Identification and Characterization of a Mammalian 39-kDa Poly(ADP-ribose) Glycohydrolase*

ADP-ribosylation is a post-translational modification resulting from transfer of the ADP-ribose moiety of NAD to protein. Mammalian cells contain mono-ADP-ribosyltransferases that catalyze the formation of ADP-ribose-(arginine) protein, which can be cleaved by a 39-kDa ADP-ribose-(arginine) protein hydrolase (ARH1), resulting in release of free ADP-ribose and regeneration of unmodified protein. Enzymes involved in poly(ADP-ribosylation) participate in several critical physiological processes, including DNA repair, cellular differentiation, and carcinogenesis. Multiple poly(ADP-ribose) polymerases have been identified in the human genome, but there is only one known poly(ADP-ribose) glycohydrolase (PARG), a 111-kDa protein that degrades the (ADP-ribose) polymer to ADP-ribose. We report here the identification of an ARH1-like protein, termed poly(ADP-ribose) hydrolase or ARH3, which exhibited PARG activity, generating ADP-ribose from poly-(ADP-ribose), but did not hydrolyze ADP-ribose-arginine, -cysteine, -diphthamide, or -asparagine bonds. The 39-kDa ARH3 shares amino acid sequence identity with both ARH1 and the catalytic domain of PARG. ARH3 activity, like that of ARH1, was enhanced by Mg2+. Critical vicinal acidic amino acids in ARH3, identified by mutagenesis (Asp77 and Asp78), are located in a region similar to that required for activity in ARH1 but different from the location of the critical vicinal glutamates in the PARG catalytic site. All findings are consistent with the conclusion that ARH3 has PARG activity but is structurally unrelated to PARG.

Mono-ADP-ribosylation of arginine appears to be a reversible process (1). A regulatory function for an ADP-ribosylation cycle was demonstrated in the photosynthetic bacterium Rhodospirillum rubrum, where it regulates dinitrogenase reductase, a key enzyme in nitrogen fixation (23). In mammals, only one ADP-ribosylarginine hydrolase, ARH1, has been identified (12,24). It cleaves the ADP-ribose-arginine bond in modified proteins, consistent with the regulation of ADP-ribosyl(arginine) protein levels by the opposing activities of transferases and hydrolases, participating in an ADP-ribosylation cycle (9,25). ARH1, a 39-kDa soluble protein, or a similar activity was found in bacterial, avian, and mammalian cells and is ubiquitous in mammalian tissues (24, 26 -28); it is conserved across mammalian species (24,27).
The amount of ADP-ribose polymer, like that of mono-ADP-ribosylated protein, appears to be regulated, in part, by enzymes that cleave the poly(ADP-ribose) chain. These include a poly(ADP-ribose) glycohydrolase (PARG), which releases free ADP-ribose from polymers (29). Only one 111-kDa PARG has been identified in the human genome (20, 30); alternative mRNA splicing gives rise to isoforms that may differ in subcellular localization (31). The critical importance of PARG is evidenced by embryonic lethality of a PARG knockout mouse (32). The presence of multiple potential PARPs in the human genome prompted us to look for other PARGs that might differ in structure and function from the known enzyme and perhaps be involved in the degradation of poly-* This work was supported by the Intramural Research Program, NHLBI, National Institutes of Health. The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked "advertisement" in accordance with 18 U.S.C. Section 1734 solely to indicate this fact. 1  (ADP-ribose) synthesized by different members of the PARP family (29). Two other proteins in the mouse and human gene data bases, the 39-kDa ARH2 and ARH3, appear to resemble ARH1 and differ in structure from the 111-kDa PARG (12). We hypothesized that these ARH1like proteins might be capable of hydrolyzing other ADP-ribose linkages. As reported here, ARH3, a 39-kDa protein with an amino acid sequence 22% identical to that of ARH1, degraded poly(ADP-ribose) to ADP-ribose monomers; it did not hydrolyze ADP-ribose-arginine, -cysteine, -diphthamide, or -asparagine. This new member of the PARG family might have different function(s) from previously studied enzymes and could play a specific role(s) in regulation of ADP-ribose metabolism.

