate of ADPr, and with specificity provided by hydrogen bonding with the distal ribose [56, 59-62]. Sequence differences among macrodomains result in their preference for different NAD+ metabolites. Interaction studies and structural work demonstrated that macrodomains can recognize O-acetyl-ADP-ribose (OAADPR) in addition to ADPr and also can recognize mono- or poly-ADP-ribosylated proteins, linking macrodomains directly to ADP-ribosylation and/or PARylation pathways. The chromatin remodelling enzyme, ALC1, for example, binds PAR via its macrodomain in response to DNA damage, whereas macroH2A.1.1 can also bind mono-ADPr and OAADPR [24, 25, 60, 62]. Importantly, OAADPR binding of macroH2A.1 is regulated by alternative splicing  and the macroH2A splice variants were shown to predict lung cancer reoccurrence [60, 63]. MacroH2A was shown to be an epigenetic regulator fine-tuning the activation of the HOXA cluster genes during neuronal differentiation . Moreover, macroH2A.1 is important for the differentiation capacity of stem cells and the loss of macroH2A isoforms increases the malignancy of melanoma cells [65, 66]. There are now at least 11 macrodomain-containing proteins present in humans, which have one to three repeats of the module. Interestingly, some positive-strand RNA viruses also carry macrodomains, suggesting that viruses might affect cellular PARylation signalling or hijack NAD+/PAR metabolism of the the host cell [67-69]. This idea is further supported by recent reports of specific PARPs involved in viral infection signalling .
Macrodomains can also exhibit catalytic activities, which makes them unique among the known PAR-binding modules. Even before the structure of any macrodomain was known, Martzen et al.  reported enzymatic activity attributed to this domain in yeast, namely the hydrolysis of ADP-ribose-1″-phosphate, a byproduct of tRNA splicing, which was also shown for the Af1521 protein . More recently, three human macrodomains, namely MacroD1, MacroD2 and C6orf130, were shown to hydrolyze OAADPR, which added to the catalytic activities and repertoire of macrodomain function [59, 61]. Furthermore, the list of macrodomains proteins was further extended by the unexpected structural discovery that the key PAR metabolizing enzyme PARG adopts the macrodomain fold . Finally, the enzymatic activities of macrodomains were further extended by the finding that MacroD1, MacroD2 and C6orf130 can fully reverse protein ADP-ribosylation by cleaving glutamate-linked ADPr (Fig. 1) [29, 42, 43]. In brief, macrodomains can function as PAR-binding modules and/or hydrolytic enzymes releasing ADPr.
PBZs cooperate in DNA damage
Zinc fingers constitute another family of structural motifs common in nucleic acid recognition, where Zn2+ ion coordination stabilizes the fold. Although zinc fingers were long considered as binding modules for DNA, RNA and even proteins, PBZs were identified only recently . The first evidence came from the identification of the zinc finger domains in the aprataxin PNK-like factor (APLF) and the checkpoint protein with FHA and RING domains (CHFR). In APLF, they are essential for the protein's localization to DNA lesions in a PARylation-dependent manner [72, 73], whereas a related zinc finger domain is also necessary for the function of CHFR . This new zinc finger family was named the PBZ domain.
PBZs are evolutionary conserved in all eukaryotes except yeast, essentially comprising all species known to have active PARylation. PBZs are exclusively found in proteins related to DNA repair or checkpoint activity. In mammals, only three proteins, APLF, CHFR and DNA cross-link repair 1A (sensitive to nitrogen mustard 1A; SNM1A), have PBZ domains. In other organisms, PBZs are more widespread, yet appear to be limited to proteins involved in DNA repair; for example Ku70, PARPs and Rad17 in Dictyostelium discoideum or DNA ligase in Caenorhabditis elegans . The PBZs mostly occur only once within a protein, with the few exceptions of proteins possessing tandem PBZs. The tandem organization of PBZs in APLF was shown to strongly increase the overall affinity for PAR, although single PBZs can recognize PAR individually .
