Nuclear ADP-Ribosylation Reactions in Mammalian Cells: Where Are We Today and Where Are We Going?

Abstract

Since poly-ADP ribose was discovered over 40 years ago, there has been significant progress in research into the biology of mono- and poly-ADP-ribosylation reactions. During the last decade, it became clear that ADP-ribosylation reactions play important roles in a wide range of physiological and pathophysiological processes, including inter- and intracellular signaling, transcriptional regulation, DNA repair pathways and maintenance of genomic stability, telomere dynamics, cell differentiation and proliferation, and necrosis and apoptosis. ADP-ribosylation reactions are phylogenetically ancient and can be classified into four major groups: mono-ADP-ribosylation, poly-ADP-ribosylation, ADP-ribose cyclization, and formation of O-acetyl-ADP-ribose. In the human genome, more than 30 different genes coding for enzymes associated with distinct ADP-ribosylation activities have been identified. This review highlights the recent advances in the rapidly growing field of nuclear mono-ADP-ribosylation and poly-ADP-ribosylation reactions and the distinct ADP-ribosylating enzyme families involved in these processes, including the proposed family of novel poly-ADP-ribose polymerase-like mono-ADP-ribose transferases and the potential mono-ADP-ribosylation activities of the sirtuin family of NAD⁺-dependent histone deacetylases. A special focus is placed on the known roles of distinct mono- and poly-ADP-ribosylation reactions in physiological processes, such as mitosis, cellular differentiation and proliferation, telomere dynamics, and aging, as well as “programmed necrosis” (i.e., high-mobility-group protein B1 release) and apoptosis (i.e., apoptosis-inducing factor shuttling). The proposed molecular mechanisms involved in these processes, such as signaling, chromatin modification (i.e., “histone code”), and remodeling of chromatin structure (i.e., DNA damage response, transcriptional regulation, and insulator function), are described. A potential cross talk between nuclear ADP-ribosylation processes and other NAD⁺-dependent pathways is discussed.

INTRODUCTION

Over 40 years ago, P. Chambon and colleagues discovered that the addition of NAD to hen liver nuclear extracts stimulated the synthesis of poly-ADP-ribose (65, 433), paving the way for research into the biology of mono- and poly-ADP-ribose. A landmark meeting on ADP-ribosylation reactions, held in October 2005 in Newcastle, United Kingdom, commemorated this important scientific anniversary. For the first 30 to 35 years, research on ADP-ribosylation reactions was a relatively esoteric field. However, the development of new approaches, such as the generation of different knockout mice, has changed the situation in the past 5 years. Recent data show that ADP-ribosylation reactions play important roles in many physiological and pathophysiological processes, including inter- and intracellular signaling, transcription, DNA repair pathways, cell cycle regulation, and mitosis, as well as necrosis and apoptosis.

ADP-ribosylation reactions are phylogenetically ancient and can be divided into four major groups: mono-ADP-ribosylation, poly-ADP-ribosylation, ADP-ribose cyclization, and formation of O-acetyl-ADP-ribose. Mono-ADP-ribosylation of proteins and generation of free ADP-ribose or O-acetyl-ADP-ribose are the most conserved evolutionarily and are common to both prokaryotes and eukaryotes. ADP-ribosyl cyclase activities occur in unicellular and multicellular eukaryotes but not in bacteria and archaea (77, 140, 256, 257, 309; reviewed in references 88, 109, and 354). Poly-ADP-ribosylation reactions occur in multicellular eukaryotes and may also be present in unicellular eukaryotes (204, 309, 434; reviewed in reference 20). Poly-ADP-ribosylation or genes encoding poly-ADP-ribosylating enzymes have not been identified in bacteria. Despite the fact that no genes encoding poly-ADP-ribosylating enzymes have been identified in the archaeal genomes analyzed so far (309), recent studies provided evidence that poly-ADP-ribosylation-like reactions may exist in archaea (116, 118).

This review focuses mainly on the nuclear enzymatic mono-ADP-ribosylation and poly-ADP-ribosylation reactions occurring in mammalian cells and on the ADP-ribosylating enzyme families involved in these processes. Cytoplasmic and extracellular membrane-associated mono-ADP-ribosylation reactions, mediated by the ecto-mono-ADP-ribosyltransferase (e-MART) family and ADP-ribose cyclases, will not be discussed in detail. The reader is referred to recent excellent reviews on this topic (57, 100, 151, 305, 356). The first part of this review offers a brief overview of the molecular mechanisms of poly-ADP-ribosylation and discusses the ADP-ribosylating enzyme families, including novel ADP-ribosylating enzymes. Studies performed exclusively with poly-ADP-ribosylation/poly-ADP-ribose polymerase (PARP) inhibitors will only be partially addressed here, due to the off-target effects of poly-ADP-ribosylation inhibitors and nonspecific inhibition of both poly-ADP-ribosylation and certain mono-ADP-ribosylation reactions (179, 447). Since the term “covalent poly-ADP-ribosylation of proteins” is currently a subject of debate, we include a special section on “covalent poly-ADP-ribosylation of proteins.” We discuss new technologies and strategies, as well as new models that might help to clarify whether poly-ADP-ribosylation is a covalent and reversible posttranslational modification of proteins. A special focus on the known and proposed physiological roles of distinct mono-ADP-ribosylation and poly-ADP-ribosylation reactions will be given in the last sections. Since it is not yet clear whether proteins are covalently poly-ADP-ribosylated or can just bind to free poly-ADP-ribose, we use the established term PARP instead of the term poly-ADP-ribosyltransferase recently suggested by Glowacki et al. and Otto et al. (140, 309).

HISTORICAL OVERVIEW

The presence of poly-ADP-ribose was first suggested in 1963 by P. Chambon and coworkers, who reported that NAD⁺ stimulated the incorporation of labeled ATP into an acid-insoluble fraction of poly(A)-containing products in hen liver nuclear extracts (65). The enzyme responsible for the synthesis of poly-ADP-ribose was named PARP. The structure of poly-ADP-ribose was later solved by three independent laboratories (111, 288, 330, 383). Several years later, mono-ADP-ribosylation reactions were discovered during studies of bacterial toxins. These enzymes turned out to be mono-ADP-ribosyltransferases (MARTs) (136, 169). Subsequently, the existence of endogenous mono-ADP-ribosyltransferases was reported (reviewed in references 151, 304, 305, and 356). At first, mono-ADP-ribosylation and poly-ADP-ribosylation were postulated to serve as reversible posttranslational modifications of proteins, acting as regulatory mechanism for proteins. During the 1970s and 1980s, several laboratories partially purified several enzymes associated with mono-ADP-ribosylation and poly-ADP-ribosylation activities. In 1971, M. Miwa and T. Sugimura discovered poly-ADP-ribose glycohydrolase (PARG), which cleaves the ribose-ribose bonds in poly-ADP-ribose (278). Eight years later, the same group described the branched structure of poly-ADP-ribose in detail (277). However, it took an additional decade to isolate the genes encoding proteins responsible for these ADP-ribosylation reactions. In the late 1980s, the gene encoding a poly-ADP-ribose synthetase (initially named PARP, poly-ADP-ribose synthetase, or poly-ADP-ribosyltransferase and now named PARP-1) was isolated (9, 217, 414). At the same time, H. C. Lee and coworkers described an additional ADP-ribosylation reaction, the cyclization of ADP-ribose, which leads to cyclic-ADP-ribose. Cyclic-ADP-ribose is formed following NAD⁺ cleavage by NAD⁺ glycohydrolases/ADP-ribosylcyclases (223), and it serves as an important second messenger involved in the regulation of calcium signaling and homeostasis (reviewed in reference 354). In the early 1990s, the groups of J. Moss and F. Koch-Nolte identified several genes encoding e-MARTs and one gene encoding an ADP-ribosylarginine hydrolase (203, 302, 303, 385, 467).

For a long time it was thought that PARP-1 was the only enzyme with poly-ADP-ribosylation activity in mammalian cells. However, after nearly 15 years of intensive characterization of PARP-1, five different genes encoding “bona fide” PARP enzymes were identified (19, 178, 185, 196, 376), indicating that PARP-1 belongs to a family of PARPs. A similar situation is found in the case of MARTs. Although the mammalian e-MARTs (e-MART1 to -5/6) so far represent the only MART family for which enzymatic activities are well characterized (88, 89, 356), several reports predicted that distinct families of mono-ADP-ribosylating enzymes with no obvious sequence similarity to the well-known e-MARTs must exist in mammalian cells (88, 89, 356). Indeed, several members of the sirtuin family of NAD⁺-dependent histone deacetylases (SIRTs) were found to posses mono-ADP-ribosyltransferase activities and thus could represent a putative novel family of intracellular MARTs (128, 132, 234, 391). Moreover, very recent reports described 11 additional novel mammalian Parp-like genes (5, 20, 140) that may be good candidates to be members of a putative large family of intracellular PARP-like MARTs (5, 6, 241, 309, 452; reviewed in references 20 and 140). Although clear biochemical evidence for protein-mono-ADP-ribosylation by the SIRTs and PARP-like ADP-ribosyltransferases has yet to be established (309, 371), growing families of MARTs and PARPs exist and may be responsible for distinct mono-ADP-ribosylation and poly-ADP-ribosylation reactions in mammalian cells (20, 140, 161, 309).

A unique reaction catalyzed by distinct SIRT family members, in which the cleavage of NAD⁺ and the deacetylation of substrates are coupled to the formation of O-acetyl-ADP-ribose (O-AADP-ribose), was recently described (46, 390). O-AADP-ribose was shown to serve as small-molecule effector, involved in the modulation of heterochromatin formation (165, 231).

NAD⁺ METABOLISM

In eukaryotic cells, NAD⁺ has been shown to play a pivotal role as an essential coenzyme/transmitter molecule in bioenergetics (reviewed in references 243, 339, and 466). The synthesis of ATP and the balance of redox potential depend directly on NAD⁺ levels in cells. The chemistry of this molecule allows it to serve both as an electron acceptor (in its oxidized form, NAD⁺) and as an electron donor (in its reduced form, NADH) in reactions catalyzed by enzymes of the mitochondrial electron transport chain, leading to the generation of ATP during oxidative phosphorylation. In addition to its well-known roles in energy metabolism, NAD⁺ also has a distinct role as a precursor or immediate substrate for multiple ADP-ribosylation reactions. Such reactions are involved in cell regulation and metabolic processes and in the formation of various metabolites, including nicotinamide, free mono-ADP-ribose, mono-ADP-ribosylated proteins, cyclic-ADP-ribose, NAADP⁺, O-AADP-ribose, and poly-ADP-ribose (reviewed in references 32, 339, 354, and 466). Hydrolysis of the high-energy bond between the nicotinamide and ribose moieties of NAD⁺ produces a free energy of −34.3 kJ/mol (−8.2 kcal/mol) (457). This energy is used by distinct NAD⁺-metabolizing ADP-ribosylation enzymes to drive the transfer of the ADP-ribose moiety to proteins and the synthesis of ADP-ribose polymers. The multiple roles of NAD⁺ in bioenergetics and production of secondary messengers as well as in protein modifications and generation of free and protein-associated poly-ADP-ribose have important physiological consequences in the regulation of multiple cellular processes, as demonstrated by various studies performed on the molecular functions of NAD⁺-dependent enzyme families (reviewed in references 32, 243, 339, 354, and 466).

The involvement of NAD⁺ in these regulatory processes as a donor of ADP-ribose requires a constant resynthesis of NAD⁺ to avoid depletion of the intracellular NAD⁺ pool. In higher eukaryotes, the biosynthesis of NAD⁺ occurs through one de novo pathway and three distinct salvage pathways (243, 245, 339). NAD⁺ can be synthesized from four distinct precursors: l-tryptophan (thought to represent the de novo pathway) and nicotinic acid, nicotinamide (Nam), and nicotinamide riboside (thought to represent the three salvage pathways) (243, 245, 339). The Nam salvage pathway, leading from Nam to NAD⁺, goes through a single intermediate, nicotinamide mononucleotide (NMN). The nicotinic acid salvage pathway, known as the Preiss-Handler pathway, goes through two intermediates, nicotinic acid mononucleotide (NaMN) and nicotinic acid adenine dinucleotide. The nicotinamide riboside salvage pathway uses nicotinamide riboside as a precursor and is connected to the Nam salvage pathway through NMN (36). The de novo pathway leads from tryptophan to quinolinate and is connected to the Preiss-Handler pathway through NaMN. The presence of these multiple NAD⁺ biosynthetic routes most likely reflects differences in tissue distribution and/or intracellular compartmentalization of NAD⁺ metabolism (31, 243, 339, 465, 466). However, because nicotinamide is a product of NAD⁺ hydrolysis by numerous NAD⁺-glycohydrolases, including ADP-ribosylating enzymes, and no nicotinamidase-producing nicotinic acid exists in vertebrates, Nam is probably the major source for the biosynthesis of NAD⁺ in most mammalian cells (243, 339). A scheme for the NAD⁺ biosynthetic pathways and metabolism is shown in Fig. 1. The common enzymes of both the de novo and salvage pathways, the family of NMN adenylyltransferases (Na/NMNATs) which catalyze the production of NAD⁺ from NMN and ATP and represent the final step in the biosynthesis of NAD⁺, also play a crucial regulatory function for ADP-ribosylation processes in the cytoplasm and nucleus (reviewed in references 32, 243, 339, and 466). The predominant form of mammalian Na/NMNATs, Na/NMNAT-1, is localized in the nucleus, whereas Na/NMNAT-2 and Na/NMNAT-3 are cytoplasmic (460), being preferentially localized to Golgi complex and mitochondria, respectively (31). Their localization suggests that local production of NAD⁺ is important for the NAD⁺-dependent processes in those compartments (31, 243, 339). It is likely that local NAD⁺ production is strictly controlled under normal physiological conditions by the recruitment of biosynthetic enzymes to sites of NAD⁺-glycohydrolase activities, such as mono- and poly-ADP-ribosylation reactions (31, 198). The Na/NMNATs could sense the level of free mono-ADP-ribose or more likely free or bound poly-ADP-ribose and may be recruited in a poly-ADP-ribose-binding-dependent manner (31, 198). In this respect, PARP-4/vault-PARP and PARP-5/tankyrase-1 are the only members of the “bona fide” PARP family that have been localized to the cytoplasm. PARP-4/vault-PARP is present in cytoplasmic ribonucleoprotein particles (vaults) and cytoplasmic clusters (vault-PARP rods) as well as in the nuclear matrix (196, 235). PARP-5/tankyrase-1 was shown to be associated, at least in part, with the Golgi complex (75). Under normal physiological conditions, all other “bona fide” PARP family members seem to be localized exclusively to the nucleus (20).

FIG. 1.

Mammalian NAD⁺ metabolic pathways. The biosynthesis of NAD⁺ occurs through both de novo and salvage pathways (339). In mammalian cells, 90% of free tryptophan is metabolized through the kynurenine pathway, leading to the de novo synthesis of NAD⁺. The three different salvage pathways start either from nicotinamide (Nam), nicotinic acid (Na), or nicotinamide riboside (NR). In mammals, the origin of nicotinic acid is mainly nutritional. Nicotinamide, a product of NAD⁺ hydrolysis, is first converted into nicotinamide mononucleotide (NMN) and then into NAD⁺ by nicotinamide phosphoribosyl transferase (NamPRT) and nicotinamide mononucleotide adenylyl transferases (Na/NMNAT-1, -2, and -3), respectively. Nicotinamide riboside was recently shown to serve as a precursor for NAD⁺ synthesis, connected to the Nam salvage pathway through NMN (36). Nicotinamide riboside is converted to NMN by the ATP-consuming nicotinamide riboside kinases 1 and 2 (NRK-1 and -2) (36). Nicotinic acid can be converted through the Preiss-Handler salvage pathway into nicotinic acid mononucleotide (NaNM) and nicotinate adenine dinucleotide by the concerted actions of nicotinic acid phosphoribosyl transferase (NaPRT) and Na/NMNAT-1, -2, and -3, respectively. Nicotinate adenine dinucleotide is directly transformed into NAD⁺ by the glutamine-hydrolyzing NAD⁺ synthetase (NADS). Na/NMNATs are ATP-consuming enzymes, using either NaMN or NMN as a substrate. Whether both NamPRT and NaPRT are also ATP-consuming enzymes in vivo is not certain. Thus, when the Preiss-Handler salvage pathway is used, the cell invests three or four molecules of ATP from Na to NAD⁺, depending on whether NaPRT is also an ATP-consuming enzyme in vivo. In mammalian cells, under the conditions where NAD⁺ is used as a glycohydrolase substrate, the Nam salvage pathway is required, since there is no nicotinamidase to produce nicotinic acid. Depending on whether NamPRT uses one ATP molecule to convert Nam into NMN, the Nam salvage pathway consumes two or three ATP molecules from Nam to NAD⁺. The de novo pathway is connected to the Preiss-Handler salvage pathway through NaMN. NAD⁺ can be hydrolyzed by various enzymatic activities, such as PARPs, MARTs, SIRTs, and ADP-ribosyl cyclases, which release the Nam moiety from NAD⁺ to produce poly-ADP-ribose, mono-ADP-ribosyl-protein, acetyl-ADP-ribose (O-AADPR), or cyclic-ADP-ribose (cADPR) and nicotinate adenine dinucleotide phosphate (NAADP), respectively. These products are then further metabolized by different hydrolase activities, yielding ADP-ribose (ADPR), which, in turn, can be transformed into 5-phosphribosyl-1-pyrophosphate (PRPP) by the ATP-consuming ADP-ribose pyrophosphatase (ARPP)/ribose phosphate pyrophosphokinase (RPPK) pathway. PRPP is used by the Nam salvage pathway enzymes NamPRT and NaPRT.

It has been suggested that the most important factor affecting the maintenance of the NAD⁺ pool is the level of poly-ADP-ribosylation in cells (32, 243, 466). The catabolism of NAD⁺ in mammalian cells occurs mainly via poly-ADP-ribosylation reactions. The concentration of NAD⁺ in undamaged, proliferating mammalian cells is approximately 400 to 500 μM, and its half-life is about 1 to 2 h (115, 329, 438). However, when cells were exposed to high doses of genotoxic agents, sustained activation of poly-ADP-ribosylation reactions, coinciding with an increase in the levels of poly-ADP-ribose polymers generated following DNA damage, was shown to rapidly decrease the half-life of NAD⁺ in a dose-dependent manner. In fact, intracellular NAD⁺ levels undergo a decrease to 10 to 20% of their normal levels within 5 to 15 min upon exposure of cells to very high doses of DNA-damaging agents (143, 369). NAD⁺ depletion also results in ATP depletion, as NAD⁺ is an essential coenzyme/transmitter for the generation of ATP. The resynthesis of NAD⁺ requires two to four molecules of ATP per molecule of NAD⁺, depending on which salvage pathway is used in the cell and whether NamPRTase or NaPRTase is the ATP-consuming enzyme in vivo (70) (Fig. 1). It should be noted, however, that several studies indicate that under moderate levels of DNA damage, intracellular NAD⁺ levels undergo a decrease of only 5 to 10%. For a more detailed description of the NAD⁺ metabolism and enzymology, the reader is referred to the recent excellent reviews on this topic (243-245, 339).

