Structural basis of autoinhibition and activation of the DNA-targeting ADP-ribosyltransferase pierisin-1

ADP-ribosyltransferases transfer the ADP-ribose moiety of βNAD+ to an acceptor molecule, usually a protein that modulates the function of the acceptor. Pierisin-1 is an ADP-ribosyltransferase from the cabbage butterfly Pieris rapae and is composed of N-terminal catalytic and C-terminal ricin B–like domains. Curiously, it ADP-ribosylates the DNA duplex, resulting in apoptosis of various cancer cells, which has raised interest in pierisin-1 as an anti-cancer agent. However, both the structure and the mechanism of DNA ADP-ribosylation are unclear. Here, we report the crystal structures of the N-terminal catalytic domain of pierisin-1, its complex with βNAD+, and the catalytic domain with the linker connecting it to the ricin B–like domains. We found that the catalytic domain possesses a defined, positively charged region on the molecular surface but that its overall structure is otherwise similar to those of protein-targeting ADP-ribosyltransferases. Electrophoretic mobility shift assays and site-directed mutagenesis indicated that pierisin-1 binds double-stranded but not single-stranded DNA and that Lys122, Lys123, and Lys124, which are found in a loop, and Arg181 and Arg187, located in a basic cleft near the loop, are required for DNA binding. Furthermore, the structure of the catalytic domain with the linker revealed an autoinhibitory mechanism in which the linker occupies and blocks both the βNAD+- and DNA-binding sites, suggesting that proteolytic cleavage to remove the linker is necessary for enzyme catalysis. Our study provides a structural basis for the DNA-acceptor specificity of pierisin-1 and reveals that a self-regulatory mechanism is required for its activity.

ADP-ribosyltransferases that target proteins as acceptor molecules are widely distributed in bacteria and eukaryotes (1)(2)(3). Hitherto known target amino acid residues are lysine, arginine, glutamate, aspartate, cysteine, diphthamide, phos-phorylated serine, and asparagine (3). ADP-ribosyltransferases are classified into poly-and mono-ADP-ribosyltransferases. Poly-ADP-ribosyltransferases are mainly found in multicellular eukaryotes and are involved in genome stability. Mono-ADPribosyltransferases are distributed throughout eukaryotes, prokaryotes, and bacteriophages. The bacterial mono-ADP-ribosyltransferases form a large family of mono-ADP-ribosylating toxins, including cholera toxin, diphtheria toxin, and pertussis toxin, which target specific cellular proteins to impair the function of the protein (1,2). 3D structures of various mono ADPribosyltransferases and their complexes with ␤NAD ϩ have been determined by X-ray crystallography and reveal that the core structures, key residues for catalysis, and ␤NAD ϩ binding motifs are conserved. Some of these core structures and key residues are also conserved in poly-ADP-ribosyltransferases (3). Mutation analysis based on the 3D structure has also been used to clarify important residues for target recognition. The catalytic mechanisms of exotoxin A and toxin have also been proposed based on their crystal structures in complex with target proteins (4 -6).
Pierisin-1 is an ADP-ribosyltransferase produced by the last stage larvae and pupae of the cabbage butterfly Pieris rapae and induces apoptosis of various types of cancer cell lines, such as HeLa and TMK-1 cells (7)(8)(9)(10). The enzyme is composed of 850 amino acid residues and consists of an N-terminal catalytic domain (amino acid residues 1-233) and a C-terminal ricin B-like HA33 domain (residues 267-850) (Fig. 1A) (11). The N-terminal domain is homologous with ADP-ribosyltransferases of various bacterial toxins and transfers the ADP-ribose moiety of ␤NAD ϩ to an acceptor molecule to modulate the molecular function. The C-terminal ricin B-like domain has binding activity toward cell surface glycosphingolipid receptors, such as globotriaosylceramide and globotetraosylceramide (12,13). Binding of the C-terminal domain to the cell surface receptors triggers the internalization of pierisin-1, where proteolytic cleavage between the N-and C-terminal domains induces apoptosis of the cells (12). Unlike the protein ADP-ribosyltransferases mentioned above, pierisin-1 is a unique mono-ADP-ribosyltransferase that targets DNA as the acceptor molecule, with the N-terminal catalytic domain transferring the ADP-ribose moiety to the N2 amino group of guanine (14). Pierisin-1 has 30% sequence identity with mosquito- The authors declare that they have no conflicts of interest with the contents of this article. This article contains supplemental Figs. S1-S7. The atomic coordinates and structure factors ( cro ARTICLE cidal toxin (MTX) 2 of Bacillus sphaericus, which also belongs to the ADP-ribosylating toxin family and has a similar domain structure to pierisin-1 (Fig. 1A) (11,15,16). Although the in vivo target of MTX has not been identified, it is likely to be a protein (17), and thus despite their similarity, the acceptor molecules of MTX and pierisin-1 are different. To date, six pierisins (pierisin-1, -1b, and -2-5) and CARP-1 have been found in the butterfly Pieridae family and in a clam, respectively, and have been identified as DNA-targeting ADP-ribosyltransferases (18 -24). To understand the acceptor recognition and catalytic mechanisms of pierisin-1, we determined the crystal structures of the N-terminal catalytic domain of the enzyme with and without a linker (residues 234 -267) (pierisin(1-267) and pierisin(1-233)) and ␤NAD ϩ -bound pierisin(1-233) and demonstrate that pierisin-1 has a binding activity toward dsDNA, not ssDNA. We also identified the amino acid residues important for dsDNA binding by mutation analysis using EMSA. We demonstrate that the linker between the N-and C-terminal domains controls catalysis by regulating how ␤NAD ϩ and DNA access their binding sites.

