Structure of human ADP-ribosyl-acceptor hydrolase 3 bound to ADP-ribose reveals a conformational switch that enables specific substrate recognition

ADP-ribosyl-acceptor hydrolase 3 (ARH3) plays important roles in regulation of poly(ADP-ribosyl)ation, a reversible post-translational modification, and in maintenance of genomic integrity. ARH3 degrades poly(ADP-ribose) to protect cells from poly(ADP-ribose)–dependent cell death, reverses serine mono(ADP-ribosyl)ation, and hydrolyzes O-acetyl-ADP-ribose, a product of Sirtuin-catalyzed histone deacetylation. ARH3 preferentially hydrolyzes O-linkages attached to the anomeric C1″ of ADP-ribose; however, how ARH3 specifically recognizes and cleaves structurally diverse substrates remains unknown. Here, structures of full-length human ARH3 bound to ADP-ribose and Mg2+, coupled with computational modeling, reveal a dramatic conformational switch from closed to open states that enables specific substrate recognition. The glutamate flap, which blocks substrate entrance to Mg2+ in the unliganded closed state, is ejected from the active site when substrate is bound. This closed-to-open transition significantly widens the substrate-binding channel and precisely positions the scissile 1″-O-linkage for cleavage while securing tightly 2″- and 3″-hydroxyls of ADP-ribose. Our collective data uncover an unprecedented structural plasticity of ARH3 that supports its specificity for the 1″-O-linkage in substrates and Mg2+-dependent catalysis.

is required for maintaining genomic integrity and cellular responses to DNA damage (1,2). The addition of poly(ADPribose) (PAR) to proteins by PAR polymerase 1 (PARP1; also known as ARTD1) plays a pivotal role in the repair of DNA single-and double-strand breaks (3)(4)(5)(6). However, excessive PARylation by PARP1 often interferes with protein function and may activate parthanatos, a PAR-dependent cell death pathway, which involves the translocation of PAR to the cytoplasm, eventually triggering the PAR-dependent release of apoptosis-inducing factor (AIF) from mitochondria (7,8). AIF released from mitochondria then goes to the nucleus, leading to DNA cleavage and cell death. The cellular level of PARylation and PAR is therefore tightly controlled by PAR synthesis and turnover.
In mammals, two enzymes, ADP-ribosyl-acceptor hydrolase 3 (ARH3; also known as ADPRHL2) and PAR glycohydrolase (PARG), function in tandem to reverse PARylation (9). These hydrolytic enzymes commonly cleave the ␣(1Љ-2Ј) O-glycosidic linkages in PAR chains (Fig. 1a) (10,11). PARG has both exoand endoglycohydrolase activities, acting at terminal and internal sites of PAR chains, respectively (9,(12)(13)(14). Thus, in addition to free ADP-ribose (ADPR), PARG generates short chains of PAR that serve as a potent cell death signal (9). In contrast, ARH3 appears to catalyze primarily exocytic cleavage of PAR, generating free ADPR (9). Consistent with this difference in biochemical activity, ARH3 protects cells from oxidative stress-induced parthanatos by lowering the cytoplasmic PAR level (9). ARH3 knockout cells are healthy under unstressed conditions. However, following H 2 O 2 -induced DNA damage, these cells show accumulation of cellular PAR, in particular in the cytoplasm where it interacts with mitochondria, leading to AIF cleavage and release, enhanced nuclear accumulation of AIF, and overall increased activity of the parthanatos pathway (9). Taken together, these findings indicate that ARH3 is a key enzyme that not only controls PAR content but also determines cell fate during the DNA damage response.
The structure of the unliganded human ARH3, lacking the N-terminal 16 residues (ARH3 ⌬N16 ), shows a compact all-␣helical fold with a binuclear Mg 2ϩ center, which constitutes an archetype of the ARH superfamily (25). This di-Mg 2ϩcontaining catalytic center (Mg A and Mg B ) is not found in PARG and is consistent with the Mg 2ϩ -dependent ADP-ribosyl-acceptor hydrolase activity of ARH3 (11,23). Consistently, mutations of Mg 2ϩ -coordinating residues of ARH3 led to a drastic decrease in PAR-and MAR-acceptor-hydrolyzing activities (22,25). However, we lack a fundamental mechanistic understanding of how ARH3 specifically recognizes and cleaves structurally diverse substrates in a Mg 2ϩ -dependent manner.
To better understand the molecular mechanism of ARH3 activities, we determined high-resolution crystal structures of full-length human ARH3 bound to Mg 2ϩ and ADPR, a product and an effective inhibitor of ARH3 activities (23,24). Coupled with computational analysis of ARH3-PAR substrate interactions, these structures reveal that substrate binding drives a large-scale conformational transition from an unliganded closed state to an open incision-competent enzyme state. Glu 41 from the "glutamate flap," which blocks the entry of substrate in the closed state, is completely ejected from the binuclear Mg 2ϩ center. The concomitant restructuring of the active site uncovers the catalytically important Mg B , widens the leaving groupbinding channel, and precisely positions the scissile 1Љ-O-linkage for hydrolysis while tightly securing the 2Љ-and 3Љ-OH groups of ADPR. Unique structural features and conformational flexibility of ARH3 strongly support ARH3's specificity for the 1Љ-O-linkage in structurally diverse substrates and Mg 2ϩ -dependent catalysis.

