Structures of RNA Complexes with the Escherichia coli RNA Pyrophosphohydrolase RppH Unveil the Basis for Specific 5′-End-dependent mRNA Decay*

Background: RppH initiates mRNA degradation in bacteria by removing pyrophosphate from the triphosphorylated 5′ terminus. Results: X-ray structures of E. coli RppH have revealed molecular features important for RNA recognition, discrimination against mononucleotides, and catalysis. Conclusion: RppH balances specificity and promiscuity. Significance: Our findings explain the broad influence of RppH on E. coli mRNA decay and suggest a catalytic mechanism for Nudix hydrolases. 5′-End-dependent RNA degradation impacts virulence, stress responses, and DNA repair in bacteria by controlling the decay of hundreds of mRNAs. The RNA pyrophosphohydrolase RppH, a member of the Nudix hydrolase superfamily, triggers this degradation pathway by removing pyrophosphate from the triphosphorylated RNA 5′ terminus. Here, we report the x-ray structures of Escherichia coli RppH (EcRppH) in apo- and RNA-bound forms. These structures show distinct conformations of EcRppH·RNA complexes on the catalytic pathway and suggest a common catalytic mechanism for Nudix hydrolases. EcRppH interacts with RNA by a bipartite mechanism involving specific recognition of the 5′-terminal triphosphate and the second nucleotide, thus enabling discrimination against mononucleotides as substrates. The structures also reveal the molecular basis for the preference of the enzyme for RNA substrates bearing guanine in the second position by identifying a protein cleft in which guanine interacts with EcRppH side chains via cation-π contacts and hydrogen bonds. These interactions explain the modest specificity of EcRppH at the 5′ terminus and distinguish the enzyme from the highly selective RppH present in Bacillus subtilis. The divergent means by which RNA is recognized by these two functionally and structurally analogous enzymes have important implications for mRNA decay and the regulation of protein biosynthesis in bacteria.

tion of mRNAs (hundreds in E. coli) (5,13). Such unexpected selectivity of RppH-dependent degradation can be explained, at least in part, by the recently discovered 5Ј-terminal requirements of these enzymes. BsRppH needs two but prefers three or more unpaired 5Ј-terminal nucleotides for activity and strictly requires guanine at the second position of its substrates (Fig.  1A) (14,15), suggesting that it triggers the decay of a distinct set of mRNAs. This sequence preference does not correlate with the nucleotide frequency at the 5Ј-end of B. subtilis transcripts (14,15). Thus, RppH can be considered a master regulator of protein biosynthesis that selectively controls mRNA translation through different rates of the 5Ј-end-dependent mRNA decay.
Although vitally important for all organisms, mRNA degradation, especially RppH-mediated 5Ј-end-dependent mRNA decay in microbes, is among the least understood mechanisms of gene regulation. The biological significance of RppH prompted determination of the NMR structure of free EcRppH (16), x-ray structures of Bdellovibrio bacteriovorus RppH (BdRppH) in the free form and bound to GTP (17), and x-ray structures of BsRppH bound to GTP and RNA (15). However, these RppH structures do not reveal whether EcRppH retains the same specificity as BsRppH or how EcRppH binds mRNA because the putative RNA-binding regions of EcRppH and BsRppH are strikingly different. Moreover, the determined structures do not feature RNA⅐RppH complexes in catalytically active conformations and do not provide sufficient details to unambiguously explain why BsRppH cleaves off two orthophosphates sequentially, although EcRppH normally removes pyrophosphate in one step (although it too sometimes catalyzes consecutive cleavage reactions) (5,6,15).
To reveal the molecular basis of mRNA recognition and catalysis by EcRppH, we have identified minimal RNA substrates, determined x-ray structures of EcRppH in the apo-and RNA-bound forms, and conducted mutational studies. Our structural and biochemical data demonstrate novel structural principles of low specificity mRNA recognition that are characteristic of innumerable proteobacterial RppHs and distinct from those used by more specific RNA pyrophosphohydrolases like BsRppH. Our results also suggest a common catalytic mechanism for many hydrolases of the Nudix superfamily and therefore will have an impact on numerous biological systems.

EXPERIMENTAL PROCEDURES
Protein Production and Purification-The E. coli rppH gene and several variants truncated at the 3Ј-end were cloned into pSUMOH10 vector (a gift of Dr. H. Li) and expressed with an N-terminal His 10 -Sumo tag from a T7 RNA promoter in E. coli BL21(DE3)CodonPlusRIL cells using auto-induction (18). Cells from a 2-liter culture were collected by centrifugation and resuspended in 20 ml of lysis buffer containing 50 mM sodium phosphate, pH 7.5, 1 M NaCl, 20 mM imidazole-HCl, pH 8.0, 0.05% (v/v) ␤-mercaptoethanol, and EDTA-free Protease Inhibitor Mixture (Roche Applied Science). Cells were lysed by EmulsiFlex C-5 (Avestin), and debris was removed by centrifugation at 20,000 ϫ g and at 4°C for 1 h. Clarified lysate was applied onto a 5-ml HisTrap FF column (GE Healthcare) and washed with the lysis buffer supplemented by 80 mM imidazole-HCl, pH 8.0. Recombinant protein was eluted by the lysis buffer that contained 0.5 M imidazole-HCl, pH 8.0. Peak fractions were combined, and the tag was cleaved off by ULP1 protease at 4°C for 14 -16 h. Digested EcRppH samples were dialyzed against a buffer composed of 50 mM sodium acetate, pH 5.0, 0.3 M NaCl, and 0.05% (v/v) ␤-mercaptoethanol, applied onto 5-ml HiTrap SP column (GE Healthcare), washed with the same buffer, and eluted with 0.3-1.0 M NaCl gradient in a buffer containing 50 mM sodium acetate, pH 5.0, and 0.05% (v/v) ␤-mercaptoethanol. Peak fractions were combined and dialyzed against a buffer prepared with 20 mM sodium acetate, pH 5.0, 150 mM NaCl, and 0.05% (v/v) ␤-mercaptoethanol. At the final step, dialyzed protein samples were concentrated to Ϸ5 ml and purified on a 2.6 ϫ 60-cm Superdex 75 column (GE Healthcare) equilibrated with the storage buffer that contained 20 mM sodium acetate, pH 5.0, 100 mM NaCl, and 1 mM DTT. Peak fractions were combined, concentrated, and stored at Ϫ80°C.