Recognition of ␤NAD ؉
The binding of ␤NAD ϩ induced no significant change around the ␤NAD ϩ binding pocket as described above. The interaction between pierisin(1-233) E165Q and ␤NAD ϩ is shown in Fig. 2. The nicotinamide moiety of ␤NAD ϩ is buried deep in the ␤NAD ϩ binding pocket, which is formed by the STS motif (residues 114 -116), the main chain of Trp 71 , and the side chains of Arg 70 , Trp 127 , Gln 163 , and Gln 165 . In the absence of ␤NAD ϩ , this pocket was occupied by water molecules and an ethylene glycol molecule used for cryoprotection in X-ray diffraction experiments at 100 K (supplemental Fig. S3A). The carboxyamide group of the nicotinamide moiety is recognized by backbone carbonyl and amide groups of Trp 71 . The 2Ј-OH group of the nicotinamide ribose moiety is bound to the side chain of Gln 165 , which replaces the catalytic Glu 165 , through a hydrogen bond. The two phosphate groups are bound to the guanidinium group of Arg 70 by ionic interactions and also to the OH group of Ser 114 in the STS motif by hydrogen bonds. The 2Ј-OH group of adenine ribose contacts the carboxyl group of Asp 72 by van der Waals interactions. The adenine moiety of ␤NAD ϩ interacts with the guanidinium group of Arg 74 viastacking, and the NH 2 group of the adenine moiety (AN6) is bound to the main chain of Val 85 by a hydrogen bond.
Many interactions observed between pierisin-1 and ␤NAD ϩ are conserved in cholera toxin (28) (supplemental Fig. S3, B and C). However, the hydrogen bond network mediated through water molecules is different in pierisin-1 and cholera toxin.

Putative DNA-binding region
We studied the interaction between pierisin-1 and DNA by EMSA. EMSA with pierisin(1-233) E165Q and dsDNA shows a clearly shifted band, but this is not the case with ssDNA ( Fig. 3A), indicating that the binding affinity for ssDNA is much lower than that for dsDNA. This result is compatible with the fact that pieri- Table 1 Data collection and refinement statistics Values in parentheses are for highest resolution shell. asu, asymmetric unit; RMSD, root mean square deviation.

Data collection
Space group
To identify the dsDNA-binding region, a series of pierisin(1-233) E165Q mutants were prepared and then subjected to EMSA to test dsDNA-binding activity. Probable candidates for mutation are positively charged residues on the molecular surface. The electrostatic surface potential of pierisin(1-233) E165Q shows a characteristic positively charged surface suitable for dsDNA binding (Fig. 4A). Another candidate is the ARTT loop that has been identified as a substrate binding site for a proteintargeting mono-ADP-ribosyltransferases (30). Likewise, an area around the ␤NAD ϩ binding pocket is also a candidate dsDNA-binding site. These candidate amino acids are mapped onto  Fig. S4). The results of the mutagenesis demonstrated that DNA-binding activity was slightly reduced by the mutations R120S, W127A, and R187A but not significantly reduced by R73A, H108A, K117A, R130A, R134A, and W160A (Fig. 4C). A single mutation of R181A considerably reduced the DNA-binding activity, and double mutagenesis of R181A and R187A completely abolished the binding activity. Furthermore, triple mutation of K122N, K123N, and K124N also completely abolished the DNA-binding activity (Fig. 4C). However, among single mutations of K122A, K123A, and K124A, K122A and K124A significantly reduced the DNA binding activity, whereas K123A did not measurably affect it (Fig. 4C, bottom). In addition, these three Lys residues in the loop are conserved among the group of all six pierisin members as described below (supplemental Fig. S5), suggesting a contribution of Lys 123 to the DNA binding. Taken together, these results indicated that the two residues Lys 122 and Lys 124 , and probably Lys 123 on the PN loop and Arg 181 and Arg 187 on a basic cleft near the PN loop, are required for dsDNA binding. Thus, pierisin-1 binds dsDNA via electrostatic interactions. This is also supported by a gel filtration chromatography analysis of the pierisin-1 interacting with dsDNA (supplemental Fig. S6). Under high-salt conditions, pierisin(1-233) E165Q and dsDNA elute separately, and an elution peak for a complex between pierisin(1-233) E165Q and