MONO-ADP-RIBOSYLTRANSFER REACTIONS

Mono-ADP-ribosylation of proteins is a phylogenetically ancient, reversible, and covalent posttranslational modification of proteins in which the ADP-ribose moiety of NAD⁺ is transferred to a specific amino acid of an acceptor protein with the simultaneous release of nicotinamide (reviewed in references 304, 305, and 356). The reaction can occur through both enzymatic and nonenzymatic mechanisms (reviewed in references 302 and 304). Enzymatic mono-ADP-ribosylation reactions, originally identified as the pathogenic mechanism of several bacterial toxins, including pertussis toxin, cholera toxin, and certain clostridial toxins, are catalyzed by MARTs. Such enzymes have been detected in many prokaryotic and eukaryotic species and in viruses (reviewed in references 88, 140, 304, and 305). The extent of posttranslational modification by mono-ADP-ribosylation depends on the activity of cellular mono-ADP-ribose-protein hydrolases (MARHs), which reverse the reaction by hydrolyzing the protein-ADP-ribose bond (reviewed in references 88, 202, 304, and 305). The simultaneous presence of mono-ADP-ribosyltransferase and mono-ADP-ribose-protein hydrolase activities in the same cell suggests that mono-ADP-ribosylation of proteins acts as a reversible regulatory mechanism (306, 374; reviewed in references 202 and 304). MARTs and MARHs are opposing arms of an ADP-ribosylation cycle (306, 374). In contrast to the case for the prokaryotic ADP-ribosylation cycle, the functional relationship between MARTs and MARHs in eukaryotes is poorly documented (304, 305). Thus, the detailed mechanisms of coupling of MARTs and MARHs in eukaryotic mono-ADP-ribosylation cycles need to be investigated further.

Mono-ADP-ribosylation occurs at a number of different amino acid residues, directed by the specificity of the individual MARTs. The best-studied mono-ADP-ribosylation reactions are catalyzed by bacterial toxins (reviewed in references 88, 304, and 356). At least six amino acid-specific subclasses of bacterial MARTs have been characterized or identified so far. The amino acid residues of crucial host cell protein acceptors modified by specific bacterial MARTs include arginine, asparagine, glutamate, aspartate, cysteine, and modified histidine (diphthamide) (reviewed in references 88, 89, 304, and 305). Mono-ADP-ribosylation of cellular proteins through nonenzymatic mechanisms usually involves the conjugation of ADP-ribose to lysine or cysteine residues (59, 60, 184, 214; reviewed in reference 174). Amino acid-mono-ADP-ribose-specific MARHs cleave the ribose-amino acid bond, leading to release of free mono-ADP-ribose and regeneration of the free reactive group of the corresponding amino acid residue. A detailed description of distinct protein mono-ADP-ribosylation products is shown in Fig. 2. In eukaryotic cells, endogenous protein-mono-ADP-ribosyltransferase activities that modify arginine, glutamate, cysteine, phosphoserine and potentially aspartate and asparagine residues of acceptor proteins have also been detected (308, 373; reviewed in references 88, 89, 304, 305, and 356). For instance, intracellular mono-ADP-ribosylation has been demonstrated for heterotrimeric GTP-binding proteins, small GTPases, endoplasmic reticulum-resident glucose regulatory protein 78, tubulin, actin, elongation factor 2, mitochondrial glutamate dehydrogenase (GDH), and histones. A detailed list of distinct intracellular protein substrates is given in Table 1. A number of mono-ADP-ribosyltransferases have also been shown to ribosylate small molecules such as free amino acids, DNA, or RNA (reviewed in references 88, 304, and 356). Pierisin-1 and its homolog pierisin-2, two unique ADP-ribosylation toxins from the cabbage butterflies (Pieris rapae and Pieris brassicae), were shown to catalyze mono-ADP-ribosylation of 2′-deoxyguanosine in DNA to form N2-(C1-ADP-ribosyl)-2′-deoxyguanosine (386, 387).

FIG. 2.

Mono-ADP-ribosylation cycle and the corresponding products of protein mono-ADP-ribosylation. (A) Mono-ADP-ribosyl transferases catalyze the transfer of the ADP-ribose moiety of NAD⁺ to an acceptor molecule (free amino acids, proteins, DNA, and RNA, etc.). The action of mono-ADP-ribosyl-amino acid hydrolases, which regenerate the corresponding free acceptor molecule, is consistent with the presence of a mono-ADP-ribosylation cycle. (B) The transferase-catalyzed reaction of protein mono-ADP-ribosylation results in a stereo-specific formation of Ω-N-(C-1-ADP-ribosyl)-l-arginine, ɛ-N-(C-1-ADP-ribosyl)-l-asparagine, ω-N-(C-1-ADP-ribosyl)-l-diphtamide, γ-S-(C-1-ADP-ribosyl)-l-cysteine, ɛ-O-(C-1-ADP-ribosyl)-l-glutamate, δ-O-(C-1-ADP-ribosyl)-l-aspartate, or O-(ADP-ribosyl)-l-phosphoserine.

TABLE 1.

Cytoplasmic/mitochondrial substrates for endogenous covalent mono-ADP-ribosylation

Substrate(s)	Amino acid modification site(s)	Functional relevance	Reference(s)
Glutamate dehydrogenase isozymes (hGDH1 and hGDH2)	C119	Inhibits substrate activity	78
Desmin (intermediate filament protein)	R48 and R68	Inhibition of assembly of desmin into 10-nm intermediate filaments	464
Tubulin	Arginine residues	Assembly inhibition and rapid depolymerization	351, 403
Alpha-actin	R95 and R372	Delay in polymerization, decreased rate of nucleation	183
Beta-actin	Arginine residues	Inhibition of actin polymerization	401
G-actin	R28 and R206	Inhibition of actin polymerization	402
F-actin	R28	Inhibition of actin polymerization	402
Myelin basic protein	Arginine residues	Stabilization of myelin?	444
Glutamine synthetase	Arginine residues	Inactivation	282
GTP-binding protein Gi alpha subunits	Cysteine residues	Attenuates inhibition of adenylate cyclase by epinephrine; regulation of hormonal control of the adenylate cyclase system	83, 393, 397
Transducin, GTP-binding protein Go alpha subunits	N343(?), D346(?), C347(?), arginine residues (R147)?	Activation of a retinal cyclic GMP-selective phosphodiesterase	45, 83, 250
GTP-binding protein Gs alpha subunits	Arginine residues	Activation of the adenylate cyclase system	172, 326
GTP-binding protein beta subunits	R129	Inactivation, prevents the inhibition of type 1 adenylyl cyclase	240
cGMP-phosphodiesterase gamma subunit	R33 and R36	Activation	43
Phosphoprotein B-50/GAP-43	C3 and C4	Cytosolic localization	320
Neurogranin	NAa	NA	81
Glucose regulatory protein 78 (GRP78/BiP)	NA	NA	221, 227

^aNA, not analyzed.

Mono-ADP-Ribosyltransferase Families

The mammalian e-MARTs (e-MART-1 to -5/6) represent the only mammalian mono-ADP-ribosyltransferase family of structurally related proteins characterized on the molecular level (reviewed in references 88, 202, 304, and 356). Human and mouse e-MARTs identified to date represent either glycosylphosphatidylinositol-anchored surface proteins or secretory proteins. e-MART-1, e-MART-2a, e-MART-2b, and e-MART-5 were found to transfer ADP-ribose to arginine residues in extracellular target proteins, on cell surfaces, or circulating in body fluids (reviewed in references 88, 140, and 356). In contrast, e-MART-3 and e-MART-4 did not display any arginine-specific enzymatic activity (reviewed in references 88, 140, and 356). Mono-ADP-ribosyltransferase activity on the surface of human monocytes correlated with the presence of e-MART-3 in unstimulated monocytes (146). Consistent with its expression in lymphatic tissues, e-MART-4 is expressed only in response to stimulation with lipopolysaccharide (146). Interestingly, cell surface mono-ADP-ribosylated proteins on human monocytes are covalently modified at cysteine residues, which strongly suggests that e-MART-3 and e-MART-4 may be cysteine-specific e-MARTs (146). Thus, it is unlikely that members of the known e-MART family are involved in mono-ADP-ribosylation of intracellular protein targets, such as heterotrimeric G proteins or histones, although one cannot exclude the possible existence of alternatively spliced or processed cytosolic isoforms (88, 109). Corda and Di Girolamo (88, 89) proposed that these novel intracellular MARTs most likely constitute different families of proteins with no obvious sequence similarity to the well-known e-MARTs (88, 89, 109). Although a preliminary study reported partial purification of an arginine-specific mono-ADP-ribosyltransferase from hen liver nuclei with activity towards histones H2A, H2B, H3, and H4 (389), no arginine-directed cytosolic or nuclear mono-ADP-ribosyltransferases from mammalian cells have been unambiguously identified and analyzed on a molecular level. Thus, most authors refer to these mono-ADP-ribose modifications as “poly-ADP-ribosylation,” presuming that these mono-ADP-ribose modifications represent remnants of poly-ADP-ribose polymers catalyzed by poly-ADP-ribosylation enzymes (see “Covalent poly-ADP-ribosylation of nuclear substrates?” below). However, at least two distinct families of putative intracellular MARTs may exist in mammalian cells. The first family is represented by the well-known sirtuin family of NAD⁺-dependent histone deacetylases and ADP-ribosyltransferases (SIRT1 to -7) (128, 132, 234, 391). The second possible family consists of 11 novel PARP-like gene products, referred to here as Pl-MARTs (Pl-MART-1 to -11) (20, 140, 309).

The family of putative PARP-like mono-ADP-ribosyltransferases.

During the last 10 years, more than 16 novel Parp-like genes were cloned or described based on thorough searches of nonredundant databases (5, 6, 20, 130, 140, 241, 309, 452). Among all 17 human and 16 mouse Parp-like genes, only the 6 human and mouse “bona fide” poly-ADP-ribose polymerase gene products contain an evolutionarily conserved catalytic glutamate residue (309). The crystal structure of chicken PARP-1 and amino acid replacement analysis of human PARP-1 demonstrated that the conserved residue, E988, is essential for poly-ADP-ribose chain elongation (255, 338, 342). The absence of this crucial residue in PARP-1 was shown to restrict its enzymatic activity to mono-ADP-ribosylation. Surprisingly, several residues important for poly-ADP-ribose chain elongation, including E988 and residues thought to be required for the PARP branching activities (255, 342, 366), seem to be missing in 11 of the Parp-like genes identified (5, 6, 20, 241, 452). Otto and coworkers recently suggested that these 11 human and 10 mouse Parp-like gene family members may possess mono-ADP-ribosyltransferase rather than poly-ADP-ribosyltransferase activity (309). However, as those authors pointed out, one cannot exclude the possibility that some of these 11 novel Parp-like gene family members are not enzymatically active or may even have acquired novel functions (309). There is preliminary experimental evidence that at least one of the proposed Pl-MARTs, BAL-1, lacks any auto-ADP-ribosylation activity (5, 6). On the other hand, most of the Pl-MART enzymes investigated on a biochemical level (PARP-10, TiPARP, BAL-2, and BAL-3) possess auto-ADP-ribosylation activity (5, 6, 241, 452). Thus, these novel PARP-like gene products represent candidates for a possible large family of intracellular PARP-like mono-ADP-ribosyltransferases.

The members of the proposed Pl-MART family can be divided according to their domain structures and the sequences of their catalytic domains into at least four subgroups. Figure 3 shows a proposed classification of the family based on recent literature and database searches (309). Although all putative Pl-MARTs share an evolutionarily conserved “PARP signature” motif in their catalytic domains (20, 140), most family members are structurally distinct from the “classical” 114-kDa PARP-1 and other previously described “bona fide” PARPs. The Pl-MARTs contain a diversity of adaptor domains and additional motifs, including WWE domains, macroH2-like domains, and ubiquitin- or RNA-binding motifs, suggesting that they possess unique properties and are involved in many biological functions in which the previously described PARP family might not participate. Subgroup I contains Pl-MART-1 (also known as PARP-6 [20]), Pl-MART-2 and possibly its alternatively transcribed or spliced short isoform Pl-MART-2b (formerly PARP-8 [20]), and Pl-MART-3 (also known as PARP-16 [20]). The functions of Pl-MART-1 to -3 are not yet known. Pl-MART-4 (also known as PARP-10), a Myc-interacting protein that inhibits transformation (452), is the only member of subgroup II. Subgroup III includes four members that contain a C3H1-type zinc finger and/or a WWE domain: Pl-MART-5 (initially described as PARP-11 [20]), which has an unknown function; Pl-MART-6 (also known as TiPARP and PARP-7 [241]), whose homolog in Arabidopsis thaliana, CEO1/RCD1, is involved in hormonal signaling and stress-response pathways (7, 30); Pl-MART-7 (ZC3HDC1, formerly PARP-12), which has an unknown function (20); and Pl-MART-8, previously described as the C3H1-type zinc finger-containing antiviral protein-1 (130). The nuclear-localized family members Pl-MART-9, Pl-MART-10, and Pl-MART-11, initially described as risk-related proteins in diffuse large B-cell lymphomas that enhance cellular migration (B-aggressive lymphoma proteins 1 to 3 [BAL-1, BAL-2, and BAL-3, also known as PARP-9, PARP-14, and PARP-15] [5, 6, 20]), belong to the macroH2A and/or WWE domain-containing Pl-MART subgroup VI. Interestingly, Pl-MART-11 may exist only in the human genome (309). Since the biochemical evidence for mono-ADP-ribosylation by Pl-MARTs has yet to be established and because several different classifications have already been proposed for Parp-like genes (20, 309), a final, consistent classification of all PARP-like proteins needs to be made in agreement with the entire PARP/MART community once their different enzymatic activities are characterized.

FIG. 3.

Domain structures of the human Pl-MART family. A new classification and a schematic comparison of protein structures of the 11 members of the Pl-MART family, based on the literature and database searches, are shown. The most significant domains detected have been indicated. The WWE domain is named after three of its conserved residues (W/W/E) and is predicted to mediate specific protein-protein interactions in ubiquitin- and ADP-ribose conjugation systems (24). Although the exact roles of the conserved macroH2A/A1pp domains remain unknown, they have been proposed to have ADP-ribose 1"-phosphate (Appr-1"p)-processing activity and may regulate mono-ADP-ribosylation (219). ZF, C3H-type zinc finger domain; RRM, RNA recognition motif (252); UIM, ubiquitin interaction motif (22); MVP-ID, M-vault particle interaction domain; TPH, Ti-PARP homologous domain; GRD, glycine-rich domain.

The SIRTs, a family of putative intracellular mono-ADP-ribosyltransferases.

The silent information regulator SIR2-like proteins, also named SIRTs, represent a family of NAD⁺-dependent deacetylases (SIRT1 to -7) (reviewed in references 37, 106, and 147). The SIRT family regulates a wide range of cellular processes, including development, metabolism, heterochromatin formation, chromosome segregation, DNA transcription, DNA repair, DNA recombination, cellular differentiation, apoptosis, and the determination of life span (reviewed in references 37, 106, and 147). The sirtuins are phylogenetically conserved in eukaryotes, prokaryotes, and archaea. The first discovered member of this protein family is the yeast SIR2 histone deacetylase of Saccharomyces cerevisiae, which is required for transcriptional silencing (reviewed in reference 37). Most of the mammalian SIRTs were found to have intrinsic histone deacetylation activity in vitro and in vivo. Of the seven mammalian SIR2-like proteins, SIRT1, SIRT2, SIRT3, and SIRT5 have been shown to have NAD⁺-dependent deacetylase activities in vitro (271; reviewed in references 37 and 106). However, SIRTs often have nonhistone substrates, and not all mammalian SIRT members are localized to the nucleus. Consequently, the SIRTs have a diversity of substrates that reflect the various biological processes in which the enzymes function. For example, human SIRT1 and its mouse homolog SIR2α were reported to deacetylate, in vivo, acetylated transcription factors such as p53 and DNA repair factors such as Ku70, while acetylated α-tubulin was found to be an in vivo deacetylation target for hSIRT2 (82, 289, 422). For a detailed description of SIRTs and their functional roles, see the recent reviews on this topic (37, 106, 147).

Based on extensive analyses of the NAD⁺-dependent deacetylation reaction, an unusual mechanism has been proposed (371, 462, 463; reviewed in reference 106). SIRTs consume one NAD⁺ cosubstrate molecule per acetyl group, which is removed from a lysine side chain. SIRTs cleave the glycosidic bond between the Nam and ADP-ribose portions of NAD⁺. The ADP-ribose intermediate is necessary for deacetylation to take place (371). Subsequently, the acetyl group removed from the target substrate can be transferred to the ADP-ribose moiety to form 2′-O-AADP-ribose and 3′-O-AADP-ribose. Several reports suggested that the mammalian SIRT family members SIRT1, SIRT2, and SIRT6 might possess intrinsic mono-ADP-ribosyltransferase activity (128, 132, 234, 284, 391). SIRT1, SIRT2, and SIRT6 were shown to transfer mono-ADP-ribose to bovine serum albumin and histones in vitro (128, 284, 391). Furthermore, a point mutation in yeast SIR2 that abolished the observed histone ADP-ribosyltransferase activities in vitro resulted in a complete loss of silencing in vivo, suggesting that the potential histone-ADP-ribosyltransferase activity of yeast SIR2 could be required for silencing (391). SIRT6 does undergo auto-mono-ADP-ribosylation via an intramolecular reaction mechanism (234). All seven human SIRTs characterized so far share a conserved histidine, which may be important not only for their NAD⁺-dependent deacetylase activities but also for their mono-ADP-ribosyltransferase activities (128; reviewed in reference 347). Whether NAD⁺-dependent deacetylation of proteins, acetylation of ADP-ribose, or the potential mono-ADP-ribosylation of proteins can occur simultaneously or depends on the conditions and the type of SIRT protein involved remains to be investigated. The mono-ADP-ribosyltransferase activity of yeast SIR2 (as well as its NAD⁺-glycohydrolase activity) has been shown to require the presence of an acetyl-lysine-containing substrate in the reaction, suggesting that it is linked to deacetylation (391). In contrast, SIRT6 and SIRT7 did not possess in vitro protein deacetylase activity, indicating that the enzymatic mechanisms are different for each SIRT member (234, 271). SIRT6 catalyzed in vitro ADP-ribosylation of nonacetylated bovine serum albumin much more efficiently than SIRT1. In contrast, histone H1 was a better substrate for SIRT1 than for SIRT6 (284). However, the possible mono-ADP-ribosyltransferase activities of the different SIRT members have not been adequately addressed on a molecular level (106). Further investigation is needed to determine if the SIRTs do indeed have bona fide protein mono-ADP-ribosyltransferase activity in vitro. The in vivo relevance of the mono-ADP-ribosylation activity of SIRTs remains a subject of debate, although it has been speculated to be involved in DNA double-strand break and base excision repair (132, 160, 284). Indeed, a recent study provided evidence that certain SIRTs may function in vivo as potential mono-ADP-ribosyltransferases in DNA damage response pathways (132). TbSIR2RP1, a SIR2-related protein from the protozoan parasite Trypanosoma brucei, has been shown to catalyze mono-ADP-ribosylation of histones in vitro, particularly H2A and H2B (132). Treatment of trypanosomal nuclei with a DNA-alkylating agent resulted in a significant increase in the level of histone mono-ADP-ribosylation, specifically that of H2A and H2B, and a concomitant increase in chromatin sensitivity to micrococcal nuclease. Both of these responses correlated with the level of TbSIR2RP1 expression (132). A possible O-AADP-ribose and mono-ADP-ribosylation metabolism is schematically drawn in Fig. 4, based on the literature (46, 106, 327, 371).

FIG. 4.

