Structural Determinants for Substrate Binding and Catalysis in Triphosphate Tunnel Metalloenzymes

Background: Triphosphate tunnel metalloenzymes carry out diverse enzymatic reactions. Results: Two metal co-factors are identified involved in substrate binding and in catalysis. Conclusion: A unified catalytic mechanism is proposed and biochemically investigated. Significance: The functional diversity of TTM enzymes is rationalized by a common mechanism that allows very different substrates to be bound and processed.

Inorganic polyphosphate (polyP) is a linear polymer of orthophosphate units joined by phosphoanhydride bonds. It occurs ubiquitously and abundantly in all life forms (1). In bacteria, polyP kinases generate polyP from ATP, but it is presently unknown how the cellular polyP stores of higher organisms are being synthesized (2,3). We previously reported a fungal polyP polymerase that is distinct from bacterial polyP kinases (4,5). The yeast enzyme resides in the vacuolar transporter chaperone (VTC) 2 membrane protein complex, which generates polyP from ATP in the cytosol and translocates the growing chain into the vacuole (4,6). The catalytic core of VTC maps to a cytoplasmic 8-stranded ␤-tunnel domain in Vtc4p, which harbors binding sites for the nucleotide substrate, for a manganese metal co-factor and for an orthophosphate priming the polymerase reaction (4).
A staggering number of different catalytic activities and substrate preferences has been reported within the TTM family (10). However, the lack of substrate-and product-bound structures has made it difficult to define the sequence-fingerprints responsible for a specific enzyme activity in individual family members.

Experimental Procedures
Protein Expression and Purification-AtTTM3 (Uniprot ID Q9SIY3) was cloned into vector pMH-HT providing an N-terminal His 6 tag and a tobacco etch virus protease cleavage site. Protein expression in Escherichia coli BL21 (DE3) RIL to A 600 nm ϭ 0.6 was induced with 0.25 mM isopropyl ␤-D-galactoside in terrific broth at 16°C for 16 h. Cells were collected by centrifugation at 4500 ϫ g for 30 min, washed in PBS buffer, centrifuged again at 4500 ϫ g for 15 min, and snap-frozen in liquid nitrogen. For protein purification cells were resuspended in lysis buffer (50 mM Tris-Cl (pH 8.0), 500 mM NaCl, 5 mM 2-mercaptoethanol), homogenized (Emulsiflex C-3, Avestin), and centrifuged at 7000 ϫ g for 60 min. The supernatant was loaded onto a Ni 2ϩ affinity column (HisTrap HP 5 ml, GE Healthcare), washed with 50 mM Tris (pH 8), 1 M NaCl, 5 mM 2-mercaptoethanol, and eluted in lysis buffer supplemented with 200 mM imidazole (pH 8.0). The His 6 tag was cleaved with tobacco etch virus for 16 h at 4°C during dialysis against lysis buffer. AtTTM3 was further purified by a second Ni 2ϩ affinity step and by gel filtration on a Superdex 75 HR26/60 column (GE Healthcare) equilibrated in 25 mM Tris (pH 7.2), 250 mM NaCl, 5 mM 2-mercaptoethanol. Monomeric peak fractions were dialyzed against 20 mM Hepes (pH 7.4), 50 mM NaCl, 0.5 mM tris(2-carboxyethyl)phosphine and concentrated to 35 mg/ml for crystallization. Protein concentrations were estimated by protein absorption at 280 nm using the calculated molecular extinction coefficient. Site-specific mutations were introduced by PCR, and mutant proteins were purified like wild type.
The coding sequence of ygiF was amplified from genomic DNA (E. coli Mach 1 cells, Life Technologies) by PCR, and a synthetic gene coding for full-length mouse ThTPase (Uniprot ID Q8JZL3) and codon-optimized for expression in E. coli was obtained from Geneart (Life Technologies). Coding sequences were cloned into plasmid pMH-HT, and expression and purification were performed as described for AtTTM3.
