Glycosylphosphatidylinositol-anchored and Secretory Isoforms of Mono-ADP-ribosyltransferases*

Mono-ADP-ribosylation, a post-translational modification of proteins in which the ADP-ribose moiety of NAD is transferred to an acceptor amino acid, occurs in viruses, bacteria, and eukaryotic cells (1). The reaction is distinct from that catalyzed by poly(ADPribose) polymerase, a nuclear protein involved in DNA repair, cell differentiation, and the maintenance of chromatin structure (2). Among mono-ADP-ribosyltransferases, the bacterial toxins, cholera toxin, pertussis toxin, diphtheria toxin, and Pseudomonas aeruginosa exotoxin A are the best characterized in molecular structure, function, and substrate specificity (reviewed in Ref. 1). Mono-ADP-ribosyltransferases from mammalian and avian cells have been cloned and characterized, and specific target proteins have been identified (3, 4). In lymphocytes, a glycosylphosphatidylinositol (GPI)-anchored transferase appears to be involved in immune modulation, whereas other isoforms in lymphocytes (5) and chicken heterophil granules (6) are membrane-associated but appear to be processed for secretion. Further, ADP-ribosyltransferases have been purified from brain, and data from several independent laboratories demonstrate that ADP-ribosylation is involved in neuronal function (7, 8). Deduced amino acid sequences of the vertebrate ADP-ribosyltransferases have similarities to those of viral and bacterial toxin transferases (9, 10) in regions that form, in part, an active site cleft, consistent with a common mechanism of NAD binding and ADP-ribose transfer (9). The majority of the eukaryotic enzymes are arginine-specific transferases. ADP-ribosylation of arginine appears to be a reversible process; free arginine can be regenerated in ADP-ribosylated proteins by ADP-ribosylarginine hydrolases (1). ADP-ribosylarginine hydrolase activity was detected in the soluble fraction of turkey erythrocytes, cultured mouse cells, and rat skeletal muscle with deduced amino acid sequences known for rat, mouse, and human brain ADP-ribosylarginine hydrolases (11, 12). ADP-ribosylation of cysteine was reported in bovine erythrocytes (13), and an NAD:cysteine ADP-ribosyltransferase that modified Gai was purified from human erythrocyte and platelet membranes (14). Consistent with this, ADP-ribosylcysteine linkages were detected in rat liver plasma membranes (15). ADP-ribosylation of cysteine can, however, occur nonenzymatically via the reaction of ADP-ribose, generated from NAD by NAD glycohydrolases, with cysteine to form an ADP-ribosylthiazolidine, a linkage distinct from the thioglycoside formed by pertussis toxin (PT)-catalyzed ADP-ribosylation of a cysteine in the heterotrimeric guanine nucleotide-binding (G) proteins (16). Nonenzymatic ADP-ribosylation of cysteine in proteins, however, yielded a product with the same chemical sensitivity as the linkage formed by PT (17). Based on these data, the ADP-ribose-cysteine produced by the human erythrocyte enzyme may have been generated nonenzymatically from free ADP-ribose. Because nitric oxide (NO) induced the noncovalent binding of the entire NAD molecule to a cysteine of glyceraldehyde-3-phosphate dehydrogenase (18), it is important to exclude NAD attachment to cysteine when assaying the radiolabeling of proteins with [P]NAD. This review summarizes information on the avian and mammalian ADP-ribosyltransferases and the recent advances in understanding their role in cellular metabolism.


Mammalian ADP-ribosyltransferases
The family of mammalian ADP-ribosyltransferases comprises five enzymes (ART1-5) based on similarities in their deduced amino acid sequences and conservation of gene structure (Fig. 1).
ART1-ART1 was extensively purified from rabbit skeletal muscle as a 36-kDa protein (19) and subsequently cloned from rabbit (19) and human (20) skeletal muscle and mouse lymphoma (Yac-1) cells (21). The human ART1 gene is on chromosome 11p15 (22). The murine sequence is 75 and 77% identical to those of the rabbit and human muscle enzymes, respectively (21), consistent with considerable conservation of structure across species.