Structure of DNA ADP-ribosylating pierisin-1
dsDNA is not observed. In contrast, in low-salt conditions, pierisin(1-233) E165Q and dsDNA associate and elute as a single complex.
Mutational analysis of DNA binding of pierisin(1-233) E165Q and its derivatives (DNA-binding mutants) revealed the amino acid residues responsible for DNA binding. However, what effect they have on the DNA ADP-ribosylating activity was not examined because they lack the DNA ADP-ribosylating activity. A useful plasmid containing a wild-type pierisin-1 (residue Glu 165 ) cDNA to express it in E. coli cells has not been established because of the instability of its toxic gene during a cloning procedure and other unknown factors.

Autoinhibition by the linker region
The linker (residues 234 -267) is positioned between the catalytic and ricin B-like domains and masks the basic cleft of the catalytic domain (Fig. 4A, center). The segment 245-256 of the linker region interacts with the PN loop and induces a ␤ strand ␤4Ј to form a ␤-sheet (␤4Ј and ␤11). The PN loop, which has a flexible nature in the structure of pierisin(1-233) E165Q , was stabilized by this interaction (Figs. 1B (right) and 5 (A and B)). The side chain of Asp 252 in the linker interacts with the side chains of Arg 67 and Gln 139 in the catalytic domain through hydrogen bonds. The main-chain carbonyl group of Phe 254 and the side chain of Asp 256 in the linker interact with the side chain of Arg 181 in the catalytic domain through hydrogen bonds. In addition to these interactions, a hydrogen bond network mediated by a number of water molecules was observed (Fig. 5, A and  B). Thus, the segment 245-256 masks the putative DNA-binding region and inhibits DNA binding.
The linker region (residues 234 -267) masks not only the putative DNA-binding region but also the ␤NAD ϩ binding pocket. The ␤NAD ϩ binding pocket was masked by a segment (residues 239 -244) of the linker in pierisin(1-267) E165Q (Fig.  5C). The crystal structure of pierisin(1-267) E165Q clearly shows that the segment is stabilized mainly by van der Waals interactions (Fig. 5, C and D). This segment is also observed in MTX in a region (residues 273-278) between the catalytic and ricin B-like domains and masks the ␤NAD ϩ binding pocket (Fig. 5, E and F) (15,31). However, the amino acid sequence of the segment is not conserved between pierisin-1 and MTX. Indeed, the interactions of the segment with pierisin-1 in the binding pocket are significantly different from those in MTX. In fact, Asp 275 of MTX makes hydrogen bonds with the guanidinium group of Arg 100 and hydroxyl group of Ser 162 . Asp 275 , Arg 100 , and Ser 162 , which correspond to Leu 241 , Arg 73 , and Asn 131 in pierisin-1, are in contact with each other by van der Waals    E165Q (center and right). DNA binding of mutants was compared with that of (1-233) E165Q as a control. The amino acid residues essential for maintaining full binding affinity are colored blue, and the residues exhibiting weak or no effect on binding are gray. The ribbon diagram (left) is drawn in the same orientation for A. C, interaction between mutants of pierisin(1-233) E165Q and 201-bp dsDNA. 10 nM 201-bp dsDNA was mixed to interact with varying concentrations (0, 25, 50, 100, 200, and 400 nM) of pierisin(1-233) and subjected to EMSA. DNA on the gels was located by SYBR Gold staining. E165Q was taken as a control. All other mutants indicated above the gels were derivatives of E165Q. The mutant K122-124N represents the triple mutation of Lys 122 , Lys 123 , and Lys 124 to Asn. R181/187A represents the double mutant R181A and R187A.