Crystallization and Data Collection-Tetragonal AtTTM3 crystals (form A) developed at room temperature from hanging drops composed of 1.5 l of protein solution and 1.5 l of crystallization buffer (2.6 M NaCl, 0.1 M citric acid/NaOH (pH 5.0)) suspended over 1.0 ml of the latter as reservoir solution. Crystals were transferred in reservoir solution supplemented with 20% (v/v) ethylene glycol and 0.5 M KI for 20 s and snap-frozen in liquid nitrogen. Single-wavelength anomalous diffraction (SAD) data were collected on a Rigaku MicroMax rotating anode equipped with a copper filament, osmic mirrors, and an R-AXIS IVϩϩ detector. Subsequently, an isomorphous native dataset was collected on a crystal originating from the same crystallization drop (see Table 1). A monoclinic crystal form (form B) developed in 20% (w/v) PEG 3350, 0.2 M NaCl, 0.1 M Bis-Tris (pH 7.0) and diffracted up to 1.3 Å resolution. Data processing and scaling were done with XDS (November 2014 version) (28). Hexagonal crystals for full-length ygiF grew at room temperature in hanging drops (1.5 l and 1.5 l) containing 0.2 M NaCl, 20 (w/v) PEG 3350. Crystals were transferred into crystallization buffer supplemented with 20% (v/v) ethylene glycol and snap-frozen in liquid nitrogen. For phasing, 0.1 M NaI was added to the cryo solution, and crystals were soaked for 5 min. SAD data were collected at a wavelength of 1.9 Å (see Table 2). Tetragonal mouse ThTPase crystals (form A) developed at room temperature from hanging drops composed of 1.5 l of protein solution and 1.5 l of crystallization buffer (27% (w/v) PEG 3350, 0.1 M Tris (pH 9.0), 0.2 M MgCl 2 ) suspended over 1.0 ml of the latter as reservoir solution. Crystals were transferred into a reservoir solution supplemented with 20% (v/v) ethylene glycol and snap-frozen in liquid nitrogen. Monoclinic crystals (form B) developed in 1.6 M sodium/potassium phosphate buffer pH 6.8 (1:1 ratio) and were cryo-protected by stepwise transfer into a solution containing 1.6 M sodium/potassium phosphate buffer pH 6.8 and 20% (v/v) ethylene glycol.
For ThTPase-ThDP-Mg 2ϩ , the product-bound structure was obtained by soaking form A crystals in a solution containing 10 mM sodium tripolyphosphate and 10 mM MnCl 2 for 30 min. Two data sets were collected, one with ϭ 1.0 Å and one close to the Mn-K edge ( ϭ 1.9 Å). No anomalous signal was found in the latter data set, possibly due to the high MgCl 2 (0.2 M) concentration present in the crystallization buffer.
Structure Solution and Refinement-To solve the structure of AtTTM3, SAD and native data were scaled together with the program XPREP (Bruker AXS, Madison, WI) for SIRAS (single isomorphous replacement with anomalous scattering) phasing. The program SHELXD (29) was used to locate 37 iodine sites. Consistent sites were input in the program SHARP (30) for site refinement and phasing at 3.0 Å resolution. Density modification and phase extension to 2.6 Å was carried out with PHENIX.RESOLVE (31). The structure was built in alternating cycles of model correction in COOT (32) and restrained refinement in refmac5 (33) against a high resolution native data set ( Table 1). The structure of crystal form B was determined by molecular replacement with the program PHASER (34).
The structure of ygiF was solved by scaling redundant SAD data using XPREP. SHELXD located 15 consistent iodine and sulfur sites, which were input into SHARP for SAD site refinement and phasing at 2.7 Å resolution. Density modification and

Crystallographic data collection and refinement statistics for AtTTM3
Highest resolution shell is shown in parenthesis.  Table 2). The structure of mouse ThTPase was solved using the molecular replacement methods as implemented in the program PHASER and using the structure of the human ThTPase (PDB ID 3BHD) as a search model. Analysis with MolProbity (35) indicated excellent stereochemistry for all refined models. Phasing and refinement statistics are summarized in Tables 1-3. Synthesis of Thiamine Triphosphate-Thiamine triphosphate was synthesized as described (36) and purified by preparative high performance liquid chromatography.

AtTTM3
NMR Time-course Experiment-A series of one-dimensional 31 P NMR experiments was acquired at 310 K with a 600-MHz Bruker Avance-III spectrometer using a QXI probe-head allowing direct detection of 31 P and equipped with a z-spoil gradient coil. 31 P spectra were recorded using a relaxation delay of 1 s and an acquisition time of 42.6 ms (spectral width ϭ 12,019.23 Hz). 128 scans were collected, resulting in a measurement time of 140 s per spectrum. 512 spectra were collected over a total acquisition time of ϳ20 h. The enzymatic reaction was performed using 50 nM AtTTM3 and 5 mM sodium tripolyphosphate in 20 mM Bis-Tris propane (pH 8.5), 250 mM NaCl, 5 mM MgCl 2 mixture. Deuterated water was added to a final concentration of 20% to the reaction mix before starting the experiment. Spectral parameters were calibrated and optimized on a 5 mM sodium tripolyphosphate sample to minimize time loss between the beginning of the reaction and the beginning of its observation by NMR. Spectra were processed using Topspin (version 2.1.6) (Bruker).