The deduced amino acid sequence of ART1 possesses hydrophobic amino-and carboxyl-terminal signal peptides that are characteristic of GPI-linked proteins (19,20,23). Rat mammary adenocarcinoma (NMU) cells, transformed with rabbit or mouse ART1 cDNAs, demonstrated membrane-associated transferase activity that was released into the medium by phosphatidylinositol-specific phospholipase C (PI-PLC), which cleaves the inositol phosphatediacyl glycerol bond and solubilizes most GPI-anchored proteins. The transferase from transformed NMU cells and transferases partially purified from rabbit and human skeletal muscle reacted on immunoblot with antibodies that recognize the inositol 1,2-cyclic phosphate moiety that remains after cleavage of the GPI anchor with PI-PLC (20), consistent with the presence of a GPI anchor on native transferases. In transformed NMU cells lacking the carboxyl-terminal signal peptide, required for attachment of the GPI anchor, transferase activity was found in the medium (20).
In C2C12 mouse myoblasts, GPI-anchored ART1 activity, which appeared with differentiation of myoblasts to myotubes, catalyzed the ADP-ribosylation of integrin ␣ 7 (3). Modification of integrin ␣ 7 did not block ␣ 7 ␤ 1 heterodimer formation or its association with the cytoskeleton or laminin. Incubation of embryonic chick myoblasts in vitro with meta-iodobenzylguanidine, an alternative substrate of NAD:arginine ADP-ribosyltransferases (24), however, inhibited proliferation and differentiation of the myoblasts (25). ADP-riboseintegrin ␣ 7 was a substrate for extracellular phosphodiesterase activity that generated phosphoribosyl-integrin and 5Ј-AMP (26). Expression of skeletal muscle ART1 in parallel with integrin ␣ 7 during muscle cell development and ADP-ribosylation of integrin ␣ 7 are consistent with a regulatory role for this modification in myogenesis.
ADP-ribosyltransferase activity with properties similar to those of the cloned ART1, was detected in mouse cytotoxic T lymphocytes (CTL) (27). Incubation of CTL with 10 mM NAD resulted in ADPribosylation of membrane proteins and inhibition of CTL proliferation. A 35-kDa protein with ADP-ribosyltransferase activity was released by incubation of intact CTL with PI-PLC, with resulting partial loss of the inhibitory effect of NAD on CTL proliferation (27). Modification by the GPI-linked lymphocyte transferase of a 40-kDa membrane protein (p40) that complexes with the tyrosine kinase p56 lck resulted in inhibition of p56 lck action (4). It is noteworthy that T cell activation and lymphokine production can be mediated, in part, by signaling through GPI-anchored molecules (28,29). Further, a GPI-linked transferase in CTL modified argi-nines in the extracellular domain of the lymphocyte function-associated molecule-1 (LFA-1), an adhesion molecule of the integrin family (30). ADP-ribosylation inhibited LFA-1-mediated generation of inositol phosphates and suppressed homotypic cell adhesion. As in C2C12 cells, in CTL, 5Ј-AMP was released from the modified LFA-1 by extracellular phosphodiesterases. ADP-ribosylation of integrins in muscle cells and CTL is consistent with the hypothesis that this modification alters cell-cell or cell-matrix interactions.
ADP-ribosyltransferase activity was detected on the surface of human neutrophils, and the partial sequence of the cloned neutrophil transferase cDNA was identical to that encoding the human muscle transferase (31). Analysis of chemotaxis of intact neutrophils and actin polymerization in permeabilized cells in response to N-formyl-peptide (fMLP) without or with inhibitors of mono-ADPribosyltransferases (e.g. novobiocin, vitamin K 1 , vitamin K 3 , nicotinamide) or the alternative substrate, diethylamino(benzylidineamino)guanidine, demonstrated a close correlation among concentrations causing 50% inhibition (IC 50 ) of ADP-ribosyltransferase activity, neutrophil chemotaxis, and actin polymerization (32). The ADP-ribosyltransferase inhibitors did not, however, affect fMLP-or platelet-activating factor-induced increases in intracellular Ca 2ϩ concentration (32).
