Crystal Structure of Bacterial Inorganic Polyphosphate/ATP-glucomannokinase

Inorganic polyphosphate (poly(P)) is a biological high energy compound presumed to be an ancient energy carrier preceding ATP. Several poly(P)-dependent kinases that use poly(P) as a phosphoryl donor are known to function in bacteria, but crystal structures of these kinases have not been solved. Here we present the crystal structure of bacterial poly(P)/ATP-glucomannokinase, belonging to Gram-positive bacterial glucokinase, complexed with 1 glucose molecule and 2 phosphate molecules at 1.8 Å resolution, being the first among poly(P)-dependent kinases and bacterial glucokinases. The poly(P)/ATP-glucomannokinase structure enabled us to understand the structural relationship of bacterial glucokinase to eucaryotic hexokinase and ADP-glucokinase, which has remained a matter of debate. These comparisons also enabled us to propose putative binding sites for phosphoryl groups for ATP and especially for poly(P) and to obtain insights into the evolution of kinase, particularly from primordial poly(P)-specific to ubiquitous ATP-specific proteins.

rylation of glucose, the first step of glycolysis, and the last catalyzes that of NAD, the last step in NADP biosynthesis. Note that poly(P)-GK is poly(P)-specific, although poly(P)/ATPtype enzymes use poly(P) and ATP. Although the primary structures of these poly(P)-dependent kinases were determined (1)(2)(3)(4), the lack of a crystal structure has prevented us from clarifying the poly(P)-utilizing mechanism of these unique kinases. Poly(P)-dependent kinases have ATP-specific partners, and a knowledge of the crystal structure of poly(P)-dependent kinase may aid in understanding the structural relationship of poly(P)-dependent kinase to the ubiquitous ATP-specific kinase, i.e. to understand structural determinants enabling poly(P)/ATP-type and poly(P)-specific type kinases to use poly(P), ATP-specific kinase to reject poly(P) and poly(P)-specific kinase to reject ATP. Such understanding would lend insights into the evolutionary relationship of poly(P)-dependent kinase to ubiquitous ATP-specific kinase, where relationships are also of interest, since poly(P) could be formed and participate in ATP synthesis under ancient prebiotic conditions and serve as a possible ancient energy carrier preceding ATP (5).
Among poly(P)-dependent kinases, GMK, isolated from the Gram-positive bacterium Arthrobacter sp. strain KM, is a monomer with 30 kDa that phosphorylates glucose and mannose with preference for glucose through the use of poly(P) and ATP (3). The primary structure of GMK shows high homology with those of poly(P)-and poly(P)/ATP-GKs and other Gram-positive bacterial GKs but little with those of Gram-negative bacterial GKs and eucaryotic hexokinases (HKs), except for a few conserved motifs (3,6). HK shows a broad specificity for hexose, whereas GK has high specificity for glucose (7,8). Although crystal structures of eucaryotic HKs from yeast (9 -13), human (14 -19), rat (20), and parasite Schistosoma mansoni (20) have been solved, no crystal structure has been reported for Grampositive or negative bacterial GKs. Hence, the crystal structural and evolutionary relationships between bacterial GK and eucaryotic HK have remained matters of debate (7,8). Crystal structures of archaeal ADP-GKs from Thermococcus litoralis (21), Pyrococcus horikoshii (22), and Pyrococcus furiosus (23) definitely show that ADP-GK is distinguished from eucaryotic HK. ADP-GKs use ADP, whereas eucaryotic HK and bacterial GK, except for poly(P)-dependent GK, are regarded as using ATP but not poly(P).
Here we show the crystal structure of GMK complexed with one glucose and two phosphate (PO 4 2Ϫ ) molecules, the first among poly(P)-dependent kinases and bacterial GKs. Crystal structural analysis for GMK provides evidence for close crystal structural and evolutionary relationships between bacterial GK and eucaryotic HK and details of substrate-binding sites, especially for putative poly(P)-binding sites, enabling us to obtain insights into the evolution of kinases, particularly from primordial poly(P)-specific to ubiquitous ATP-specific proteins.