To re-engineer crystal packing interactions, we predicted mutations on the surface of EcRppH using the surface entropy reduction algorithm (19,20) and introduced double Ala mutations into the EcRppH(1-160)-coding construct by megaprimer PCR (21) and single mutations in the EcRppH-coding construct by inverse PCR (22).
Preparation of RNA Oligonucleotides-Short RNA oligonucleotides were prepared by in vitro transcription with T7 RNA polymerase and DNA templates that comprise a double-stranded T7 promoter and a single-stranded extension encoding for RNA (23). DNA templates were prepared by annealing two oligonucleotides in a 100-l volume at 10 M concentration. Transcription reactions were performed in 1 ml of buffer containing 100 mM Tris-HCl, pH 8.0, 20 mM MgCl 2 , 2 mM spermidine, 40 mM DTT, 1 M DNA template, and 0.1 mg/ml T7 RNA polymerase at 37°C for 4 h and quenched by 0.1 ml of 0.5 M EDTA. The total concentration of NTPs was kept within 15-16 mM, and concentrations of each NTP were adjusted according to oligoribonucleotide sequence. For example, to prepare ppp-AGU, we used 5 mM each of ATP, GTP, and UTP, and we used 8 mM each of ATP and GTP to transcribe pppAG. RNAs were purified on a 5-ml HiTrap Q column (GE Healthcare), precipitated by ethanol, washed with 80% ethanol, dried, and dissolved in water.
Crystallization of EcRppH and EcRppH Complexes-Crystals of EcRppH(1-156) were prepared by mixing 2 l of 0.5 mM protein and 2 l of reservoir solution containing 0.1 M Tris-HCl, pH 8.0, 1.5 M potassium acetate, 5% (v/v) glycerol, and 20 mM CaCl 2 . Crystals of EcRppH(1-158)AA were grown from a solution typically prepared by mixing 1 l of 1 mM protein solution and 2 l of reservoir solution composed of 0.4 M (NH 4 ) 2 SO 4 , 12% (v/v) PEG3350, and 12% (v/v) glycerol. All crystals were grown at 295 K for several days. To obtain crystals of EcRppH⅐RNA complexes, RNA was added at 1 mM to a drop of the reservoir solution prior to adding EcRppH(1-158)AA.
Cross-linking of EcRppH(1-158)AA⅐RNA Crystals-To remove sulfate ions from the EcRppH⅐RNA crystals, the crystals were stabilized by cross-linking with glutaraldehyde. First, to avoid inhibition of cross-linking, NH 4 ϩ cations were substituted by Na ϩ cations by transferring a crystal into a 1-l droplet of the stabilizing solution containing 20 mM sodium acetate, pH 5.0, 200 mM Na 2 SO 4 , 15% (v/v) PEG 3350, 15% (v/v) glycerol, and supplemented by 0.5 mM ppcpAGU. After 30 min of incubation, the crystal was transferred to a fresh 1-l hanging drop of the same solution and incubated for 1 h over a 2-l droplet of 50% (w/w) glutaraldehyde placed on a sitting drop bridge in the same well (24). During all procedures, drops were kept in the closed well with 400 l of stabilizing solution. After cross-linking, the crystal was transferred into a 2-l drop containing 50 mM MOPS, pH 7.0, 25 mM MgCl 2 , 15% (v/v) PEG 3350, and 25% (v/v) pentaerythritol propoxylate 5/4 PO/OH for 30 min. Crystals were frozen in cryostream prior to data collection.
Data Collection and Structure Determination-Diffraction data were collected at 100 K on an in-house Rigaku x-ray source and beamline X25 of the National Synchrotron Light Source, Brookhaven National Laboratory. Data were processed using XDS suite (Table 1) (25). The crystal structure for EcRppH(1-156) was solved using single wavelength anomalous diffraction method implemented in PHENIX.AUTOSOL (26). The initial model was built automatically by PHENIX.AUTOBUILD (26) and then adjusted manually in COOT (27). Structure refinement was conducted by PHENIX.REFINE (26) and REFMAC 5 (28) within CCP4 (29). Structures of EcRppH(1-158)AA in the apo-state and with RNA were solved by molecular replacement using PHENIX.AUTOMR (26) with EcRppH(1-156) as a search model. RNA and ions were added at the late stages of refinement based on the F o Ϫ F c and 2F o Ϫ F c electron density maps. Ca 2ϩ and Cl Ϫ ions were visualized by anomalous electron density maps.
