The first indications that ADP-ribosylation may be involved in the regulation of transcription were studies that identified ADP-ribosylation as a modification of core histone [17, 18]. Much later, publications on the role of ARTD8 (formerly PARP14 or BAL2), a mono-ADP-ribosyltransferase, in regulation of STAT6 activity appeared. STAT (signal transducer and activator of transcription) factors are typically recruited to activated cytokine receptors. Upon ligand binding these are phosphorylated by Janus kinases (JAKs) creating binding surfaces for signaling molecules, including STATs. These are then phosphorylated, enabling them to form dimers and translocate into the nucleus to initiate transcription [19, 20]. ARTD8 was reported to activate STAT6-dependent reporter gene constructs, mediated by its macrodomains and catalytic activity [21, 22]. Upon interleukin 4 stimulation, ARTD8 ADP-ribosylates HDAC2 and HDAC3, leading to their dissociation from the promoter, thereby allowing the STAT6 dimers to bind . ARTD8 furthermore ADP-ribosylates the transcriptional co-activator p100 with thus far unknown consequence and directly interacts with STAT6 to enhance transcription [22, 24].
ARTD14 (formerly PARP7 or TiPARP) expression is upregulated by 2,3,7,8-tetrachlorodibenzo-p-dioxin through activation of the aryl hydrocarbon receptor (AHR) . Later, ARTD14 was shown to interact with AHR leading to decreased AHR transcriptional activity , indicating a negative feedback loop. The zinc finger and the catalytic activity are necessary for ARTD14-dependent repression of transcription, but it remains elusive through which substrates this is mediated. Thus far, only MARylation of histones by ARTD14 could be demonstrated , but other target proteins remain to be identified.
Finally, ARTD10 (formerly PARP10) possibly plays a role in regulating MYC-induced transcription. ARTD10 was identified as interaction partner of MYC  and this interaction was later demonstrated to take place in the nucleus . How this interaction influences MYC activity has not been further investigated, but one might speculate that it has similar effects on transcription as the ARTD14 interaction with AHR, possibly depending on MARylation of histones or as a direct consequence of physical interaction.
Immunity and inflammation
Several links between the expression of different ARTD family members and inflammatory processes have been established, in particular in relation to various pathogens. During infection by Venezuelan equine encephalitis virus (VEEV), an alphavirus, one of the genes that is upregulated is the long isoform of ARTD12 (formerly PARP12 or ZC3HDC1). ARTD12 exhibits an inhibitory effect on the replication of VEEV as well as on other alphaviruses and RNA viruses in this study . These findings are in accordance with a publication wherein the role of interferon-stimulated genes in the cellular defense against invading viral pathogens was investigated and ARTD12 was found as one of the genes upregulated to counteract infections , although not investigated mechanistically. Also for ARTD13 (formerly ZAP or ZC3HAV1) a role has been proposed in viral immunity. The long isoform of ARTD13 contains in addition to the ART domain also a WWE domain and a CCCH zinc-finger-containing domain, a structure shared with ARTD12 and ARTD14 . ARTD13 binds to different viral RNAs through its zinc fingers as reviewed in . ARTD12's reported function in viral immunity might also be mediated through its zinc fingers.
Additionally, ARTD10 was reported to MARylate NEMO and reduces its poly-ubiquitination, leading to increased I-κB stability and less p65 translocation into the nucleus. ARTD10 thus functions as a repressor of nuclear factor-κB (NF-κB) signaling . Stimulation by proinflammatory chemokines results in ubiquitination of NEMO, which is part of the IKK complex. Ubiquitination is essential for signal propagation leading to activation of the IKK complex. As a consequence, I-κBα, an inhibitor of NF-κB, is phosphorylated and degraded enabling NF-κB transcription factors such as p65 to translocate into the nucleus and drive target gene expression . Moreover, several studies aimed at defining genes influenced by certain inflammatory stimuli have identified ARTD10 as being upregulated, implying that the protein is required for the immune response [35-37]. Indeed, ARTD10 inhibits VEEV replication, although with a lower efficiency than ARTD12 .
Together, these reports strongly suggest that some mono-ARTDs are involved in immunity but leave the description of the exact mechanisms influenced by MARylation open for future investigations.
Several lines of evidence imply that MARylation is involved in the regulation of stress responses. ARTD5 (formerly PARP5a or Tankyrase 1), ARTD7 (formerly PARP15 or BAL3), ARTD8, ARTD12 and ARTD13 localize to stress granules (SGs) in response to stress conditions such as heat shock, glycogen deprivation or proteasome inhibition . In these SGs, mRNA-binding proteins such as Ago2 and TIA-1 are ADP-ribosylated, depending on their mRNA-binding domain, indicating that modification takes place within these domains or that the responsible ARTDs are associated via mRNA. Overexpression of ARTD12 and ARTD13 results in a relief of microRNA silencing, a phenomenon that usually occurs upon stress . These ARTDs may thus function to regulate the cellular stress response.