Phosphohydrolase Activities of AtTTM3 and ygiF Mutant
Proteins-For the determination of the phosphohydrolase activity 2.5 nM AtTTM3 were incubated with 0.5 M concen-  trations of the different phosphate-containing substrates in reaction buffer (150 mM NaCl, 20 mM Bis-Tris propane (pH 8.5), 5 mM MgCl 2 ) at 37°C. The reaction was stopped after 10 min, and the amount of free inorganic phosphate released was measured using the malachite green assay with minor modifications (37). 100 l of reaction solution were mixed with 28 l of dye mix (3 mM malachite green, Tween 20). After 5 min of incubation with the dye, the absorption at 595 nm was measured using a synergy H4 plate reader (Biotek). For each substrate a blank reaction was prepared lacking the enzyme. Controls either contained EDTA to a final concentration of 5 mM, or the enzyme was heat-inactivated at 95°C for 5 min. To compare wild-type and mutant versions of AtTTM3 and ygiF, enzyme concentrations were increased to 180 nM to detect residual activity for some of the mutant proteins. Experiments were performed in triplicate; average values are plotted ϮS.D. Thermal shift assays were performed as previously described (38). Sypro Orange (Sigma) was added to wild-type and mutant AtTTM3 proteins in 25 mM Tris (pH 8.0), 250 mM NaCl, 5 mM 2-mercaptoethanol mixed with Sypro Orange to a final protein concentration of 10 M. The protein-dye mixtures were loaded into a 96-well RT-PCR plate (Thermo Scientific), and measurements were performed using a C1000 thermal cycler (Bio-Rad). The fluorescence of SYPRO Orange was continously monitored at 570 nm, as a temperature gradient was applied (0.05°C/s from 10°C to 95°C). Data were analyzed using the

Results and Discussion
Our initial aim was to identify a polyP polymerase in plants using a combined structural and biochemical screen. We located three putative TTM proteins in Arabidopsis thaliana. AtTTM3 shares 12% sequence identity with Vtc4p (r.m.s.d. is 2.6 Å, comparing 149 corresponding C ␣ atoms and 1.1 Å comparing 77 C ␣ atoms in the tunnel center) and contains a conserved ␤-tunnel domain (26). We determined co-crystal structures of AtTTM3 with tripolyphosphate and Mg 2ϩ or Mn 2ϩ metal co-factors in two independent crystal lattices (Fig. 1). Crystal form B corresponds to the previously reported structure of AtTTM3 (PDB ID 3v85) in complex with a citric acid molecule (r.m.s.d is 0.3 Å comparing 203 corresponding C ␣ atoms) (26). Our structures reveal a conserved mode of substrate and metal co-factor binding in AtTTM3 and Vtc4p and an invariant arrangement of residues in the tunnel center ( Fig.  2A) (4). We thus tested whether AtTTM3 can polymerize polyP from ATP or other nucleotide substrates but could not detect such an activity (Fig. 2B). Instead, AtTTM3 has specific, Mg 2ϩ ion-dependent, short-chain polyphosphatase activity (Fig. 2C) as previously reported (26). We find that AtTTM3 hydrolyzes tripolyphosphate (PPP i ) into pyro (PP i )-and orthophosphate (P i ) with a turnover rate of ϳ10/s (Fig. 2, D and E). The P i release from PPP i amounts to 27 Ϯ 6/s when assayed by malachite green (see "Experimental Procedures"). AtTTM3 is not able to hydrolyze PP i but appears to catalyze the asymmetric cleavage of inorganic polyP n (n ϭ 3-15), releasing P i (Fig. 2, C and D).