In the EcRppH⅐3Mg⅐ppcpAGU complex structure, the tetrahedrally shaped density map was assigned to a noncleaved ␥ phosphate because of the following: (i) the Ϸ2.9 Å distance from the ␤ phosphate is somewhat short for positioning of a cleaved-off phosphate; (ii) a water molecule was not sufficient to account for the size of the map; (iii) the ␣ phosphate in this position would require at least the sugar atoms C5Ј and O5Ј to be shifted and generate a new density map, which was not observed at reasonable levels. The ␥ phosphate was refined at 50% occupancy to reflect the residual density map.
In Vitro Assay of EcRppH Activity-Pyrophosphohydrolytic assays were performed in 10 -50-l reactions containing 50 mM HEPES-Na, pH 7.5, 10 mM MgCl 2 , 0.1% (v/v) Triton X-100, 100 M to 1 mM triphosphorylated mono-or oligoribonucleotides, and 100 nM to 10 M RppH. In control reactions, dilution buffer or alkaline phosphatase at 0.001-0.1 units/l was added to the reaction mixtures. Reactions were incubated at 37°C for 10 -60 min and quenched by a 10-fold dilution with 50 mM sodium acetate, pH 5.0, for chromatographic and pyrophosphate analysis, or a 2-fold dilution with 40 mM EDTA for thin layer chromatography.
Chromatographic separation of reaction products was performed on 5 ϫ 50-mm Mono Q column (GE Healthcare) by 10 column volumes of a 0 -0.5 M NaCl gradient in 10 mM Tris-HCl, pH 8. Elution profiles were recorded at 260 nm, and peaks corresponding to monophosphorylated products or triphosphorylated substrates were integrated using UNICORN software (GE Healthcare). Absolute amounts of the reaction products were calculated based on the values of integrated peaks for substrates and products.
Inorganic pyrophosphate assay was performed colorimetrically according to the published method (30,31). Briefly, inorganic pyrophosphate from the quenched reactions was precipitated in the presence of 1 mM CaCl 2 and 100 mM NaF. Precipitate collected by centrifugation at 20,000 ϫ g for 10 min was washed with acetone, dried, and re-dissolved in solution containing 1.25 N H 2 SO 4 , 10 mM ammonium heptamolybdate, and 40 mM ␤-mercaptoethanol. Intensity of green color developed after 1 h of incubation at room temperature was measured at 700 nm, and the amount of the pyrophosphate liberated during the reaction was calculated from the calibration curve prepared with standard solutions of sodium pyrophosphate.
Thin layer chromatography was performed on PEI-cellulose plates and 0.3 M potassium phosphate buffer, pH 7.5, at room temperature. Chromatography samples were EDTA-quenched EcRppH reaction products that contained 5 nmol of RNA oligonucleotides in 10-l volumes. RNAs were visualized by UV 254 shadowing.

RESULTS
EcRppH Is Active on Short RNAs-Our x-ray crystallography experiments were driven by the unexpected finding, presented in the accompanying article (32), that EcRppH and BsRppH have similar length requirements but distinct sequence preferences for RNA 5Ј-ends. Like BsRppH, EcRppH favors guanine at the second position but, unlike BsRppH, is also active on RNAs that have adenine or a pyrimidine there (Fig. 1B) different specificities of EcRppH and BsRppH have important functional consequences for gene regulation mechanisms in bacteria. For instance, BsRppH, which is highly specific, appears to be supplemented by another RNA pyrophosphohydrolase that may target a distinct subset of mRNAs (14,15), whereas EcRppH, which is less selective, appears to be the only enzyme of its kind in E. coli. Because EcRppH orthologs are present in innumerable bacterial species and flowering plants, although BsRppH orthologs are much less common (32), we set out to investigate the selectivity of EcRppH-like RNA pyrophosphohydrolases by crystallizing EcRppH bound to target RNAs.