Structure of human ARH3 FL -ADP-ribose-Mg 2؉ complex
The crystal structure of apo-ARH3 ⌬N16 explained the lack of ADPR binding (25) (Fig. S1). We reasoned that the full-length ARH3 (ARH3 FL ) might provide opportunities to capture ARH3 in an ADPR-bound active form. The purified ARH3 FL is likely metal-free given its comparable basal activity without the addition of Mg 2ϩ and in the presence of EDTA (Fig. 1b). The addi-tion of Mg 2ϩ markedly enhances ARH3 FL -mediated PAR hydrolysis (Fig. 1b), which is consistent with a previous report (11). ARH3 FL shows maximal activity with Mg 2ϩ followed by Mn 2ϩ and Ca 2ϩ (Fig. 1b). Unlike DraG, the closest functional homolog that has a strong preference for Mn 2ϩ over Mg 2ϩ (25)(26)(27), ARH3 exhibits rather comparable activity with either Mg 2ϩ or Mn 2ϩ . We crystallized ARH3 FL in complex with ADPR and Mg 2ϩ and determined its crystal structure at 1.7-Å resolution (Fig. 1d). Thus, the numbering of amino acid residues in this report deviates by 16 from that used for the apo-ARH3 ⌬N16 structure (25), but it is consistent with those used by Oka et al. (11) and Abplanalp et al. (22). ARH3 FL binds to one ADPR as evidenced by a clear and strong difference electron density at the active site (Fig. 1c).
ARH3 FL adopts a compact all-␣-helical fold with a central deep ADPR-binding cleft, a signature of the ARH3 superfamily ( Figs. 1 and 2). The ARH3-ADPR-Mg 2ϩ complex structure identifies two unique structural elements of ARH3, the "adenine cap" and glutamate flap, which undergo a structural rearrangement upon ADPR binding and strongly contribute to the specific substrate recognition of ARH3 (Figs. 1d and 2). The binuclear Mg 2ϩ catalytic center lies at the heart of a long, J-shaped substrate-binding channel. This structural feature enables ARH3 to bind and align both ADPR and the leaving group, e.g. two ADPR units (n ADPR and the n Ϫ 1 ADPR leaving group) in PAR substrates, for specific cleavage (Fig. 1,  a and d).
The N-terminal 13 amino acids in the N-terminal extension are disordered, which is consistent with their dispensable role in ARH3 function (25). However, unexpectedly, Arg 18 in the N-terminal extension, previously replaced by alanine in the apo-ARH3 ⌬N16 structure (25), contributes to ARH3 folding. Arg 18 makes helix-capping interactions with the main-chain carbonyls of Ser 161 and Leu 162 in ␣7 to stabilize ␣7 (Fig. S2). Arg 18 further stabilizes ARH3 folding by forming a hydrogen bond with the side-chain carbonyl of Gln 361 of ␣19 and a van der Waals contact with the side chain of Phe 23 of ␣1. However, given the fully functional PAR hydrolysis activity of ARH3 ⌬N16 (25) and a remarkably long distance between Arg 18 and Glu 41 (ϳ 32 Å), it is unlikely that Arg 18 contributes to the observed conformational changes.
Structural comparison of the unliganded and ADPR-bound forms of ARH3 reveals dramatic conformational changes in the Glu 41 -containing flap motif, which we named the glutamate flap (Glu 41 -flap) (r.m.s.d. of 2.5 Å, comparing 52 C␣ atoms in the Glu 41 -flap and its flanking residues of the unliganded ARH3 ⌬N16 and complex D of the ARH3-ADPR-Mg 2ϩ complex) (Fig. 1d). The Glu 41 -flap is composed of the end of ␣1 that exists as a 3 10 -helix in the unliganded ARH3, ␣2, and a flexible loop connecting ␣1 and ␣2 (L1) (Figs. 1d and 2). This substrateinduced conformational transition fully exposes the bimetallic catalytic center for substrate engagement and to allow efficient hydrolysis to occur as described below. Together, these findings imply that ARH3 can exist in at least two states: a substratebound "open" state and an unliganded "closed" state.