Based on kinetic and mutational studies, one (23) and twometal (40, 41) mechanisms have been proposed for TTMs, but only a few substrate complexes in the presence of metal co-factors have been reported thus far (4,18). As AtTTM3 efficiently hydrolyzes PPP i in the presence of Mg 2ϩ ions only at neutral or basic pH, we used two crystal forms grown at pH 5.0 (form A) and pH 7.5 (form B) to crystallize substrate-metal complexes under non-hydrolyzable and hydrolyzable conditions, respec-  (Tables 1 and 2). At pH 5.0 (form A) we find a triphosphate moiety bound in the AtTTM3 active site. Here, PPP i coordinates a Mg 2ϩ /Mn 2ϩ ion that acts as a bridge between the substrate and the invariant Glu-169 AtTTM3 (Fig. 3A). The same metal co-factor coordina- . An acidic patch in TTM proteins allows for the coordination of two metal ions. A, structural superposition of AtTTM3 (in yellow) and Vtc4p (in gray) (r.m.s.d. is 1.1 Å comparing 77 corresponding C ␣ atoms in the tunnel center) identifies an acidic patch in AtTTM3 (in orange, in bond representation), which is absent Vtc4p. The AtTTM3 tripolyphosphate substrate and a Mn 2ϩ ion (magenta sphere) are shown alongside. B, structure of a AtTTM3 post-catalysis complex reveals two orthophosphates and three Mn 2ϩ ions bound to the tunnel center. C, ribbon diagram of full-length ygiF from E. coli. The tunnel domain (residues 1-200) is shown in yellow (␤-strands) and blue (␣-helices), the linker region (residues 201-220) is highlighted in cyan, and the four-helix bundles of the C-terminal CHAD domain are depicted in red (residues 221-324) and orange (residues 383-433), respectively. D, structural superposition (r.m.s.d. is 0.65 Å comparing 75 corresponding C ␣ atoms) of the ygiF tunnel core (in yellow, the acidic patch is in orange) bound to PPP i and Mn 2ϩ ions (magenta spheres) with the AtTTM3 product complex (in gray). A phased omit difference density map contoured at 25 is shown alongside (black mesh). Note that two of the three AtTTM3 metal coordination sites are also found in ygiF. E, table comparison of acidic patch residues in different TTM proteins. RTPase, RNA triphosphatase. tion has been previously found in a Vtc4p AppNHp-Mn 2ϩ complex (4) (Figs. 3A and 4A). A short soak at pH 7.0 (form B) shows the same arrangement of PPP i and Mn 2ϩ in the active site (Fig. 1). However, when we co-crystallized AtTTM3 with its substrate at neutral pH, we found PPP i hydrolyzed and the tunnel center occupied by two orthophosphates and three Mn 2ϩ ions (Fig. 3B). One of these Mn 2ϩ ions is again found coordinated by the two phosphates (which correspond to the ␣ and ␥ positions in PPP i ; Fig. 4B) and by Glu-169 AtTTM3 , whereas the remaining metal coordination centers are formed by Glu-2 AtTTM3 , Glu-4 AtTTM3 , Glu-90 AtTTM3 , and Glu171 AtTTM3 from the acidic patch conserved among many TTMs (Fig. 3, B and E).
To assess the functional relevance of the observed metal cofactors in AtTTM3, we studied substrate binding and metal ion coordination in the evolutionary distant TTM ygiF from E. coli. YgiF shares ϳ18% sequence identity with AtTTM3 and has the same catalytic activity and substrate specificity (25,26). The structure of full-length ygiF was solved using iodide/sulfur SAD phasing and reveals the conserved TTM fold (r.m.s.d. with the AtTTM3 tunnel domain is 2.0 Å compared with 172 corresponding C ␣ atoms, and 0.65 Å comparing 75 corresponding C ␣ in the tunnel center) (Fig. 3C). ygiF contains a second, ␣-helical domain, which is connected to the TTM by a well ordered linker (residues 201-221, cyan in Fig. 3C) and which has been previously annotated as CHAD domain (8). The domain folds into two four-helix bundles of similar architecture and connectivity forming a V-shaped assembly (shown in red and orange in Fig. 3C). A structural homology search with the program DALI (42) returned no significant hits, and the molecular function of this domain, which is often found in bacterial metallophosphoesterases (8), remains to be elucidated.