Possible metabolism of acetyl-ADP-ribose (O-AADPR) and mono-ADP-ribosylation of proteins by SIRTs. SIRTs cleave the glycosidic bond between the Nam and ADP-ribose portions of NAD⁺. The ADP-ribose intermediate is necessary for the deacetylation reaction. Following hydrolysis of the glycosidic bond, nicotinamide is released, and ADP-ribose binds the acetyl-peptide, forming of an O-alkylamidate intermediate. The acetyl group removed from the target substrate is transferred to the ADP-ribose moiety to form 2′-acetyl-ADP-ribose (2′-O-AADPR) and then subsequently released together with the deacetylated protein from the enzyme-intermediate complex. 2′-O-AADPR spontaneously equilibrates with the regioisomer 3′-acetyl-ADP-ribose (3′-O-AADPR) through trans-esterification. In mammalian cells, acetyl-ADP-ribose can be deacetylated by esterases to the ATP precursor ADP-ribose or can function as an acetyl donor to acetylate unknown substrates by nuclear trans-acetylases. Transformation of acetyl-ADP-ribose into AMP and acetyl-ribose-5-phosphate by ADP-ribose hydrolases of the Nudix family is not clearly established. The possible deacetylation-dependent and -independent transfers of mono-ADP-ribose to proteins catalyzed by SIRTs are shown in the lower part of the figure.

Nuclear substrates for covalent mono-ADP-ribosylation of proteins.

In eukaryotes, mono-ADP-ribosylation of arginine residues occurs on extracellular, cytoplasmic, and nuclear target proteins, whereas mono-ADP-ribosylation of cysteines occurs on extracellular, cytoplasmic, and mitochondrial target proteins. Mono-ADP-ribosylation of asparagine residues may be restricted to the cytoplasm, while mono-ADP-ribosylation of glutamate residues and potentially of the phosphate group of phosphoserine may be restricted to the nucleus (Tables 1 and 2). During the last two decades, several studies indicated that histones are covalently modified by mono-ADP-ribose in response to genotoxic stress, while other reports proposed that the extent of mono-ADP-ribosylation of histones varied depending on the cell cycle stage, proliferation activity, and degree of terminal differentiation (1, 142, 211, 212, 373, 379, 395, 435). For instance, when cells were exposed to damage by · OH radicals or methylating/alkylating agents, total covalent mono-ADP-ribosylation of histones increased 3 to 15 times, whereas the levels of histone H1-linked mono-ADP-ribosyl groups were even elevated by more than 30-fold (211, 212). Initial reports suggested that histone H1 (H1.1/H1.2/H1.3/H1.4/H1.5) was covalently mono-ADP-ribosylated on glutamate residues E2, E15, and E114/E115/E117 and arginine residue R33 (H1.3) and histone H2B on E2 (55, 292, 293, 335, 416). These modified sites were identified by “in vivo” ADP-ribosylation of histones, using radiolabeled NAD⁺ and subsequent high-pressure liquid chromatography-coupled Edman sequencing analysis of the radiolabeled single peptides (55, 292, 293, 416). Other studies indicated that mono-ADP-ribosylation also occurs on glutamic acid residues of H2A; on arginine residues of H2A, H2B, H3, and H4; and on the phosphate group of phosphoserine in histones H1, H2A, H2B, H3, and H4 (141, 308, 373, 389). Several reports indicated that the nonhistone, high-mobility group (HMG) proteins HMGA1a, HMGA1b, HMGA2, HMGB1, HMGB2, HMGN1, and HMGN2 might also serve as targets for mono-ADP-ribosylation in intact cultured cells (394, 396, 398). However, the specific amino acid residues in HMG proteins that were thought to be mono-ADP-ribosylated were not identified. To date, endogenous mono-ADP-ribosylation of HMGB1 and HMGB2 has been observed following DNA damage in intact cells (394, 396, 398). The same group also reported that treatment of cells with glucocorticoids quickly decreased endogenous mono-ADP-ribosylation on HMGN1 and HMGN2 proteins, while mono-ADP-ribosylation on HMGB1, HMGB2, and histone H1 was less susceptible to hydrolysis during glucocorticoid treatment (394). Mono-ADP-ribosylation of chromosomal proteins may influence the regulation of human myeloid cell maturation (239). A detailed list of distinct nuclear protein substrates for mono-ADP-ribosylation is given in Table 2.

TABLE 2.

Nuclear substrates for covalent mono-ADP-ribosylation

Substrate(s)	Amino acid modification sites	Functional relevance	References
Histone H1 (H1.1, H1.2, H1.3, H1.4, H1.5)	Glutamic acid residues E2 (all), E15 (all), E114 (H1.2), E115 (H1.3/H1.4), E117 (H1.5); arginine residue R33 (H1.3); phosphoserine residues(?)	NAa	293, 308, 335, 373, 389, 416
Histone H2A	Glutamic acid residues, arginine residues, phosphoserine residues (?)	NA	141, 308, 373, 389
Histone H2B	Glutamic acid residue E2, arginine residues, phosphoserine residues (?)	NA	55, 292, 308, 373, 389
Histone H3	Arginine residues, phosphoserine p-S57 (?)	NA	141, 308, 373, 389
Histone H4	Arginine residues, phosphoserine residues (?)	NA	141, 308, 373, 389
High-mobility-group proteins HMGA1a, HMGA1b, HMGA2, MGB1, HMGB2, HMGN1, HMGN2	Arginine and glutamic acid residues (?)	NA	117, 148, 394-396, 405
Low-mobility-group proteins (?)

^aNA, not analyzed.

Substrate specificities of the SIRT and putative Pl-MART families.

The enzymatic activities and bond specificities of SIRTs and Pl-MARTs have not yet been experimentally determined. Thus, it is quite difficult to predict any substrate specificities for these enzymes. Based on the assumption that the SIRTs have arginine-specific and cysteine-specific MART activities, while the Pl-MARTs have glutamate- and aspartate-specific MART activities, one may propose that mono-ADP-ribosylation on arginine residue R33 of histone H1.3 and at arginine residues of histones H2A, H2B, H3, and H4 is mediated by nuclear members of the SIRT family, such as SIRT1, SIRT6, or SIRT7, whereas mono-ADP-ribosylation on glutamate residues E2, E15, and E114/115/117 of H1 (H1.1/H1.2/H1.3/H1.4/H1.5) and on E2 of histone H2B is mediated by nuclear members of the Pl-MART family. The mitochondrial SIRT3, SIRT4, or SIRT5 (271) may be responsible for the cysteine-specific mono-ADP-ribosylation of GDH on cysteine 119 (Table 1) (78). SIRT2, a predominantly cytoplasmic protein associated with microtubules and acting as a bona fide tubulin deacetylase (289), could be responsible for the arginine-specific mono-ADP-ribosylation of both alpha and beta chains of tubulin. Several studies provided evidence that both deacetylation and arginine-specific mono-ADP-ribosylation of tubulins are involved in depolymerization of the microtubules (260, 289, 351, 403). The heterotrimeric GTP-binding proteins and small GTPases may represent substrates for the predominantly cytoplasmic members of the Pl-MART family (Table 1).

The diversity of substrate and amino acid specificity is reflected by the diverse biological activities of specific SIRTs and Pl-MARTs. Additionally, substrate specificity could also be regulated by cofactors, such as small GTPases and ADP-ribosylation factors (ARFs). ARFs are 20-kDa guanine nucleotide-binding proteins that play a critical role in many vesicular trafficking events and were initially identified as stimulators of bacterial toxin-catalyzed ADP-ribosylation of GTP-binding proteins (45, 179, 280, 281, 421). ARFs were shown to activate, allosterically, bacterial toxin mono-ADP-ribosyltransferases (45, 307).

Mono-ADP-Ribose-Protein Hydrolases

Several reports demonstrated the existence of distinct intracellular mono-ADP-ribose-protein hydrolase activities (reviewed in references 202, 304, 305, and 306). The best-characterized intracellular mono-ADP-ribose-protein hydrolase activity is represented by mono-ADP-ribose-arginine-hydrolase-1 (MARH-1), which specifically hydrolyzes ADP-ribose-arginine bonds, leading to the release of free mono-ADP-ribose and regeneration of the guanidino group of arginine (385; reviewed in reference 283). In mammalian cells, most ADP-ribose-arginine hydrolase activities are cytosolic, although some are located on the cell surface. Besides the well-characterized MARH-1, additional proteins exhibiting hydrolase activity towards mono-ADP-ribose linked to glutamates or cysteines were identified and partially purified. A mono-ADP-ribosyl protein-lyase, catalyzing the cleavage of the bond between mono-ADP-ribose and glutamate residues in histone H2B or H1, was purified from rat liver and characterized (296, 300). The purified enzyme exhibited a single protein band at the position of 83 kDa on sodium dodecyl sulfate-polyacrylamide gel electrophoresis. This putative enzyme was hypothesized to act as a mono-ADP-ribose-glutamate-lyase, whose cleavage product is 5′-ADP-3"-deoxypent-2"-enofuranose, a dehydrated form of ADP-ribose (296). In addition, hydrolysis of linkages between mono-ADP-ribose and cysteine residues in the alpha subunits of heterotrimeric GTP-binding proteins was found in the cytosol of human erythrocytes (392). This putative mono-ADP-ribose-cysteine-hydrolase was tentatively named ADP-ribose-protein hydrolase C. Moreover, mitochondrial GDH was recently identified as a specific target for enzymatic cysteine-specific mono-ADP-ribosylation (78, 163). This covalent mono-ADP-ribosylation was removed by an ADP-ribose-cysteine hydrolase activity present in mitochondria (163).

Until recently, the only cloned and characterized mammalian gene encoding a soluble, intracellular cytosolic mono-ADP-ribose-protein hydrolase was the Marh1 gene, whose product hydrolyzes mono-ADP-ribose-arginine bonds (385). Two additional ADP-ribosyl-hydrolase-like genes, Arh2 and Arh3, have recently been identified by screening the human genome sequence and cDNA databases for homologues of the Marh1 gene (140). A recent report demonstrated that ARH-2 and ARH-3 did not hydrolyze ADP-ribose-arginine, -cysteine, -diphthamide, or -asparagine bonds (297). ARH-3 may possess intrinsic poly-ADP-ribose-ribose-glycohydrolase activity, generating free ADP-ribose from PARP-1-bound poly-ADP-ribose (see “PARGs” below) (297). Thus, ARH-2 may be a candidate for a glutamate- or aspartate-specific mono-ADP-ribose-protein hydrolase. The presence of distinct intracellular mono-ADP-ribose-protein hydrolase activities which are not connected to the identified Arh-like genes in the human genome suggests that mono-ADP-ribose-protein hydrolases might encompass different families.

POLY-ADP-RIBOSYLATION REACTIONS

Poly-ADP-Ribosylation Cycle

Poly-ADP-ribose is a homopolymer of ADP-ribose units linked by glycosidic bonds and synthesized by members of the PARP family (20, 93, 161). Like mono-ADP-ribose synthesis, poly-ADP-ribose synthesis requires NAD⁺ as a precursor and immediate substrate of the reaction. The constitutive levels of poly-ADP-ribose are usually very low in unstimulated cells (122, 164, 212; reviewed in reference 93). However, in response to mitogenic stimuli or genotoxic stress (i.e., in the presence of DNA strand breaks), the PARP activity and the levels of poly-ADP-ribose may increase 10- to 500-fold, while cellular NAD⁺ levels are correspondingly reduced. Both constitutive and activated levels of poly-ADP-ribose are functions of the concentration of NAD⁺ in cells (15, 164, 436, 437). However, most free or protein-associated poly-ADP-ribose polymers synthesized upon genotoxic stress are rapidly degraded in vivo, with a half-life of >40 s to 6 min, which accounts for their transient nature in living cells (15, 173, 175, 435). There may also be a biphasic decay: 85% of poly-ADP-ribose polymers synthesized due to genotoxic stress have a half-life of less than 40 s, while the residual fraction is catabolized with a half-life of approximately 6 min (15, 435). This rapid turnover contrasts with the slow catabolism of the constitutive fraction of poly-ADP-ribose, with a much longer half-life of 7.7 h (15). It is likely that the degradation starts immediately upon initiation of poly-ADP-ribose synthesis, suggesting that poly-ADP-ribose-metabolizing enzymes are tightly regulated.

The structure of poly-ADP-ribose is well characterized. The ADP-ribose units in the polymer are linked by glycosidic ribose-ribose 1′-2′ bonds. The chain length of polymers is heterogeneous and can reach 200 to 400 units in vitro and in vivo (16, 181, 182, 186, 276). Long polymers are branched in an irregular manner. Branching occurs in vitro and in vivo with a frequency of approximately one branch per linear section of 20 to 50 units of ADP-ribose (16, 181, 182, 186, 276). The chemical structure of the branching site of poly-ADP-ribose was determined by nuclear magnetic resonance and mass spectroscopy (MS) and found to be the same as in the linear regions of the polymer (276). The branching site of poly-ADP-ribose is O-d-ribofuranosyl-(1′-2′)-O-d-ribofuranosyl-(1′-2′)-adenosine-5′,5′,5-triphosphate, commonly known as Ado-(P)-Rib-(P)-Rib-P (276). The global structures of the branched poly-ADP-ribose can be very complex. T. Minaga and E. Kun postulated that the structures of certain types of long poly-ADP-ribose chains can have helicoidal secondary structures and thus may have some similarity to the structures of RNA and DNA (272, 273). Interestingly, many antibodies against poly-ADP-ribose can recognize RNA and DNA and vice versa (188, 362, 363). The functional significance of the structural heterogeneity of poly-ADP-ribose is not known, but it could play a crucial role in determining specific functional outcomes in vivo (i.e., stress-dependent signaling in regard to survival or cell death).

At least four distinct enzymatic activities were postulated to be required for the synthesis of free or PARP-associated linear and branched poly-ADP-ribose (17, 93, 264): (i) initiation or covalent auto-mono-ADP-ribosylation of the corresponding poly-ADP-ribose polymerase; (ii) elongation of the polymer, whereby the covalently bound mono-ADP-ribose serves as a starting unit; (iii) branching of the polymer; and (iv) release of the enzyme-bound branched poly-ADP-ribose, either through poly-ADP-ribose-ribose-glycohydrolase activities (intrinsic or by PARG [see below]) or through a putative intrinsic poly-ADP-ribosyl-protein-hydrolase activity. Based on experimental evidence in vitro, it was suggested that the classical PARP enzyme, PARP-1, possesses (i) auto-mono-ADP-ribosylation, (ii) elongation, and (iii) branching activities. Whether poly-ADP-ribose polymerases also possess intrinsic poly-ADP-ribose-ribose-glycohydrolase or poly-ADP-ribosyl-protein-hydrolase activities that release the polymers remains to be investigated.

To date, two different enzymes or enzymatic activities are known or hypothesized to degrade free (non-protein-bound) or protein-bound linear or branched poly-ADP-ribose (reviewed in references 93 and 340). The major enzymatic activity is the well-characterized poly-ADP-ribose glycohydrolysis carried out by PARGs (53). Poly-ADP-ribose phosphodiesterase/ADP-ribose pyrophosphatase was suggested to possess pyrophosphatase activity and to cleave the pyrophosphate linkages to release 5′-AMP from chain termini, phosphoribosyl-AMP from internal residues, and diphosphoribosyl-AMP from branching points (49). Of these two enzymatic activities, only the Parg genes and their gene products have been identified and characterized to date (18, 230, 297). The major mammalian poly-ADP-ribose-ribose-glycohydrolase, PARG, has both endoglycosidase and exoglycosidase activities (18, 52, 53), which are responsible for the hydrolysis of glycosidic ribose-ribose bonds internal to and at the ends of ADP-ribose polymers, respectively. The endoglycosidase activity releases free poly-ADP-ribose from PARPs and is of particular physiological importance because it may provide a mechanism for the generation of various types of free poly-ADP-ribose. These products may be important signaling molecules involved in distinct cellular processes, such as cell death or cell growth. In addition, branched and short polymers are degraded more slowly by PARGs than long and linear poly-ADP-ribose polymers (18, 52, 53). This mode of action of PARGs may explain the very short half-life of poly-ADP-ribose synthesized in the presence of DNA damage in vivo (<40 s) compared with the far longer half-life (≤7.7 h) of constitutively synthesized poly-ADP-ribose in unstimulated cells (15, 435, 437). Thus, the biphasic degradation of poly-ADP-ribose in vivo clearly indicates that two major types of polymers (linear↔branched) with different structures and distinct half-lives exist in vivo (16). The complexity and concentration of each distinct type and structure of poly-ADP-ribose may vary not only depending on the cellular context and stimuli, but also depending on specific branching activities of different PARPs in vivo. An overall view of poly-ADP-ribosylation reactions and metabolism is shown in Fig. 5.

FIG. 5.

Poly-ADP-ribose metabolism. Steps 1 to 3 and steps 4 to 7 of the poly-ADP-ribose cycle represent the anabolic and catabolic reactions, respectively, in the metabolism of poly-ADP-ribose. The synthesis of poly-ADP-ribose requires three distinct PARP activities: step 1, initiation or mono-ADP-ribosylation of a specific glutamic (?) acid residue(s) in the corresponding PARP enzyme (acceptor); step 2, elongation of the polymer; and step 3, branching of the polymer. The degradation requires at least four (alternative) PARG and (P/M)ARH activities: step 4, exoglycosidase and endoglycosidase (PARG) activities, respectively, that hydrolyze the glycosidic linkages between the ADP-ribose units; step 5, potential poly-ADP-ribosyl-protein hydrolase activities; and step 6, MARH, or step 7, mono-ADP-ribosyl-protein lyase activities. Chemical structures in this figure were drawn with MarvinSketch, version 4.0.4 (ChemAxon, Budapest, Hungary).

In addition to the well-established model of synthesis of free or PARP-associated poly-ADP-ribose, several groups proposed that poly-ADP-ribosylation may also serve as a covalent posttranslational modification of proteins (reviewed in references 93, 160, and 161). It was suggested that different poly-ADP-ribose polymerases covalently attach poly-ADP-ribose to the side chain carboxyl groups of glutamic or aspartic acid residues of putative acceptor proteins (reviewed in reference 14). Similar to the synthesis of free poly-ADP-ribose, ADP-ribose units may be added successively to acceptor proteins to form branched protein-bound polymers (reviewed in reference 14). More than 30 years ago, several groups proposed that this putative covalent posttranslational modification is very transient but extensive in vivo, with the poly-ADP-ribose chains reaching more than 200 units on protein acceptors (reviewed in reference 14). The observed mono-ADP-ribose groups covalently bound to proteins in vivo were suggested to be remnants of poly-ADP-ribose polymers, and the major regulatory step of poly-ADP-ribosylation of proteins was proposed to be catalyzed in vivo by ADP-ribose-protein hydrolases (435, 436). More recently, it was postulated that PARG itself has the predicted ADP-ribose-protein hydrolase activity responsible for the hydrolysis of the most proximal unit of ADP-ribose on the protein acceptor (108). Thus, PARG was thought to modulate the level and complexity of poly-ADP-ribose on the different acceptor proteins, thereby preventing hypermodification of nuclear proteins with very long poly-ADP-ribose chains (108). However, no convincing in vitro data have been published, and to date, no ADP-ribose-protein hydrolase mutants of PARGs exist.

The PARP Family

For a long time, the best-investigated PARP protein, PARP-1, has been thought to be the only enzyme with poly-ADP-ribosylation activity in mammalian cells. However, this view has been challenged recently by the development of mice deficient in the Parp1 gene (431, 432) and the identification of novel poly-ADP-ribosylating enzymes. Primary cells derived from Parp1^−/− mice can still synthesize poly-ADP-ribose following treatment with DNA-damaging agents (360). Five new genes encoding “bona fide” PARP enzymes have been identified (19, 178, 185, 196, 376), indicating that PARP-1 belongs to a family of poly-ADP-ribose polymerases.