Structure of human ARH3 bound to ADP-ribose and Mg 2؉
A-D) and the unliganded ARH3 is summarized in Fig. S3b. Complexes B and D have nearly identical structures with a large-scale conformational switch in the Glu 41 -flap (Fig. S3a). Complex D shows the largest degree of conformational transition, which defines a fully open state, and therefore was primarily used for structural analyses. By contrast, complexes A and C show an intermediate conformational step between the unliganded ARH3 and complex D. A part of L1 in the Glu 41 -flap is disordered in complexes A and C, further supporting its structural flexibility (Fig. S3a).

ARH3 specifically exposes 1؆-OH of ADP-ribose
ADPR is located at the deep ADPR-binding cleft of ARH3. All three parts of ADPR (adenosine, diphosphate, and the distal riboseЉ) make extensive contacts with ARH3 ( Figs. 1d and 3). ADPR has a surface area of 694 Å 2 of which ϳ80% (555 Å 2 on average in complexes A-D) is buried by direct contacts with ARH3. Overall, this matrix of ARH3-ADPR-Mg 2ϩ interac-tions specifically exposes 1Љ-OH, corresponding to the scissile O-linkage in substrates (Figs. 1a and 3c), toward the catalytic Mg B , strongly supporting its specificity for the 1Љ-O-linkage for substrate cleavage.
The distal riboseЉ of ADPR lies adjacent to the binuclear Mg 2ϩ center where a group of acidic and polar residues extensively contact Mg 2ϩ ions and ADPR. The interactions between the riboseЉ and two Mg 2ϩ ions are asymmetrical with more extensive contacts on Mg B . 3Љ-OH of the riboseЉ is directly coordinated by Mg B , and it is additionally hydrogen-bonded with the side chain of Asn 151 (Fig. 3, b and c). A water molecule (-aqua ligand) that bridges Mg A and Mg B simultaneously engages 2Љ-OH of the riboseЉ with an unusually short distance (2.2 Å). These Mg 2ϩ -mediated ARH3 interactions with 2Љ-and 3Љ-OH appear to secure tightly the riboseЉ to facilitate efficient cleavage of substrates at the 1Љ-O-linkage. In support of this hypothesis, in contrast to 2Љ-and 3Љ-OH groups, 1Љ-OH of ADPR that corresponds to the 1Љ-O-linkage in substrates is sol- n ADPR (n-1) ADPR  between n and n Ϫ 1 ADP-ribose, releasing ADP-ribose as a product. b, left, Mg 2ϩ enhances the ADP-ribosyl-acceptor hydrolase activity of ARH3. The ARH3-mediated hydrolysis of PAR on PARylated PARP1C was monitored in the presence and absence of Mg 2ϩ using a gel-based assay. PARG has stronger PAR turnover activity than ARH3 and was used as a positive control. Right, metal preference of ARH3. ARH3 prefers Mg 2ϩ for catalysis followed by Mn 2ϩ and Ca 2ϩ . c, difference electron density maps (F o Ϫ F c ) for ADPR and Mg 2ϩ ions contoured at 3.0 (blue, ADPR; orange, Mg 2ϩ ). d, structure of the ARH3-ADPR-Mg 2ϩ complex revealing two unique and flexible structural elements, adenine cap (green) and glutamate flap (Glu 41 -flap) (red), that undergo conformational changes and strongly contribute to specific substrate recognition. The 1Љ-OH of ADPR (blue circled), corresponding to the scissile 1Љ-O-linkage in substrates, is exposed to solvent, consistent with ARH3 specificity for 1Љ-O-linkage for cleavage. A putative binding site for the leaving group is highlighted with a black ellipse. The N and C termini of ARH3 are indicated by N and C, respectively.

Structure of human ARH3 bound to ADP-ribose and Mg 2؉
vent-accessible and exposed to the leaving group-binding site (Figs. 1d and 3c), consistent with cleavage at the C1Љ position (11,20,24). Another water molecule (W1) is axially liganded to Mg B and makes a hydrogen bond with 1Љ-OH (Fig. 3c). This W1 ligand is located proximal to the anomeric C1Љ and appears well aligned for nucleophilic attack of C1Љ during catalysis. At the other side of the binuclear metal center, the carboxyl group of Asp 77 is coordinated by Mg A and further secures 2Љ-OH (Fig. 3c).
Despite the structural similarity between ARH3 superfamily members, the binding orientation of ADPR in ARH3 is remarkably different from that in DraG (28) (r.m.s.d. of 2.6 Å, comparing all C␣ positions of complex D and the DraG-ADPR complex (Protein Data Bank code 2WOE)) (Figs. 3a and S4a). This distinct mode of ADPR binding can be accounted for by a unique adenine-binding module of ARH3, the adenine cap, which consists of ␣6, ␣12, and a loop flanking the C terminus of ␣6 (L6). Notably, ␣6 and ␣12 are missing in DraG (Figs. 2 and 3a). Two aromatic amino acids from the adenine cap, Phe 143 and Tyr 149 , sandwich the adenine base through extensive -stacking interactions (Fig. 3a). In line with this finding, the ARH3-Y149A mutant showed a drastic decrease in PAR-and MAR-acceptor-hydrolyzing activities (22,25). Leu 235 from the flexible ␣12 in the adenine cap moved ϳ2.7 Å toward the adenine ring relative to the unliganded ARH3, forming a van der Waals contact with N6 of the adenine base (Fig. 3, a and b). The chemical selectivity of the adenine base is further enhanced by hydrogen bonds between the backbone carbonyl of Gly 147 and N6 and between the backbone nitrogen of Tyr 149 and N7 (Fig. 3,  a and b). The diphosphate moiety interacts with the side chains of Ser 148 and His 182 and the main-chain nitrogen of Gly 119 . Consistently, substitution of Ser 148 and His 182 with alanine led to loss of ADP-ribosyl-acceptor hydrolase activities (22,25).