To define the minimal RNA substrate sufficient for targetspecific EcRppH activity, we compared pyrophosphate removal from 5Ј-triphosphorylated mononucleotides and RNAs of various lengths. In these RNAs, the first nucleoside was adenosine, preferred by EcRppH in long mRNAs (32); the second nucleoside was varied, and the third nucleoside, less important for selectivity, was uridine for convenience of RNA preparation. The reaction of these substrates with purified EcRppH was monitored by analyzing the products by anion-exchange (  Red, no EcRppH; blue, reaction products with 0.1 M EcRppH; black, reference compounds. Reactions were carried out with 100 M substrate for 10 min. B, EcRppH activity on the 5Ј-triphosphates of mononucleotides and oligonucleotides, as determined by chromatography (A) and graphed as the concentration of hydrolyzed substrates (mean Ϯ S.D., n ϭ 3). C, TLC of 5Ј-triphosphorylated and modified RNAs treated with 1 M EcRppH (R lanes) or 0.1 unit/l alkaline phosphatase (P lanes). pcppAGU and ppcpAGU contain a nonhydrolyzable methylene group (letter "c") that replaces a bridging oxygen atom. RNAs at 1 mM concentration were incubated with or without EcRppH for 1 h and visualized by UV 254 shadowing. D, EcRppH activity determined by colorimetric measurement of liberated inorganic pyrophosphate. Reactions were carried out as in A and B. E, Mg 2ϩ dependence of EcRppH activity with pppAG. pppAG hydrolysis by EcRppH is expressed as a percent of maximal activity observed for the enzyme. Reactions were performed at 100 nM EcRppH and 100 M pppAG at 37°C for 10 min. AC 50 is concentration of MgCl 2 , at which half-maximum of enzyme activity is achieved; n is Hill coefficient. Presented values are mean Ϯ S.D., n ϭ 3. F, inhibition of EcRppH activity with (NH 4 ) 2 SO 4 . Activity measurements and determination of IC 50 , at which half-maximum of enzyme activity is achieved, were conducted at 10 mM MgCl 2 as in E. 2D). With triphosphorylated mononucleotide substrates at physiological pH, EcRppH slowly cleaves off just the ␥ phosphate and removes both the ␥ and ␤ phosphates only at elevated enzyme concentrations (Fig. 2B). Under the same conditions, EcRppH is 10 -50-fold more active on di-and trinucleotide substrates (Fig. 2, B and D), from which it predominantly removes pyrophosphate in a single step (Fig. 2C). The highest activity was observed with the oligonucleotides pppAG and pppAGU, both of which bear guanosine at the second position ( Fig. 2, B and D). Adenosine was slightly less preferred at that position, whereas uridine and cytidine decreased activity by a factor of 1.6 -2.3 and 3.4 -3.7, respectively. Because the selectivity of EcRppH with oligonucleotides resembled its sequence preferences with long RNA substrates (32), trinucleotides were used as mimics of natural mRNAs in crystallization trials with EcRppH.
Crystal Structure of EcRppH in the Apo-form-Because initial efforts to co-crystallize EcRppH with various RNA ligands were unsuccessful, we focused on crystallizing EcRppH in the free form and soaking short RNAs into the crystal. Because full-length EcRppH did not crystallize, we progressively removed unstructured C-terminal amino acid residues (16) dispensable for enzymatic activity and identified two constructs, EcRppH(1-156) and EcRppH(1-160), that yielded crystals and crystalline needles. The structure of EcRppH(1-156) was determined at a 1.7-Å resolution (Table 1). Because a symmetry-related molecule blocked access of RNA in these crystals, we re-engineered crystal contacts by replacing residues that have flexible surfaceexposed side chains with alanines (20). One truncated mutant, bearing two alanine residues at the C terminus and hereafter referred as EcRppH(1-158)AA, crystallized with the front surface opened for RNA binding (Fig. 3).
Unlike BdRppH (17), EcRppH crystallized with one molecule in the asymmetric unit. EcRppH comprises a Nudix hydrolase domain composed of a mixed ␤-sheet (␤3, ␤1, ␤6, ␤4, and ␤5) attached to an antiparallel ␤-sheet (␤2 and ␤7) (Fig. 3A). The ␤-sheets are sandwiched between ␣-helices on the left (␣1) and right (␣4 and ␣5). The x-ray structure agrees well with the EcRppH NMR structure (16) and is reminiscent of the BdRppH and BsRppH structures overall (Fig. 3B) (15, 17), although it is dissimilar in many important details, consistent with the low sequence identity between the proteins (Fig. 3D). The catalytic site is primarily composed of ␣1, the ␣1-␤3 loop, and ␤3, the three structural elements that contain amino acid residues characteristic of the Nudix motif (Fig. 3A). Positively charged surfaces that might participate in RNA binding are located below and to the right of the catalytic site. Two sulfate anions bind in this area, possibly mimicking bound RNA phosphates (Fig. 3C). EcRppH⅐2Mg⅐ppcpAGU Structure Reveals Specifically Bound RNA-To determine a biologically relevant structure of EcRppH bearing an RNA ligand, we preserved the integrity of the enzyme and crystallized catalytically active EcRppH(1-158)AA bound to the trinucleotide ppcpAGU. This RNA substrate contains a nonhydrolysable methylene group (designated "c") substituted for an oxygen atom between the ␣ and ␤ phosphorus atoms of the 5Ј-triphosphate moiety (Fig. 2C). In an attempt to further reduce EcRppH activity and prevent minor cleavage between the ␤ and ␥ phosphates (Fig. 2C) (5), the crystals were grown in the absence of Mg 2ϩ cations, which are essential for catalysis. Enzymatic activity measurements revealed that EcRppH reaches almost full activity at ϳ10 mM Mg 2ϩ (Fig. 2E). To obtain the catalytically relevant structure, the concentration of (NH 4 ) 2 SO 4 was lowered to 50 mM to reduce inhibition of catalysis by the high concentration of this salt, and the crystals were stabilized by pentaerythritol propoxylate and soaked at neutral pH in a solution containing 50 mM Mg 2ϩ so as to allow Mg 2ϩ cations to interact with amino acids in the catalytic site. At this lower (NH 4 ) 2 SO 4 concentration, EcRppH retains ϳ40% of its maximal activity determined in the absence of (NH 4 ) 2 SO 4 at 10 mM Mg 2ϩ (Fig. 2F).