ART2-The ART1 enzymes have significant amino acid sequence identity to the RT6 (ART2) family of rodent T cell alloantigens (10,19). In rats, RT6 a and RT6 b are alleles of a single gene that encodes for the corresponding alloantigens RT6.1 and RT6.2, respectively (33). The expression of RT6 proteins appears in postthymic lymphocytes and is restricted to peripheral T cells and intestinal intraepithelial lymphocytes. RT6.1 and RT6.2 are 25-30-kDa GPI-anchored proteins that differ by only 10 amino acids. In the mouse, there are two functional copies of the RT6 gene (Rt6 -1 and Rt6 -2) located on chromosome 7. The deduced amino acid sequences of mouse Rt6 -1 and Rt6 -2 are 79% identical, whereas those of mouse Rt6 -1 and rat RT6.2 proteins are 71% identical. In humans and chimpanzees, the ART2 gene contains three premature stop codons corresponding to amino acids 47, 141, and 193 of the rat RT6 protein (33) and appears not to be expressed. In humans, the role of ART2 may be assumed by related ADP-ribosyltransferases.
In rats, the absence of RT6 ϩ T cells is associated with lymphopenia and the development of an autoimmune-mediated diabetes (34). Diabetes can be prevented in these diabetes-prone (DP BB/ Wor) rats by transfusion and engraftment of RT6 ϩ lymphocytes (35). In contrast, diabetes-resistant (DR BB/Wor) rats, depleted of RT6 ϩ T cells following infusion of anti-RT6 monoclonal antibody, have an increased incidence of diabetes (36). In agreement, the nonobese diabetic mouse has reduced levels of Rt6-specific mRNA and is prone to develop an immune-mediated diabetes. Likewise, in (NZB ϫ NZW)F1 hybrid and BSXB mice, reduced levels of Rt6 mRNA are associated with an autoimmune lupus-like glomerulonephritis (33).
NMU cells transformed with RT6.2 cDNA expressed a GPIlinked NAD glycohydrolase (37). Both RT6.2 and RT6.1 were capable of auto-ADP-ribosylation but not ADP-ribosylation of exogenous substrates (33,38,39). In contrast, mouse Rt6 -1, expressed in insect cells using the baculovirus system (40) or in NMU cells (41), catalyzed ADP-ribosylation of histones and agmatine. Unlike the rat RT6 proteins, the mouse Rt6 -1 was primarily an ADPribosyltransferase, with a relatively low level of NAD glycohydrolase activity (41). As in the rat, although reduced levels of mouse ART2 have been associated with immune-mediated disease, it has not been proven that pathogenesis of disease is related to enzymatic activity or lack thereof. Because mouse ART2 is capable of ADP-ribosylating exogenous proteins, however, this activity may result in effects similar to that observed in CTL.
ART3 and ART4 -ART3 and ART4 were recently cloned from human testis (42) and spleen (43), respectively. The deduced amino acid sequences of ART3 and ART4 possess several regions of sequence similarity with ART1 and are 14 and 31% identical, respectively, to ART1. The hydropathy profiles of the amino-and carboxyl-terminal sequences of ART3 and ART4 demonstrate hydrophobic signal sequences consistent with the possibility that ART3 and ART4, like ART1, may be GPI-linked. The ART3 gene was localized to chromosome 4q13-q21 (42), and ART4 is on chromosome 12q13.2-q13.3 (43). On Northern analysis, a 1.8-kb band in poly(A) ϩ RNA from human testis and skeletal muscle and a weaker 1.6-kb band in poly(A) ϩ RNA from heart hybridized with an ART3 cDNA probe (42). An ART4 probe hybridized with 1.4-, 2.4-, and 5.5-kb bands in poly(A) ϩ RNA from spleen, ovary, and intestine (43).