The six “bona fide” PARP family members can be divided into at least three subgroups according to their domain structures, the sequences of their catalytic domains, and their enzymatic activities. Figure 6 shows a classification and schematic comparison of protein structures of the PARP family based on the literature and on database searches. Subgroup I includes PARP-1; PARP-1b, which seems to be a product of an alternative transcription initiation site within the Parp1 gene (previously described as short PARP-1 [343]); PARP-2; and PARP-3 (19, 178). Experimental data suggest that both PARP-1 and PARP-2 play a major role in distinct stress response pathways (reviewed in references 20 and 161). Subgroup II contains a single member, PARP-4 (vault-PARP), which is the largest of the family (192.6 kDa) and was identified as a component of the vault complex. The vault complex is a cytoplasmic ribonucleoprotein complex of unknown function associated with an untranslated vault RNA and two other highly conserved proteins, major vault protein and telomerase-associated protein-1 (196). Tankyrase 1, tankyrase 2a, and perhaps its alternatively spliced or transcribed isoform tankyrase 2b, which are referred to in this review as PARP-5 and PARP-6a/b, respectively, belong to subgroup III (20, 185, 376). Both PARP-5 and PARP-6a were identified as components of the telomeric complex (185, 376).

FIG. 6.

Domain structures of the human PARP family. A classification and a schematic comparison of protein structures of the six members of the bona fide PARP family, based on literature and database searches, are shown. The most significant domains detected have been indicated. The PRD domain is called the PARP regulatory domain and may be involved in regulation of the PARP-branching activity. The WGR domain is named after the most conserved central motif (W/G/R) of the domain. This motif is found in a variety of poly(A) polymerases and other proteins of unknown function. The BRCT domain is named after the breast cancer suppressor protein-1 (BRCA1) carboxy-terminal domain and is found within many DNA damage repair and cell cycle checkpoint proteins (446). The unique diversity of this domain superfamily allows BRCT modules to interact by forming homo- or hetero-BRCT multimers and phosphorylation-dependent BRCT-non-BRCT interactions (139, 446). The sterile alpha motif (SAM) is a widespread domain in signaling and nuclear proteins and mediates homo- or heterodimerization in many cases (reviewed in reference 20). The ankyrin repeat domains (ARD) mediate protein-protein interactions in very diverse families of proteins (279). The number of ankyrin repeats in a protein can range from 2 to over 20 (279). The vault protein inter-alpha-trypsin (VIT) and von Willebrand type A (vWA) domains are conserved domains found in all inter-alpha-trypsin inhibitor (ITI) family members (261). Although the exact roles of these domains remain unknown, they are presumed to be involved in mediating protein-protein interactions (261). ZF-I and ZF-II, PARP-1-type zinc finger domains (they can act as DNA nick sensors and general DNA-binding domains [161]); SAP, SAF/Acinus/PIAS-DNA-binding domain; LZM, putative leucine zipper-like motif; MVP-ID, major-vault particle interaction domain; NLS, nuclear localization signal; CLS, centriole-localization signal; HPS, His-Pro-Ser region.

All “bona fide” PARP enzymes (PARP-1, PARP-1b, PARP-2, PARP-3, PARP-4, PARP-5, and PARP-6a) have automodification activity and most likely covalent auto-ADP-ribosylation activity (19, 26, 185, 196, 343, 376). PARP-1 showed the strongest automodification activity in vitro. Based on this unique property, PARP-1 was identified as the main acceptor of radioactively labeled ADP-ribose in isolated nuclei and permeabilized cells (294). It was proposed that the covalently or noncovalently automodified forms of the enzyme do not serve as an intermediate in the synthesis of poly-ADP-ribose but play some biological role(s) as structural elements (294). The automodified domains were mapped for PARP-1 and PARP-2. Automodification takes place in the DNA-binding domains of PARP-1 and PARP-2 and in the so-called automodification domain of PARP-1 (107, 353). Earlier studies suggested that the auto-ADP-ribosylation activity targets the 25 to 30 glutamic acid residues in the automodification domain of PARP-1 (166). Moreover PARP-1 and PARP-2 were proposed to be able to trans-ADP-ribosylate each other at multiple sites, although it is not clear whether the modification is covalent or noncovalent (353; our unpublished observations). PARP-1 was shown to form different heteromers with PARP-2 and PARP-3 (26, 353). Another interesting aspect is the role of the PARPs in synthesis and branching of distinct types of poly-ADP-ribose. PARP-1 was suggested to synthesize complex, branched poly-ADP-ribose (reviewed in references 20 and 93). Further investigation is needed to see if the other PARPs of subgroup I, PARP-2 and PARP-3, have the same ability to catalyze all the reactions necessary to produce branched polymers or whether they can synthesize only linear polymers. Future studies will certainly clarify whether the PARP family could be subdivided into three classes of enzymatic activities: branched-polymer-synthesizing PARPs (i.e., PARP-1, PARP-2, and PARP-3), linear-polymer-synthesizing PARPs (PARP-4?), and linear-oligomer-synthesizing PARPs (PARP-5 and PARP-6) (334). The type of branched polymers might be characteristic for each branched-polymer-synthesizing PARP enzyme.

The enzymatic activities of PARP-1 and PARP-2 were initially proposed to be exclusively dependent on the presence of single-strand breaks and double-strand breaks in DNA (19, 145, 171). Recent studies have demonstrated that PARP-1 may be activated not only by DNA breaks induced by peroxynitrite, gamma radiation, and alkylating agents such as N-methyl-N′-nitro-N-nitrosoguanidine or methylnitrosourea but also by other stimuli, such as d-myo-inositol-1,4,5-triphosphate, which does not directly involve DNA damage (167, 198, 215). These studies indicate that PARP-1 might be involved in new DNA damage-independent, poly-ADP-ribosylation-dependent signaling pathways. The PARP activities of the very closely related PARP-1b (short PARP-1) and PARP-3 are not stimulated by DNA strand breaks (26, 343). PARP-1 was thought to be the major anabolic activity responsible for poly-ADP-ribosylation in living cells. Activation of poly-ADP-ribose polymerases was proposed to be one of the earliest responses of mammalian cells to genotoxic stress (reviewed in reference 93). Poly-ADP-ribose formation following DNA damage in primary mouse embryos and mouse embryo fibroblasts from Parp1 knockout mice was observed at 2 to 50% of wild-type values, depending on the tissue and cell type (149), and was drastically reduced only in Parp1^−/− brain, pancreas, liver, small intestine, colon, and testis. Moderate levels of residual poly-ADP-ribose formation were seen in Parp1^−/− stomach, bladder, thymus, heart, lung, kidney, and spleen (149). Only limited data are available regarding the physiological roles and poly-ADP-ribosylation activity of the novel PARP family members. Generally, the contribution of each PARP member to the total cellular poly-ADP-ribosylation activity depends on tissue, cell type, and stimuli, and the new PARPs are likely involved in specific nuclear and cytoplasmic functions requiring limited levels of poly-ADP-ribosylation.

PARGs

The “classical” poly-ADP-ribose glycohydrolase, PARG, is known to represent the major PARG activity catalyzing the hydrolysis of poly-ADP-ribose polymers to free ADP-ribose in the cell (reviewed in references 44 and 269). While at least six genes encode different “bona fide” PARPs synthesizing ADP-ribose polymers, only one single gene coding for the “classical” PARG activity has been detected in mammalian cells until recently (268, 297). The mammalian Parg gene encodes at least four isoforms: the low-abundance nuclear isoform PARG-110/111 and the three high-abundance cytoplasmic isoforms PARG-102, PARG-99 (characterized mainly in humans), and PARG-59/60 (characterized mainly in mice) (54, 110, 268). Whether all four isoforms simultaneously exist in all mammalian species remains to be investigated. PARG-110/111 represents the full-length 110- to 111-kDa PARG protein in human, rat, and mouse, while PARG-102 and PARG-99, so far detected only in human cells, are alternative splice variants leading to the expression of the 102-kDa and 99-kDa PARG isoforms (268). These PARG isoforms differ from PARG-110/111 by the lack of exon 1 (PARG-102) or exons 1 and 2 (PARG-99). They are generated by the use of ambiguous splice donor sites in the 5′ untranslated region of the Parg gene (268). The 59-kDa isoform (PARG-59) in mice, potentially corresponding to the 60-kDa PARG isoform observed in human cells, lacks exons 1, 2, and 3 and half of exon 4 and is either an additional splice variant or a product of an internal transcriptional regulation leading to a translation start located in exon 4 (starting at M514). The latter hypothesis is partially supported by different Parg knockout studies (90, 92, 206). The PARG-110/111 full-length isoform is localized primarily in the nucleus, whereas PARG-102 and PARG-99 seem to be exclusively cytoplasmic, under nonstimulated conditions (152). The localization of the PARG-59 isoform was suggested to be exclusively cytoplasmic (92) but has not been investigated in detail. The nuclear targeting of PARG-110/111 is due to two strong classical nuclear localization signals in exon 1 (268). The putative nuclear localization signal sequences found at amino acid positions 421 to 446 and 838 to 844 may not be functional (268). Several studies demonstrated that the cytoplasmic isoforms of PARG account for most of the PARG activity in cells in absence and presence of genotoxic stress. Two putative nuclear export signals have been localized to amino acids 126 to 134 and 881 to 888, which may explain the preferential cytoplasmic localization of the PARG isoforms lacking exon 1 (268). The predominantly cytoplasmic location of mammalian PARG isoforms is surprising, as most known cellular PARPs have a nuclear localization, especially PARP-1, which is thought to represent the major PARP activity in mammalian cells. A recent study, using green fluorescent protein-tagged PARG isoforms, provided evidence that the cytoplasmic 102-kDa PARG isoform translocates into the nucleus, while the nuclear PARG-110 isoform relocalizes to the cytoplasm in response to DNA damage (152). However, further investigation is needed to determine whether the alternative splicing and shuttling of PARG could be highly regulated in a cell type- and/or stimulus-specific manner (44, 152). Since Na/NMNAT-2 and Na/NMNAT-3 are preferentially localized to Golgi complex and mitochondria, respectively (31), and poly-ADP-ribosylation catabolism was also shown to be associated with both compartments (reviewed in references 31, 44, and 291), it was suggested that a subpopulation of the cytoplasmic PARG-102, PARG-99, or PARG-59 isoforms may also be localized in these compartments (44, 291). The structures of the isoforms of the “classical” PARG enzyme are schematically drawn in Fig. 7.

FIG. 7.

Domain structures of the human PARG isoforms. A classification and a schematic comparison of protein structures of the four PARG isoforms is shown (data are from references 267, 268, and 315). NLS, nuclear localization signals (amino acids 10 to 16, 32 to 38, 421 to 446, and 838 to 844); NES, nuclear export signals (amino acids 126 to 134, 421 to 446, and 881 to 888). Active sites: E728, E738, E756, E757, and T995.

Mice with a targeted deletion of exons 2 and 3 of the Parg gene, resulting in depletion of the nuclear PARG-110 protein and the cytoplasmic isoforms PARG-102 and PARG-99, are viable and phenotypically normal but show an increased sensitivity to alkylating agents and ionizing radiation (90). The mice were also susceptible to streptozotocin-induced diabetes or endotoxic shock and showed an enhancement of ischemic brain injury, most likely due to dysregulation of the nuclear and cytoplasmic poly-ADP-ribosylation metabolism and accumulation of poly-ADP-ribose (90, 92). The PARG activity was greatly reduced in the cytoplasmic and nuclear fractions of Parg-Δ2-Δ3/Δ2-Δ3-knockout cells. However, the 59-kDa isoform PARG-59 was still expressed in the Parg-Δ2-Δ3/Δ2-Δ3-knockout mice (90, 92), indicating that this isoform may be responsible for the residual poly-ADP-ribose-degrading activity present in these mutant cells. Surprisingly, ATP depletion was similar in Parg-Δ2-Δ3/Δ2-Δ3-knockout and wild-type mice after ischemia, indicating that impairment of PARG-110- and potentially PARG-102- or PARG-99-dependent poly-ADP-ribosylation catabolism does not significantly affect the brain's energy dynamic during hyper-poly-ADP-ribosylation. A recent report demonstrated that mice with a targeted deletion of exons 3 and 4 of the Parg gene, resulting in a complete depletion of all isoforms, show early embryonic lethality, and primary mouse embryo fibroblast cells derived from these embryos were hypersensitive to alkylating agents and ionizing radiation (206). Moreover, these Parg null cells accumulate very high levels of poly-ADP-ribose and undergo increased cell death (206). These data strongly suggest that only the PARG-59 isoform is required for embryonic development. The PARG-59 isoform may be constitutively active due to the lack of the N-terminal PARG regulatory domain (268) and also required for the degradation of the constitutively synthesized poly-ADP-ribose in unstimulated cells. However, the recent finding that the ADP-ribose-protein-hydrolase-like gene product, ARH-3, possesses intrinsic PARG activity indicates that mouse cells may contain several additional Parg genes that have not yet been detected (297). Indeed, in the genome of the cress plant, Arabidopsis thaliana, which contains fewer “bona fide” Parp genes than mammals (309), several uncharacterized genes (96) are similar to the gene for the TEJ protein, a plant homolog of the mammalian PARG (311). Interestingly, the 39-kDa ARH-3 shares amino acid sequence identity with both MARH-1 and the catalytic domain of the “classical” PARG but is structurally unrelated to the “classical” PARG (309).

Covalent poly-ADP-ribosylation of nuclear substrates?

As mentioned above, several investigators in the PARP field proposed that proteins can be covalently mono-ADP-ribosylated and poly-ADP-ribosylated in vitro and in vivo on glutamate and aspartate residues. Until recently, more than 200 nuclear proteins, most of them chromatin associated, have been proposed to be covalently modified by poly-ADP-ribose in vitro. Target proteins include classical PARPs, topoisomerases I and II, histones, p53, and the HMG proteins. An updated list of nuclear proteins suggested to be associated with poly-ADP-ribose in vivo is given in Table 3. In intact organisms, PARP-1 itself was postulated to be the predominant acceptor of poly-ADP-ribose (294). However, the data on covalent poly-ADP-ribosylation of proteins have to be cautiously interpreted. Despite many studies over the past 40 years, no specific glutamate or aspartate residues functioning as poly-ADP-ribose acceptor sites could be identified in vitro and confirmed in vitro or in vivo by amino acid exchange analysis. Many investigators in the PARP field argue that this lack of proof could be explained by the instability of the ester bond between glutamic or aspartic acid residues and poly-ADP-ribose. Indeed, a characteristic feature of glutamic and aspartic acid ester bonds is their instability with respect to fragmentation at alkaline pH (reviewed in reference 348). However, the problem of instability could be circumvented by carrying out the experiments under mildly acidic or neutral pH conditions. Thus, it is still not clear whether poly-ADP-ribose is covalently attached to the acceptor protein or simply associated in a noncovalent manner. Several cautionary remarks are necessary.

TABLE 3.

Postulated nuclear poly-ADP-ribose-associated proteins in vivo

Substrate(s)	Proposed functional relevance	Reference(s)
Histones H1 (H1.1, H1.2, H1.3, H1.4, H1.5), H2A, H2B, H3, H3.1, H4	Inhibition of DNA binding activity	129, 213, 322, 324, 367
High-mobility-group proteins HMGA1a, HMGA1b, HMGA2, HMGB1, HMGB2, HMGN1, HMGN2	NAa	117, 135, 324, 325
Low-mobility-group protein	NA	117
Poly-ADP-ribose polymerases PARP-1, PARP-1b, PARP-2, PARP-3, PARP-4, PARP-5, PARP-6a, PARP-6b	Inhibition of DNA binding activity?	20
Topoisomerase I, topoisomerase II	Inhibition of catalytic activity	121, 190, 355, 372
A24 protein	NA	213, 322, 368
Heterogeneous ribonucleoproteins hnRNP A1, hnRNP A2/B1, hnRNP C1/C2, hnRNP G, hnRNP H, hnRNP K, hnRNP M	NA	129
Lamin A/C	NA	3, 377
Numatrin/B23	NA	226
Nucleolin/C23	NA	226
Centromere-binding proteins CENPA, CENPB, Bub3	NA	349
MARCKS F52/MacMARCKS/MRP50/GAP-43	NA	69, 352
p53	Inhibition of DNA binding	249, 321, 372
DNA polymerase alpha	Inhibition of catalytic activity	114, 364
PCNA	Inhibition	114, 364
Telomeric repeat binding factor-1	Inhibition of DNA binding	376
CTCF	Enhancement of DNA binding	454
NuMA	NA	68

^aNA, not analyzed.

(i) In contrast to the case for other histone modifications such as acetylation, methylation, phosphorylation, biotinylation, or SUMOylation (40, 41, 74, 125, 137, 160, 201, 420), there have been no mass spectrometric studies confirming in vitro or in vivo the hypothesized covalent poly-ADP-ribosylation of proteins, including the different classical PARPs themselves (40, 41, 160).

(ii) Several reports demonstrated that free poly-ADP-ribose, including PARP-1 and PARP-2, could noncovalently bind to proteins in a salt-, acid-, and detergent-resistant manner (247, 248, 312). Many of the postulated in vitro and in vivo poly-ADP-ribosylation substrates tested in these studies bound noncovalently to free highly charged poly-ADP-ribose in vitro (247, 248, 312, 321). This noncovalent interaction turned out to be very stable and highly specific (247, 321). Long and branched poly-ADP-ribose molecules were found to be preferentially involved in the interaction with the histone variants (247).

(iii) Studies suggesting covalent poly-ADP-ribosyl-histone modifications were performed only with crude histone/nucleosomal extracts or partially purified proteins. In contrast, several groups showed that highly purified PARP-1 was unable to covalently poly-ADP-ribosylate highly purified histones (298, 451). The only observed acceptor protein for poly-ADP-ribose was PARP-1. The same authors also reported that H1 and H2B were covalently mono-ADP-ribosylated, but not poly-ADP-ribosylated, when rat liver nuclei were incubated with radioactive NAD⁺ (299, 301). More recently, using a defined chromatin assembly system containing highly purified recombinant and DNA-free proteins, M. Y. Kim and coworkers confirmed that nucleosomal core histones cannot be covalently poly-ADP-ribosylated by PARP-1 in vitro (198).

(iv) The data are even conflicting for the PARPs themselves. Earlier reports have suggested that the majority of the 25 to 30 glutamic acid residues in the automodification domain of PARP-1 are covalently poly-ADP-ribosylated by PARP-1 (107, 166). However, to date, no amino acid acceptor sites for poly-ADP-ribose have been identified in this domain.

(v) All coimmunoprecipitations and chromatin immunoprecipitation studies claiming poly-ADP-ribose modification of histones and chromatin-associated factors were performed with monoclonal or polyclonal anti-poly-ADP-ribose antibodies against free poly-ADP-ribose. Highly specific monoclonal and polyclonal antibodies recognizing only peptide-bound mono-ADP-ribose, oligo-ADP-ribose, or linear or branched poly-ADP-ribose have yet to be generated. Other studies using different anti-poly-ADP-ribose antibodies reported a direct cross-reactivity of the antibodies towards mono-ADP-ribose, oligo-ADP-ribose, and even RNA, single-stranded DNA, and Z-DNA (188, 224, 362, 363). Thus, the potential cross-reactivity of available anti-poly-ADP-ribose antibodies has to be carefully reevaluated.

(vi) It was clearly demonstrated that covalent ADP-ribosylation at cysteines and lysines, known as glycation and glycoxidation, can occur nonenzymatically via the reaction of free ADP-ribose with lysines through Schiff bases or with cysteines to form an ADP-ribosyl-thiazolidine. The thioglycoside ADP-ribosyl-cysteine linkage, however, is enzymatically formed by cysteine-specific mono-ADP-ribosyltransferases (59, 60, 174, 214).