The resulting 1.5-Å crystal structure of the complex revealed the majority of the protein chain, two Mg 2ϩ cations, and a large part of RNA, only missing the ␥ phosphate and the third RNA nucleoside ( Fig. 4A and Table 1). The 5Ј-end of the RNA is positioned in the catalytic site, whereas the remainder of the RNA binds EcRppH in the region to the right of the catalytic site. The first nucleobase, Ade1, defined by a fragmented electron density map (Fig. 4B), is rotated outward toward solution, whereas the well defined nucleobase of Gua2 sits in a cleft formed mostly by the ␤2-␤3 loop and ␣5.
The structure revealed two Mg 2ϩ cations bound to several glutamates and a glutamine in the catalytic site. To coordinate the Mg 2ϩ cations, the side chains of these glutamates change their conformation, and we tracked these changes in a single crystal. The 5Ј-end of the RNA is bound by direct coordination of its ␤ phosphate to Mg1 and Mg2 (Fig. 4C). In addition, both the ␣ and ␤ phosphates form hydrogen bonds with Arg-8 and Gly-39. The ␥ phosphate was not modeled in the structure because a tetrahedrally shaped electron density adjacent to the ␤ phosphate (Fig. 4B) was ϳ0.6 Å further away than expected for a covalently linked ␥ phosphate. Although this density might belong to the cleaved off ␥ phosphate, the density most likely corresponds to a SO 4 2Ϫ anion capable of binding in the same   region in the apo-structure and present in the crystals at the much higher concentration. Despite the loss of the ␥ phosphate, the diphosphorylated RNA remains a substrate for EcRppH in solution (Fig. 2C). Because none of the ordered water molecules were aligned well for cleavage between ␣ and ␤ phosphates, we hypothesize that this structure reflects one of the states on the reaction pathway either before or after the catalytic step.
Distinct Multiple Interactions Contribute to Binding of the Second Nucleobase-The EcRppH⅐2Mg⅐ppcpAGU structure showed that the base of Ade1 apparently rotates freely along the glycosidic bond and does not form specific hydrogen bonds with EcRppH. Contacts between the protein and Ura3 are limited to a nonspecific hydrogen bond between a phosphate and Arg-27 (Fig. 4D). By contrast, Gua2 forms multiple interactions with the protein. First, the sugar-phosphate moiety is hydrogen-bonded to side chains of Gln-37 and Tyr-77. Second, the guanine base is sandwiched between a side chain of Arg-27 and the hydrophobic platform formed by Val-137 and Phe-139. In this position, the base forms cationinteractions with the guanidinium group of Arg-27. Third, atoms on the Hoogsteen edge are hydrogen-bonded with the side chain of Lys-140. Fourth, on the Watson-Crick edge, N2 makes a direct hydrogen bond to Ser-32 in one of its alternative conformations, whereas N1 forms water-mediated bonds with Ser-32, Trp-3, and Val-137. These multiple contacts of EcRppH with Gua2 strongly suggest that Gua2 is critical for RNA binding and explain the observed preference for guanine in the second position.
Although direct hydrogen bonding to the second nucleobase favors guanine, it neither precludes binding of adenine and pyrimidines in a semi-open RNA-binding site nor explains the much stronger preference for adenine over pyrimidines. The structure suggests that the other factors contributing to higher affinity for purines include hydrophobic and cationinteractions. Thus, the semi-specific substrate preference of EcRppH is not likely to be determined solely by hydrogen bonding and probably involves an unusual combination of different types of interactions (see "Discussion").
EcRppH⅐3Mg⅐ppcpAGU Structure Reveals Changes in Protein and RNA Conformation-We reasoned that the sulfate anion bound in the catalytic site prevented the EcRppH⅐2Mg⅐ppcpAGU complex from adopting a catalytically active conformation. Because sulfate anions participate in crystal packing interactions and cannot be replaced during crystallization, we stabilized the crystals by glutaraldehyde cross-linking and removed loosely bound sulfate anions by soaking in solutions that were sulfate-free and contained Mg 2ϩ . This procedure resulted in a 1.6-Å crystal structure of the EcRppH⅐3Mg⅐ppcpAGU complex, which contained EcRppH in the same global conformation as before, three Mg 2ϩ cations, and RNA positioned in the RNA-binding site slightly differently than in the EcRppH⅐2Mg⅐ppcpAGU structure (Fig. 5, A-C).
In addition to Mg1 and Mg2, another Mg 2ϩ cation, Mg3, binds the catalytic site in place of the sulfate anion and is coordinated by the ␤ phosphate and Glu-53 (Fig. 5D). Glu-56 returns to the apo conformation, although the ␣ phosphate is located far from Arg-8 and in the vicinity of Gln-37 and Mg1. The ␤ phosphate remains in the same position but directly coordinates all three magnesium ions. We observed residual density for the ␥ phosphate linked to ␤ phosphate (Fig. 5C), likely because our crystal treatment decreased EcRppH activity and prevented complete hydrolysis of the ␥ phosphate. Importantly, Mg3 binding leads to the formation of a binuclear Mg2⅐Wat-1⅐Mg3 complex that stabilizes water molecule Wat-1 in position for an in-line nucleophilic attack on the ␤ phosphorus atom (Fig. 5D). Therefore, the EcRppH⅐3Mg⅐ppcpAGU structure appears to be closest to the "catalytic" conformation of the complex.