ART5-An ART5 cDNA was cloned from Yac-1 murine lymphoma cells (5). Its deduced amino acid sequence has similarities to other ART proteins in regions believed to be involved in catalytic activity and is 32% identical to that of mouse ART1, approximately 30% identical to that of mouse ART2, and 29 and 25% identical to the human ART3 and ART4 proteins, respectively. Unlike ART1, ART5 had significantly more NAD glycohydrolase than ADP-ribosyltransferase activity, and although it catalyzed auto-ADP-ribosylation, ADP-ribosylation of other proteins was relatively poor. The membrane-associated ART5 enzyme activities were not solubilized by PI-PLC. Consistent with this, ART5 possesses a hydrophobic amino-, but not carboxyl-terminal, signal sequence and may be secreted instead of GPI-linked. On Northern analysis, an ART5 cDNA probe hybridized with 1.6-and 2.0-kb bands in poly(A) ϩ RNA from testis, where it was most abundant, and a 1.6-kb band in poly(A) ϩ RNA from cardiac and skeletal muscle (5). In muscle, therefore, three ART proteins (ART1, -3, and -5) are expressed whereas ART3 and ART5 are present in testis and ART1, ART2, and ART5 are expressed in lymphocytes.
The structures of the ART1, ART2, and ART5 genes are strikingly similar (Fig. 1). The amino-terminal signal peptide, a region containing the catalytic site, and the carboxyl-terminal signal sequence in the GPI-anchored transferases are encoded in separate exons. 2 In ART1 and ART5, but not ART2, a short exon containing a basic amino acid-rich region is separated from the catalytic core of the protein, consistent with the presence of a unique structural domain in these ART proteins.
FIG. 1. Gene structure of ART proteins. Structure of the ART genes has been determined for ART1, ART2 (RT6), and ART5. Exons (boxes) encode domains common to the ART. The transferase catalytic core contains conserved amino acids critical for the formation of the active site. In ART1 and ART5, an arginine-and lysine-rich region is encoded by a separate exon, whereas in ART2, this region is included in the exon that contains the catalytic core. The GPI-anchored enzymes (ART1 and ART2) possess a carboxyl-terminal signal sequence, whereas ART5, which may be secreted, does not contain a hydrophobic signal sequence in its carboxyl terminus (black rectangle). Lines represent introns. strate MARCKS (48), GAP-43, and the GAP-43-related protein neurogranin (49), as well as the 20-kDa myelin basic protein (50). Incubation of hippocampal slices with vitamin K 1 and nicotinamide, inhibitors of mono-ADP-ribosyltransferases, blocked long term potentiation without affecting basal excitatory synaptic transmission, inhibitory postsynaptic potential, or N-methyl-D-aspartate receptor-mediated transmission (8), consistent with the possibility that ADP-ribosylation of brain proteins may affect function.
ADP-ribosyltransferase activity has been detected in a variety of cell systems. Examples include the endogenous ADP-ribosylation of G␣ s with resulting stimulation of adenylyl cyclase activity in cardiac muscle membranes, platelets, and NG108 -15 cells (51-53). ADP-ribosyltransferases from RBL (rat basophilic leukemia) and FRTL5 (Fischer rat thyroid line 5) cells modified glyceraldehyde-3-phosphate dehydrogenase and BARS-50, a novel heterotrimeric GTP-binding protein in a brefeldin A (BFA)-stimulated reaction (54). BFA inhibits protein trafficking and disrupts Golgi by blocking activation of ARF-GDP by guanine nucleotide exchange proteins (47). ADP-ribosylation of BARS-50 was correlated with BFAinduced disassembly of the Golgi (54).

Avian ADP-ribosyltransferases
The first vertebrate NAD:arginine mono-ADP-ribosyltransferases (transferases A, B, C, and AЈ) were identified and purified from turkey erythrocytes (55)(56)(57)(58). Transferases A and B were from the erythrocyte cytosol, and transferases C and AЈ from the plasma membrane and nucleus, respectively. In addition to differences in localization, the transferases had different kinetic and physical properties.