We next solved structures of ygiF in complex with PPP i and in the presence of either MgCl 2 or MnCl 2 ( Table 2). The ygiF triphosphate substrate is bound in the same conformation as observed in AtTTM3, and we could identify two metal co-factors. One Mg 2ϩ /Mn 2ϩ ion is again found coordinated by tripolyphosphate and by Glu-160 ygiF , which corresponds to Glu-169 in AtTTM3 (position 1 in Fig. 3D). The second Mg 2ϩ /Mn 2ϩ 3 Å comparing 203 corresponding C ␣ atoms) reveals that the two orthophosphates in the product complex align with the terminal phosphates of the PPP i substrate. One of the three Mn 2ϩ ions (bound to site 1) is found consistently in both structures. C, structural superposition (r.m.s.d. is 2.2 Å comparing 138 corresponding C ␣ atoms) of ygiF bound to PPP i and to 2 Mn 2ϩ ions located in sites 1 and 2 with the RNA triphosphatase Cet1p (PDB ID 1D8H, in gray) reveals that the Cet1p manganese ion previously reported maps to metal binding site 2. D, comparisons of the ygiF-PPP i -Mn 2ϩ complex with the product-bound state of the adenylate cyclase cyaB (PDB ID 3N10, in gray) again reveals two conserved metal binding sites. The Mn 2ϩ ion bound to site 2 in the case of cyaB is coordinated by His-122.
ion is coordinated by three glutamate residues from the acidic patch in ygiF (Glu-4 ygiF , Glu-6 ygiF , Glu-162 ygiF ) and by a water molecule that is positioned close to a terminal phosphate of the tripolyphosphate substrate (position 2, Fig. 3, D and E). When we calculated phased anomalous difference maps from diffraction data collected near the Mn-K edge, we found that the ygiF metal binding site 1 shows a higher occupancy compared with the second site (peak heights are 125 and 45 , respectively), whereas we cannot detect difference density at the third metal position found in AtTTM3 (Fig. 3D). Taken together, our experiments define two consistent Mg 2ϩ /Mn 2ϩ coordination centers in plant and bacterial TTM tripolyphosphatases.
We compared our structures to other TTM enzymes for which metal ion-bound complexes have been reported. In the RNA triphosphatase Cet1p, a Mn 2ϩ ion is bound by the acidic patch, and its position corresponds to site 2 in ygiF and AtTTM3 (Figs. 4C and 3E) (11). In a bacterial TTM protein with adenylate cyclase activity, both positions 1 and 2 are occupied by Mn 2ϩ ions (Figs. 4D and 3E) (18). The acidic patch in mouse ThTPase (corresponding to site 2 in AtTTM3 and ygiF) again allows for Mg 2ϩ ion binding, as concluded from NMR titration experiments (20). These findings together indicate that many TTM proteins contain two metal ion binding sites, as previously proposed (40,41). It is likely that site 1 is generated by a triphosphate substrate-divalent metal ion complex binding to the tunnel center (21,23). Here, the octahedral coordination of the Mg 2ϩ /Mn 2ϩ ion is completed by an glutamate residue (Glu-169 AtTTM3 , Glu-160 ygiF ), which is invariant in all TTM proteins characterized thus far (Figs. 3D and 4). A second binding site is formed by three glutamate residues originating from an acidic patch conserved among many but not all TTM enzymes (Figs. 3, A, D,  and E, and 4) (4,12,24). This metal ion is positioned close to a terminal phosphate in our AtTTM3 and ygiF substrate complex structures (Fig. 3D) and may thus be involved in catalysis rather than in substrate binding.