Concurrently with the conformational changes at the 5Ј-end of the RNA, Ade1 and Gua2 shift their positions in the EcRppH⅐3Mg⅐ppcpAGU structure (Fig. 5, A and B). In comparison with the EcRppH⅐2Mg⅐ppcpAGU structure, the 5Ј-phosphate of Gua2 moves toward Lys-140, whereas the guanine base slides slightly out of the cleft and loses direct hydrogen bonds with Lys-140 (Fig. 5, B and E). In this new position, the Gua2 nucleobase forms only water-mediated hydrogen bonds and weakened cationinteractions with EcRppH and is therefore more mobile and less defined in the electron density map.
EcRppH Mechanism Involves Conformational Rearrangements-On the basis of these structures, we propose the following catalytic cycle for EcRppH. Initially, EcRppH binds Mg 2ϩ cations Mg1 and Mg2 either directly or likely in complex with the triphosphate moiety of an RNA substrate. Although a Ca 2ϩ cation binds between the Mg1 and Mg2 sites in the EcRppH(1-156) structure, so far our studies failed to identify Mg 2ϩ cations bound to the protein without RNA. Protein-bound Mg1 and Mg2 facilitate binding of the RNA 5Ј terminus through electrostatic interactions and coordination of the ␤ phosphate. The Gua2 nucleobase binds deeply and specifically in the protein cleft, resulting in the conformation found in the EcRppH⅐2Mg⅐ppcpAGU complex (Fig. 5B)  Mg3 and positioning of the ␣ phosphate closely to Gln-37 stabilizes water molecule Wat-1 between Mg2 and Mg3 and close to Glu-56 and the ␤ phosphorus atom (Fig. 6A). In this location, Wat-1 forms a nearly ideal angle (170.6°) for in-line attack on the ␤ phosphorus atom. Interestingly, placement of the ␣ phosphate in the catalytically relevant conformation near Gln-37 causes the Gua2 nucleobase to slide out of the cleft, weakening interactions with the protein, as observed in the EcRppH⅐3Mg⅐ppcpAGU complex.
As suggested for other Nudix hydrolases (33), Mg 2ϩ cations can function as Lewis acids and assist in deprotonating Wat-1 by a catalytic residue of RppH. Our structural studies show that Glu-56 is the most plausible candidate for accepting a proton from Wat-1 (Fig. 6, A and B). This residue apparently switches between two conformations during catalysis and may serve as a ligand sensor and reaction trigger. Other adjacent residues and Mg 2ϩ cations hold the substrate in place and stabilize the transition state, although Arg-8 likely assists the departure of the pyrophosphate leaving group, presumably in complex with Mg3. The monophosphorylated RNA product loses ␤ phosphate-mediated interactions with EcRppH and should easily dissociate from the protein if tight interactions with the Gua2 nucleobase have been weakened by the movement of this base out of the cleft.
To evaluate the contribution of Nudix motif residues to catalysis, we replaced the Nudix glutamates with alanines and observed significantly decreased (E53A and E56A) or abolished (E57A) EcRppH activity (Fig. 6D). However, substitution of alanine for Glu-120, which functions as a lid covering Mg1 and Mg2, reduces activity by only ϳ2.5-fold. Our data also show that Arg-52 of the Nudix motif holds Glu-53 and Glu-56 in the catalytic site and is absolutely required for catalysis. Arg-8, which coordinates the ␤ phosphate and likely stabilizes the leaving product, is also essential for EcRppH activity. These mutagenesis data demonstrate the important contribution of Nudix amino acids, including a candidate catalytic residue Glu-56, to catalysis and the less critical role of non-Nudix Glu-120 in EcRppH activity.

DISCUSSION
To understand the molecular basis of the bacterial reaction functionally analogous to decapping of eukaryotic mRNAs, we determined the crystal structure of EcRppH in the apo-form and bound to RNA and Mg 2ϩ cations. Although crystallization of low specificity RNA⅐protein complexes often yields poor electron density maps because of the mobility of the loosely bound RNA and the dissociation of the weak complexes under harsh crystallization conditions, we developed crystal treatment procedures that allowed us to visualize the first two nucleotides of EcRppH-bound RNA almost in their entirety. The EcRppH⅐RNA structure is more complete than the previously reported BsRppH⅐RNA structure, which mapped only the 5Ј-terminal triphosphate moiety, the second nucleotide, and two Mg 2ϩ cations (15).
The EcRppH⅐RNA structures reveal that EcRppH binds in a bipartite manner to the 5Ј-phosphates and to a nucleobase at the second position, as well as to adjoined sugar-phosphate moieties (Figs. 7A and 8A). As in the BsRppH⅐RNA complex (Fig. 8C) (15), the first and third nucleobases are not specifically recognized but must be unpaired to place the second nucleobase into the RNA-binding site (15,32). Recognition of the second nucleotide by EcRppH and BsRppH allows discrimination between mRNAs and mononucleotides and prevents hydrolysis of cellular NTPs by RppH. Too short to simultaneously reach both the triphosphate and nucleobase two binding sites, NTPs are expected to be poor substrates, as confirmed by the very slow cleavage of the ␥ phosphate of ATP. This conclusion raises questions about the specificity of Bdellovibrio RppH in vivo because GTP binds to BdRppH at a position equivalent to where the first nucleotide of RNA binds to EcRppH and BsRppH (Fig. 8, A-C) (17). Although the triphosphate of GTP reaches the catalytic center of BdRppH, it was not hydrolyzed in the crystals, suggesting that the published structure of the BdRppH⅐3Mg⅐GTP complex may not be representative of how the enzyme functions with an RNA substrate.  Our structural studies uncovered markedly dissimilar modes of RNA binding by EcRppH and BsRppH despite their structural and functional similarities. Although both prefer guanine at the second position of their RNA substrates, the enzymes bind the second nucleobase differently, in either a deep cavity (BsRppH) or a cleft (EcRppH), each lined by a distinct set of nonequivalent amino acids (Fig. 7). The most notable sequence-specific contact in the EcRppH⅐RNA complex is two hydrogen bonds from Lys-140 to the Hoogsteen edge of guanine. These interactions would be expected to play a large part in determining the specificity of mRNA recognition by EcRppH because only guanine in an anti conformation can accept two hydrogen bonds from the lysine side chain, whereas adenine and uracil can accept only one hydrogen bond and cytosine cannot accept a hydrogen bond. Our structural prediction is corroborated by the reduced selectivity of the K140M mutant reported in the accompanying article (32). An additional hydrogen bond between Ser-32 and N2 of guanine might not contribute much to the specificity of EcRppH because only one of two alternative conformations of this residue allows it to contact the RNA. Nevertheless, mutations of Ser-32 can enhance selectivity, suggesting that this residue contributes to the promiscuity of EcRppH (32).