Structure of DNA ADP-ribosylating pierisin-1
The electron density of a segment (residues 231-238), which precedes amino acids 239 -244 and which contains three C-terminal residues (Ser 231 -Arg 233 ) of the catalytic domain, is miss-ing in pierisin(1-267) E165Q , most likely due to the disordered nature of the segment (Fig. 1B, right). Because the flexible unstructured segment protrudes from the protein surface, pro-

Structure of DNA ADP-ribosylating pierisin-1
teases can target and cleave the site. It is reasonable to propose that the segment is cleaved by proteolysis and is followed by dissociation of the C-terminal ricin B-like domain, together with the linker region from the catalytic domain, resulting in enzyme activity in vitro and in vivo (12). Pierisin-1 is specifically cleaved at Arg 233 -Ser 234 by trypsin in vitro. In vivo proteolytic cleavage might occur in intracellular lysosomes after pierisin is ingested into cells, and then the excised catalytic domain is transported to the nucleus to bind to DNA (Fig. 6). However, little is known about these processes. A similar activation mechanism is also suggested for MTX (15,16,31). In fact, the segment 263-269 of MTX is disordered and is subject to protease digestion in the activation of its catalytic activity (32,33).

Discussion
In this study, we determined the crystal structures of pierisin(1-233) E165Q and a ␤NAD ϩ -bound pierisin(1-233) E165Q . These structures share a common fold with such mono-ADP-ribosylating toxins as cholera toxin, diphtheria toxin, MTX, and others, despite high sequence diversity. ␤NAD ϩ is also recognized in a similar manner to the mono-ADP-ribosylating toxins. The mutation analysis based on the 3D structure of the catalytic domain of pierisin-1 shows that the PN loop and the basic cleft are important for dsDNA binding. Modeling dsDNA onto pierisin(1-233) E165Q shows that dsDNA fits well between the PN loop and the basic cleft (Fig.  4A). The width of the basic cleft (the distance between C␣ atoms of Lys 123 and Arg 187 ) was 19.2 Å, which is suitable for dsDNA but too large for ssDNA. Our results from DNA binding assays substantiated the previous finding that as an acceptor molecule for ADP-ribosylation, dsDNA is much better than ssDNA. The PN loop exhibits structural flexibility (Fig. 1D). It is possible that this flexibility is necessary for dsDNA recognition to fit the DNA into the binding surface. The basic residues of the PN loop and the basic cleft nearby are conserved between pierisin-1, -1b, -2, -3, -4, and -5, all of which ADP-ribosylate DNA (supplemental Fig. S5), suggesting a common feature of all pierisins. Despite the high similarity in 3D structure between pierisin-1 and MTX, their electrostatic surface potentials are distinctively different (Fig. 4A). This difference may reflect acceptor specificity of the enzymes. In fact, the surface that corresponds to the basic cleft of pierisin-1, is also masked by the linker region in MTX, although the substrate binding region of MTX has not been identified.
Recently, Lyons et al. (34) determined the crystal structure of scabin, a DNA-targeting ADP-ribosyltransferase from Streptomyces scabies. They showed that scabin has a positively charged surface suggested to be a DNA-binding surface (supplemental Fig. S7A). The core structure is similar to those of pierisin-1 (Z score ϭ 16.7) as well as MTX and other ADP-ribosyltransferases (supplemental Fig. S7). However, scabin has neither a ricin B-like domain nor an autoinhibitory linker, indicating that activity control is different between scabin and pierisin-1. Furthermore, arrangement of ␣-helices and loop regions around the core structure are different, and the PN loop of scabin is significantly shorter than that of pierisin-1 (supplemental Figs. S5 and S7B). The key residues for DNA binding are also not conserved between pierisin-1 and scabin (supplemental Fig. S5). These observations indicate that the binding mode for dsDNA and preference for dsDNA, ssDNA, and mononucleotide differ between scabin and pierisin-1. Thus, the DNAtargeting mechanisms seem to be divergent among ADPribosyltransferases. We are currently determining the crystal structure of pierisin(1-233) E165Q in complex with dsDNA and with dsDNA and ␤NAD ϩ to obtain further structural insight into the ADP-ribosyltransferase DNA-targeting mechanism.