ARTD15 (formerly PARP16) was reported to localize to the endoplasmic reticulum (ER) [39, 40]. Here it plays an important role in regulation of the unfolded protein response (UPR) that serves to signal ER stress and can ultimately lead to apoptosis . During ER stress, ARTD15 activity is upregulated, leading to auto-MARylation and MARylation of the stress sensors IRE1α and PERK1, stimulating their activities, which appears necessary for a proper execution of the UPR . Additionally, the C-terminal part of ARTD15 that is localized within the ER lumen seems to influence binding of the inhibitory BiP protein to IRE1α and PERK1, as knockdown of ARTD15 leads to stabilized BiP association and thus inhibited IRE1α and PERK1 activity . Interestingly, MARylation of BiP has also been demonstrated, although on arginine residues by a thus far unknown transferase [42, 43]. One possibility is that ARTCs are involved . These enzymes are located at the outer cell membrane and thus are located in the ER lumen prior to being transported to the cell membrane. ARTCs are mono-ADP-ribosyltransferases and are arginine specific; however, whether they are active in the ER has not been analyzed. The MARylation of BiP destabilizes its closed-lid formation, predicted to lead to a decreased ability to bind substrates based on the location of the modified arginines within the protein. This was confirmed by ADP-ribosylation mimicry mutants that indeed show lower substrate binding. The result of MARylation of BiP is thus inhibition of its activity, leading to an increased UPR and promoting protein folding. The MARylation of BiP appears to be highest in fasting mice and disappears rapidly in response to feeding, indicating that enzymes exist in the ER that mediate this arginine modification but also enzymes capable of removing the modification . It is not known which enzyme might reverse arginine MARylation in the ER. However, recent progress has revealed hydrolases specific for MARylated glutamates.
DNA damage response
SIRT6 is probably best known for its deacetylase activity and its biological functions are diverse . In addition SIRT6 functions also as a nuclear ART . MARylation by SIRT6 appears to be involved in DNA damage repair under oxidative stress . The overexpression of SIRT6 resulted in higher non-homologous end-joining and homologous recombination efficiency, indicating a role for SIRT6 in double-strand break repair. Both enzymatic activities, MARylation as well as deacetylase activities, were necessary for this property. ARTD1 has a well-documented role in the DNA damage response  and could be identified as substrate for MARylation catalyzed by SIRT6. This resulted in elevated ARTD1 activity, leading to the conclusion that SIRT6 exerts its function in DNA damage repair by stimulating ARTD1 . Further supporting that MARylation occurs during the DNA damage response is the recruitment of specific macrodomain-containing proteins that recognize MARylation sites as discussed below [48, 49]. Thus SIRT6 appears to regulate genomic stability.
The second sirtuin with ART activity is SIRT4, which is localized in mitochondria. It has been suggested that SIRT4 ADP-ribosylates and inactivates glutamate dehydrogenase [50, 51]. This enzyme, which converts glutamate to α-ketoglutarate, thereby suppressing insulin secretion from pancreatic β-cells, has been shown to be ADP-ribosylated by an unidentified mitochondrial enzyme . Of note is that the rate constants for ADP-ribosylation of SIRT4 and SIRT6 are very slow and thus it was argued that the physiological relevance of their ART activities requires further investigation .
Reversibility of MARylation
Underlining the relevance of MARylation as dynamic PTM is the recent identification of both dedicated readers for MAR  as well as hydrolases capable of removing MAR from modified proteins (Fig. 1) [48, 49, 55]. Modules that bind to MAR specifically allow cells to read this signal and probably respond to it through the proteins linked to the recognition modules. An example thereof is the recognition of PARylated Axin by the WWE domain of ubiquitin E3-ligase Iduna (also known as RNF14) that leads to Axin ubiquitination and subsequent proteasomal degradation [56-58]. The identification of three proteins that reverse MARylation [48, 49, 55] indicates that MARylation serves as a transient signal that can be switched on and off, e.g. to regulate glycogen synthase kinase beta (GSK3β) activity [55, 59]. MacroD1, MacroD2 and C6orf130 were previously characterized as O-acetyl-ADP-ribose deacetylase [60, 61] and appear to also share the capacity to hydrolyze MAR from proteins. A mutation of the C6orf130 gene was described in patients with a severe neurological disorder, arguing for non-redundant functions of these three proteins . MacroD1 appears to be predominantly mitochondrial and might thus encounter different target proteins than MacroD2 and C6orf130 . More work is needed to dissect their specific functions in the regulation of MARylation. These reports thus provide firm evidence of MARylation as a reversible PTM that is relevant in diverse biological processes. Further investigation of MARylation has been complicated by the lack of antibodies against MARylation and by the technical challenge in the mapping of modification sites as discussed in the following.