The first structural insights into the recognition of PAR by PBZs came from the study of the APLF PBZs . The adenine of an ADPr unit of PAR stacks with conserved tyrosine residues in the PBZ, and the proximal ribose of the same ADPr is recognized mostly via main-chain interactions. However, the specificity for PAR is mostly achieved through the recognition of the distal ribose and the pyrophosphate of the adjacent ADPr unit (Fig. 1). For some PBZs, a second adenine binding pocket was observed, allowing binding of two consecutive ADPr units, contributing to an even higher specificity for the PAR polymer [75, 76]. The structure of the PBZ domain from CHFR was also determined in the context of a bigger cysteine rich domain, where the PBZ, in addition to PAR binding, was also found to contribute to the overall fold, despite the PAR-binding residues residing solely within the PBZ .
PBZs serve as a recruitment domain for proteins to the sites of PARylation (e.g. upon DNA damage) [77-79]. Additionally, they might recruit the proteins to the vicinity of active PARPs to be PARylated (CHFR, APLF) , or even link PARylation and ubiquitylation pathways  in a similar fashion to that shown for the WWE domains (see below). Although the third protein bearing PBZ in mammals, SNM1A, is also involved in DNA repair, its PBZ domain lacks aromatic residues critical for PAR recognition and is predicted not to bind PAR . Instead, this specific PBZ might carry a different function.
WWE domains link PARylation and ubiquitylation
The WWE domain, named after its three most conserved residues, is the latest addition to the PAR-binding motifs. Despite the first structure of a WWE domain showing a role for WWE in protein–protein interactions (Deltex–Notch in the fruit fly Drosophila) , WWE domains have also been shown to specifically recognize PAR. The presence of WWE domains in proteins bearing E3 ligase or PARP catalytic domains suggested an involvement in ubiquitylation and PARylation pathways . The proof that WWE domains participate in PAR recognition came from the findings that Iduna (RNF146), a PAR-directed E3 ligase, regulates PARylated protein levels by ubiquitylation-mediated proteasomal degradation (notably the targets of tankyrase 1 and 2, as well as proteins involved in DNA repair that are PARylated by PARP1) [83, 84]. Interestingly, Iduna confers a protective role against PAR-induced cell death in a PAR-binding-dependent manner .
PAR binding by Iduna was mapped within the WWE domain, which was initially considered to contain a PBM . However, the proposed PBM is not conserved in other WWE domains, which prompted Wang et al.  to structurally investigate the WWE domain and its specificity for PAR. The iso-ADPr bound structure resolves the specificity of WWE domain for PAR over mono-ADPr. Although the adenine ring inserts into a pocket, charged surfaces on both sides of the pocket make important contacts to the phosphate groups that belong to different ADPr units within the same PAR molecule. Thus, the β-phosphate of one ADPr unit and the α-phosphate of the following ADPr unit are required for binding by the WWE domain (Fig. 1). Based on the structure, multiple sequence alignments allow the prediction of WWE domains that can bind PAR. Particularly, according to experimentally validated predictions, the WWE domain of PARP11 binds PAR, whereas the WWE domain of PARP14 cannot [86, 87].
interpro searches (http://www.ebi.ac.uk/Tools/pfa/iprscan/) show 11 WWE domain proteins in humans: six are associated with an E3 ubiquitin ligase domain, five with the PARP catalytic domain and one with the phospholipase domain. Essentially, WWE domains appear to provide a link between PARylation and ubiquitylation pathways. The importance of such connection is illustrated by the molecular cause of Cherubism, which is a disease where a mutation in SH3 domain-binding protein 2 disrupts its PARylation and, in turn, ubiquitylation-mediated degradation. This leads to SH3 domain-binding protein 2 accumulation, which, in turn, is the molecular basis for the pathomechanism .