EXPERIMENTAL PROCEDURES
Materials-[adenine-U- 14 C]NAD (252 mCi/mmol) was purchased from Amersham Biosciences; [U-14 C]L-arginine (50 mCi/mmol) and [adenylate-32 P]NAD were from PerkinElmer Life Sciences; ␤-NAD was from Sigma; Affi-Gel boronate was from Bio-Rad; the plasmid DNA isolation QIAprep Spin Miniprep Kit was from Qiagen (Valencia, CA); the ABI PRISM Big Dye Cycle Sequencing Ready Reaction kit was from Applied Biosystems (Foster City, CA); glutathione-Sepharose 4B was from Amersham Biosciences; and cholera toxin A subunit, pertussis toxin, Pseudomonas exotoxin A, and Botulinum C3 enzyme were from List Biological Laboratories (Campbell, CA). Restriction enzymes were purchased from Roche Applied Science; bovine PARP and PARG were from BioMol (Plymouth Meeting, PA); and poly(A) ϩ RNA blots with reagents for Northern analysis were from Ambion (Austin, TX). For Western blotting, HepG2 human liver carcinoma cells were purchased from the American Type Culture Collection (Manassas, VA); goat antirabbit IgG conjugated to horseradish peroxidase was from Southern Biotech (Birmingham, AL); SuperSignal chemiluminescent substrate was from Pierce; and the QuikChange site-directed mutagenesis kit was from Stratagene (La Jolla, CA). Other reagents were of analytical grade. Custom primers were ordered from Invitrogen.
Mouse and Human ADP-ribosyl-Acceptor Hydrolase Constructs and Mutagenesis-To clone members of the ARH family, we used DNA sequences (GenBank TM ) of mouse ARH1 (mARH1; L13290), mouse ARH2 (mARH2; AJ427360), mouse ARH3 (mARH3; AJ427296), human ARH1 (hARH1; L13291), human ARH2 (hARH2; AJ313429), and human ARH3 (hARH3; AJ313333) to design forward and reverse PCR primers with unique restriction enzyme sites (Table 1). Mouse and human hydrolase cDNAs were amplified from a Marathon-Ready brain cDNA library (Clontech, Mountain View, CA) using the Advantage 2 PCR enzyme system (Clontech). PCR products were subcloned using a Zero Blunt TOPO PCR cloning kit (Invitrogen). Plasmid cDNAs were isolated (QIAprep Spin Miniprep kit; Qiagen). Mutations were generated using the Stratagene QuikChange site-directed mutagenesis method, according to the manufacturer's protocol. Complementary mutant primers used to generate ARH3 mutants are shown in Table 1. The entire coding regions were ligated into pGEX-2T expression vector (Amersham Biosciences) for transfection into Escherichia coli BL21 Rosetta supercompetent cells (Novagen, Madison, WI). Positive clones were confirmed by DNA sequencing (ABI PRISM 377; PerkinElmer Life Sciences) of the entire open reading frames in both directions. Proteins synthesized as glutathione S-transferase fusion products were purified using glutathione-Sepharose 4B according to the manufacturer's instructions (Amersham Biosciences).
Anti-ARH3 Antibodies-Rabbits were immunized with a peptide (CTDVLAQSLHRVFQESS) representing amino acids 355-370 of mouse ARH3 with cysteine added at the N terminus to facilitate coupling to keyhole limpet hemocyanin. Antibodies were purified from sera of two rabbits, using a peptide affinity column.
To prepare pure nuclei, the crude nuclear fraction was washed once with TKMS buffer, incubated at 37°C for 45 min in 2 ml of TKMS buffer, washed twice with TKMS buffer, and applied to the top of a sucrose gradient (2-ml layers of TKMS buffer containing 2.5, 2.25, 2.0, 1.75, and 1.5 M sucrose), which was then centrifuged at 100,000 ϫ g for 90 min at 4°C. Pure nuclei were collected at the 1.75-2 M interface and washed twice with TKMS buffer (33).