A conformational switch of ARH3 enables specific substrate recognition
The most striking feature in the ARH3-ADPR-Mg 2ϩ complex is the dramatic conformational change of the Glu 41 -flap (Fig. 4). The end of ␣1, which is at the beginning of the Glu 41 -  L2 wall (DraG) Figure 2. Structure-based alignment of ARH3 and DraG with structural elements and residues that are important for function. The Glu 41 residue of the Glu 41 -flap of ARH3 that undergoes a large conformational change upon ADPR binding is indicated by a red box. A part of L2 of DraG (L2 wall) completely blocks the conformational change of ␣1 and restricts its activity to cleavage of mono(ADP-ribosyl)ated substrates. The end of ␣1 in ARH3, which exists as 3 10 -helix in the unliganded form and undergoes 3 10 -to-␣ transition upon ADPR binding (Fig. 4a), is indicated by a blue bar. Two aromatic residues in DraG that stabilize the L2 wall are indicated by a red triangle. Asp 97 in DraG that is essential for the cleavage of MARylated arginine is indicated by a blue triangle.

Structure of human ARH3 bound to ADP-ribose and Mg 2؉
flap, exists as a kinked 3 10 -helix structure in the unliganded ARH3 (Fig. 4a). ADPR binding drives a 3 10 -to-␣ transition, leading to ϳ27°of rotation from the helical axis of the 3 10 -helix. This rotation induces a concomitant ϳ8.5-Å displacement of L1 away from the active site, which accompanies ϳ24°of rotation in ␣2 (Fig. 4a). Consequently, ADPR binding results in an ϳ4.5-Å movement of the carboxylate of Glu 41 of the Glu 41 -flap away from the catalytic Mg B . The straightened conformation of ␣1 is stabilized by new intrahelical stacking interactions within ␣1 among residues Phe 39 , Tyr 40 , and His 43 (Fig. 4a). Mechanistically, the observed conformational switch of the Glu 41 -flap is required for specific substrate recognition. In the unliganded state, Glu 41 of the Glu 41 -flap completely masks Mg B from access to substrate (Fig. 4c). This closed state is therefore enzymatically inactive, and Glu 41 must be ejected away from Mg B to allow substrate to enter the bimetallic catalytic center. In support of this hypothesis, the ARH3-ADPR-Mg 2ϩ complex shows a conformational switch in the Glu 41 -flap from a closed state to an open state. This substantial active-site restructuring unmasks Mg B and significantly widens the leaving group-binding site (Figs. 1d and 4, b and c), which now allows substrate entrance to the dimetallic catalytic center. This open conformation of ARH3 provides an optimal alignment between the scissile 1Љ-O-linkage and catalytic groups (Mg 2ϩ and catalytic residues) and therefore constitutes an "incisioncompetent" enzyme state (Figs. 3d and 4b). Collectively, the Glu 41 -flap controls substrate access to the binuclear Mg 2ϩ center, and its closed-to-open conformational switch enables specific binding and cleavage of substrates.
To gain further mechanistic insights into substrate hydrolysis, we modeled di-ADPR, a substrate with the largest leaving group (Fig. 1a), to the active site of ARH3 in catalytic position (Fig. 4, b and c). In this ARH3-di-ADPR-Mg 2ϩ model, the n Ϫ 1 ADPR leaving group snugly fits into the curved J-shaped substrate binding channel (Fig. 4b), whereas n ADPR occupies the identical site as that seen in the ARH3-ADPR-Mg 2ϩ com-  and DraG (gray) reveals a distinctive ADPR-binding mode in ARH3. The adenine cap of ARH3 (green) grasps the adenine ring and is essential for ARH3 activities. Hydrogen bonds contributed by the main-chain atoms of the adenine cap to N6 and N7 of the adenine ring impart specificity. b, diagram showing interactions between ADPR and ARH3. c, close-up of the binuclear Mg 2ϩ catalytic center and ADPR binding in ARH3. A matrix of Mg 2ϩ -mediated coordination and hydrogen-bonding interactions secure ADPR in the active site, which is consistent with Mg 2ϩ -dependent catalysis by ARH3. The 1Љ-OH of ADPR (blue circled), corresponding to the scissile 1Љ-O-linkage, is specifically exposed to solvent, strongly supporting ARH3 specificity for the 1Љ-O-linkage as a site for cleavage. d, close-up of the ARH3-di-ADPR-Mg 2ϩ complex model. The energy-minimized computational model for the ARH3-di-ADPR-Mg 2ϩ complex was generated using the ARH3-ADPR-Mg 2ϩ complex as a starting model (see "Experimental procedures"). In this model, 3Ј-OH of the n Ϫ 1 ADPR leaving group is directly coordinated by Mg B , replacing the W1 water ligand, and Mg B tightly secures both n and n Ϫ 1 ADPR units for efficient cleavage to occur.