Although the structures with adenosine and pyrimidines in the second position have not been determined yet, our structures with Gua in this position suggest that, in contrast to many other RNA⅐protein complexes, direct hydrogen bonding alone cannot explain the preference of EcRppH for guanosine. Most likely, hydrophobic and cationinteractions also contribute to the specificity of RNA binding. Compared with pyrimidines, purines have a larger surface area that promotes stronger interactions with the pedestal formed by the hydrophobic side chains of Val-137 and Phe-139. Purines also form stronger cationinteractions with the guanidinium group of Arg-27 according to the calculated strength of cation-interactions in DNA⅐protein complexes (G Ͼ AϾ T Ͼ C) (34). Thus, loss of cationinteractions explains the reduced ability of the R27A mutant to discriminate between purines and pyrimidines (32). Although cation-interactions with ligands are not common among Nudix hydrolases, they may be involved in adenine binding by the plant Ap 4 A hydrolase (35).
The structures illustrate that multiple features of EcRppH that favor binding of purines are compatible with weaker interactions with pyrimidines. Therefore, an unusual combination of low specificity interactions and an open RNA-binding site defines the relaxed specificity of EcRppH. In contrast to EcRppH, BsRppH binds guanine in a tight pocket where the base is surrounded by amino acid side chains that recognize the base through a set of direct hydrogen bonds incompatible with other nucleobases (Fig. 7). Structural comparison reveals that the only sequence-specific contact with nucleobase 2 common to both RNA⅐protein complexes has remarkably resulted from convergent evolution that positioned nonequivalent lysines to recognize the Hoogsteen edge of guanine.
OurstructuresoftheEcRppH⅐RNAcomplexespresenttwoimportant pieces of information that encourage rethinking the catalytic mechanism of Nudix proteins. The EcRppH⅐2Mg⅐ppcpAGU structure demonstrates that Glu-56 makes a hydrogen bond with water molecule Wat-1. The EcRppH⅐3Mg⅐ppcpAGU structure reveals that Wat-1 is flanked by two Mg 2ϩ cations and is ideally positioned for the catalytic reaction. These structures hint at the direct involvement of Glu-56 in catalysis, namely in activating Wat-1 for in-line nucleophilic attack on the ␤ phosphorus atom. This hypothesis is supported by the severe effect of the E56A mutation on activity. In addition, in several structures of Nudix hydrolases, the counterpart of Glu-56 is positioned in proximity to a water molecule analogous to Wat-1 (36,37). Glu-56 is one of the most conserved residues of the Nudix motif and is thought to be essential for coordinating metal cations (33). However, in the EcRppH⅐3Mg⅐ppcpAGU and several other structures (36 -38), this residue does not coordinate divalent cations. Thus Glu-56 may be more important for activating a water molecule for catalysis than for binding Mg 2ϩ cations involved in catalysis.
Previous studies proposed that another glutamate of the Nudix motif (Glu-53 in EcRppH) sometimes acts as a catalytic residue in Nudix hydrolases (33). However, this residue coordinates two Mg 2ϩ cations in the EcRppH⅐RNA complex, BdRppH (17), and other Nudix enzyme structures (36,37) and therefore may have difficulty deprotonating the nucleophilic water molecule. To alleviate this contradiction, it was suggested that Glu-53 could function as a catalytic residue in Nudix hydrolases if coordinated to only a single Mg 2ϩ cation (33). Our structural data and the residual activity of the Glu-53 mutants of EcRppH and BdRppH (17) do not support the idea that Glu-53 is the sole catalytic residue of these RNA pyrophosphohydrolases. Nevertheless, our data do not exclude the possibility that Glu-53 is responsible for the residual activity observed for the E56A mutant at high EcRppH concentrations.
The E57A mutation abolished EcRppH activity entirely, in agreement with the severe effects of Glu-57 mutations on the activity of other Nudix hydrolases, such as MutT (33). Our structures do not reveal any water molecule that could be activated for nucleophilic attack by this residue, but its side chain does undergo a conformational change upon Mg 2ϩ binding. Because Glu-57 coordinates two key Mg 2ϩ cations, this amino acid residue is essential for properly positioning the enzymebound metal cations.