Samples (25 g) of homogenate proteins and recombinant ARH3 (25 ng) were subjected to SDS-PAGE in 4 -12% gels and transferred to nitrocellulose membranes, which were reacted with antibodies against ARH3 (0.5 g/ml). Secondary goat anti-rabbit IgG antibodies conjugated to horseradish peroxidase (Southern Biotech) were detected using SuperSignal chemiluminescent substrate (Pierce), followed by exposure to x-ray films (Eastman Kodak Co., Rochester, NY).
Reactions were stopped by the addition of 20% trichloroacetic acid (1 ml), and after 1 h on ice, precipitated proteins were sedimented by centrifugation (16,000 ϫ g, 4°C, 30 min), washed three times with icecold acetone, and dissolved with 50 mM potassium phosphate, pH 7.5, to be used as substrates for ARH assays.
Assay of ARH Activity Using Autoradiography-Samples (50 g) of [ 32 P]ADP-ribosylated proteins synthesized by bacterial toxins were incubated with the indicated ARH (5 M) in 50 mM potassium phosphate, pH 7.5, 10 mM MgCl 2 , and 5 mM DTT (total volume, 25 l), at 37°C for 2-2.5 h or at 30°C for 2 h or overnight. After termination of the reactions with the addition of 6 l of 5ϫ Laemmli buffer, samples of proteins (30 g) were separated by SDS-PAGE, in 12 or 4 -20% polyacrylamide gels, and transferred to nitrocellulose membranes, which were exposed to x-ray films for 10 h (Kodak).
Protein-free poly(ADP-ribose) was prepared using dihydroboronyl-Bio Rex affinity resin as described (38)  Poly(ADP-ribose) Hydrolysis by Autoradiography-[ 32 P]Poly(ADPribose)PARP (500 ng of PARP with poly(ADP-ribose)) was incubated (37°C, 2 h) with 2 M ARH in 25 l of 50 mM potassium phosphate buffer, pH 7.5, with or without 10 mM MgCl 2 and/or 5 mM DTT. Reactions were stopped by adding 6 l of 5ϫ Laemmli buffer. Samples (20 l) of proteins were separated by SDS-PAGE in 4 -20% gels and transferred to nitrocellulose membranes, which were exposed to x-ray film (Kodak) for 10 h.
Hydrolysis of Poly(ADP-ribose) by PARG and ARH3-Purified [ 32 P]poly(ADP-ribose) (5.5 ϫ 10 5 cpm, ϳ300 nM ADP-ribose) was incubated for the indicated time at 37°C in 25 l of 50 mM potassium phosphate, pH 7.5, containing 10 mM MgCl 2 and 5 mM DTT with enzyme or other additions as indicated; the reaction was terminated by the addition of 25 l of 2ϫ electrophoresis buffer (Invitrogen).   JANUARY 13, 2006 • VOLUME 281 • NUMBER 2 trophoresis at 400 V was stopped when bromphenol blue had moved 9 cm from the origin. Gels were exposed to x-ray films (Kodak). HPLC Analysis-HPLC (with a Hewlett-Packard series 1100 high pressure liquid chromatograph equipped with a diode array spectrophotometric detector set at 254 nm) was used for separation of reaction products. For anion exchange perfusion chromatography, a Zorbax SAX column (4.6 ϫ 250 mm; DuPont) was washed with 20 mM potassium phosphate, pH 4.5, for 30 min, followed by a linear gradient of 0 -1 M NaCl in the same buffer for 10 min (30 -40 min) and then the same buffer with 1 M NaCl for 10 min (40 -50 min) at a flow rate of 1 ml/min. Nicotinamide was eluted at 3 min, NAD at 10 min, and ADP-ribose at 39 min. For reverse phase, a Discovery C18 column (4.6 ϫ 250 mm; SUPELCO, Bellefonte, PA) was used. Samples were separated isocratically with 100 mM potassium phosphate, pH 6.0, containing 7% methanol at a flow rate of 1 ml/min. ADP-ribose was eluted at 6 min, NAD at 8 min, and nicotinamide 9 min.  The sources of amino acid sequences are given in Table 2. A, amino acid sequences of human ARH1, -2, and -3, with amino acid acids critical for ARH1 activity indicated by asterisks (41). B, amino acid sequences of human PARG catalytic domain and human ARH3. *, mutated amino acids in ARH3 (D77N/D78N); ϩ, amino acids reported critical for activity based on mutagenesis of PARG (42); #, mutated amino acids in ARH3 (E238Q/E239Q); carat, mutated amino acids in ARH3 (E261Q/ E262Q). Deduced amino acid sequences of human ARH1, -2, and -3 or human ARH3 and human PARG catalytic domain were aligned by the T-Coffee method (43) using the original program available at the author's site on the World Wide Web (www.ch.embnet.org/software/TCoffee.html). Identical sequences are in white letters on black. Conserved amino acids (see legend to Table 2) are shaded gray. Additional details are in the legend to Table 2.