The enzymatic removal of covalently linked PAR from proteins
PARG: a diverged macrodomain hydrolysing PAR
Removal of PAR by PARG was first recognized four decades ago in calf thymus and rat liver extracts; however, as a result of a low abundance of the enzyme, as well as a lack of specific antibodies and inhibitors, the majority of scientific interest was initially focused on PAR production rather than its removal, with the first PARG from Bos taurus cloned only in the late 1990s [37, 39, 89]. Despite its lower abundance compared to PARP1, PARG cleaves unique O-glycosidic ribose-ribose bonds within PAR very rapidly and PAR is degraded in a timescale of minutes [36, 90]. There are at least five PARG isoforms in humans, resulting from alternative splicing. After the initial cloning of PARG, increasingly, more evidence began to emerge supporting the localization of different PARG isoforms to different cellular compartments [89, 91]. The only PARG localized to the nucleus is the full-length isoform of 110 kDa, whereas other splice variants are found in the cytoplasm and mitochondria [92-96]. Yeast and bacteria were historically considered to be devoid of PAR metabolism (i.e. to lack both PARP and PARG orthologues). However, recent data indicate that several bacterial species contain close PARP1 homologues . Over fifty currently sequenced bacterial species were noted to also contain a short, highly diverged form of PARG, now annotated as bacterial-type PARG (bactPARG). BactPARG is not only present in bacteria, but also is found in eukaryotic organisms where PARPs (but no PARGs) were previously identified, such as filamentous fungi, Aspergillus fumigatus, Neurospora crassa and mushrooms (Fig. 2) [12, 97, 98]. In addition to filamentous fungi, several other eukaryotic organisms also contain this diverged bactPARG, such as the rotifer Adineta vaga (where PARG is fused to several repeats of the PBZ PAR-binding motif) and several protozoan species. The genomes of some eukaryotes, such as C. elegans, Arabidopsis thaliana or Tetrahymena thermophila encode more than one canonical PARG homologue [99, 100]. Similarly, several species were found to contain more than one bactPARG (e.g. Aspergilus fumigatus) or even a combination of both bacterial and canonical PARG orthologues (e.g. sea anemone Nematostella vectensis and fungus Gibberella).
These bacterial-type PARGs are as efficient as canonical PARGs in degrading PAR and rely on the same catalytic mechanism (see below). Effectively, the bactPARGs can be seen as the minimal catalytic PARG architecture (Fig. 2). The available structure of the bactPARG (from the thermophilic bacterium Thermomonospora curvata) revealed that the catalytic region of PARG consists of a macrodomain module with a small N-terminal extension. The diphosphate binding loop (GVFG motif) present on one side of the ADPr binding cavity is highly conserved among other macrodomain proteins. The opposite side of the ligand-binding cavity contains a PARG specific loop with signature sequence GGG-X6–8-QEE, bearing the two vicinal glutamate residues essential for PARG catalytic activity [12, 101]. ADP-α-ribose occupies a position similar to other macrodomain proteins; however, PARG-like macrodomains specifically position the PARG signature loop, in particular the second essential glutamate, allowing direct hydrogen contact with the 2′-OH group of the distal ribose.
The bactPARG: the elucidation of PARG exo-glycohydrolase activity
Similar to the human histone macroH2A.1.1 variant, as a result of steric constraints, the bacterial PARG only appears to be capable of binding the terminal unit (referred to as N ADP-ribose) of the PAR chain (Fig. 3A,D) [12, 62]. Structural and mutational analysis of bactPARG allowed a catalytic mechanism to be proposed for the exo-glycohydrolase activity of PARGs. Binding of N ADP-ribose positions the unique O-glycosidic ribose-ribose bond between N and N-1 ADP-ribose units in direct hydrogen contact with the two essential glutamates from the PARG signature sequence. The second glutamate residue (Glu115 in T. curvata PARG; Protein Data Bank code: 3SIG) protonates the N-1 ADP-ribose 2′-OH leaving group, attached to the rest of the PAR chain, which results in the formation of a positively-charged oxocarbenium intermediate. This intermediate is stabilized by the steric constrains imposed by the conserved phenylalanine residue (Phe227; Phe875 in human PARG), which allows proximal positioning to the negatively-charged phosphate groups. This highly conserved Phe residue is present on the macrodomain-wide GVFG motif. The unstable oxocarbenium intermediate is then nucleophilically attacked by a water molecule, ultimately resulting in the release of ADPr and the remainder of the PAR chain .