Two distinct chicken ADP-ribosyltransferases (AT1 and AT2) localized to heterophil granules were cloned from a bone marrow cDNA library (6). The deduced amino acid sequence contained an amino-terminal signal peptide but lacked a carboxyl-terminal GPIattachment signal, compatible with AT1 and AT2 being in granules and secreted (Fig. 2). To some extent, the structure of the heterophil transferases resembled that of ART5, which also appeared to be missing the carboxyl-terminal signal (Fig. 2). A third transferase from a chicken erythroblast cDNA library had a deduced sequence 50 -52% identical to those of the heterophil transferases (59). Several in vitro substrates of the heterophil transferase were identified (60) including nonmuscle actin and a 33-kDa heterophil granule protein, which is similar in amino acid sequence to the myb-induced myeloid protein-1, Mim-1. As with turkey transferase A, it was not demonstrated that in vitro reactions catalyzed by the heterophil transferase occur in cells.

Regions Conserved among ADP-ribosyltransferases
Despite lack of overall identity of deduced amino acid sequences, the bacterial toxin ADP-ribosyltransferases possess regions of sequence similarity that appear to form, in part, the catalytic site (9). Based on crystallography and computer modeling, the NAD-binding cleft of the bacterial toxins and that of eukaryotic transferases appear to be composed of an ␣-helix bent over a ␤-strand. An arginine or histidine and an active site glutamate, which is critical for enzymatic activity, are located on two ␤-strands flanking the NAD-binding cavity (9).
Two types of active sites have been proposed for bacterial toxin transferases (9). In the first, found in DT and ETA, the catalytic cleft is formed by a histidine-containing region, a segment rich in aromatic and hydrophobic amino acids, and the region of the active site glutamate. In region I, His-21 of DT is important for both NAD binding and ADP-ribose transfer to elongation factor-2, whereas in ETA, His-440 is involved in the binding of NAD but appears too distant from the N-glycosidic bond of NAD to affect ADP-ribose transfer directly (61,62). In region II of DT and ETA, a tyrosine on a ␤-strand is separated in the linear sequence by 10 amino acids from a second tyrosine on an ␣-helix (9,61). Photoaffinity labeling, site-directed mutagenesis, and crystallography have identified a region III active site glutamate (Glu-148 and Glu-553 in DT and ETA, respectively) that is critical for enzyme activity (9,61,63).
In another group of bacterial toxins exemplified by CT, LT and PT, region I contains an Arg located on a ␤-strand, region II is identified by the consensus sequence aromatic residue-X-Ser-Thr-Ser-hydrophobic residue that is positioned on the floor of the cat-alytic cleft, and region III contains the active site Glu on a ␤-strand (9). Arg-7 in region I of LT potentially binds the AMP-ribose or phosphate portions of NAD and also forms a hydrogen bond with the backbone carbonyl of Ser-61, which is located in region II (64). Ser-61 apparently maintains proper conformation of the active site of LT and does not play an essential role in catalysis (9). In region III, the carboxylate group of Glu-112 of LT and CT superimposes on that of Glu-148 of DT and Glu-553 of ETA and is in close proximity to the N-glycosidic bond of NAD (61).
The regions of amino acid sequence similarity among bacterial toxin transferases are also apparent in alignments of the mammalian ADP-ribosyltransferases (10,65). In computer modeling studies of the mouse ART2, Arg-126 on a ␤-strand (region I), Ser-147 on a ␤-strand followed by an ␣-helix (region II), and the active site Glu-184 on a ␤-strand (region III) are positioned in the catalytic cleft in a manner similar to that found in the crystal structure of the bacterial toxins, in particular, LT, CT, and PT (9,65). In the alignment of deduced amino acid sequences of ART1, ART4, and ART5, a region I arginine and region II serine similar to those in LT and PT appear to be conserved. Although His-114 of rabbit ART1 aligns with His-21 of DT, its replacement with Asn did not abolish enzymatic activity (10). A recombinant mouse ART1 protein from which the first 121 amino acids were deleted possessed NAD glycohydrolase, but not ADP-ribosyltransferase, activity (66). The truncated transferase lacks His-110 (analogous to His-114 of rabbit ART1) but possesses Arg-174, which is consistent with the hypothesis that, the transferase is similar to LT not DT and that as in LT, the conserved Arg may play a role in maintaining active site conformation and NAD binding.