RESULTS
Amino acid sequences of human ARH1, -2, and -3 and the catalytic domain of PARG are aligned for comparison in Fig. 1. Amino acid sequences of human ARH1 and ARH2 are 45-47% identical but only ϳ22% identical to that of ARH3 (Table 2). Overall, ARH3 is more identical to the catalytic region of the 111-kDa PARG (19%) than to ARH1 (10%) or ARH2 (13%).
ARH3 mRNA and protein in tissues and cells were detected using Northern and Western blot analyses, respectively. ARH3 appeared to be widely expressed. The full-length ARH3 cDNA hybridized with a single ϳ1.6-kb band in 10 mouse tissues tested ( Fig. 2A). Similarly, ϳ39-kDa immunoreactive ARH3 was seen on Western blots of the same tissues (Fig. 2B). Prominent second bands were present in brain and liver with minor secondary bands in heart and kidney (Fig. 2B). The anti-ARH3

. Hydrolysis by ARH1, -2, or -3 of proteins mono-ADP-ribosylated by bacterial toxins.
A, samples (50 g) of mouse brain membranes [ 32 P]ADP-ribosylated by incubation with cholera toxin were incubated at 30°C overnight with 5 M recombinant ARH1, -2, or -3 or BSA and analyzed as described under "Experimental Procedures." Cont, brain membranes that had been incubated with 32 P-labeled NAD without CTA. B, samples of [ 32 P]ADP-ribosylated G␣ i /G␣ o (20 g) synthesized by pertussis toxin (PT), [ 32 P]ADP-ribosylated EF-2 in mouse brain cytosol (total 100 g of protein) synthesized by P. aeruginosa exotoxin A (ExoA), or [ 32 P]ADP-ribosylated Rho in mouse brain cytosol (total 50 g protein) synthesized by C. botulinum C3 toxin, all generated as described under "Experimental Procedures," were incubated at 30°C overnight with a 5 M concentration of the indicated ARH, BSA, or glutathione S-transferase and analyzed as described under "Experimental Procedures." Cont, substrate that had been incubated with 32 P-labeled NAD without bacterial toxin.

39-kDa Poly(ADP-ribose) Glycohydrolase
antibodies did not react with ARH1 or ARH2 on Western blots (data not shown). Immunoreactive ARH3 was present in both cytosolic and nuclear, but not membrane, fractions from mouse brain and liver. In the brain, the cytosolic protein migrated slightly faster than that in the nuclear fraction, whereas in the liver, two proteins of similar size were present in both fractions (Fig. 2C). The cytosol fraction from HepG2 cells contained a band of ϳ38 kDa, apparently corresponding to the smaller of the immunoreactive proteins in brain and liver cytosol (Fig. 2C). No ARH3 was detected in the purified HepG2 cell nuclei, in contrast to its presence in the crude and pure nuclear fractions from tissues.
HPLC established that ADP-ribose was the product of the ARH3catalyzed reaction, based on its co-elution with authentic ADP-ribose with a retention time of 39 min and its co-migration with the product of the PARG-catalyzed reaction. Release of ADP-ribose from poly(ADPribose) catalyzed by ARH3 was markedly enhanced by 10 mM MgCl 2 (Fig. 4B), consistent with its effect in the experiment in Fig. 4A.