Towards identifying the ADP-ribose acceptor site
For most PTMs it is clear which amino acids are the acceptor sites; however, for ADP-ribosylation this remains controversial (Table 1). To distinguish between the different amino acids as acceptor residues, neutral hydroxylamine treatment has been a prominent tool as it disrupts not only the ester bond between acidic residues and ADPr but also the ketamine bond between arginine or lysine and ADPr, although with different kinetics . ADP-ribosylated proteins with a half-life of about 3 min in neutral hydroxylamine represent ester linkages between acidic residues and ADPr, whereas the half-life of the ketamine bond formed by arginine-linked ADPr is approximately an hour . Both modified arginines and glutamates were found in total protein extracts from rat liver, with MARylation rather than PARylation being the major modification . In rat liver histone preparations, glutamate-linked H2B ADP-ribosylation was also measured . Analysis of nuclear proteins of the slime mold Physarum polycephalum revealed that H2A and H2B are mainly glutamate-ADP-ribosylated whereas H3 and H4 are mainly arginine-ADP-ribosylated as determined with hydroxylamine treatment . ARTD15 supposedly modifies neither acidic nor basic residues but threonine or serine of karyopherin-β1 . This was concluded because the automodification was stable in hydroxylamine and mercuric chloride, which would have cleaved cysteine-linked ADPr. Instead, the linkage seems destabilized by HCl treatment, which disrupts serine- or threonine-ADPr bonds .
Table 1. Overview of identified modification sites of different ARTDs. PDE, phosphodiesterase; CID, collision-induced dissociation; ETD, electron transfer dissociation; HCD, higher energy collisional dissociation
|ARTD1||ARTD1||K498, K521 and K524||Mutagenesis||  |
|ARTD1||ARTD1||E190, E456, E461, E488, E491, D578, K579, E807, E809, E883||PDE treatment followed by LC-MS/MS (CID)a||  |
K37 and K16
|ARH3 treament followed by LC-MS/MS (ETD)a and mutagenesis||  |
|ARTD1-E988Q||ARTD1-E988Q||D387, E488 and E491||LC-MS/MS (ESI)a and mutagenesis||  |
|ARTD1-E988Q||ARTD1-E988Q||E3, E147, E168 or E169, E190, E471, E484, E488, E491||PDE treatment followed by LC-MS/MS (HCD)a||  |
|ARTD2||ARTD2||K36 and K37||Mutagenesis||  |
|ARTD10||ARTD10||Glutamate or aspartate||Sensitivity to neutral hydroxylamine|| [1, 55] |
|ARTD10||ARTD10||E882||Mutagenesis||  |
|ARTD10||H2B||E2||Mutagenesis||  |
|ARTD15||Karyopherin-β1||Serine or threonine||Sensitivity to HCl||  |
Both mass spectrometry and mutagenesis approaches have also been employed to determine modification sites. Mutation of glutamate 882 in ARTD10 leads to decreased automodification  and mutation of glutamate 2 of H2B results in reduced modification by ARTD10 , both papers defining glutamates as acceptor sites for ARTD10. An ARTD10 mutant in which all lysines were exchanged for arginines could be trans-MARylated, indicating that lysines are not the sites targeted by ARTD10 . Structural considerations suggested that MacroD1, MacroD2 and C6orf130 are glutamate specific, further strengthening the evidence for glutamate modification by ARTD10 [48, 49, 55]. In ARTD1 mutation of K498, K521 and K524 cause a significant decrease in automodification, implying that these lysines are sites automodified by ARTD1 . Contradicting these findings is a report on glutamates within ARTD1 as automodification sites , although here an ARTD1 mutant was used that predominantly MARylates. This ARTD1 E988Q mutant was also used in a recent report where glutamates were identified as automodification sites using phosphodiesterase digestion prior to mass spectrometry analysis . Additionally, both a lysine and several glutamates/aspartates were found as automodification sites of ARTD1 by performing LC-MS/MS analysis on automodified ARTD1 treated with phosphodiesterase . Possible in vitro artifacts can occur, in which for example mutation of a certain amino acid leads to a structural change in the protein in such a way that a modification site becomes covered or a binding surface is disturbed. This could result in the false assumption that the mutated site is the modification site. Glycation, the process in which lysines become modified non-enzymatically through the formation of a Schiff base and subsequent Amadori rearrangement [73-75], complicates matters further. Summarizing, these data indicate that no consensus has been reached yet concerning amino acid specificity of the ARTDs.