Structures of Mammalian Thiamine Triphosphatase
Reveal the Location of the ␥-Phosphate-We investigated the contributions of the two metal ion centers to TTM substrate binding and catalysis. A conceptual problem with the analysis of our plant and bacterial tripolyphosphatases is that they catalyze the asymmetric cleavage of a symmetric substrate (Fig. 2) (25,26). It is thus difficult to assess in crystal structures, which terminal phosphate represents the ␥-phosphate that is being hydrolyzed (Fig. 3, A and B). We thus structurally characterized a mammalian TTM ThTPase, which was previously shown to specifically hydrolyze thiamine triphosphate (ThTP) into ThDP and P i (19 -21, 43). We synthesized ThTP from ThDP and produced co-crystal structures of mouse ThTPase with its substrate at pH 6, where ThTPase catalytic activity is minimal (20). Consequently, we found an intact ThTP molecule bound in the tunnel center of ThTPase (Fig. 5A). The thiamine portion of the substrate binds to a pocket generated by the tunnel walls and the C-terminal plug helix, with the thiazole ring making a stacking interaction with Trp-53 and with Met-195 from the plug helix (Fig.  5A). The ThTP triphosphate moiety binds in the same conformation as outlined for the PPP i -bound structures of AtTTM3 and ygiF above. Our substrate-bound mouse ThTPase structure supports an earlier docking model of human ThTPase (21). We next solved a crystal structure of mouse ThTPase in the presence of ThTP and Mg 2ϩ in a second crystal form grown at pH 9.0, where substrate hydrolysis can occur (20). Indeed, we found a product complex trapped in the active site of the enzyme, with a ThDP molecule and an orthophosphate located in the tunnel center (Fig.  5A). ThDP is coordinated by Arg-55 and Arg-57 in the substrate binding site but no longer allows for the coordination of a Mg 2ϩ / Mn 2ϩ ion in metal binding site 1, possibly because the missing ␥-phosphate would be required for Mg 2ϩ /Mn 2ϩ coordination (Fig. 5A). The ␥-phosphate in our structure apparently has been hydrolyzed, and the resulting P i has slightly moved away from the tunnel center (Fig. 5A). It is now found coordinated by Arg-125 and in direct contact with a Mn 2ϩ ion located in metal binding site FIGURE 5. TTM proteins use a two-metal catalytic mechanism. A, close-up view of the mouse ThTPase tunnel center with either the ThTP substrate bound (transparent gray) or the ThDP/P i products post-catalysis (in yellow, in bond representation). The thiamine part of ThTP is buried in a pocket close to the C-terminal plug helix of the TTM domain formed by Tyr-79 and Met-195; the thiazole ring makes a stacking interaction with Trp-53 (in yellow, in bond representation). The product P i is coordinated by a Mn 2ϩ ion bound to site 2 and by Arg-125. B, close-up of the ygiF active site (in yellow, in bond representation) bound to PPP i and two Mn 2ϩ ions (magenta spheres) in sites 1 and 2. The Mn 2ϩ ion in site 2 coordinates a water molecule (red sphere), which is well positioned to act as nucleophile. Structural superposition with a product-bound class IV adenylate cyclase (PDB ID 3N10) reveals the O3Ј of cAMP in the same position as the water molecule in ygiF. This position is also occupied by an oxygen atom of the product P i located in the AtTTM3 post-catalysis complex. C, the suggested mechanism for acidic-patch containing TTM proteins. The metal ion in site 1 coordinates the triphosphate moiety of the substrate to the tunnel center by interacting with a conserved Glu residue. Three additional glutamates form metal binding site 2, which coordinates and polarizes a water molecule to attack the ␥-phosphate of the substrate. Conserved basic residues in the tunnel center are involved in substrate binding and potentially stabilize the pyrophosphate leaving group.
2, reinforcing the notion that this metal ion may play a crucial role in catalysis (Fig. 5A).
TTM Proteins Use a Two-metal Catalytic Mechanism-The structural features surrounding metal binding site 2 in our structures allow proposing a unified catalytic mechanism for triphosphate tunnel metalloenzymes. In our PPP i -bound ygiF structure we find a water molecule coordinated by the second Mn 2ϩ ion, which is in an ideal position to serve as the activated nucleophile (Fig. 5, B and C). Indeed, structural superposition with a bacterial adenylate cyclase (18) reveals that its cAMP O3Ј group, which acts as the nucleophile in the cyclization reaction, is located in the same position as the water molecule in our ygiF structure (Fig. 5B). Consistently, this position also is occupied by an oxygen atom of a product orthophosphate, which we located in our AtTTM3 postcatalysis structure (see above, Figs. 3B and 5B). Based on these findings, we propose that metal binding site 1 is required for proper substrate coordination in TTM proteins (21,23), and metal ion 2 activates a water molecule to allow for a nucleophilic attack on the triphosphate substrate (Fig. 5C). This would rationalize why the glutamate residues from the acidic patch, which are involved in the coordination of the second metal ion, are well conserved among TTM proteins (Figs. 3E and 4). The basic residues in the tunnel center appear to be mainly involved in substrate coordination ( Figs. 2A and 3A); however, the invariant Arg-56 ygiF (Arg-52 AtTTM3 ) possibly activates the substrate for the nucleophilic attack by forming a hydrogen bond with the oxygen atom connecting the ␤and ␥-phosphate of the substrate (Fig. 5, B and C). The suggested TTM reaction mechanism is reminiscent of the one found in mammalian type V adenylate cyclases (44) and in nucleic acid polymerases (45,46) as previously speculated (8).