Based on site-directed mutagenesis (10) and amino acid sequence alignment, the region containing Glu-X-Glu present in ART1 and ART5 is postulated to be analogous to Glu-110 and Glu-112 of CT and LT. In the mouse ART2 sequences (Rt6 -1 and Rt6 -2), Glu-207 in region III is replaced by Gln in the rat ART2 sequences (RT6.1 and RT6.2) (65). Replacement of Gln-207 of the rat ART2 (RT6.1) enzyme, which possesses predominantly NAD glycohydrolase activity, with Glu resulted in the generation of an arginine-specific ADP-ribosyltransferase and increased auto-ADPribosyltransferase activity, with little effect on NAD binding or NAD glycohydrolase activity (67,68). After introduction of a Glu-X-Glu motif, the rat ART2 protein exhibited enzymatic activity like the mouse ART2 or ART1 transferases. Thus, the first glutamate in the Glu-X-Glu motif appears to be responsible for arginine-specific ADP-ribosylation.

FIG. 2. ADP-ribosylation pathways in eukaryotic cells.
Proteins are ADP-ribosylated by intracellular NAD:arginine ADP-ribosyltransferases (ART i ), and free arginine (protein) is regenerated by ADP-ribosylarginine hydrolases. In ADP-ribosylated proteins generated by GPI-anchored transferases (ART g ), ribosylarginine is the product of sequential processing by extracellular phosphodiesterase and phosphatase activities. Other transferases such as the chicken heterophil enzyme appear to be processed for secretion (ART s ).
The deduced amino acid sequences of human poly(ADP-ribose) polymerase (PARP) (9,69) and perhaps ART3 appear to have regions of similarity that align with DT and ETA. Moreover, crystal structure of the chicken PARP (70) and mutagenesis of human PARP (69,71) demonstrated that Glu-988, which is essential for ADP-ribose chain elongation, is positioned in a cleft similar to that found in the bacterial toxins. These data are consistent with the hypothesis that several of the bacterial toxin and vertebrate transferases possess a common mechanism of NAD binding and ADPribose transfer and that differences observed in three-dimensional structures may reflect differences in substrate proteins.

Summary and Future Directions
Eukaryotic ADP-ribosyltransferase activity has been detected in diverse sources, including turkey erythrocytes, rabbit skeletal muscle, and mouse testis. The mammalian ART proteins appear to be expressed in a tissue-specific manner. In muscle cells and lymphocytes, GPI-linked ART enzymes modify integrins, consistent with a role in regulation of cell-cell or cell-matrix interactions (Fig. 2). In lymphocytes, the GPI-anchored ART1 and ART2 transferases are associated with modulation of immune function, which is intriguing and an area of active investigation. A subgroup of the ADPribosyltransferases (including ART5 and the heterophil enzymes) lacks a carboxyl-terminal signal sequence and thus may not be GPI-linked. Conceivably, because they have an amino-terminal signal sequence and have been found in chicken heterophil granules, these proteins may be secreted (Fig. 2).
An intracellular ADP-ribosylation cycle in eukaryotic cells had been proposed (72) based on the presence of NAD:arginine ADPribosyltransferases and ADP-ribosylarginine hydrolases, which remove ADP-ribose, regenerating free arginine (protein) (Fig. 2). Proteins modified by GPI-anchored transferases are processed by extracellular phosphodiesterases and phosphatases (Fig. 2). Thus, three discrete families of transferases may exist, GPI-anchored, secreted, and intracellular.