Paradoxical with the current data on amino acid specificity of the ARTDs, where thus far no arginine specificity has been identified, is the presence of an intracellular arginine-ADP-ribosylhydrolase (ARH1)  and the identification of ADP-ribosylated arginines in a phospho-proteome mass spectrometry data set, where only one modified glutamate was identified versus 87 arginines . One possible explanation is that the more stable arginine-linked ADP-ribosylation is measured, whereas the ester bond through which glutamates are modified is too labile for efficient detection in conventional mass spectrometry approaches. Eight of 88 identified ADP-ribosylated peptides were modified by ribose-phosphate , indicating that the ADPr moieties are unstable and difficult to measure as summarized before . One could argue that an intracellular ARH1 forms a protection mechanism against arginine-modifying bacterial toxins, which is supported by the finding that mice lacking ARH1 are more sensitive towards the toxic effects of cholera toxin , which specifically ADP-ribosylates arginine as reviewed in . However, it remains open whether one of the ARTDs or ARTCs might be able to modify intracellular proteins on arginine or whether another protein family is responsible, such as some sirtuins that have been reported to MARylate arginines . The dynamic arginine-ADP-ribosylation of BiP within the ER also remains unclear concerning the relevant transferase and hydrolase involved .
Evidence for intracellular MAR
Due to the lack of antibodies specific against MARylated proteins, direct evidence of intracellular MARylation is difficult to obtain. Multiple reports provide indirect support for the occurrence of MARylation, thereby highlighting its relevance. The MAR-binding macrodomains of ARTD8 have been used to precipitate MARylated proteins from cells such as RAN and NEMO, which was only possible when active ARTD10 was coexpressed [33, 54], strongly suggesting the occurrence of MARylation in cells. That macrodomains can be used as a tool to pull down ADP-ribosylated proteins was also demonstrated with the macrodomain Af1521 from the thermophilic Archaeoglobus fulgidus . Additionally, GSK3β activity could be lowered by coexpressing ARTD10 and, conversely, knockdown of the endogenous ARTD10 led to higher GSK3β activity, thereby also indicating that endogenous intracellular MARylation occurs and regulates GSK3β activity . Overexpression of the hydrolase MacroD2, which could be shown to remove MAR from GSK3β in vitro, also leads to increased GSK3β activity, further supporting the presence of endogenous intracellular MARylation as MacroD2 is MAR specific . Underlining the importance of strictly regulated MARylation is the identification of a mutation in the MAR-hydrolase C6orf130 in patients with severe neurodegeneration . Finally, recruitment of MacroD2 to DNA damage occurs in a pattern differing from the PAR-binding macroH2A1.1 . The hypothesis posed is that MacroD2 is recruited to MARylation occurring at the onset of DNA damage, and then disappears as PARylation becomes dominant, to be recruited again when poly-ADP-ribose glycohydrolase (PARG) mediated degradation of PAR chains has occurred and MARylated residues are left . Although all these reports provide indirect evidence of MARylation, together they make a strong argument for the presence of endogenous intracellular MARylation.
As there are only a few macrodomain-containing proteins encoded in the human genome , it has to be considered that further readers of MARylation may exist with unrecognized macrodomain folds. This is not unlikely, as the macrodomain in PARG could not be deduced from its sequence but was only recognized upon solving of its crystal structure . Moreover, the possible existence of other motifs recognizing MARylated proteins has to be considered. It would not be unexpected that additional domains exist that can read MARylation, similar to PARylation or other PTMs that are recognized by multiple domains [9, 83, 84]. Using protein microarrays, approximately 200 targets were identified for ARTD8 and ARTD10 . Assuming that the mono-ARTDs each have approximately 100 substrates and that hydrolysis of PAR by glycohydrolases leads to MARylated proteins, a rough estimation would be that around 1000 MARylated proteins exist. We do not know the exact number of MARylated proteins in cells; one reason is that no robust mass spectrometry protocols have been developed to faithfully screen for MARylated proteins. Although this number is an estimate, it is worth comparing it with other PTMs and the ratio of enzymes to modification sites. For example, acetylation is mediated by roughly 20 enzymes  and a mass spectro-metry based approach revealed about 3600 lysine acetylation sites on 1750 distinct proteins . Thus, assuming a comparable complexity between these two PTMs, it is reasonable to suggest the existence of 1000 MARylated proteins. The question that then arises is whether the few macrodomains that have been identified are sufficient to read and erase this PTM. Most probably additional macrodomain folds will be discovered in proteins where the sequence does not reveal a macrodomain, as is the case for PARG , but where solving of the structure will provide more insight. This will increase the complexity of readers and erasers; however, we think it is appropriate to hypothesize that further domains exist to deal with MARylation.