On the other hand, based on the known molecular mechanisms of similar enzymatic polymerization reactions (i.e., synthesis of glycogen), one could assume that a covalent poly-ADP-ribose-protein linkage could exist, possibly as an intermediate. There are many similarities between the synthesis of poly-ADP-ribose and glycogen synthesis. Apo-glycogenin is postulated to serve as the primer for glycogen synthesis by utilizing UDP-glucose to transfer a first glucose residue to Tyr194 (11), subsequently elongating this into maltosaccharide. The proglycogen synthase then utilizes maltosaccharide and, together with a branching enzyme, synthesizes proglycogen (237). Macroglycogen is then synthesized by macroglycogen synthase along with a separate form of the branching enzyme (237). For further information the reader is referred to reviews on the search for and discovery of the primer for glycogen synthesis (10, 337) and especially to the recent review by Lomako et al. (237).

From the data presented above and the fact that covalent poly-ADP-ribosylation of proteins has not yet been confirmed, it is possible that the PARPs may only covalently poly-ADP-ribosylate themselves during the initiation phase and that the resulting poly-ADP-ribose chains are immediately released by the putative intrinsic ADP-ribose-protein-hydrolase activities of PARPs or by a PARP-associated PARG(s). The synthesis of cyclic-ADP-ribose by CD38 was proposed to go through a covalent glutamate-ADP-ribose intermediate, where the highly conserved Glu226 serves as the acceptor amino acid (345, 346). Thus, it is possible that the suggested covalent trans-poly-ADP-ribosylation of proteins by PARPs could correspond to poly-ADP-ribose noncovalently associated with proteins in vitro or to nonenzymatic, covalent ADP-ribosylation of proteins as a potential side effect of in vitro ADP-ribosylation reactions. In vivo, the proposed covalent poly-ADP-ribosylation activities of PARPs may be highly regulated, i.e., through additional cofactors, and thus may not be clearly detectable under the nonphysiological conditions in most in vitro poly-ADP-ribosylation assays.

Models of and methods for characterizing possible covalent poly-ADP-ribose modifications in vivo.

Based on the assumption that covalent poly-ADP-ribosylation of proteins, including PARPs, does indeed exist in vivo, three different scenarios can be postulated.

(i) Covalent poly-ADP-ribosylation occurs on preformed mono-ADP-ribosylated substrates, mediated through mono-ADP-ribosylation of protein substrates by distinct Pl-MARTs or SIRTs. These modifications might occur simultaneously through heteromerization of PARPs with Pl-MARTs and SIRTs or stepwise without heteromerization events. Over 20 years ago, Tanigawa et al. showed in vitro that covalent mono-ADP-ribose adducts of histones, generated by an arginine-specific DNA-dependent mono-ADP-ribosyltransferase(s), can serve as initiators for poly-ADP-ribose synthesis (389).

(ii) Heterodimerization, tetramerization, or heterotetramerization of distinct PARPs may be required for covalent attachment activity. Several reports have already demonstrated that a feature of the PARP family is the capacity of its members to interact with each other (PARP-1 with PARP-2 and PARP-3 or PARP-5 with PARP-6a) (reviewed in reference 20). These types of combinations/associations could completely change the enzymatic properties of each PARP.

(iii) A third possibility is that small-molecule cofactors or activator proteins may be required to switch the enzymatic properties to covalent poly-ADP-ribosylation activity. Similar to the case for bacterial toxins, the activity of PARPs and Pl-MARTs could also be regulated by ADP-ribosylation factors. Moreover, analogous to kinases, unknown adapter proteins interacting with both PARPs and their putative substrate could serve as bridging and modulatory factors.

(iv) Finally, it is possible that posttranslational modifications of PARPs induce covalent poly-ADP-ribosylation activities. Previous reports showed that the enzymatic activity of PARP-1 could be regulated by protein kinase C- or calmodulin-dependent protein kinase kinase IIδ-dependent phosphorylation (23, 180). Whether other kinases, such as extracellular signal-regulated kinases and AKT/protein kinase B (PKB), or other posttranslational protein modification enzymes, such as histone acetyltransferases, could regulate the enzymatic activities of different PARPs remains to be investigated. It was recently shown that PARP-1 could be posttranslationally modified through acetylation (159).

An ideal method to overcome the limitations of conventional approaches is the use of novel MS approaches, which are also suitable for less stable amino acid-ADP-ribose-ester bonds. These techniques are extremely powerful tools to detect new types of modifications or combinations of modifications, and the hypotheses mentioned above could be investigated. Chemical substitution of amino acid-ADP-ribose-ester linkages with nucleophilic compounds such as hydrazine and hydroxylamine and MS analysis of the corresponding amino acid derivatives such as γ-glutamyl hydroxamate are suitable for less stable amino acid-ADP-ribose-ester bonds. Another, more elegant approach, recently established by Sawatzki et al. for the identification of proteins containing lipid-modified glutamic acid residues, is the selective cleavage of the peptide bond N terminal to glutamic acid esters (348), through an approach related to the affinity-cleavage technique (84). Sawatzki et al. used sodium methoxide for the transesterification-methanolysis-based site-specific cleavage reaction, which led to formation of peptides with an N-terminal pyroglutamyl residue (348). Subsequent electrospray ionization-MS screening for the occurrence of peptides containing N-terminal pyroglutamate residues was performed to identify the initial lipid-modified glutamic acid residues (348). For a general overview and detailed description of new, highly sensitive MS methods, as well as affinity chromatography-based posttranslational modification trapping methods (including those used for protein glycosylation and glycation analysis), the reader is referred to excellent reviews on this topic (192, 193, 220, 225, 262, 265, 290, 441).

The use of mass spectrometry in combination with immunopurification analysis of single proteins or genome-wide chromatin immunoprecipitation studies will be of great interest to confirm the previously published data about mono-ADP-ribosylation and to investigate whether histones or chromatin-associated nonhistone proteins are covalently poly-ADP-ribosylated in vivo. Furthermore, the development of new in vitro reconstitution systems in combination with mass spectrometry could shed light on the types of poly-ADP-ribose modification and covalent versus noncovalent attachment under defined conditions. S. He and coworkers recently described a native chemical ligation method for the in vitro preparation of full-length histone proteins containing site-specific modifications in their amino-terminal histone tails (162). Those authors synthesized peptide thioesters corresponding to histone N termini. Various synthesized histone N-terminal peptides, containing particular site-specific modifications, were then ligated to recombinantly produced histone C-terminal globular domains containing an engineered N-terminal cysteine residue. Those authors showed that the synthetic histones generated by this method are fully functional, as verified by the self-assembly into higher-order heterotetramers, their deposition into nucleosomes by chromatin assembly/remodeling factors, and enzymatic modifications (162). Such a method, combined with mass spectrometry and amino acid exchange analysis of potential target proteins, could be utilized for studies of poly-ADP ribosylation. New in vitro reconstitution systems based on this method will not only elucidate whether a potential target can be covalently mono- or poly-ADP-ribosylated under specific conditions (i.e., screening of heterodimers and cofactors/coenzymes or coregulatory proteins) but also shed light on the epigenetic signaling mechanisms involved in the stepwise recruitment of chromatin-associated factors involved in transcription or DNA repair processes. Data obtained from these proposed studies would need to be confirmed with genetic approaches, i.e., complementation of knockout cell lines or generation of knock-in mice with enzymatic mutations of specific ADP-ribosylation enzymes or modification mutations of the identified mono- or poly-ADP-ribosylation targets.

PHYSIOLOGICAL RELEVANCE OF MONO- AND POLY-ADP-RIBOSYLATION REACTIONS IN THE NUCLEUS

There is growing evidence that nuclear ADP-ribosylation reactions are critical for many physiological and pathophysiological outcomes, such as cellular differentiation and proliferation, genome integrity, carcinogenesis, cell survival and cell death, inflammation, and neuronal functions. However, most of the studies investigating the specific functions of mono- and poly-ADP-ribosylation reactions in these processes either used nonspecific PARP inhibitors, targeting both mono- and poly-ADP-ribosylation reactions (179, 447), or were performed in vitro. Thus, the exact molecular mechanisms behind these exciting observations can be addressed only by extensive genetic studies, using mice with single or combinations of double and triple knockouts of distinct Parp, Pl-Mart, and Sirt genes. Generation of single and combinations of double or triple knock-in mice, expressing enzymatic mutations, will certainly give insight into the particular contribution of the enzymatic activity of each PARP, Pl-MART, and SIRT member in these mono- and/or poly-ADP-ribosylation-dependent processes. The major cellular processes in which nuclear ADP-ribosylation reactions were shown to be involved are discussed in detail below.

Role of ADP-Ribosylation in the Regulation of Mitosis

A recent report suggested that poly-ADP-ribose is a nonproteinaceous, nonchromosomal component of the spindle that is required for bipolar spindle assembly and chromosome segregation (67). Accurate chromosome segregation is performed by a dynamic and complex microtubule-based superstructure known as the spindle and is essential for the viability of future cell generations. The organization of microtubules into a symmetric bipolar spindle is driven by microtubule-associated proteins, which are regulated by posttranslational modifications, mainly phosphorylation, in a cell cycle-dependent manner (reviewed in references 86 and 285). Chang et al. provided evidence that both poly-ADP-ribose and PARG localize to spindles in cells, preferentially during mitosis (67). They showed that poly-ADP-ribose is enriched at spindle poles and centromere/kinetochores and throughout the body of the spindle (67). When PARG was added in excess to frog extracts, the reduction of the poly-ADP-ribose level did not alter spindle pole organization but inhibited both the formation and maintenance of bipolar spindles by disrupting the association between half-spindles (67). These observations indicate that a certain type of poly-ADP-ribose structure might be critical for establishing and maintaining spindle bipolarity in vivo, perhaps through stabilizing antiparallel microtubule interactions in the central spindle (67). There are many candidate poly-ADP-ribose-regulated spindle proteins, including the centromere proteins CENP-A, CENP-B, Bub3, and NuMA, that are known to associate with poly-ADP-ribose (68, 349, 350). In contrast to spindle proteins, which have half-lives of only seconds to minutes, almost no dynamic exchange of poly-ADP-ribose in the spindles could be observed. This is consistent with a low rate of turnover in the spindle (67). These observations strongly suggest that certain types of poly-ADP-ribose play structural rather than regulatory roles in the spindle. For example, it has been postulated that poly-ADP-ribose could serve as a nonmicrotubule lattice or matrix that contributes to force generation within the spindles (67).

Although these in vitro and ex vivo observations are very exciting, the exact physiological relevance and molecular mechanisms remain to be carefully investigated. How does constitutive or cell cycle-dependent poly-ADP-ribosylation affect the function of centromere and spindle pole proteins? Which PARP member is responsible in vivo for the observed spindle-associated poly-ADP-ribosylation? Due to the viability of the corresponding knockout mice, it would appear that neither PARP-1, PARP-2, PARP-4, nor PARP-6/tankyrase-2 alone is required for these processes, (76, 104, 170, 235, 258, 259, 266). Considering that mouse knockouts of the Parp3 and Parp5/tnks1 genes have not been made, it is too early to make a clear prediction regarding which PARP family member(s) is involved in synthesis of spindle-associated poly-ADP-ribose.

On the other hand, several knock-down studies with the PARP-5/tankyrase-1 protein using small interfering RNA (siRNA) identified PARP-5/tankyrase-1 as a factor required for mitosis in human cells (66, 68, 112). Treatment of immortalized human cells with siRNA against Parp5/tnks1 resulted in mitotic arrest, with aberrant chromosome configurations and abnormal spindle structures (66, 68, 112). Sister chromatids were unable to separate at their telomeres (66, 68, 112). Complementation of siRNA-treated cells with an siRNA-resistant Parp5/tnks1 wild-type cDNA, but not an enzymatically dead mutant, could rescue this phenotype (112), suggesting that the PARP-5/tankyrase-1-catalyzed oligo-ADP-ribosylation products are required for sister telomere resolution and mitotic progression (112). Additionally, several PARPs were shown to localize to the centromeres in a cell cycle-dependent manner. PARP-1 localized to the centrosomes and the chromosomes at the cell division phase and interphase (187). PARP-2 was reported to accumulate at the centromeres during prometaphase and metaphase, to disassociate during anaphase, and to disappear from the centromeres by telophase (349, 350). PARP-5/tankyrase-1 localized to multiple subcellular sites, including the telomeres and mitotic centrosomes. At mitosis, PARP-5/tankyrase-1 was found to relocate around the pericentriolar matrix of mitotic centrosomes (68, 375). It is possible that the observed effects are mediated through a combination of local and cell type-specific actions of PARP-1, PARP-2, PARP-3, or PARP-5/tankyrase-1. Future genetic studies will certainly clarify this point. The role of mono-ADP-ribosylation reactions, mediated by Pl-MARTs or SIRTs, also needs to be investigated.

Nuclear ADP-Ribosylation in Cellular Differentiation and Proliferation

In addition to the suggested roles of nuclear poly-ADP-ribosylation in mitosis, nuclear mono- and poly-ADP-ribosylation reactions were proposed to play major roles in cellular proliferation and differentiation. However, only the roles of plasma membrane-associated cytosolic and ecto-ADP-ribosylation reactions (mono-ADP-ribosylation activities and ADP-ribose-cyclase activities) in cellular differentiation and proliferation processes (mainly in the immune system) have been analyzed on a functional level to date (150, 194, 286, 295, 388, 406). The roles of nuclear ADP-ribosylation reactions (mono- and poly-ADP-ribosylation) in these processes have yet to be investigated. Almost all studies on the role of nuclear mono- and poly-ADP-ribosylation in differentiation and proliferation processes used nonspecific PARP inhibitors exclusively, targeting both mono- and poly-ADP-ribosylation reactions (179, 447). Although the majority of these studies suggested that high levels of mono- and poly-ADP-ribosylation activity may be required for proliferation and differentiation, the opposite effects were also observed in some studies, depending on the cell type, growth conditions, inhibitor, and inhibitor concentration used (51, 85, 98, 126, 216, 400). These discrepancies may reflect a lack of specificity in the experimental systems. Alternatively, the results may indicate the participation of distinct MARTs and PARPs that modify different substrates with opposing roles or that produce distinct types of poly-ADP-ribose structures with opposite activities. When poly-ADP-ribose levels were directly measured ex vivo, a correlation between high levels of poly-ADP-ribosylation activity and proliferation was observed (56, 61, 62, 251, 336). For instance, nuclear poly-ADP-ribosylation activity was shown to increase when quiescent 3T3 cells or B and T lymphocytes were stimulated to proliferate (56, 251, 336). The correlation between differentiation and high levels of poly-ADP-ribosylation activities may be dependent on the cell type. In primary cultures of rat astrocytes, a biphasic increase in nuclear poly-ADP-ribosylation activities was observed (62). The first rise in poly-ADP-ribosylation occurred at the beginning of cell proliferation, and the second coincided with the start of cell differentiation (after treatment with fibroblast growth factor β or nerve growth factor). However, in primary cultures of neurons, an initial high level of poly-ADP-ribosylation activities was followed by a decrease. Differentiation was progressively achieved concomitant with only a limited increase of poly-ADP-ribosylation activity during this phase (62). Poly-ADP-ribosylation was recently shown to be upregulated in spermatids during spermiogenesis (270). The highest levels of poly-ADP-ribose formation were observed in spermatids during spermiogenesis steps 12 and 13, correlating with the highest rates of chromatin nucleoprotein exchanges (270). Recent studies by J. Moss and coworkers showed that primary mouse embryonic fibroblasts isolated from Marh1 knockout mice proliferated faster than wild-type cells and when injected into nude mice generated tumors (presented at the 53rd Fujihara International Seminar [“New Challanges in Research on ADP-Ribose Metabolism”], 26 to 29 July 2004, Hokkaido, Japan; reviewed in reference 359). These preliminary data indicate that arginine-specific mono-ADP-ribosylation is required for proliferation and that mono-ADP-ribose-arginine-specific hydrolization is an important regulatory mechanism in controlling cell growth and cancer development (359). Future genetic studies will certainly clarify the potentially distinct roles of different nuclear ADP-ribosylation activities in cellular differentiation and proliferation processes.

Role of ADP-Ribosylation in the Regulation of Telomere Length and Longevity

Telomeres, the tandem-repeated hexamers at the termini of eukaryotic chromosomes, form protective complexes in association with specific proteins, such as the telomere-binding proteins Rap1, TRF-1 and TRF-2, POT-1, and PARP-5/PARP-6/tankyrase-1 and tankyrase-2, which are thought to regulate telomere length together with telomerase. In humans, telomere length maintenance is partly controlled by a feedback mechanism in which telomere elongation by telomerase is limited by the accumulation of the negative regulator, TRF-1 complex, at chromosome ends. TRF-2 serves as a protective factor at telomeres to prevent end-to-end chromosome fusions (reviewed in reference 38). Telomeres are (or are thought to be) essential for genome stability in all eukaryotes. Telomere shortening correlates with cellular senescence ex vivo; short telomeres were shown to activate replicative senescence (reviewed in reference 38). Moreover, changes in telomere length and telomerase activity have been proposed to be important factors in human aging and in the pathobiology of human disease. Telomeres shorten in most human cells during aging in vivo. In addition, short telomeres, which might compromise cell viability in vivo, correlate with age-related diseases and premature aging syndromes. Conversely, many tumor cells can prevent telomere loss by aberrantly upregulating telomerase expression levels and activity (reviewed in reference 38).

Both PARP-5/tankyrase-1 and PARP-6/tankyrase-2 localize to human telomeres and are thought to poly-ADP-ribosylate specifically the telomeric repeat binding factor TRF-1 in vitro and in vivo (87, 376). TRF-1 does not serve as an acceptor of poly-ADP-ribosylation by PARP-1 (87, 376). Whether TRF-1 is directly and covalently poly-ADP-ribosylated by tankyrase-1 and tankyrase-2 or binds specifically but noncovalently to free poly-ADP-ribose produced by these enzymes has not been established definitively either in vitro or in vivo. TRF-2 physically interacts with PARP-2 and can bind noncovalently to free poly-ADP-ribose products generated by PARP-2 in vitro (94). TRF-2 plays a key role in the protection of chromosome ends (reviewed in reference 38). Poly-ADP-ribosylation of TRF-1 by tankyrase-1 and tankyrase-2 diminished the ability of TRF-1 to bind to telomeric DNA in vitro (87, 376). Overexpression of wild-type but not enzymatic mutant forms of tankyrase-1 and tankyrase-2 in the nucleus removed the endogenous TRF-1 complex from telomeres and promoted telomere elongation in vivo, suggesting that poly-ADP-ribosylation by tankyrase-1 and tankyrase-2 regulates access of telomerase to the telomeric complex in vivo (87). A similar observation was made for TRF-2, where its DNA-binding activity was shown to be negatively regulated by PARP-2 in vitro (94).

Taken together, these data suggest that telomere extension by telomerase in human cells might be positively regulated by poly-ADP-ribosylation, mediated through PARP-5/tankyrase-1 and PARP-6/tankyrase-2. Since none of the single-knockout mice generated so far (Parp1^−/−, Parp2^−/−, Parp4^−/−, and Parp6/tnks2^−/−) showed any abnormal telomerase activity or aging-related phenotypes (76, 104, 170, 235, 258, 259, 266), genetic studies using combinations of knockouts as well as knock-in mice expressing enzymatic dead mutants of different PARPs (Parp5/tnks1^−/− and Parp6/tnks2^−/−) are needed to elucidate potential roles of poly-ADP-ribosylation in telomere-related and nonrelated processes associated with aging and cancer in vivo. Such approaches should also clarify whether mono-ADP-ribosylation reactions, mediated by Pl-MARTs or SIRTs, might be involved in these processes.