Although the structure of the bactPARG revealed the catalytic mechanism of the exo-glycohydrolase mode of PARG, it could be argued that, as a result of their higher complexity, canonical PARGs may have a different catalytic mechanism; for example, allowing them to carry out endo-glycohydrolytic cleavage. The possibility that PARG enzymes may carry out endo-glycohydrolase activity has been debated even before the enzyme had been cloned but, until now, no general consensus has been reached [102-104]. Although the bactPARG structure shows a lack of intra-chain ADPr binding, the subsequently published structures of several canonical PARGs suggest the potential for such binding [105-107].
The canonical PARG: the potential function of PARG as an endo-glycohydrolase
Canonical PARGs are larger and more complex than their bacterial counterparts. These more composite PARGs show very high evolutionary conservation, especially in the catalytic region (Figs 2 and 3B,C,E,F) and are conserved throughout eukaryotic species, from vertebrates to protozoans. Within the catalytic region, in addition to the macrodomain, canonical PARGs also contain a short C-terminal extension and an accessory domain of unknown function. This accessory domain is less conserved than the macrodomain; nevertheless, this domain is retained by all canonical PARGs (Fig. 2) [92, 105, 107]. Although the function of the accessory domain is yet to be determined, it appears to structurally stabilize the PARG active site. Further towards the N-terminus, vertebrate PARGs also bear a short segment containing a mitochondrial targeting sequence (MTS) and a large regulatory region (1–456 in rat and 1–460 in human PARG). The latter region is poorly characterized and is dispensable for PARG activity in vitro [12, 92, 105, 107]. Recently, a proliferating cell nuclear antigen-binding motif was identified in the regulatory region of PARGs that acts as an additional recruitment platform to DNA damage sites, further tightening the control between DNA repair and cell death pathways . The MTS motif was shown to be important for catalytic activity [92, 105] and was proposed to play a role in stabilizing a tyrosine clasp element, which is present in the active sites of all canonical PARGs [106, 107].
All three canonical PARG structures reveal additional interactions with ADPr. One of these is the previously reported base stacking interaction between the adenine moiety of ADPr and a tyrosine residue (Tyr795 in human PARG and Tyr296 in protozoan PARG), which allows for a more intricate recognition of the substrate and binding selectivity for adenine nucleosides [105-107, 109]. This tyrosine residue is located in the aforementioned tyrosine clasp (YTGYA; residues 792–796 in human PARG), which is absent in bactPARG. It was proposed that this tyrosine clasp is flexible. However, comparison with other available structures reveals a low likelihood for such mobility. Despite more significant re-arrangements of the active site upon binding of ADPr (i.e. to remove the steric block found in unliganded canonical protein to form a more intricate bonding network with the ligand), the reaction mechanism appears to be highly similar for the bacterial and canonical PARGs. The binding of ADPr to the active site ideally positions the O-glycosidic ribose–ribose bond in close proximity to the catalytic glutamates. Asp737 (absent in bactPARG) and Glu755 (residue numbering of human PARG) act in a proton relay network to reduce the pKa of Glu756, acting as a catalytic acid/base, which allows it to protonate the N-1 proximal ribose 2′-OH group (presuming linear PAR chains). Glu755 also accepts a hydrogen bond from the distal ribose 2″-OH group, contributing to the stabilization and orientation of the ADPr conformation. The positively-charged oxocarbenium intermediate is subsequently stabilized as previously reported by the close spatial proximity of the di-phosphate group, as a result of the steric restrains enforced by the presence of conserved phenylalanine residue. This oxocarbenium intermediate is then nucleophilically attacked by either of the two observed water molecules in the canonical PARG active site, water 1 or water 2, resulting in an inverting or retaining mechanism, respectively [12, 105-107].