Structure of human ARH3 bound to ADP-ribose and Mg 2؉
plex. Notably, 3Ј-OH of the adenosine riboseЈ of n Ϫ 1 ADPR is directly coordinated to Mg B , replacing the axially coordinated water ligand (W1), and it is positioned within hydrogen-bonding distance to Asp 314 and Glu 41 (Figs. 3d and 4b). These findings suggest that Mg B is important for catalysis by securing ADPR and the leaving group to promote the subsequent cleavage at the C1Љ position. Structural superposition of ADPR-bound forms of ARH3 and DraG reveals that the observed 3 10 -to-␣ transition of ␣1 and the conformational flexibility of the Glu 41 -flap are unique in ARH3. Analysis of secondary structures using Dictionary of Secondary Structure of Proteins (DSSP) (29) indicates that DraG lacks the flexible 3 10 -helix structure at the end of ␣1 (Fig.  2). Furthermore, a part of L2 of DraG (named the "L2 wall") tightly caps ␣1 through an edge-stackinginteraction between the side chains of Trp 50 from L2 and Phe 29 from ␣1 (Figs. 3a and S4). This structural arrangement of DraG effectively restricts the conformations of ␣1 and L1 (corresponding to the Glu 41 -flap in ARH3) to the closed state in which substrate access to the leaving group-binding site is prevented (Fig. 4a). The DraG mechanism instead involves a ring opening of the riboseЉ, which positions the scissile 1Љ-N-linkage in close proximity to Asp 97 of DraG for cleavage (28). This structural feature would enforce DraG substrate specificity to MARylated arginine (28). In ARH3, Phe 29 and Trp 50 of DraG are replaced with alanine and serine, respectively, relieving the constraining interaction. Consequently, the L2 of ARH3, equivalent to the L2 wall of DraG, is disordered. Furthermore, Asp 97 of DraG is replaced with glycine in ARH3 (Fig. 2). Taken together, the conformational flexibility of the Glu 41 -flap, the lack of the L2 wall, metal preference for Mg 2ϩ , and the lack of the catalytic residue equivalent to DraG Asp 97 in ARH3 (25) (Fig. 2), all suggest a distinct catalytic mechanism for ARH3 and explain different substrate specificity in ARH3.

Binuclear metal center of the ARH3-ADPR-Mg 2؉ complex
In the ARH3-ADPR-Mg 2ϩ complex, two Mg 2ϩ ions (Mg A and Mg B ) have an octahedral coordination geometry and are fully occupied with only 3.3-Å intermetal distance (3.8 Å in the unliganded ARH3 structure). They are bridged by the bidentate carboxyl group of Asp 316 as well as by the -aqua ligand (Fig.  3c). The bridging -aqua ligand is nearly symmetrically coordinated by two Mg 2ϩ ions (1.9 and 1.8 Å to Mg A and Mg B , respectively) and fixes 2Љ-OH of the riboseЉ of ADPR (Fig. 3c).
In contrast to Mg A , whose coordination ligands remain identical (including Thr 76 , Asp 77 , Asp 78 , and Asp 316 ), the Mg B coordination setup is dynamically changed. In the unliganded ARH3, Glu 41 of the Glu 41 -flap is coordinated by Mg B . Upon PAR substrate binding, 3Ј-OH of the adenosine riboseЈ of the n Ϫ 1 ADPR leaving group replaces Glu 41 and directly coordinates Mg B . Thus, in the substrate-bound state, Mg B engages both ADPR and the leaving group to precisely expose the scissile 1Љ-O-linkage for cleavage (Figs. 3d and 4b). Asp 314 and Glu 41 further aid in the correct positioning of the n Ϫ 1 ADPR leaving group by interacting with 3Ј-OH of the adenosine riboseЈ. Finally, in the ARH3-ADPR (product) complex, the leaving group departs, and a new water ligand (W1) is axially coordinated by Mg B (Figs. 3c and 4b). This dynamic switching of the metal-site makeup in each catalytic step of ARH3 is consistent with Mg 2ϩ -dependent ARH3 activity.