We next performed a mutational analysis of substrate-and metal co-factor-interacting residues in AtTTM3 and ygiF (Fig.  6). Mutation of Glu-2 AtTTM3 , Glu-4 AtTTM3 , and Glu-169 AtTTM3 to Asp or Ala strongly reduces the PPP i hydrolysis of the plant enzyme (Fig. 6, A and B; see Fig. 7 for mutant protein stability). Consistently, changing the corresponding Glu-6 ygiF and Glu-160 ygiF to Ala impairs the enzymatic activity of ygiF, suggesting that the proper arrangement of metal binding sites 1 and 2 in the tunnel center is essential for catalysis (Fig. 6, A-C). Mutation of Arg-52 AtTTM3 or Arg-56 ygiF , but not of the neighboring Arg-54 AtTTM3 , to Ala again strongly inhibits catalysis, highlighting its potential role as proton donor (Fig. 6, A-C, see above). In addition, we find moderately reduced enzymatic activities upon the mutation of Lys-76 AtTTM3 to Leu or Ala (Fig.  6, A and B). This residue and the corresponding Lys-69 ygiF appear to be involved in substrate binding and orient Glu-2 AtTTM3 /Glu-4 ygiF to coordinate metal ion 2 (Fig. 6, A and B). Glu-2 AtTTM3 is also in contact with Glu-90 AtTTM3 , mutation of which to Ala again reduces the enzymatic activity of the Arabidopsis enzyme (Fig. 6, A and B). Taken together, our and previous mutational analyses (22,24,26) consistently suggest that the proper establishment of two metal centers appears to be critical for catalysis in TTM tripolyphosphatases and other TTM proteins (40,41). The basic residues, with the exception of the catalytic Arg-52 AtTTM3 appear to be mainly involved in substrate coordination in the tunnel center (Fig. 6, A-C). Further experimentation will be required to rationalize the specific effects of certain point mutations on substrate/metal co-factor binding or catalysis itself. Comparison with Vtc4p suggests that the first but not the second metal binding site are present in the yeast polyphosphate polymerase (Figs. 2, A and B, and 4A) (4). Importantly, mutation of Lys-200 AtTTM3 , which corresponds to the catalytic Lys-458 Vtc4p , to either Leu or Ala has little effect on tripolyphosphatase activity of the plant enzyme (Fig. 6, A and B) (4). This suggests that there are significant mechanistic differences between TTM polyphoshatases and polyphosphate polymerases despite their strong structural homology (Figs. 2, A and B,  and 3A).
Directionality of Substrate Binding Defines TTM Catalytic Activity-To better understand how acidic-patch containing TTMs are able to carry out very different enzyme reactions, we superimposed our substrate-bound structures of AtTTM3, ygiF, and mouse ThTPase with an ATP-analog complex of a bacterial TTM adenylate cyclase (18) (Fig. 8). We found that although the triphosphate parts of all ligands closely align in the tunnel center, their "tail" moieties can enter the tunnel domain from opposite sites in different enzymes (Fig. 8). The unique modes of substrate binding in ThTPases and adenylate cyclases allow these enzymes to carry out rather different reactions and to produce different leaving groups (ortho-and pyrophosphate, respectively) while maintaining a unified cleavage site in close proximity of metal binding site 2. Both the N-and C-terminal sides of the tunnel domain have evolved to recognize specific substrates, as exemplified by our ThTPase structure and by the class IV adenylate cyclase complexes (Figs. 5, A and B, and 8) (18). The observed substrate binding mode in mouse ThTPase reinforces the notion that to bind their substrates, TTMs require opening of their closed tunnel domains into a cup-shaped hand, as previously shown by NMR spectroscopy (20).
Members of the ancient triphosphate tunnel metalloenzyme family can be found in all kingdoms of life. Our comparative analysis defines that most of these enzymes share a common catalytic mechanism in their tunnel centers, yet they have evolved different substrate recognition modes on their tunnel entries. The observed substrate plasticity apparently allows TTM proteins to act on a wide array of enzyme substrates and to perform very different reactions, many of which likely remain to be discovered.
Author Contributions-J. M. and M. H. designed the study and wrote the paper. V. T. performed NMR titrations, J. M. performed all other experiments. All authors analyzed data and approved the final version of the manuscript.