The canonical PARG structures revealed that, in contrast to bactPARG, the 2′-OH PAR attachment point of the proximal N ADP-ribose is accessible to solvent (i.e. canonical PARGs have the potential to bind intra-chain ADPr and not only the terminal ADPr unit) (Fig. 3D–F) [12, 105-107]. Molecular dynamics modelling suggests that the canonical PARG active sites could indeed accommodate additional ADPr groups within linear PAR chains, or bind the ADPr group adjacent to (but not at) PAR branch points, thus accounting for the previously observed endo-glycohydrolase activity of the protein [12, 105-107]. Although this could explain the observed biphasic kinetics of PAR degradation, structural evidence has so far been difficult to obtain because of the lack of a source for large quantities of homogeneous, defined length fragments of PAR. Nevertheless, a recent study suggests that obtaining the quantities of defined length ADPr oligomers for structural studies is now possible .
The recently published human PARG structure also reveals a small hydrophobic cavity capable of binding adenine on a diametrically opposite side away from the ADPr binding pocket . It was suggested that this secondary binding site, V-X-I-X-VD-X10-F (located shortly after the MTS), conserved among vertebrates, and perhaps also the presence of other small binding cavities, may potentially accommodate adenine from long PAR chains. It could bind the remainder of the PAR chain in cases where such longer fragments were also bound by the primary active site or potentially serve a regulatory role. Further insight could again be provided by structural data with the substrate PAR.
PARG therapeutic implications: acting in synergy with PARP
Given the importance of PAR metabolism in cellular processes, targeting PARP1 in human disease has attracted significant scientific effort, especially in cases of hereditary breast and ovarian cancers [111-113]. PARPs belong to a large family of proteins, some of which have redundant functions. However, modulating the activity of PARG, a relatively low abundance protein that is encoded by a single gene, represents an attractive alternative strategy for achieving effects similar to the targeting of PARPs.
Because PARPs and PARG have ‘antagonistic’ roles in the cell, it would appear that inhibition or silencing of these proteins should in turn result in opposing biological effects. By contrast, most cellular models illustrate that PARP1 and PARG act in synergy with each other and are both required for efficient DNA repair [114-117]. Inhibition of PARG also results in a similar synthetic lethal phenotype in a homologous recombination deficient background and targeting both PARP1 and PARG does not increase the chemosensitivity any more than does targeting each of the proteins separately [115, 118]. As a result of the embryonic lethality of PARG−/− mice, where all PARG isoforms are absent most of the in vivo data originate from animal studies in a PARG110−/− background, where mice are lacking only the nuclear isoform of PARG . Consistent with cellular studies, several studies have reported the similarities of phenotypes after targeting PARP1 or PARG. Similar to PARP1−/− mice, PARG110−/− mice show increased sensitivity to ionizing radiation and alkylating agents, protection against ischaemia-reperfusion injury, and splanchnic artery occlusion and reperfusion [120-123]. However, in contrast to PARP1 targeting, they also show increased susceptibility to post-ischaemic brain damage . These opposing effects show that the interplay between PARP1 and PARG is more intricate than previously assumed, further illustrating the need for a cell permeable, potent and specific PARG inhibitor.
Subsequent to the discovery of an ADPr analogue, adenosine diphosphate-(hydroxymethyl)-pyrrolidinediol, a non cell-permeable PARG inhibitor, progress in the field has been slow . Recently, a series of rhodanine-containing compounds that also show activity in cell lysates was reported . Although still in need of optimization, the structure of one of these compounds in conjunction with the canonical protozoan TTPARG revealed the inhibitor binding mode, hopefully speeding up further development . In addition to cellular data indicating the potential of PARG targeting in neurodegeneration, inflammatory diseases and cancer, the emergence of small, cell-permeable PARG inhibitors would also provide an important tool for further elucidation of the biological roles of PARP1 and PARG.