Asp 314 is essential for the formation of binuclear metal center
The structure of the ARH3-ADPR-Mg 2ϩ complex shows that the dimetallic catalytic center of ARH3 is tightly packed against ADPR and appears ideally optimized for cleavage of the scissile 1Љ-O-linkage in substrates. Therefore, a subtle change in the active-site arrangement may result in a detrimental effect on ARH3 functions. Consistently, a conservative substitution of Asp 314 with glutamate led to the loss of enzymatic activity (22,25) (Fig. S5a). To gain further insights into this structurefunction relationship, we determined the crystal structure of ARH3 D314E bound to ADPR and Mg 2ϩ at 1.

Structure of human ARH3 bound to ADP-ribose and Mg 2؉
Mg B is completely missing in the ARH3 D314E -ADPR-Mg 2ϩ complex, and the side chain of Glu 314 is flipped out from the binuclear metal center, presumably due to a steric clash caused by the one-carbon-longer side chain in Glu 314 (Fig. 5). The distance between Glu 314 and the corresponding Mg B in the ARH3 WT -ADPR-Mg 2ϩ complex is beyond the range of direct coordination (3.7 Å), explaining the loss of Mg B and the loss of enzymatic activity. Consistent with the important role of Mg B in securing ADPR (Fig. 3c), the overall B factors for atoms in the riboseЉ of ADPR in the ARH3 D314E -ADPR-Mg 2ϩ complex are significantly higher than those in the ARH3 WT -ADPR-Mg 2ϩ complex. Collectively, these findings indicate that Asp 314 is required for the formation of the binuclear Mg 2ϩ center and suggest a critical role of Mg B for catalysis.

Discussion
Our structures of the ARH3-ADPR-Mg 2ϩ complex and a computational model of the ARH3-di-ADPR-Mg 2ϩ complex reveal the previously unknown conformational plasticity of the Glu 41 -flap, which strongly supports ARH3 specificity for the 1Љ-O-linkage for cleavage. Our studies also explain the published mutational analysis data for ARH3. These structural findings are consistent with Mg 2ϩ -dependent hydrolysis and provide important clues to the catalytic mechanism of ARH3. First, the observed conformational switch in ARH3 is required to specifically recognize substrates. In the closed unliganded state, Glu 41 coordinates Mg B and prevents substrates from entering into the dimetallic catalytic center (Fig. 6). To engage substrates, the Glu 41 -flap is completely moved away from Mg B , inducing substantial active-site rearrangement from a closed to an open state. This conformational transition now allows substrates to enter the active site of ARH3. Second, our structures strongly support the observation that ARH3 favors the 1Љ-Olinkage in substrates for cleavage (24). In the ARH3-ADPR-Mg 2ϩ complex, 2Љ-and 3Љ-OH of ADPR are secured by Mg 2ϩ ions, the bridging -aqua ligand, and active-site residues, whereas 1Љ-OH corresponding to the scissile 1Љ-O-linkage in substrates is specifically exposed to the open leaving group-binding site (Fig. 3c). Substrates containing 2Љ-or 3Љ-O-linkage therefore would result in a serious steric clash with the tightly packed binuclear metal center of ARH3, leading to inefficient substrate cleavage. Consistently, IC 50 values for ARH3 inhibition by 2Љ-or 3Љ-N-acetyl-ADPR, analogs of O-acetyl-ADPR, were significantly higher than that for ADPR (24). In our ARH3-di-ADPR-Mg 2ϩ model, the di-ADPR is slightly bent due to the simultaneous engagement of 3Љ-OH of n ADPR and 3Ј-OH of n Ϫ 1 ADPR by Mg B (Figs. 3d and 6). This structural feature would likely further reinforce the correct positioning of the scissile 1Љ-O-linkage for efficient cleavage (Fig. 6). Third, the substrate-induced widening of the leaving group-binding site enables ARH3 to specifically bind and cleave substrates with a structurally diverse leaving group (Figs. 1a and 4), supporting ARH3's broad substrate specificity.
The flexible Glu 41 -flap dynamically switches between closed and open conformations and plays important roles in substrate recognition (Fig. 6). In the unliganded ARH3, Glu 41 of the Glu 41 -flap serves as the key residue that constitutes the closed enzyme state by masking Mg B . In the PAR substrate-bound state, Glu 41 is released from Mg B and instead interacts both with ADPR and the leaving group (Fig. 6), which presumably contributes to the precise alignment of the scissile O-linkage for subsequent cleavage and constitutes the open enzyme state. Therefore, we propose the Glu 41 -flap as the gate that controls substrate entrance to the active site. Given the structural flexibility and solvent accessibility of the Glu 41 -flap, it is also possible that the Glu 41 -flap functions as a protein-protein interaction module. Proteins that specifically interact with the closed conformation of the Glu 41 -flap of ARH3 might conformationally lock ARH3 in an inactive state and thereby regulate ARH3 function.
A notable ARH3 inhibition by ADP-HPD (Fig. S6a) (30), an analog of the oxocarbenium ion intermediate, raises the possibility that the ARH3 mechanism might involve an oxocarbenium intermediate in the distal riboseЉ of ADPR in a similar way to PARG (15,31), which is followed by water-mediated nucleophilic attack at the anomeric C1Љ position. The observed 18 O incorporation to C1Љ during hydrolysis of O-acetyl-ADPR (24) is also consistent with this model. In glycosidase superfamily members, a catalytic acidic residue is typically found near the scissile bond, e.g. Glu 756 in rat PARG (Glu 752 in human PARG) (31,32), to activate a water molecule for nucleophilic attack on the anomeric carbon (33). In the ARH3-ADPR-Mg 2ϩ complex, Asp 314 is located proximal to both 1Љ-OH (corresponding to the scissile 1Љ-O-linkage) and the axial water ligand (W1) of Mg B (Fig. 3c). It is plausible that Asp 314 might function as a general acid or base to protonate the leaving group and then activate W1 for backside attack of the anomeric C1Љ in a manner similar to Glu 756 in rat PARG (Fig. S6b) (31). In support of this catalytic role of Asp 314 , the ARH3-D314N mutant shows a dramatic reduction in ARH3 activities (22,25). Although more work is needed to determine the catalytic mechanism of ARH3, it is unlikely that ARH3 has a redox chemistry step for catalysis given its preference for Mg 2ϩ over Mn 2ϩ (Fig. 1b).
ARH3 is a unique multitasking enzyme that regulates cellular concentrations of both PAR, either free or attached to proteins, and mono(ADP-ribosyl)ated substrates. Elevated PARylation levels are often found in many human cancers, including