ADP-Ribosylation Reactions in Cell Death Processes

Although a specific role for nuclear mono-ADP-ribosylation reactions in cell death processes remains to be established, numerous inhibitor and genetic studies during the last two decades provide clear evidence that overstimulation of poly-ADP-ribosylation reactions plays a crucial role in cell death pathways under lethal DNA damage conditions. Both mice deficient in PARP-1 or PARP-2 and wild-type mice treated with PARP inhibitors are protected from several DNA damage-dependent pathophysiological conditions, such as postischemic damage, that cause aberrant cell death in animals proficient in PARP-1- or PARP-2-mediated poly-ADP-ribosylation (reviewed in references 176, 232, 358, and 425). Several models have been proposed to account for induction of cell death by overstimulation of poly-ADP-ribosylation: (i) energy failure-induced “programmed necrosis,” (ii) an increase in susceptibility of chromatin to cellular endonucleases and internucleosomal DNA fragmentation by poly-ADP-ribosylation of chromatin in the early stage of apoptosis, and (iii) poly-ADP-ribosylation-dependent induction of apoptosis-inducing factor (AIF)-dependent apoptosis (see below).

Mono-ADP-ribosylation reactions in apoptosis.

The roles of mono-ADP-ribosylation in cell death processes, especially in the nucleus, have not been thoroughly characterized. Several reports have demonstrated that extracellular NAD⁺ induces rapid apoptosis in naive T cells through mono-ADP-ribosylation of surface proteins mediated by the ecto-mono-ADP-ribosyltransferase e-MART-2a or e-MART-2b (4, 21, 153). Knockout mice lacking functional eMart2a or eMart2b genes were resistant to NAD⁺-mediated T-cell apoptosis (4). Further studies revealed that the ADP-ribosylcyclase/cyclic-ADP-ribose hydrolase CD38 is specifically mono-ADP-ribosylated on cysteine and arginine residues during exposure of activated T cells to NAD⁺ (21, 153). Arginine-mono-ADP-ribosylation resulted in inactivation of both cyclase and hydrolase activities of CD38, whereas cysteine-mono-ADP-ribosylation resulted only in the inhibition of the hydrolase activity. The arginine-mono-ADP-ribosylation causes a decrease in intracellular cyclic-ADP-ribose and a subsequent decrease in Ca²⁺ influx, resulting in apoptosis of the activated T cells (21, 153). Another recent study provided evidence that arginine-mono-ADP-ribosylation of actin during the execution phase of apoptosis could induce apoptotic body formation (236). Cytoskeletal features of apoptosis, leading to apoptotic body formation and release, were inhibited by meta-iodobenzylguanidine, an inhibitor of arginine-mono-ADP-ribosyltransferases. Suppression of mono-ADP-ribosylation as late as 2 h after UV irradiation blocked the depolymerization of actin and release of apoptotic bodies (236). Further investigation is needed to determine if cell death processes could also be regulated or induced by nuclear mono-ADP-ribosylation reactions mediated by Pl-MARTs or SIRTs.

Poly-ADP-ribosylation-mediated “programmed necrosis.”

In 1983, N. A. Berger et al. suggested that overstimulation of poly-ADP-ribosylation reactions might be linked to necrotic cell death (33, 34). According to their “PARP suicide” model, lethal levels of DNA damage lead to overactivation of PARP(s) and a rapid decline of cellular NAD⁺, which in turn affect the activities of the enzymes involved in glycolysis, the pentose phosphate shunt, and the Krebs cycle. In an attempt to restore the NAD⁺ pools, NAD⁺ is resynthesized, utilizing two to four molecules of ATP per molecule of NAD⁺ (depending on which salvage pathway is used in the cell). Consequently, cellular ATP levels become depleted, leading to subsequent cellular energy failure, which results in cellular dysfunction and finally in necrotic cell death (33, 34). This hypothesis was confirmed on a cellular level in numerous studies by using novel PARP inhibitors or cells from Parp1 knockout mice (reviewed in references 176, 358, and 425). Pharmacological inhibition of the enzymatic activity of PARPs or the complete absence of PARP-1 significantly improved cellular energetic status and cell viability after exposure to necrosis-inducing agents. DNA damage-induced NAD⁺ depletion is a poly-ADP-ribosylation-dependent process that can be completed within 15 min and therefore precedes the execution of the apoptotic process (reviewed in reference 93). Since ATP is required for the execution of apoptosis, overstimulation of poly-ADP-ribosylation reactions in vivo can result in necrotic cell death, even in a situation where the stimulus is proapoptotic (reviewed in references 47, 48, 99, and 449). However, NAD⁺ depletion (60 to 95% of normal levels) occurs only in the presence of very high levels of DNA damage, whereas intracellular NAD⁺ levels undergo a decrease of only 5 to 10% under moderate levels of DNA damage.

The contribution of poly-ADP-ribosylation reactions to necrotic cell death depends on the cell type and cellular metabolic status. Poly-ADP-ribosylation reactions may play an important role in necrotic cell death of various endothelial and epithelial cells, as well as several types of neuronal cells. However, necrotic cell death caused by oxidative damage in other cell types, such as hepatocytes, does not depend on poly-ADP-ribosylation reactions (reviewed in references 176, 419, and 425). Recent studies demonstrated that the cellular metabolic status is a key factor in determining how ATP levels are affected by overstimulation of poly-ADP-ribosylation reactions (449, 468). The massive generation of poly-ADP-ribose in the nucleus by overactivation of PARP-1 and, to a lesser extent, PARP-2 was suggested to preferentially deplete the nuclear and cytosolic pools of NAD⁺ but not the mitochondrial pools, thereby inhibiting glycolysis but not oxidative phosphorylation (205, 468). Actively proliferating cells use glucose almost exclusively through aerobic glycolysis for the production of ATP and die from NAD⁺ and ATP depletion resulting from overactivation of poly-ADP-ribosylation. Under similar conditions, nonproliferating cells can catabolize a mixture of metabolic substrates (including amino acids and lipids), can maintain ATP levels through oxidative phosphorylation in the mitochondria, and are resistant or less sensitive to ATP depletion and cell death (468). These nonactively growing cells are therefore more sensitive to inhibition of the mitochondrial respiratory chain. The decision between cell death and survival after exposure to necrosis-inducing agents is regulated mainly by the availability of metabolic substrates and the expression levels or enzymatic activities of PARP-1 and, in part, PARP-2 (47, 205, 468). Thus, there is clear evidence that poly-ADP-ribosylation reactions play a central role in “programmed necrosis” pathways (47, 113, 449, 468).

Poly-ADP-ribosylation reactions and apoptosis-inducing factor-dependent cell death.

Genetic studies using Parp1 knockout mice provided preliminary evidence that energy depletion alone may not be sufficient to mediate poly-ADP-ribosylation-dependent cell death (144). Moreover, ATP depletion was found to be similar in Parg-Δ2-Δ3/Δ2-Δ3-knockout and wild-type mice after ischemia, indicating that impairment of PARG-110-, PARG-102-, and PARG-99-dependent poly-ADP-ribosylation catabolism does not significantly affect the brain's energy dynamic during hyper-poly-ADP-ribosylation. Thus, other mechanisms are required for poly-ADP-ribosylation-dependent cell death (92). Several groups demonstrated that hyper-poly-ADP-ribosylation during a very high insult of genotoxic stress initiates a nuclear signal that spreads to the cytoplasm and triggers the release of AIF from the mitochondria and its translocation to the nucleus, where AIF induces apoptosis-like cell death. Pharmacological inhibition of the enzymatic activity of PARPs or the complete absence of PARP-1 in different cells derived from Parp1 knockout mice blocked the release of AIF and its translocation to the nucleus (72, 429, 442, 453).

Mitochondria play important roles in cell death through the release of proapoptotic factors such as cytochrome c and AIF, which activate caspase-dependent and caspase-independent cell death, respectively (reviewed in reference 419). AIF, located in the mitochondrial intermembrane space, is a phylogenetically conserved 57-kDa flavoenzyme which possesses both death-promoting and protective functions (199, 384, 415, 417). AIF exhibits both reactive oxygen species (ROS)-generating NAD(P)H oxidase and monodehydroascorbate reductase activities (384, 415, 417). Under normal physiological conditions, the presence of AIF is restricted to mitochondria in almost all tissues and several cancer cell lines (reviewed in references 95 and 238). Recent data indicate that the redox-active enzymatic region of AIF is associated with antiapoptotic functions, while its DNA-binding region possesses proapoptotic/necrotic activities (415, 418, 448; reviewed in references 156 and 233). Induction of cell death by high doses of genotoxic agents or N-methyl-d-aspartate causes the opening of the mitochondrial transition pore and release of AIF from mitochondria to the cytoplasm. There, it combines with cyclophilin A to form an active DNase, which in turn translocates to the nucleus and contributes to nuclear DNA fragmentation into 50-kbp fragments and chromatinolysis (429, 448, 453). AIF was shown to participate in both caspase-dependent and -independent cell death processes (429, 448, 453). Translocation of AIF has been shown to occur quickly after overactivation of poly-ADP-ribosylation reactions and precedes cytochrome c release and caspase activation (72, 429, 453). It is not clear if activation of caspases is required for the poly-ADP-ribosylation-mediated cell-death program. A recent study, based on genetic approaches and pharmacological inhibition, demonstrated that c-Jun N-terminal kinase-1 (JNK-1), but not other groups of mitogen-activated protein kinases, is required for PARP-1-induced mitochondrial dysfunction, AIF translocation, and subsequent cell death (443). Furthermore, those authors showed that receptor-interacting protein 1, tumor necrosis factor receptor-associated factor 2, and JNK constitute a pathway that mediates the effect of PARP-1 overactivation on mitochondrial dysfunction (443). AIF thus appears to be an essential downstream executor of the cell death program initiated by poly-ADP-ribosylation. Even though poly-ADP-ribosylation-mediated shuttling of AIF and cell death have so far been demonstrated in the central nervous system, this type of cell death program could also operate in other organ systems (reviewed in references 97 and 419).

However, shuttling of AIF does not depend exclusively on poly-ADP-ribosylation. The activities of other factors, such as p53 and caspase-2, were also shown to be required for this process (28, 313, 357, 461; reviewed in reference 419). Moreover, the production of ROS may be a key factor in the presence or absence of poly-ADP-ribosylation (314, 361). The observed effects may be highly cell type and condition dependent. No studies to date report that poly-ADP-ribosylation-mediated shuttling of AIF occurs in vivo in animal tissues. In addition, the role of different PARGs and Na/NMNATs in AIF shuttling needs to be carefully studied. One could expect that overexpression or deficiency of PARGs or NMNATs may modulate this process in opposing manners. Moreover, it is not clear whether the other isoforms of AIF (102) and a related factor, the AIF-homologous mitochondrion-associated inducer of cell death (AMID) (254, 439, 440), are partially poly-ADP-ribosylation dependent. AMID is a proapoptotic flavoprotein similar to AIF, with NAD(P)H oxidase activity that is localized to the outer mitochondrial membrane and the cytosol (254, 439, 440). Several studies suggested that the mechanism of cell death induction by AMID is similar to that by AIF but independent of caspases and p53, making it distinguishable from most caspase-dependent apoptosis pathways (254, 439, 440). Finally, PARP-2 was recently shown to function as a novel executor of cell death pathways in focal cerebral ischemia (205). Thus, further investigation is needed to determine if poly-ADP-ribosylation reactions mediated by family members other than PARP-1 are also required for AIF shuttling.

(i) Molecular mechanisms underlying the poly-ADP-ribosylation-mediated shuttling of AIF.

Another question is how PARP-1 activation triggers the release of AIF from mitochondria to the nucleus. The published data suggest two mechanisms: (i) NAD⁺ and ATP depletion following overactivation of poly-ADP-ribosylation reactions or (ii) poly-ADP-ribosylation products serving as a potential signal. Several studies support a model in which NAD⁺ depletion, as well as ROS-induced mitochondrial dysfunction, may lead to mitochondrial permeability transition (MPT) and trigger AIF-induced cell death. These studies demonstrated that, despite an exclusive localization of PARP-1 and poly-ADP-ribose in the nucleus, ATP levels decreased first in mitochondria and then in the cytoplasm of cells undergoing hyper-poly-ADP-ribosylation (8, 79). Treatment of cells with PARP inhibitors or submicromolar concentrations of cyclosporine, an inhibitor of MPT, and injection of NAD⁺ rescued ATP levels and blocked translocation of AIF from mitochondria to nuclei and subsequent cell death in cells undergoing hyper-poly-ADP-ribosylation (8, 79). These observations strongly suggest that NAD⁺ depletion and MPT are necessary intermediary steps linking poly-ADP-ribosylation to AIF translocation and cell death (8, 79). Furthermore, a recent report provided preliminary evidence that the ATPase domain of heat shock protein 70 (HSP-70) is critical for sequestering AIF in the cytosol under conditions of ATP depletion (341). HSP-70 antagonizes AIF-mediated cell death by both inhibiting mitochondrial AIF release and retaining leaked AIF in the cytosol (58). Although the relative importance of the ATPase domain of HSP-70 for sequestering leaked AIF in the cytosol remains controversial (58), it is possible that the interaction between HSP-70 and AIF may be regulated through the ATP/NAD⁺ levels, thus implying that HSP-70 may act as an ATP sensor under these conditions.

The second proposed mechanism is partially supported by the observation that cytoplasmic accumulation of free or protein-associated poly-ADP-ribose in Drosophila melanogaster PARG loss-of-function mutants resulted in severe neurodegeneration (154). The presence of free or protein-associated poly-ADP-ribose in the cytoplasm may be essential for proper cell death signaling. It would be interesting to investigate whether pre- or posttreatment of cells with leptomycin B, an inhibitor of chromosomal region maintenance protein 1-dependent nuclear protein export, could inhibit the cell death process under pronecrotic/apoptotic conditions. A recent report provided the first evidence that PARP-1 may translocate to the cytoplasm under cytotoxic conditions (287). When cells were exogenously treated with high levels of purified human immunodeficiency virus type 1 (HIV-1) Vpr proteins, PARP-1 shuttled to the cytoplasm in a glucocorticoid receptor complex-dependent manner (287). Several reports demonstrated that high levels of extracellular HIV-1 Vpr exhibit cytotoxicity to uninfected bystander cells through apoptotic or necrotic mechanisms (reviewed in reference 445). HIV-1 Vpr was also shown to disrupt the nuclear envelope architecture and integrity (105). Furthermore, the linker histone variant H1.2 was recently identified as an apoptogenic factor released from the nucleus to the cytosol, exclusively in response to double-strand DNA breaks (207). In the cytosol, H1.2 promotes the activation of proapoptotic Bcl-2 family proteins, mitochondrial cytochrome c release, and ultimately cell death (207). Certain poly-ADP-ribose-associated chromatin proteins, such as histones, high-mobility-group box proteins, or p53, may be used as AIF-releasing signals during apoptosis and programmed necrosis, but further studies are needed. It is quite possible that both proposed mechanisms, i.e., NAD⁺ depletion and distinct/specific poly-ADP-ribosylation products, could simultaneously serve as a trigger, depending on the stimuli, metabolic status, and cell type. This might occur in parallel or sequentially in waves.

(ii) Are poly-ADP-ribosylation reactions required for the release of HMGB1 during necrosis?

Several studies demonstrated that high-mobility-group B1 protein, a chromatin component, can be secreted by activated monocytes and macrophages and functions as a late mediator of inflammation (35, 42, 427, 428). Activation of monocytes and macrophages by inflammatory signals shifts the balance towards chromatin acetylation and leads to hyperacetylation of HMGB1, which in turn induces its relocalization to the cytosol. Acetylation of HMGB1 at two specific lysines was shown to interfere with nuclear import but not with nuclear export of HMGB1 (42). Hyperacetylated cytosolic HMGB1 is then concentrated into secretory lysosomes and secreted when monocytic cells receive an appropriate second signal (42). In addition, necrotic cells, which lose the integrity of their membranes, also leak HMGB1, which triggers inflammation by acting as a messenger of death, thereby signaling the necrotic status of cells to the surrounding tissue (35, 42). Conversely, apoptotic cells bind HMGB1 irreversibly to their chromatin (42). As with AIF release, distinct poly-ADP-ribosylation reactions may also be required for the nuclear export and release of HMGB1 during necrosis.

(iii) Does poly-ADP-ribosylation negatively regulate antiapoptotic kinases?

Several recent studies indicate that PARP-1-catalyzed poly-ADP-ribosylations may affect signaling pathways through negative or positive modulation of distinct kinase activities, which play a significant role in cell survival and cell death (310, 399, 423, 424). The first indication is from inhibitor studies, which demonstrated that several different nonspecific PARP inhibitors could enhance the endotoxin-induced or ischemia-reperfusion-induced activation of phosphatidylinositol 3-kinase (PI-3-kinase)/AKT/PKB, extracellular signal-regulated kinase 1/2, and p38 mitogen-activated protein kinase in ex vivo models. The rapid activation of the JNK-1 cascade was blocked in the presence of nonspecific PARP inhibitors (310, 423, 424). As mentioned above, a recent report provided indirect evidence that activation of JNK-1 is required for PARP-1-dependent mitochondrial dysfunction, AIF translocation, and subsequent cell death (443). More importantly, subsequent studies using PARP-1-deficient cells suggested that PARP-1-catalyzed poly-ADP-ribosylation reactions may negatively affect the cytoprotective PI-3-kinase/AKT/PKB pathway and mitogen-activated protein kinase signaling cascades (399). It is not yet clear whether the observed effect is mediated directly through PARP-1-catalyzed poly-ADP-ribosylations or indirectly through other, nonenzymatic activities of PARP-1 in the nucleus. The specificity of PARP inhibitors is in general questionable due to the nonspecific inhibition of different PARPs and MARTs, as well as their non-ADP-ribosylation-related, off-target activities. A recent report demonstrated that activation of the cytoprotective PI-3-kinase/AKT/PKB pathway is not affected in neuronal cells of Parg-Δ2-Δ3/Δ2-Δ3 knockout mice compared with wild-type mice after ischemia (92), indicating that the observed effects might be highly cell type, stimulus, or growth condition specific. The question of whether poly-ADP-ribosylation directly affects these pathways can therefore be addressed only by genetic analysis in vivo or by using knock-in mice of enzymatic mutants of distinct PARP members and subsequent poly-ADP-ribose-binding and kinase activity assays in vitro.

The current literature clearly demonstrates that poly-ADP-ribosylation reactions play an important role in apoptosis and “programmed necrosis” pathways. The PARP/poly-ADP-ribosylation system and AIF may function together as a sensor that integrates, in a “yin/yang”-like fashion, information from the mitochondria and nucleus on the metabolic and oxidative states of cells, acting as a double-edged sword in ROS-dependent death/survival pathways. On the one hand, poly-ADP-ribosylation reactions protect the animal from the development of tumors by turning off the antiapoptotic functions of certain kinases, such as AKT/PKB, and switching the dual functions of AIF towards apoptosis, thereby limiting the development of cancer. On the other hand, uncontrolled poly-ADP-ribosylation reactions can result in massive necrosis and tissue damage, which in turn often lead to severe inflammatory or neurodegenerative disorders.

Cross Talk of ADP-Ribosylation and Other NAD⁺-Dependent Reactions

Another interesting feature is the potential cross talk of mono- and poly-ADP-ribosylation reactions and other NAD⁺-dependent reactions, either directly through trans-ADP-ribosylation or indirectly through modulation of the NAD⁺ levels. The first indication for such cross talk was raised from studies with the ecto-mono-ADP-ribosylation system. A recent report demonstrated that during the exposure of activated T cells to NAD⁺, the ADP-ribosyl cyclase/cyclic-ADP-ribose hydrolase CD38 is modified by ecto-mono-ADP-ribosyltransferases specific for cysteine and arginine residues. e-MART-mediated mono-ADP-ribosylation of CD38 on arginine residues inactivates both cyclase and hydrolase activities and causes a decrease in intracellular cyclic-ADP-ribose and a subsequent decrease in Ca²⁺ influx, resulting in apoptosis of the activated T cells (153). A recent study provided evidence that e-MART-2 can sense and translate the local concentration of ecto-NAD⁺ into corresponding levels of mono-ADP-ribosylated cell surface proteins, while CD38, through its enzymatic activity, can control the level of e-MART-2-catalyzed ADP-ribosylation of cell surface proteins by limiting the substrate availability for e-MART-2 (210). These results suggest that the interaction of two classes of ecto-ADP-ribose transferases plays an important role in immune regulation by the selective induction of apoptosis in activated T cells and that cyclic-ADP-ribose-mediated signaling is essential for the survival of activated T cells. In this process, ecto-NAD⁺ functions as a signaling molecule following its release from cells by lytic or nonlytic mechanisms (153, 210).