ARH3: a back-up pathway for PAR hydrolysis
Another enzyme reported to cleave PAR is ARH3, which belongs to the dinitrogenase reductase-activating glycohydrolase-related protein family. ARH3, found in the nucleus, mitochondria and cytosol, shows no similarity to PARG either structurally or in sequence: it relies on the presence of a binuclear metal (Mg2+) cluster to assist acid/base catalysis, and has a predominantly α-helical fold (Fig. 2) [12, 38, 127, 128]. ARH3 was reported to mainly cleave OAADPR, a product of NAD+-dependent protein deacetylases of the sirtuin family . Recently, using embryonic fibroblasts from ARH3−/− mice, Niere et al. [95, 128] provided new insight into the controversial topic of whether mitochondrial PARG isoforms exist and are active. They proposed that ARH3 and not mitochondrial PARG accounts for the degradation of PAR in the mitochondrial matrix.
Mono-ARHs: macrodomains that finish the work started by PARG
The fact that PARG activity generates a mono-ADP-ribosylated protein product left the question of how the terminal mono-ADP-ribose is removed. Although ARH1, a mono-ADP-ribosyl-arginine hydrolase was already known, an enzyme capable of reversing glutamate-linked mono-ADP-ribose substrates remained elusive until recently. In humans, the enzymes MacroD1, MacroD2 and terminal ADP-ribose glycohydrolase 1 (TARG1) (C6orf130), previously shown to hydrolyze OAADPR, were identified as the long-sought enzymes that are able to reverse PARP-dependent mono-ADP-ribosylation. The proposed catalytic mechanisms, however, are different from those of OAADPR hydrolysis [29, 42, 43, 59, 61] or PAR hydrolysis by PARG  reported previously. It appears that the TARG1 catalytic mechanism involves a conserved lysine residue forming a covalent lysyl-ADPr intermediate, which is resolved by a closely positioned, catalytic aspartic acid residue . For MacroD1 and MacroD2, the suggested mechanism is that of substrate-assisted catalysis, where the macrodomain accommodates the substrate in a conformation where the pyrophosphate of ADPr can activate a positioned water molecule for nucleophilic attack on the glutamate-ADPr bond . By contrast, the ‘inactive’ macrodomains, such as that of macroH2A.1.1, have a closed groove in place of the pyrophosphate-accessible water molecule located in active macrodomains. Indeed, structure-based alignment predicts the activity of the macrodomains based on the accessibility of this particular groove above the pyrophosphate group. Enzymatic activity was first predicted and then experimentally confirmed for macrodomains from yeast (Poa1P) and archaea (Af1521), which also possess mono-ADP-ribosyl hydrolysis activity .
The exact biological roles of MacroD1, MacroD2 and TARG1 are not yet known, studies report MacroD1 acting as a cofactor of androgen and oestrogen receptors [130, 131], the potential involvement of MacroD2 in Kabuki syndrome [132, 133] and MacroD1, MacroD2 and TARG1 being OAADPR hydrolases, suggesting an involvement in sirtuin signalling. A recent study describes the effects of MacroD2 on GSK3β signalling via the reversal of its inhibitory mono-ADP-ribosylation [43, 134]. Significantly, the loss of TARG1 function was identified in a familial autosomal recessive neurodegenerative disorder, expanding the relevance of ADP-ribosylation pathways with respect to disease . Additionally, TARG1 knockdown in cells confers sensitivity to DNA-damaging agents, reduces cell proliferation and induces cellular senescence .
The enzymatic release of full PAR chains
Protein-free PAR has been identified as a death-signalling molecule that is able to translocate from the nucleus to the mitochondria and trigger apoptosis [135, 136]. This type of cell death, named parthanatos, has been identified in neuronal cells upon excitotoxicity and in other cell types when exposed to DNA-damaging agents [49, 85]. However, the mechanism by which free PAR is released from the nucleus has remained elusive.