Structure of human ARH3 bound to ADP-ribose and Mg 2؉
BRCA-deficient breast cancers (34,35) and triple-negative breast cancers (36,37). ARH3 Ϫ/Ϫ cells undergo enhanced PARdependent cell death upon genotoxic stresses but remain healthy under unstressed conditions. We suggest that pharmacological intervention in ARH3-dependent pathways could be a safe and efficient therapeutic strategy for cancers with up-regulated PARylation by increasing the cytoplasmic PAR level and triggering parthanatos, either alone or in combination with current chemotherapeutic agents. Our structures reveal extensive adenine interactions with the adenine cap and a large conformational switch of the Glu 41 -flap that are crucial for specific engagement of structurally diverse substrates. These unique structural elements of ARH3 can be exploited for the development of specific ARH3 inhibitors, which have potential therapeutic applications, as well as the ability to advance our understanding of the role of ARH3 in regulation of PARylation. A concern is that ARH3 also catalyzes hydrolysis of O-acetyl-ADPR, which may be involved in other cellular pathways such as chromatin remodeling. This activity also relies on the ability of ARH3 to act at the C1Љ position of ADPR. ARH3's contribution to the O-acetyl-ADPR-dependent pathways, in contrast to other proteins in the ARH-macrodomain superfamily, is not known. Thus, some ARH3 inhibitors may affect multiple signaling pathways due to its multiple enzymatic activities and substrates.

Plasmids and protein purification
Human ARH3 FL was cloned into a modified pET21b vector with an N-terminal His 6 tag and a following cleavage site for PreScission protease (pET21b-His 6 -pps). The pET21b-His 6pps-ARH3 FL plasmid was introduced into Escherichia coli Rosetta 2 (DE3) cells, and ARH3 FL was induced by adding 1 mM isopropyl ␤-D-thiogalactoside overnight at 16°C. ARH3 FL was purified by affinity capture on a nickel-nitrilotriacetic acid (GE Healthcare) column. After elution with imidazole (250 mM), the protein was loaded onto a heparin column (GE Healthcare) and eluted with a NaCl gradient (0.1-1 M). Fractions with ARH3 FL were pooled, and PreScission protease was added to cleave the N-terminal histidine tag. Finally, ARH3 FL was loaded to a Sephacryl 200 (GE Healthcare) size-exclusion column using a buffer containing 25 mM Tris-HCl, pH 7.5, 150 mM NaCl, 1 mM DTT, and 5% glycerol. The purified ARH3 FL was concentrated to ϳ30 mg/ml and then stored at Ϫ80°C. A gene encoding the ARH3-D314E mutant was synthesized (GeneUniversal, Inc.) and cloned into pET-21b-pps vector. The ARH3-D314E mutant was purified using the same protocol as the WT protein. The human PARG catalytic domain (residues 448 -976) was cloned in pET21b-His 6pps, expressed in E. coli Rosetta (DE3) cells, and purified by nickelnitrilotriacetic acid-affinity chromatography, heparin chromatography, and Sephacryl 200 size-exclusion chromatography as described previously (32). For preparation of PARylated PARP1, the DNA-binding domain (residues 1-374) of human PARP1 and the PARP1C catalytic domain (residues 375-1014) were purified as described previously (6,31).