Expression and purification of pierisin(1-233) E165Q and pierisin(1-267) E165Q
The plasmid vector pET-32a containing a cDNA encoding pierisin(1-233) E165Q was transformed into E. coli strain BL21(DE3) pLysS. Pierisin(1-233) E165Q was expressed as an N-terminal thioredoxin and His tag fusion protein. The cells were grown at 37°C in LB medium and induced with 1 mM isopropyl-thio-␤-D-galactopyranoside when they reached an optical density of 0.4 -0.6 at 660 nm and were further incubated at 16°C for 3 h. The cells were harvested; resuspended in a lysis buffer containing 50 mM Tris-HCl, pH 7.5, 0.5 M NaCl, and 1  Gb3) and globotetraosylceramide (Gb4) through its four ricin B-like domains (13) and is internalized into cells. iii, the pierisin incorporated into lysozomes is cleaved at a C-terminal site of the catalytic domain and then followed by the catalytic domain dissociation from the rest of the linker and the ricin B-like domains, resulting in the activation of its catalytic domain. Thus, the catalytic domain is exposed at a catalytic site and freely accessible by a substrate ␤NAD ϩ and an acceptor DNA. iv, the active catalytic domain (residues 1-233) released into the cytosol migrates into the nucleus and binds to DNA, where it transfers the ADP-ribose moiety of ␤NAD ϩ to the N2 amino group of guanine.

Structure of DNA ADP-ribosylating pierisin-1
mM PMSF; and disrupted by sonication on ice. After centrifugation, the supernatant was applied to a nickel-Sepharose column (GE Healthcare). Thioredoxin and His tag fusion protein was eluted with 50 mM Tris-HCl, pH 7.5, containing 0.5 M NaCl, 200 mM imidazole. After digestion of the fusion protein with thrombin, samples were diluted with 50 mM Tris-HCl, pH 7.5, and loaded onto a HiTrapSP HP column (GE Healthcare). The protein was eluted by a linear gradient of 0 -1 M NaCl. The eluted protein was concentrated to 1.5-2.0 mg/ml, and the buffer solution was exchanged with 20 mM HEPES-NaOH, pH 7.0, containing 400 mM NaCl. The purified protein was stored at 4°C. All of the mutant proteins of pierisin(1-233) E165Q used for EMSAs were purified as described above.
Crystals of pierisin(1-233) E165Q (form 2) were obtained using a reservoir solution containing 0.1 M NaF and 6 -10% (w/v) PEG 3350. The crystal was transferred to a reservoir solution containing 5-25% (v/v) ethylene glycol stepwise and flashfrozen at Ϫ180°C. For the preparation of the ␤NAD ϩ -bound form, the crystal was soaked in a reservoir solution containing 5 mM ␤NAD ϩ without NaF for 10 min. Then the crystal was transferred to 5-25% (v/v) ethylene glycol containing 10 mM ␤NAD ϩ , 6% (w/v) PEG 3350 stepwise and flash-frozen at Ϫ180°C.
Crystals of pierisin(1-267) E165Q were obtained by using a reservoir solution containing 22% (v/v) t-butanol and 0.1 M Tris-HCl, pH 8.0. The crystal was transferred to a reservoir solution containing 22.5% (v/v) glycerol and flash-frozen at Ϫ180°C.
X-ray diffraction data were collected on a Quantum315 CCD detector (ADSC) in beamline BL5A or BL17A at PF (Tsukuba, Japan) and processed using the HKL-2000 program suite (35). The structure of pierisin(1-267) E165Q was solved by molecular replacement with the program MOLREP (36) using the MTX-1 catalytic domain (PDB code 2CB4) as a search model. In the same way, the structure of pierisin(1-233) E165Q was solved using the structure of pierisin(1-267) E165Q . The structures were manually improved with the program COOT (37) and refined with the program REFMAC (38) and PHENIX (39). Geometries of the final structures were validated with molprobity (40,41). Data collection and refinement are summarized in Table 1. Coordinates and structure factors have been deposited in the Protein Data Bank (PDB entries 5H6K, 5H6J, 5H6M, 5H6L, and 5H6N).
For analysis of DNA-binding activity of pierisin(1-233) E165Q and other mutants, 201-bp dsDNA (a sequence between the T7 promoter and T7 terminator of pET-32a plasmid) was used. The 201-bp dsDNA was amplified by PCR using KOD plus DNA polymerase (TOYOBO). The amplified 201-bp dsDNA was purified by electrophoresis on a 1.5% agarose gel with a TAE buffer and extracted from the gel using the Wizard SV Gel and PCR Clean-Up System (Promega). Sample solutions (