Significantly, TARG1 can remove the whole PAR chain from PARylated proteins, although it does not hydrolyze free PAR and, as such, may contribute to parthanatos (along with the endo-glycohydrolase activity of PARGs) (Fig. 1) . Although TARG1 appears to be less efficient than PARG, the nature of the activity and the product are very different, with TARG1 releasing the full PAR chain and PARG releasing mono-ADPr and oligo-ADPr. For such macrodomain activity, it is necessary to be able to bind PAR at the amino acid attachment site. The structure of TARG1 fits with such a prerequisite: the 2′ OH group of TARG1 bound ADPr is not obstructed and could accommodate the next 2′→1″ linked ADPr unit in the PAR polymer. Just by looking at the macrodomain structures, it is not obvious whether a particular macrodomain could bind PAR. Based on reported pull-down assays, some but not all macrodomains were found to bind PAR, whereas others were capable of binding only mono-ADP-ribosylated proteins but not PAR . Although for PAR-binding macrodomains, it is commonly assumed that binding occurs at the end of the PAR chain, it is not always true, as demonstrated by the behaviour of TARG1 and suggested for canonical PARGs. Therefore, there are in principle four documented ways for macrodomains to recognize ADP-ribosylated substrates: (a) binding only mono-ADP-ribosylated protein but not PARylated (e.g. human MacroD2); (b) binding at the end of the PAR chain (with or without ability to bind mono-ADP-ribosylated substrate) (e.g. macroH2A.1.1); (c) binding at the root of the PAR chain (e.g. TARG1); or (d) PARG-binding along the PAR chain.
Although many macrodomain structures are available in complexes with small ligands, no structure to date reveals the interactions with the PAR polymer or mono-ADP-ribosylated protein. Therefore, the exact sequence and structural requirements for macrodomain-binding PAR substrates are not defined. Yet, the comparison of macrodomain structures with regards to the accessibility of 1″OH of the distal ribose or the 2′OH of the proximal ribose could help explain the differential binding capacities. More flexible loops and open conformations would be expected at the critical 1″OH and 2′OH sites if macrodomains were to bind PAR either at the end of the chain or at the protein attachment site, respectively. This is supported by the fact that MacroD2 positions the 2′OH site of ADPr in a more closed region than does TARG1 and therefore cannot bind PAR. Recent structures of mono-ADP-ribosylated substrates binding macrodomains of PARP14 should add to our understanding of ADP-ribosylated protein recognition .
Interestingly, in addition to the reported MacroD1, MacroD2 and TARG1 (C6orf130) ARHs, other human macrodomains are predicted to have ARH activities, such as the first macrodomain of PARP14 , which might be difficult to test because this macrodomain has low affinity for ADPr . All in all, the presence of at least three mono-ADP-ribosyl hydrolyzing macrodomains leaves the question of their specificity and functional overlap open and further studies are required to determine whether remaining macrodomains carry such catalytic activities, as well as what their precise functions are. In addition to ARH characterization, further understanding of PARG activity, in particular the prevalence of the endo-glycohydrolase catalytic mode and insight into how the enzyme is regulated, as well as definition of the roles played by ARH3 and PARG in the turnover of PAR, would broaden our knowledge of the life cycle of this essential post-translational modification.
The vibe in the ADP-ribosylation field is growing, with promising new developments in our understanding of how PAR is both recognized and removed. The appreciation that ADP-ribosylation signalling is involved in diverse disease phenotypes is interesting and provides added impetus to further dissect this elusive post-translational modification.
We apologize to all our colleagues whose work could not be cited because of space limitations. We would like to thank the members of our laboratories for their discussions and comments on the manuscript. We acknowledge support from the Ludwig Maximilian University of Munich to A.G.L., G.J. and G.T.; the Center for Integrated Protein Science Munich to A.G.L. and G.J.; the Deutsche Forschungsgemeinschaft (LA 2489/1-1) and the EU FP7 Marie Curie Initial Training Network ‘Nucleosome4D’ to A.G.L.; and the Deutsche Akademische Austauschdienst (#55934632) to G.T. Work in the laboratory of Ivan Ahel is funded by Cancer Research UK and the European Research Council.