Crystallization and data collection
The WT and the D314E mutant of ARH3 (10 mg/ml) were cocrystallized with 5 mM ADPR by hanging-drop vapor diffusion. ADPR binding is required for crystallization as the unliganded ARH3 FL did not yield any crystals. The protein solution was mixed with an equal volume of well solution (22% PEG 4000, 0.

Structure determination
The full-length human ARH3 WT -ADPR complex structure was determined by molecular replacement using MolRep (40) in the CCP4 suite with the apo-ARH3 ⌬N16 structure (25) as a search model. The asymmetric unit contains four ARH3 molecules, and they all show a strong difference electron density for the bound ADPR at the active site, which was supported by polder map calculations. The restraints for ADPR were generated using Monomer Library Sketcher in the CCP4i suite (41). The model was manually rebuilt using Coot (42) and refined with PHENIX (43) to an R factor of 18.4% and an R free of 21.8%. The ARH3 D314E -ADPR complex structure was determined by molecular replacement using MolRep (40) with the ARH3 WT -ADPR-Mg 2ϩ complex structure as a search model. The ARH3 D314E -ADPR-Mg 2ϩ complex model was rebuilt using Coot (42) and refined with PHENIX (43) to an R factor of 16.7% and an R free of 19.9%. Crystallographic data statistics are shown in Table S1. The Ramachandran plot shows that Ͼ98% of the residues in both ARH3 WT -ADPR-Mg 2ϩ and ARH3 D314E -ADPR-Mg 2ϩ complexes are in the favored regions, and all the others are in the allowed regions. No outlier residue was observed.

ARH3 activity assay
ARH3 activity was measured against PARylated PARP1 using a method similar to that for PARG activity measurement as described previously (31). The C-terminal catalytic domain of PARP1 (PARP1C; residues 375-1014) containing the automodification domain was PARylated in the presence of the N-terminal DNA-binding domain of PARP1, a nicked DNA, and NAD ϩ as described previously (6,31). Briefly, PARylation of PARP1 was performed at 37°C in a reaction buffer containing 50 mM Tris-HCl, pH 7.5, 100 mM NaCl, and 2 mM DTT. To PARylate PARP1, PARP1C (2 M), DNA-binding domain (residues 1-374) (2 M), and a nicked DNA (2 M) were preincubated for 10 min on ice. 200 M NAD ϩ was then added to the reaction, and it was then incubated at 37°C for 30 min. After desalting with a buffer containing 50 mM Tris-HCl, pH 7.5, 100 mM NaCl, and 5% glycerol, purified human ARH3 proteins (WT and D314E) or human PARG was treated with PARylated PARP1 substrates in the presence or absence of 5 mM EDTA or divalent metals (Mg 2ϩ , Mn 2ϩ , and Ca 2ϩ ) and incubated for 60 min at 37°C. The level of modification of PARP1 was visualized by Coomassie Blue staining of SDS-polyacrylamide gels.

Computational modeling of the ARH3-di-ADPR-Mg 2؉ complex
The di-ADP-ribose model was generated by covalently linking 2Ј-OH of the adenosine riboseЈ of n Ϫ 1 ADPR to the anomeric C1Љ of the distal riboseЉ of n ADPR in the ARH3-ADPR-Mg 2ϩ complex using YASARA (44). The n ADPR molecule, two Mg 2ϩ ions, and the bridging -aqua ligand remained anchored to the experimentally identified site in the ARH3-ADPR-Mg 2ϩ complex. Then n Ϫ 1 ADPR was docked to the putative n Ϫ 1 ADPR-binding site (leaving group-binding site) of the J-shaped substrate-binding channel of ARH3. The ARH3-di-ADPR-Mg 2ϩ model was then subjected to energy minimization using the AMBER ff14SB protein force field in the Chimera package (45). All atoms in the protein and di-ADPR ligand were allowed to move during energy minimization. All structural figures were prepared using PyMOL.