Identity of the main product of the ARH3-catalyzed reaction was confirmed by HPLC using C18 and SAX columns. Products eluted from C18 HPLC (Fig. 5A) (fractions 5-7) were concentrated by an evaporator to 200 l, which were applied to a Zorbax SAX column, where it was eluted with a retention time of 39 min, corresponding to that of ADPribose (Fig. 5B). The rate of release of ADP-ribose from [ 14 C]poly(ADP-
To obtain reaction products for characterization, protein-free [ 32 P]poly(ADP-ribose) purified by DHB-Bio-Rex affinity resin was used as the substrate. After incubation with human ARH3, mouse ARH3, or PARG, products were analyzed by high resolution PAGE and autoradiography. Purified [ 32 P]poly(ADP-ribose) was hydrolyzed to lower molecular weight species in a time-dependent manner with all three enzymes (Fig. 7). Migration of the smallest molecular product (and with time the most abundant) corresponded to that of ADP-ribose, not phosphoribosyl-AMP, which is generated by phosphodiesterase cleavage of poly(ADP-ribose). Long polymers of ADP-ribose appear to be hydrolyzed first, similar to what is seen with PARG (40).
To determine whether the catalytically inactive ARH3 (D77N/D78N) protein was structurally intact, binding of ADP-ribose was measured. After incubation of ARH3 with [ 14 C]ADP-ribose, unbound ADP-ribose was removed by binding to Affi-gel boronate (Bio-Rad), and ARH3 with ADP-ribose bound was collected for radioassay. In the absence of magnesium, ADP-ribose binding by D77N/D78N and wild-type ARH3 was increased somewhat by DTT (Fig. 9). Binding was much lower in the presence of magnesium; DTT increased [ 14 C]ADP-ribose binding by wild-type ARH3 but not by the D77N/D78N mutant (Fig. 9). Heat inactivation abolished binding. Binding by wild-type or mutant ARH3 (D77N/D78N) was similar, although the mutant enzyme was catalytically inactive.  15,000 cpm, ϳ850 nM ADP-ribose) were incubated at 37°C (total volume, 100 l) with human ARH3 as indicated. Reactions were terminated with the addition of 5 l of o-phosphoric acid (final pH 2-2.5) and placed on dry ice. Just before HPLC analysis, 100 l of 100 mM potassium phosphate buffer, pH 6.0, containing 7% methanol, were added to each sample, and 200 l of the mixture were applied to HPLC on a Discovery C18 column as described under "Experimental Procedures." A, assays containing the indicated concentration of ARH3 were incubated for 60 min. B, assays containing 40 nM ARH3 were incubated for the indicated time. Data in A and B are means Ϯ one-half of the range of values from two experiments with triplicate assays, and findings were similar in four experiments with human or mouse ARH3. FIGURE 7. Size of ADP-ribose polymers after incubation with mouse or human ARH3 or PARG from calf thymus. Samples (5.5 ϫ 10 5 cpm, ϳ300 nM ADP-ribose; 20 l) of [ 32 P]poly(ADP-ribose) were incubated for 2, 10, or 60 min with 1 M human or mouse ARH3 (hARH3 and MARH3, respectively) or 1.5 nM PARG, as described under "Experimental Procedures." Products were separated by high resolution PAGE in 20% gel and quantified by autoradiography. Cont, reaction without enzyme incubated for 60 min; on the right, 32 P-labeled standards are ADP-ribose (ADPR), ␤NAD (PerkinElmer Life Sciences) (NAD), AMP, and phosphoribosyl-AMP (PRAMP). Bromphenol blue (BPB) and xylene cyanol (XC) co-migrated with (ADP-ribose) 8 and (ADP-ribose) 18 , respectively. Data are representative of four experiments. JANUARY 13, 2006 • VOLUME 281 • NUMBER 2

DISCUSSION
We have shown that the recombinant ϳ39-kDa ARH3 protein, encoded in the human and mouse genomes, exhibited poly(ADP-ribose) glycohydrolase activity. Ubiquitous expression of ARH3 in mouse and human tissues was shown by both Northern and Western analyses. The protein appeared to be specific in its cleavage of poly(ADP-ribose), since ADP-ribose-arginine, -cysteine, -asparagine, and -diphthamide synthesized enzymatically by different bacterial toxin ADP-ribosyltransferases were not hydrolyzed. Thus, the substrate specificity of ARH1, which cleaves the ADP-ribose-arginine linkage, and that of ARH3 were clearly different. ARH1 and ARH3 are otherwise similar in molecular size (39 kDa), in amino acid sequence, and in ability to bind free ADP-ribose.