The crystal structures of the EcRppH⅐RNA complexes revealed a conformational change in the side chain of Glu-120 that allows it to coordinate Mg1 and Mg2. Despite the seemingly strategic position of this side chain, which "covers" Mg1 and Mg2 and stabilizes their enzyme-bound positions, Glu-120 has a relatively low impact on catalytic activity, emphasizing the fact that it assists catalysis but does not play an essential role in the reaction mechanism. This glutamate residue not only is absent from many Nudix hydrolases but also is not conserved among RppHs. For example, it is replaced by His in BdRppH.
Mutational analysis showed that replacement of positively charged Arg-8 or Arg-52 significantly reduces EcRppH activity. This effect is easily explained for Arg-52, a conserved residue that orients the side chain of Glu-53 to facilitate its function. The strong reduction in activity observed for the R8A mutant was less expected because Arg-8 is not conserved among RppHs. This residue contacts the ␥ and ␤ phosphates and likely stabilizes the leaving group. Although Arg-8 is not conserved, BdRppH and BsRppH contain lysines that are situated in proximity to the 5Ј-terminal phosphates and that may have a function similar to that of Arg-8 in EcRppH (Fig. 8, D-F).
The catalytic mechanism proposed for EcRppH may also pertain to BdRppH (17) and BsRppH (15), which have similarly organized catalytic centers despite the presence of different numbers of bound Mg 2ϩ cations (Fig. 8, D-F). In all of these complexes, Glu-56 adopts the same conformation and therefore has the potential to deprotonate the catalytic water molecule Wat-1. The EcRppH and BdRppH structures show that Wat-1 is aligned for nucleophilic attack on the ␤ phosphorus atom and hydrolysis between the ␣ and ␤ phosphates. Although the BsRppH structure does not reveal Wat-1, such a water molecule would attack the ␥ phosphorus atom and hydrolyze the bond between the ␥ and ␤ phosphates. Thus, the preferential initial cleavage between the ␥ and ␤ phosphates reported for BsRppH can be explained by the differential binding of the triphosphate moiety in the catalytic site so that the ␥ phosphate in the BsRppH⅐RNA complex is positioned in the location of the ␤ phosphate in the EcRppH⅐RNA complex. As suggested in the earlier model of the BsRppH catalysis (15), the flexibility of the first nucleotide may facilitate movement of the ␤ phosphate closer to the metal cations for the second round of catalysis after the ␥ phosphate has left.
Species ranging from bacteria to mammals can contain over 20 Nudix enzymes involved in distinct cellular activities. Therefore, our mechanistic study to gain insights on the catalytic mechanism of EcRppH will have an impact on many prokaryotic and eukaryotic systems. At the moment, there is no catalytic mechanism accepted as universal for Nudix hydrolases, and several amino acid residues have been proposed to play a catalytic role in different enzymes. However, the Nudix motif is the only set of residues common to all Nudix hydrolases, and because an important catalytic residue is expected to be preserved in evolution, it is likely that such a residue would reside in this motif in the majority of Nudix enzymes. The difficulty in understanding Nudix catalysis likely stems from the abundance of noncatalytic enzyme structures that have been reported and the strong effects on catalysis of various residues that form the catalytic site. The structure of the E. coli ADP-ribose pyrophosphatase (39) indirectly supports our hypothesis that Glu-56 directly participates in catalysis by the majority of Nudix hydrolases. This enzyme has an extra glutamate that can activate the nucleophilic water molecule. However, this glutamate is not conserved, and many Nudix proteins do not have a specialized non-Nudix amino acid residue that could play a similar catalytic role.
An intriguing observation in our study is a slightly different position of the Gua2 nucleobase in the binding cleft in two EcRpH⅐RNA complexes. Surprisingly, specific hydrogen bonding and deep positioning of this nucleobase in the cleft were observed in the catalytically inactive complex although shallow positioning and loss of specific hydrogen bonds were seen in the catalytically relevant complex. We also noticed that placement of the ␣ phosphate in the catalytically relevant conformation appears to be structurally incompatible with deep binding of the Gua2 nucleobase in the cleft. To explain two distinct means of the Gua2 nucleobase binding, we hypothesize that EcRppH initially forms base-specific interactions with the nucleotide at the second position of RNA, although the triphosphate moiety and Mg 2ϩ cations are not fully bound to the enzyme. Weak initial binding of the triphosphate moiety may constitute a specific adaptation of RppH to reject nucleotides and other phosphate-containing noncognate substrates so that productive interactions are formed only with RNA. Following initial binding, movement of the triphosphate to the catalytic conformation would then result in tighter binding, hydrolysis, and sliding of the second nucleobase partway out of the cleft, thereby facilitating release of the monophosphorylated RNA product from the protein.
The crystal structures of the EcRppH⅐RNA complexes reveal how a Nudix hydrolase that does not possess an established RNA-binding fold has nevertheless evolved for RNA binding and catalysis. The bipartite mRNA recognition mechanism involving the 5Ј-terminal phosphates and the second RNA nucleotide distinguishes EcRppH from Argonaute (40), RIG-I (41,42), and other proteins (43,44) that recognize RNA ends. The unique combination of molecular features revealed by the EcRppH⅐RNA complexes expands the known repertoire of principles that govern RNA binding and demonstrates that functionally and structurally analogous enzymes can employ very different structural principles for the recognition of similar substrates.