Although there are no data on intracellular cross talk of distinct ADP-ribosylation/NAD⁺-utilizing systems, J. Zhang recently proposed that NAD⁺/nicotinamide levels could serve as converging points for interactions of PARP/poly-ADP-ribosylation reactions and SIRT-dependent pathways (459). Different poly-ADP-ribosylation reactions could modulate the NAD⁺-dependent deacetylation of proteins by SIRTs via the NAD⁺/nicotinamide connection. The decline of NAD⁺ levels and the rise of nicotinamide on activation of poly-ADP-ribosylation reactions may downregulate the activity of SIRTs due to deacetylation of Sir2 being dependent on high concentrations of NAD⁺ and inhibited by low physiological levels of nicotinamide (50% inhibitory concentration of <50 μM) (reviewed in references 106 and 459). As mentioned above, the SIRT deacetylase family has been implicated in mediating cell survival and growth, longevity, and genome stability. It was suggested that caloric restriction could extend life span by inducing SIRT1 expression and promoting the long-term survival of irreplaceable cells (82). For example, SIRT1-deficient cells exhibited p53 hyperacetylation after DNA damage and increased ionizing radiation-induced thymocyte apoptosis (73). Acetylation of p53 at lysine residues K320 and K373 was shown to be essential for the upregulation of p53-dependent proapoptotic genes (404). Moreover, SIRT1 deacetylates the DNA repair factor Ku70, causing it to sequester the proapoptotic factor Bax from mitochondria and thereby inhibiting stress-induced cell death (82). Ku70 binds to Bax in an acetylation-sensitive manner (82). Inhibition of SIRT1 enhances acetylation of Ku70 and induces release of Bax, allowing it to translocate to mitochondria and trigger cytochrome c release, leading to caspase-dependent cell death (381).

Since poly-ADP-ribosylation reactions consume the same substrate and function in the same processes but in opposite ways, it is important to test whether these SIRT1-dependent mechanisms can be modulated by poly-ADP-ribosylation. For example, how would mice with knockouts of different Parp or Pl-Mart genes respond to calorie restriction and other SIRT1-dependent conditions, e.g., during aging? Cross talk of SIRT1 and poly-ADP-ribosylation reactions may provide balance between cell survival and cell death, longevity, and senescence (459). Moreover, SIRTs and mono- or poly-ADP-ribosylation pathways may provide a unified network for multicellular eukaryotes to deal with nutritional supply and environmental stress (459). The net result, such as survival or death, proliferation or terminal differentiation, will depend on the equilibrium between specific pathways and the local cellular environment (459).

PROPOSED MOLECULAR MECHANISMS OF MONO- AND POLY-ADP-RIBOSYLATION REACTIONS

There is currently no consensus on the molecular mechanisms of mono- and poly-ADP-ribosylation reactions in the physiological processes described above. During the last two decades, various studies suggested different molecular mechanisms for how mono- and poly-ADP-ribosylation reactions are integrated in diverse physiological processes. Mono- and poly-ADP-ribosylation reactions may act on the level of signaling, modulation of chromatin structure, and epigenetic histone code.

Signaling

Several cellular pathways produce ADP-ribose-based metabolites, such as the second messengers cyclic-ADP-ribose and ADP-ribose. The functional roles of cyclic-ADP-ribose and free ADP-ribose in different signaling processes in the cytoplasm (mainly calcium signaling) are well documented. However, signaling functions of the nuclear ADP-ribose products cyclic-ADP-ribose, free mono-ADP-ribose, O-AADP-ribose, and free poly-ADP-ribose need to be further investigated (for a detailed description of the many roles of free ADP-ribose and cyclic-ADP-ribose in cytoplasmic signaling processes, the reader is referred to several excellent reviews [57, 100, 354]).

Signaling through O-AADP-ribose.

O-Acetyl-ADP-ribose production among SIR2-like enzymes is evolutionarily conserved in yeast, Drosophila, and human (reviewed in references 147 and 347). O-Acetyl-ADP-ribose may serve as a signaling molecule (46, 147). Recent studies provided evidence that O-AADP-ribose is involved in maturation of frog oocytes (46). By using a quantitative microinjection assay, the authors demonstrated that low concentrations of O-AADP-ribose caused a delay in oocyte maturation, while high doses blocked it completely (46). This effect could be mimicked by injection of low-nanomolar levels of active human SIRT2 but not a catalytically impaired mutant, suggesting that the enzymatic activity of hSIRT2 is essential for the observed effects (46). Another recent report demonstrated that O-AADP-ribose can promote the association of multiple copies of ySIR3 with ySIR2/ySIR4 and can induce a dramatic structural rearrangement in the yeast SIR complex (231). In addition, the cellular concentrations of O-AADP-ribose could be controlled by distinct mechanisms in response to appropriate conditions.

Free mono-ADP-ribose and cyclic-ADP-ribose in nuclear signaling pathways.

Nuclear calcium signaling has been a controversial field for many years. Several recent studies indicated that an inner nuclear membrane-associated and functionally active CD38/ADP-ribose-cyclase is present in mammalian cells (2, 195). Those studies also suggested that the generation of intranuclear cyclic-ADP-ribose could induce Ca²⁺ release from the luminally continuous ER/Ca²⁺ store in the nuclear envelope, independent of d-myo-inositol-1,4,5-trisphosphate (2, 195, 318). Those authors suggested that NAD⁺ levels regulate the altered nuclear Ca²⁺ homeostasis, which occurs during metabolic stress (2, 195). Thus, the amount of cellular metabolic activity may influence critical and highly Ca²⁺-sensitive nuclear processes, such as gene expression and apoptosis, via this pathway (2, 318). It was previously shown that the CREB-binding protein, CBP, is regulated by nuclear Ca²⁺ as well as by calmodulin kinase IV (71).

In addition to its role in cytoplasmic signaling processes, free mono-ADP-ribose is likely involved as a crucial second messenger in the regulation of the cellular NAD⁺ homeostasis. Free mono-ADP-ribose is the major turnover product of poly-ADP-ribose, NAD⁺, and cyclic-ADP-ribose metabolism. PARG-mediated breakdown of the high levels of poly-ADP-ribose generated due to DNA damage leads to a dramatic increase of free mono-ADP-ribose in the cell. High levels of poly-ADP-ribose can lead to the formation of advanced glycation end products or nonenzymatic glycation of protein targets (59, 60, 174, 214). It was suggested that the intracellular levels of free mono-ADP-ribose are tightly controlled by specific ADP-ribose hydrolases/pyrophosphatases, which hydrolyze ADP-ribose to AMP and d-ribose 5-phosphate, thereby acting as protective factors and limiting free mono-ADP-ribose accumulation and protein glycation (120, 332, 333).

Free mono-ADP-ribose, present in low to moderate levels, could serve as a specific messenger molecule signaling the cell's metabolic states. Specific regulatory proteins may recognize free mono-ADP-ribose through their distinct ADP-ribose-binding modules and PARPs, and PARG function may be connected to cellular NAD⁺ homeostasis.

Free poly-ADP-ribose and the “poly-ADP-ribose code.”

Several reports suggested that many types of poly-ADP-ribose structures observed in vitro also exist in vivo. Although the functional relevance of this heterogeneity is not yet known, it could play a crucial role in determining specific functional outcomes. It was suggested that certain types of free poly-ADP-ribose are involved in stress-dependent signaling processes in vivo (13, 97, 168, 246). Free or protein-associated poly-ADP-ribose may recruit or regulate the activities of signaling proteins (i.e., in DNA repair pathways or for cell cycle progression) or may activate proapoptotic/necrotic factors (i.e., nuclear translocation of AIF in cell death programs) (13, 97, 168, 246). The complexity and concentration of each structural type of poly-ADP-ribose may vary, depending on the cellular context and stimuli and also on the specific branching of different poly-ADP-ribose polymerases in vivo. Given the complexity of poly-ADP-ribose structures and the existence of at least six PARP enzymes, this heterogeneity of poly-ADP-ribose structures likely reflects the different signaling functions of PARP family members and the specificity of poly-ADP-ribose signaling pathways. Analogous to the “glyco code” of the highly diverse oligosaccharide moieties of glycoproteins, glycolipids, proteoglycans, and polysaccharides, a putative “poly-ADP-ribose code” may exist in vivo and could dictate the outcome of distinct poly-ADP-ribose signaling pathways.

Theoretically, in silico structures of poly-ADP-ribose need to be experimentally confirmed by in-depth analyses of poly-ADP-ribose structures generated in vitro and in vivo. This would be the first step in deciphering the potential “poly-ADP-ribose code.” Direct-binding experiments with each type of poly-ADP-ribose will be needed to detect poly-ADP-ribose-binding proteins, as well as the poly-ADP-ribose structures recognized by them. Direct-binding experiments will also be required to understand the cross-reactivity between related types of poly-ADP-ribose structures. These studies can be performed only with sensitive high-throughput methods. New technologies are therefore needed for the detection of poly-ADP-ribose-protein interactions. Recently, T. Feizi and W. Chai described a microarray platform for deciphering the glyco code for oligosaccharide moieties of glycoproteins, glycolipids, proteoglycans, and polysaccharides (119). Their technology involves the generation of oligosaccharide microarrays from entire glycomes, which could then be probed for protein binding. The authors suggested that these microarrays could also be coupled to techniques that allow determination of the range of proteins in proteomes that interact with carbohydrates and identification of the oligosaccharide sequences recognized by these proteins (119). Such a microarray technology adapted for poly-ADP-ribose would allow the large-scale identification of poly-ADP-ribose-binding proteins and poly-ADP-ribose-binding motifs and would enable the molecular identification of specific poly-ADP-ribose-recognition systems in whole organisms. This would allow the elucidation of the putative “poly-ADP-ribose code” for specific poly-ADP-ribose structures. Knowledge of these poly-ADP-ribose-recognition systems, including the cross-reactivity between related poly-ADP-ribose structures, would provide the tools to develop therapeutic drugs which would act through manipulation of poly-ADP-ribose-protein interactions.

ADP-ribose-binding modules.

Several reports suggested that free mono-ADP-ribose, poly-ADP-ribose, and O-AADP-ribose are recognized by specific ADP-ribose-binding motifs or modules (247, 248). Recently, Gagne et al. and Pleschke et al. provided evidence that poly-ADP-ribose can preferentially bind in a noncovalent manner to proteins carrying 20-amino-acid-long stretches containing a cluster rich in basic amino acids and a pattern of hydrophobic amino acids interspersed with basic residues (129, 321). To date, no precise binding modules functioning similarly to the 14-3-3 isoforms or bromo or chromo domains have been identified for mono-ADP-ribose, poly-ADP-ribose or O-AADP-ribose (253). Karras et al. and Kustatscher et al. recently provided strong biochemical and structural evidence that distinct human macro domains found in some proteins, including Pl-MARTs and histone variants, can serve as high-affinity binding modules for different classes of free ADP-ribose (i.e., mono-ADP-ribose, poly-ADP-ribose, and O-AADP-ribose) (189, 218). Human macroH2A1.1 could bind the SIRT metabolite O-AADP-ribose through its macro domain (218). The histone variant macroH2A contains two distinct macrochromatin domains that have been shown to be independently capable of directing macroH2A to the inactive X chromosome and macro chromatin body formation (63, 64). Moreover, some macro domains seem to recognize free poly-ADP-ribose as a ligand (189). These data strongly suggest an important regulatory role of the macro domains found in PARPs and Pl-MARTs. As mentioned above, the macro domain-containing PARPs and Pl-MARTs may be regulated through O-AADP-ribose, suggesting a cross talk between SIRTs and the macro domain-containing PARPs or Pl-MARTs. A. Ladurner and coworkers proposed that macro domains may even regulate ADP-ribosylation-dependent processes, such as necrosis or SIRT-mediated gene silencing, through the depletion of mono-/poly-ADP-ribose or O-AADP-ribose, respectively (189, 219). Whether some macro domains could also serve as binding modules for covalently protein-bound mono-ADP-ribose or protein-associated poly-ADP-ribose remains to be investigated. Because macro domains are associated with such a diverse range of biological processes, the above studies further support the important role of distinct forms of ADP-ribose in mediating different biological responses. Macro domains may therefore provide a molecular link between specific forms of ADP-ribose and key biological pathways.

Epigenetic Modification of Histones

Thirty years ago, studies suggested that the large quantity of NAD⁺ synthesis taking place in the nucleus is involved in modulation of the chromatin structure (25, 103, 127, 323; reviewed in reference 93). Chromatin or histone components, and to a lesser extent the nonhistone HMG proteins, are subject to a wide variety of covalent, reversible posttranslational modifications, such as acetylation; mono-, di-, and trimethylation on lysine residues; symmetric or asymmetric mono- and dimethylation on arginine residues; phosphorylation on serine and threonine residues; ubiquitination, biotinylation, and SUMOylation on lysine residues; and mono-ADP-ribosylation on arginine and glutamate residues (reviewed in references 40, 74, 137, 160, 177, 201, 413, and 420). Although more examples of modifications within the central domains of histones are being identified, the majority of these posttranslational modifications occur on the amino-terminal tails and carboxy-terminal domains, with a clear preference for the amino-terminal tails due to their protruding position from the nucleosome core. A detailed modification map of core and linker histones is shown in Fig. 8. The primary effect produced by modifications of amino-terminal histone tails, such as ubiquitinylation, SUMOylation, or poly-ADP-ribosylation, was thought to be through the disruption of histone-DNA interactions and/or nucleosome-nucleosome interactions, leading to open or active chromatin structures (420). However, there is no significant evidence in vivo and in vitro that posttranslational modifications, such as phosphorylation, acetylation, methylation, or mono-ADP-ribosylation, may alter nucleosomal dynamics by themselves. There is strong evidence showing that histone modifications regulate the binding of chromatin-associated nonhistone proteins to the chromatin fiber (reviewed in references 125, 177, 380, and 410). Given the number of modification sites, the fact that particular sets of modifications occur concomitantly on the same histone tail, and the fact that different types of modifications can occur on the same amino acid residues in the amino-terminal histone tails, histone modifications are likely to control the local surface structure and function of chromatin (reviewed in references 124, 160, 177, 380, 410, and 413).

FIG. 8.

Summary of known histone modifications in the human linker and core histones. The covalent modifications on histones include acetylation, phosphorylation, methylation, ubiquitination, biotinylation, and mono-ADP-ribosylation (data are from references 27, 74, 80, 91, 131, 201, 208, 209, 331, 344, and 430).

Epigenetic code.

The initial observations that acetylation may influence the initiation and/or elongation phases of transcription in a chromatin context led B. M. Turner to suggest that acetylation of histone tails could provide an epigenetic code and may be involved in long-term regulation of transcription, by which states of genetic activity or inactivity are maintained from one cell generation to the next (411, 412). B. D. Strahl and C. D. Allis established a general histone code model for the regulatory function of specific posttranslational histone modifications (380). Their histone code hypothesis predicts (i) that distinct histone modifications on a specific histone tail can occur sequentially or in combination to form a “histone code” that is read by other chromatin-associated nonhistone proteins, thereby serving as a signaling/binding platform to recruit nuclear factors that can mediate distinct downstream functions, and (ii) that specific sets of multiple covalent modifications on different histone tails may be interdependent and result in various combinations of different modifications on specific nucleosomes in the same chromatin region. Thus, the histone code hypothesis also implies the existence of a “nucleosome code.” The local concentration and combination of differentially modified nucleosomes could strongly influence certain qualities of higher-order chromatin, such as euchromatic or heterochromatic regions. This “nucleosome code” could permit the establishment of different epigenetic states, resulting in distinct “readouts” of genetic information, which may be maintained and stably inherited through mitosis and meiosis (reviewed in references 380, 411, 412, and 413).

During the past decade, many of the histone-modifying enzymes which form the basis of this code, as well as several code-reading proteins have been identified (reviewed in references 101 and 253). Some protein modules in histone-modifying enzymes are known to interact with these covalent marks in the histone tails, thereby “reading” the histone code (reviewed in references 101 and 253). The bromodomain was found to interact selectively with acetylated lysines in the histone tails (reviewed in references 101 and 253). More recently, it has been demonstrated that not only the chromodomain and chromoshadow domains but also tudor and MBT domains serve as methyl-lysine-binding protein modules that can be targeted to lysine methylation marks in the amino-terminal tails of histones H3 and H4 (197; reviewed in references 101 and 253). Finally, 14-3-3 isoforms were shown to represent a class of protein modules that recognize the phospohorylated and phosphoacetylated amino-terminal tails of histone H3 (242).

Mono-ADP-ribosylation and the “histone code.”

As mentioned above, mono-ADP-ribosylation of histones is thought to be linked to DNA repair processes and cell proliferation (50, 212). During the last two decades, several studies indicated that histones are covalently modified by mono-ADP-ribose in response to DNA damage. When cells were exposed to damage by OH radicals or methylating/alkylating agents, total covalent mono-ADP-ribosylation of histones was increased by factors of 2 to 12, while the levels of histone H1-linked mono-ADP-ribosyl groups were even elevated by more than 30-fold (1, 211, 212). Mono-ADP-ribosylation on H4 seems to occur preferentially when H4 is hyperacetylated (39, 141), suggesting a potential cross talk of histone mono-ADP-ribosylation and histone acetylation. T. Ushiroyama and coworkers provided evidence that mono-ADP-ribosylation of histone H1.3 on arginine residue R33 may reduce cyclic AMP-dependent phosphorylation of histone H1.3 on serine residue 36 (416). That study also supports the hypothesis of cross talk between mono-ADP-ribosylation and other posttranslational modifications of histones, such as acetylation or phosphorylation.

There is clear evidence that histone/chromatin modifications also play a crucial role in DNA repair processes (reviewed in reference 160). Similar to transcription, histone modifications could serve as signal amplifiers, marking the specific positions of DNA damage and providing an interaction/landing platform for the corresponding repair machinery. Since DNA damage occurs anywhere in the genome and is recognized by specific DNA damage sensor complexes, specific histone modifications may be required for signaling the presence and identifying the type of damage and/or to regulate the recruitment of specific subsets of DNA repair complexes (reviewed in reference 160). Thus, it is possible that mono-ADP-ribosylation, along with other modifications of histone tails, may regulate subsequent steps in DNA damage response pathways: mono-ADP-ribosylation could act with other initial modifications, such as acetylation and phosphorylation, as a DNA damage signal to recruit additional signaling factors and chromatin modifiers. This in turn may lead to the specification of the DNA damage type and regulating the recruitment of the appropriate repair machinery, thus determining the DNA repair pathway of choice (i.e., between nonhomologous end joining and homologous recombination repair or between short-patch and long-patch base excision repair). Since the cellular response can result, depending on the level of DNA damage, in cell cycle arrest and subsequent DNA repair, or even apoptosis/necrosis, certain levels and/or combinations of mono-ADP-ribosylation with other modifications may serve as a particular marker for the severity of DNA damage and thus regulate the pathway of choice between normal proliferation, cell cycle arrest, or apoptosis/necrosis signaling cascades (reviewed in reference 160). The recent finding that specific macro domains found in several proteins recognize different types of free ADP-ribose, thus serving as ADP-ribose-binding modules (189, 218), indicates that the covalently histone-bound mono-ADP-ribose could be read by other, not-yet-identified macro domain-containing proteins.