Although human genes encoding 18 different PARP-related proteins are known (16), only one PARG has been found (20). Similarly, ARH1 is the only gene thus far known that encodes an enzyme capable of hydrolyzing ADP-ribose-arginine (12,26,27). Our data are consistent with the existence of a second protein, ARH3, with PARG-like activity.
Although PARG and ARH3 appear to be structurally very different, with sizes of 111 and 39 kDa, respectively, in fact, their catalytic domains exhibit some similarities, with identities of amino acid sequences similar to those of ARH3 and ARH1. ARH1, ARH3, and the PARG catalytic domain all contain pairs of vicinal acidic amino acids, aspartate or glutamate (41,42). It has been demonstrated (41) that the first pair of vicinal aspartates, seen in ARH1, are conserved among ADP-ribose-(arginine)protein hydrolases from bacteria (R. rubrum) to humans and are required for activity. Replacement by alanine decreased activity to 10 Ϫ5 -fold that of the wild-type molecule (41). In PARG, although three sets of vicinal acidic amino acids are present, it is the last pair of glutamates, rather than the first two, that are critical (42). Among the three pairs of acidic residues, in ARH3, the first set, Asp 77 and Asp 78 , is the one necessary for activity. Replacement of the third pair (Glu 261 and Glu 262 ) with alanine did not affect hydrolase activity significantly. Thus, in this regard, the 39-kDa ARH3 appears to be more similar in structure to ARH1 than to the PARG catalytic site.
Activities of mammalian ADP-ribose-(arginine)protein hydrolases (ARH1) exhibit a dependence on Mg 2ϩ and, in some species, also require thiol (24,27). ARH3 activity appears to need Mg 2ϩ , but not thiol, for cleavage of poly-ADP-ribose and generation of mono-ADP-ribose. In this respect, ARH3 and ARH1 from the same species are different. Mutagenesis had shown that replacement of a critical cysteine with serine in the rat ARH1 resulted in loss of the thiol dependence (24); human ARH1 has a serine at that position and is not thiol-dependent (24). ARH3 from mice and humans contain cysteines, but not in the position that determines thiol sensitivity in ARH1; thus, not all cysteines are conserved across the ARH family, and there is no reason to expect ARH3 to exhibit a thiol dependence.
Eighteen PARP-like enzymes with both nuclear and cytoplasmic localizations have been identified in the human genome (16,22). To date, however, only one PARG has been described (20, 29). By alternative splicing, however, both nuclear and cytoplasmic PARG isoforms could be generated (31). In addition to PARG, other enzymes acting on FIGURE 8. Effect of mutagenesis of human ARH3 on its hydrolysis of poly(ADP-ribose). A, assays without enzyme (Cont) or with 1 M ARH3 (wild-type (WT) or mutant) or 1.5 nM PARG were incubated for 10 or 60 min, as described under "Experimental Procedures," before separation of products and autoradiography. Cont, reaction without enzyme. Positions of standards are indicated as in Fig. 7. Data are representative of four experiments. B, assays carried out for 1 h as in Fig. 7 with [ 14 C]poly(ADP-ribose)PARP (600 ng, 35,000 cpm, ϳ2 M ADP-ribose; 100 l) replacing [ 32 P]poly(ADP-ribose) were incubated without enzyme (Cont) or with 50 nM human ARH3 (wild-type or mutant) or 1 nM PARG before radioassay of ADP-ribose separated by HPLC on a Discovery C18 column as described under "Experimental Procedures." *D77N/D78N 5 M samples were incubated overnight. Data are means of values Ϯ S.D. from four experiments in duplicate.

39-kDa Poly(ADP-ribose) Glycohydrolase
the pyrophosphate bond or the ADP-ribose-amino acid linkage could degrade the ADP-ribose polymer (45)(46)(47). Thus, regeneration of unmodified target would not necessarily be PARG-dependent. The studies reported here show that there is an alternative enzymatic activity for the degradation of poly(ADP-ribose), involving an enzymatic reaction similar to that catalyzed by PARG.