Poly-ADP-ribosylation and the “histone code.”

Based on the large size of poly-ADP-ribose and on the fact that covalent poly-ADP-ribosylation of histones could not be confirmed, the modification is unlikely to play a direct role in the “histone code.” Poly-ADP-ribosylation was suggested to indirectly contribute to the “histone code” by dictating the levels of local chromatin compaction (340) (see the next section). Protein-associated poly-ADP-ribose may also directly participate in the “histone code” to a minor extent.

ADP-Ribosylation-Mediated Changes in Chromatin Structure

Over 20 years ago, high-resolution electron microscopy elegantly demonstrated that poly-ADP-ribosylated chromatin adopts a more relaxed structure than its native counterpart (25, 103, 127, 222, 323). When isolated polynucleosomes of interphase chromatin were poly-ADP-ribosylated in vitro by a highly purified preparation of PARP-1 at low and moderate ionic strengths, the solenoid structure (30-nm fiber) unwound into the 10-nm fiber and adopted the fully extended “beads-on-a-string” structure characteristic of H1-depleted chromatin (25, 103, 323). Nonmodified histone H1, as well as its poly-ADP-ribose-modified form, remained associated with relaxed chromatin (127, 323). As expected, poly-ADP-ribosylation of polynucleosomes rendered chromatin more susceptible to micrococcal nuclease digestion (103, 198, 317). The high negative charge of poly-ADP-ribose may prevent interaction of poly-ADP-ribose-associated proteins with other anionic molecules such as DNA (123, 455). Most DNA-binding proteins interact with the phosphodiester backbone of DNA through cationic amino acid residues via electrostatic interactions. In contrast to all classes of covalent posttranslational modifications, noncovalent (and/or covalent?) attachment of highly anionic poly-ADP-ribose polymers to proteins severely changes the ionicity of the target protein (123, 455). Chromatin relaxation, induced by poly-ADP-ribosylation, was fully reversible following degradation of poly-ADP-ribose by exogenous PARG (133, 198). The dual action of PARPs and PARG in chromatin was suggested to result in reversible relaxation of chromatin in vivo.

However, it is still not clear if the PARP-associated poly-ADP-ribose (auto-modified PARPs) or the histone-associated poly-ADP-ribose is responsible for the relaxation of chromatin. Over 10 years ago, F. Althaus and coworkers showed that even short oligomers of poly-ADP-ribose (no longer than 40 ADP-ribose units) associated with PARP-1 are sufficient to induce relaxation of the chromatin structure without any subsequent noncovalent modification of histones (12, 328). M. Y. Kim and coworkers recently confirmed these observations, although they demonstrated that the PARP-1-associated poly-ADP-ribose could open the chromatin structure without dissociation of nucleosomes from DNA (198). These studies suggest that the association/modification of nucleosomes is not required for PARP-1-dependent regulation of chromatin structure, at least under the tested conditions (12, 198, 328). Further investigation is needed to determine if free poly-ADP-ribose, in the absence of PARP-1, can promote the relaxation of chromatin and, under certain conditions, the dissociation of nucleosomes from DNA. Based on observations in vitro, several models regarding the molecular mechanism and functional role of poly-ADP-ribosylation of chromatin in vivo were proposed. The induction of chromatin relaxation by protein-associated poly-ADP-ribose was suggested to be the key biochemical mechanism which unfolds the chromatin structure in vivo to facilitate pivotal genomic activities that are important for transcription, cellular recovery from DNA damage, or “programmed” cell necrosis (93, 340).

Regulation of DNA repair pathways.

The observed affinity of histones for free poly-ADP-ribose, especially on automodified PARPs, led to a proposal of the “histone shuttle” mechanism for chromatin relaxation and recondensation, occurring in response to DNA damage and involving PARPs and PARG (12, 246, 328). According to this hypothesis, after activation by DNA strand breaks at the site of damage, poly-ADP-ribose synthesized by PARP-1 and to a lesser extent by PARP-2 could dissociate histones/nucleosomes from DNA, thus granting the DNA repair machinery access to damaged DNA (93, 246, 328, 353). The free large and branched poly-ADP-ribose would form an interaction matrix, serving as a scaffold onto which histones could be sequestered in order to render DNA accessible to the repair machinery, facilitating repair of damaged DNA (12, 246, 328). Subsequent degradation of free and protein-associated poly-ADP-ribose by PARG would then allow histone-DNA complexes to reform, resulting in the refolding of the chromatin structure (12, 246, 328). Automodified PARP-1, PARP-2, and potentially PARP-3 could act within relaxed chromatin domains as major poly-ADP-ribose scaffolds for the transient and local sequestration of histones and for the recruitment of enzymes and cofactors involved in DNA repair, replication, and chromosome segregation (26, 93, 246, 266, 340). Based on these data, M. Rouleau and colleagues have proposed a model in which, under low physiological levels of PARPs activation, poly-ADP-ribosylation is restricted to the nucleosome core and results in establishment of a local area of structural plasticity. Overactivation of PARPs, i.e., after accumulation of DNA strand breaks, leads to hypermodification of the nucleosomes and also linker histone H1, concomitant with a full relaxation of the chromatin structure (340).

Regulation of transcriptional processes.

Only a few studies on the participation of poly-ADP-ribose in the regulation of transcription initiation have been reported during the past two decades (198, 263, 274, 370, 382, 426, 456). The experiments in that work were, for the most part, carried out in vitro, using cell-free systems and not fully chromatinized templates or permeabilized cells (263, 274, 370, 382, 426, 456). Recent genetic studies using PARP-1 enzymatic inactive mutants suggested that the presence or absence of poly-ADP-ribosylation in mammalian cells did not influence transcriptional processes ex vivo under the tested conditions (157, 158, 319, 365). On the other hand, recent studies with Drosophila melanogaster, an organism containing only two Parp genes (one Parp-1-like gene, encoding three isoforms, and one Parp-5-like gene [155, 191, 275, 409]), suggested that poly-ADP-ribose could also play a role in these processes. Disruption of gene expression of all three PARP-1-like isoforms caused larval lethality, whereas, unexpectedly, inhibition of poly-ADP-ribosylation activity by high doses of 3-aminobenzamide, a nonspecific PARP inhibitor and ROS scavenger, was not lethal. However, both blocked the accumulation of poly-ADP-ribose polymers, relaxation, and transcription at loci containing highly inducible genes (408). The dramatic increase in the micrococcal nuclease sensitivity (relaxation) was specific for heterochromatic but not euchromatic regions (409). However, one has to be cautious in comparing the poly-ADP-ribosylation systems of Drosophila melanogaster and mammals. In addition, the disruption of gene expression of all three PARP-1-like isoforms in Drosophila melanogaster was generated by P-element insertion mutagenesis and, though this is unlikely, may have resulted in additional lethal insertion mutations independent of the PARP-1 isoforms, due to the instability of this insertion mutagenesis system (138, 229).

Based on previous observations by F. Althaus and coworkers, M. Y. Kim and colleagues recently proposed a new model that automodification of PARP-1 alone is sufficient to open the chromatin structure in vitro (198, 328). In this scheme, PARP-1, when incorporated into chromatin structures and enzymatically inactive due to the local high levels of ATP, promotes the formation of compact, transcriptionally repressed and nuclease-resistant chromatin structures (198). Inhibition of enzymatic activity of PARP-1 is not observed at low concentrations of ATP (<3 mM) but is observed at higher concentrations (>6 mM) (29, 198). They proposed that PARP-1 is acutely sensitive to small changes in ATP concentration in vivo. Thus, reduction of local ATP concentrations by many transcription-related factors (i.e., chromatin-remodeling factors) utilizing ATP during transcription could increase the enzymatic activity of PARP-1. This in turn results in the automodification and subsequent release of PARP-1 from chromatin, facilitating chromatin relaxation and transcription by RNA polymerase II, without any further noncovalent modifications of nucleosomes and dissociation of nucleosomes from DNA (198). The authors suggested that poly-ADP-ribosylation, catalyzed by PARP-1, acts as a general chromatin structure-remodeling mechanism, allowing access to specific areas of the genome especially in transcriptional processes (198).

Regulation of chromatin insulator and imprinting.

A recent study by Yu et al. suggested that the chromatin insulator protein CTCF might be constitutively associated with poly-ADP-ribose (454). Chromatin insulators are defined as regulators of large chromatin domains that demarcate expression domains by blocking the cis effects of enhancers or silencers in a position-dependent manner (reviewed in references 134, 228, and 458). CTCF is currently the only known factor common to all vertebrate chromatin insulators (228). In their ex vivo study, W. Yu and coworkers used chromatin immunoprecipitation analysis with an anti-poly-ADP-ribose antibody against free non-protein-bound poly-ADP-ribose to show that poly-ADP-ribose is constitutively associated with CTCF or the surrounding nucleosomes and segregates preferentially with the maternal allele of the insulator domain in the H19 imprinting control region, containing CTCF target sites (454). They also reported that CTCF could associate with PARP-1, most likely indirectly through poly-ADP-ribose. Based on chromatin immunoprecipitation-on-chip analysis, using the same antibody, they claimed that more than 140 mouse CTCF target sites are poly-ADP-ribosylated in vivo (200, 454). This suggested that the association of poly-ADP-ribose with CTCF did not impede the binding of CTCF to most of its target sites. Using an in vitro insulator trap assay, they also showed that the insulator function of most of the CTCF target sites was sensitive to very high doses of 3-aminobenzamide, a nonspecific PARP inhibitor and ROS scavenger. However, primary cells isolated from wild-type and different Parp knockout mice were not used in their studies (454).

In their model, constitutive poly-ADP-ribosylation of CTCF stabilizes the long-range interaction between the H19 imprinting control region locus and the differentially methylated region 1 locus on the maternal chromosome, leading to insulation at the Igf2 gene, which is essential for the manifestation of the imprinted state of the Igf2 gene in vivo (200, 454). The poly-ADP-ribose may serve as “chemical glue” between CTCF and chromatin or chromatin-associated interaction partners to render these interactions constitutive. In contrast to the other models and experimental data mentioned in previous sections, the authors proposed that poly-ADP-ribosylation of CTCF is required to establish higher-order chromatin structures in vivo and will not result in chromatin relaxation. They also suggested that poly-ADP-ribosylation of CTCF is tightly linked to the various functions of CTCF and is essential for developmental processes and tumor suppression. The loss of poly-ADP-ribosylation of CTCF may, therefore, lead to imprinting imbalances in the entire genome and may have severe consequences, such as developmental defects and an increase in development of tumors (200, 454).

Changes of chromatin structure during apoptosis.

Another interesting aspect is the influence of chromatin poly-ADP-ribosylation on internucleosomal DNA fragmentation associated with apoptosis. Increased poly-ADP-ribosylation of chromatin correlated with internucleosomal DNA fragmentation, mediated by apoptosis/necrosis inducers such as DNA-damaging agents (i.e., UV light and chemotherapeutic drugs) (450). These processes could be prevented when cells were treated with nonspecific PARP inhibitors (450). Thus, poly-ADP-ribosylation of chromatin in the early stages of apoptosis could facilitate internucleosomal DNA fragmentation by increasing the susceptibility of chromatin to cellular endonucleases. Poly-ADP-ribosylation-induced relaxation of the chromatin structure, observed in vitro, may explain the internucleosomal DNA fragmentation that occurs during apoptosis.

Physiological relevance of the proposed modulation of chromatin structure.

These models are very exciting and provide potential molecular mechanisms of how poly-ADP-ribosylation, and poly-ADP-ribose in particular, could influence the chromatin structure. However, there are several questions regarding their physiological relevance. There is evidence, provided by in vivo genetic analysis, supporting a role for poly-ADP-ribosylation of chromatin in genome maintenance and apoptosis, or “programmed necrosis,” pathways (reviewed in references 160, 161, 407, and 468). Under normal physiological conditions, the proposed fundamental role for poly-ADP-ribosylation of chromatin by PARP-1 in the global organization of chromatin in mammals is to date not based on genetic analysis of PARP-1, which represents the major poly-ADP-ribosylation activity in the cell. All three different Parp1 knockout mice show no developmental abnormalities or major functional deficiencies expected from an essential chromatin regulator under normal physiological conditions (104, 258, 259, 431, 432). Thus, the possibility that the observations by M. Y. Kim and colleagues (198) may reflect the physiological situation during early stages of apoptosis or “programmed necrosis” cannot be excluded (450, 468). R. Pavri and coworkers recently demonstrated that poly-ADP-ribosylation catalyzed by PARP-1 is not required for RAR-dependent transcription in vivo. Using a chromatin reconstitution system in the presence of the whole transcription machinery, they reached the same conclusions in their in vitro studies (316). The suggested model in which poly-ADP-ribosylation of CTCF by PARP-1 should be linked to the functions of CTCF and should be essential for developmental processes (200, 454) is not supported by the different phenotypes observed in studies using Parp1, Parp2, Parp4, or Parp6/tnks2 single-knockout mice and the lethal Parp1 and Parp2 double knockouts (76, 104, 170, 235, 258, 259, 266). Neither developmental defects linked to the functions of CTCF nor any CTCF-related imprinting disorders, such as the Beckwith-Wiedemann syndrome (378), could be observed in these different Parp knockout mouse models. The only evidence so far supporting that poly-ADP-ribosylation promotes the formation of more compact chromatin structures was obtained from genetic disruption of Parp1-like gene expression in Drosophila melanogaster. This causes hyperexpression of the copia retrotransposon (up to 50-fold elevation) and enhances the variegation of certain transgenes (409). In addition, the possibility that under certain conditions, poly-ADP-ribosylation of CTCF may be linked to the tumor suppression function of CTCF cannot be excluded and remains to be investigated.

Based on the published data, poly-ADP-ribosylation of chromatin, mediated by PARP-1 alone, may be required for the modulation of chromatin structures under stress conditions but likely not under normal physiological conditions. On the other hand, given the fact that more than six different “bona fide” PARPs exist in mammals, poly-ADP-ribosylation of chromatin mediated by PARP-1 is only a part of the whole chromatin poly-ADP-ribosylation process, and thus, under normal physiological conditions, poly-ADP-ribosylation of chromatin could be mediated by PARP-1 in concert with other PARPs in vivo. Such a model would be supported, in part, by the lethal phenotype of the double-knockout Parp1 and Parp2 mice (266). Future genetic studies are needed to investigate this model.

CONCLUSIONS AND FUTURE PERSPECTIVES

A tremendous amount of work has been done over the last decade to decipher the physiological and pathophysiological roles of ADP-ribosylation reactions on the molecular level. ADP-ribosylation reactions, initially an esoteric field involving only a small community of researchers, is currently a hot topic. Many groups with a wide range of expertise have become involved in mono- and poly-ADP-ribosylation research. However, despite the progress made in recent years in the biochemistry, molecular biology, physiology, and pathophysiology of ADP-ribosylation, no unified picture of the physiological and pathophysiological roles of distinct poly-ADP-ribosylation reactions has yet emerged. Despite the development of new genetic tools and the availability of new techniques such as mass spectrometry, in vitro chromatin reconstitution systems, or in vivo chromatin immunopreciptation technologies, dozens of the most basic questions remain unanswered, including the following. Can proteins really be covalently modified by poly-ADP-ribosylation, or are the poly-ADP-ribose polymers just noncovalently associated with proteins? Do the PARPs possess intrinsic poly-ADP-ribose glycohydrolase or (mono-/poly)-ADP-ribose-amino acid hydrolase activities? What are the exact structures of distinct types of poly-ADP-ribose in vivo, and how many different structural types of poly-ADP-ribose exist in vivo? By what mechanisms are chromatin structures modulated through poly-ADP-ribosylation? What is the functional relevance of poly-ADP-ribosylation in transcription and DNA repair? Can poly-ADP-ribose have an influence on the histone code? How is the histone code modulated by mono-ADP-ribosylation of histones? Could mono-ADP-ribose serve as a histone modification marker for DNA repair?

Obviously, we are just beginning to gain insight into the biochemistry of ADP-ribosylation reactions and the ways in which cellular processes are (or may be) regulated by ADP-ribosylation. In the future, effort will have to be made to answer the following questions. How are the distinct types of the many poly-ADP-ribosylation reactions regulated in vivo? Which are the exact regulatory mechanisms underlying the distinct roles of poly-ADP-ribosylation between “programmed necrosis” and apoptosis pathways, especially in regard to AIF and HMGB1 shuttling? Is the outcome of poly-ADP-ribosylation reactions regarding these two distinct processes determined by the proposed “poly-ADP-ribose code” or, alternatively, by a shift in the homeostasis of the poly-ADP-ribosylation metabolism? Do NAD⁺/nicotinamide levels serve as a converging point for interactions of poly-ADP-ribosylation reactions and other NAD⁺-metabolizing pathways?

Although technically difficult, the question of whether proteins are covalently or just noncovalently modified by poly-ADP-ribosylation has to be urgently addressed by biochemical approaches combined with mass spectrometry techniques. The answer will undoubtedly change the field, and if the term “covalent modification” could be confirmed in vitro and in vivo, it will certainly provide opportunities for exciting new research. Moreover, deciphering the exact structures of the different types of poly-ADP-ribose in vitro and in vivo will allow the elucidation of the putative “poly-ADP-ribose code” and distinct poly-ADP-ribose recognition systems. The use of novel mass spectrometry approaches in combination with newly developed antibodies which specifically recognize distinct free or protein-bound poly-ADP-ribose structures would allow large-scale identification of poly-ADP-ribose-binding proteins and thus decipher the putative “poly-ADP-ribose-glycomes” potentially existing in mammals. Additionally, the combined application of new in vitro reconstitution systems with mass spectrometry and genome-wide chromatin immunoprecipitation methods used to operate at a resolution of less than one nucleosome will provide insight into the putative epigenetic road map of histone ADP-ribosylation for specific DNA damage pathways and transcriptional processes.

Another important aspect is the identification of yet-unknown Mart genes with non-ART-related structures, which might represent the previously identified and characterized arginine-, asparagine-, glutamate-, cysteine-, or phosphoserine-specific cytoplasmic and nuclear MART activities in mammalian cells. In addition, the enzymatic properties and substrate specificities of the novel PARP-like mono-ADP-ribosyltransferases, Pl-MART-1 to -11, need to be thoroughly investigated. The recent identification of new PARP-like ADP-ribosylating enzymes in eukaryotes has revealed a novel level of complexity in the regulation of the mono- and poly-ADP-ribose metabolism and will certainly represent an exciting new field of research.

Finally, a powerful method to establish the physiological relevance of the current and future findings is the generation of knockout mice with distinct Parp and Pl-Mart genes or knock-in mice expressing enzymatic mutants of PARP and Pl-MART family members. In addition, conventional or conditional knock-in mice (single or combinations) expressing distinct ADP-ribosylation target proteins (i.e., histones and their variants) with a single or a combination of mutated ADP-ribosylation sites are suitable tools for in vivo studies of the roles of mono- and, potentially, poly-ADP-ribosylation in transcription and DNA repair. Such studies would verify ex vivo and in vitro data and can be used to identify the sequential order of particular signaling and recruitment cascades in vivo. Targeted genetic approaches will also provide suitable in vivo platforms to study apoptosis and “programmed necrosis,” memory and neurodegenerative disorders such as Parkinson's or Alzheimer's disease, cancer, aging, or premature aging disorders and can also serve as a platform for pharmaceutical applications and preclinical trials of novel PARP/MART inhibitors.