EXPERIMENTAL PROCEDURES
Purification, Crystallization, and X-ray Diffraction-GMK was expressed from the GMK gene at NdeI-XhoI sites of pET-21b (Novagen) as a recombinant-enzyme in Escherichia coli BL21(DE3) cells (Novagen) in LB medium by inducing isopropyl-␤-D-thiogalactopyranoside (0.4 mM) as described elsewhere (4) but at 20°C for 24 h with sufficient aeration. The expressed recombinant GMK was purified using the same procedure as for native-GMK (3), except that, after ammonium sulfate precipitation, the supernatant was directly applied to a butyl Toyopearl (Tosoh) column and eluted using the gradient in ammonium sulfate from 30 to 0%. After dialysis, eluted GMK was purified by a hydroxyl apatite (Nacalai Tesque) column using the gradient in potassium phosphate from 10 to 600 mM. Active fractions contained the purified GMK and were homogeneous on SDS-PAGE (24). GMK was crystallized by hanging-drop vapor diffusion method as described elsewhere (25). The solution of a crystallization drop was prepared on a siliconized coverslip by mixing 3 l of protein solution (10 mg/ml GMK and 10 mM glucose in 10 mM potassium phosphate, pH 7.0) with 3 l of mother liquor composed of 2.0 M ammonium sulfate, 2% (v/v) polyethylene glycol 400, 0.1 M Hepes, pH 8.0.
Crystals were soaked in several heavy atom derivative solutions composed of 1 mM UO 2 Ac 2 , 10 mM AgNO 3 , 10 mM GdCl 3 , 1 mM HgCl 2 , 10 mM K 2 Pt(CN) 4 , 2 mM TmCl 2 , and 1 mM EuCl 3 for 20 -60 min at 20°C. Derivative solutions were prepared in a mother liquor, but 0.1 M Tris-HCl, pH 7.0 (for UO 2 Ac 2 ) and 0.1 M sodium acetate, pH 4.9 (for the remainder), were included instead of 0.1 M Hepes, pH 8.0. Diffraction data for the native crystal of GMK up to 2.8 Å and derivative crystals at about 3.0 Å were collected with a Bruker Hi-Star multiwire area detector at 20°C using CuK ␣ radiation generated by a MAC Science M18XHF rotating anode generator and were processed with SADIE and SAINT software packages (Bruker). Diffraction data for further refinement up to 1.8 Å was collected using synchrotron radiation at a wavelength of 0.9 Å at the BL-41XU station of SPring-8 in Hyogo, Japan. Obtained data was processed, merged, and scaled using program package HKL 2000 (DENZO and SCALEPACK) (26) and truncated with the CCP4 program package.
Structure Determination and Refinement-The crystal structure of GMK was determined by multiple isomorphous replacement (MIR). Major sites of heavy atoms were determined by interpretation of difference Patterson maps calculated at 3.0 Å of resolution. Additional heavy atom sites were determined from difference Fourier maps. The phase refinement was performed with program PHASES (27). Results of heavy atom refinement and phasing by MIR at 3.0 Å of resolution are listed in Table I. An electron density map was made with the solventflattened (28) MIR phase using data from 15.0 to 3.0 Å, and the map was averaged using noncrystallographic symmetry. The initial model built using an averaged MIR map at 3.0 Å of resolution was refined by stimulated annealing with molecular dynamics using X-PLOR (29) and CNS (30). Then, several rounds of restrained least squares refinement followed by manual model building were conducted to improve the initial model to the R-factor of 19.2% (free R-factor ϭ 26.2%) at 2.8 Å of resolution using the TURBO-FRODO (AFMB-CNRS) program on a Silicon Graphics Octan computer. The model was further refined by using diffraction data up to 1.8 Å with simulated annealing with mo-lecular dynamics using a CNS program package. Several rounds of restrained least squares refinement to a resolution of 1.8 Å followed by manual model building were conducted. Water molecules were incorporated when the difference in density was more than 3.0 above the mean and the 2 F o Ϫ F c map showed a density of more than 1.0 . The final model was determined with an R-factor of 19.1% (free R-factor ϭ 22.0%) at 50.0 -1.8 Å resolution.
The stereochemical quality of the model was assessed using PRO-CHECK (31) and WHAT-CHECK (32) programs. Molecular models were prepared using MOLSCRIPT (33) and Raster3D (34) programs. Homology for the crystal structure was searched for in DALI (www. ebi.ac.uk/dali). Alignment of the primary structure was constructed by ClustalW (35). SCOP (36) was used to classify the crystal structure. Coordinates of crystal structures were taken from the Protein Data Bank (www.rcsb.org). Molecular models were superimposed by a fitting program implemented in TURBO-FRODO (Bio-Graphics). Coordinates and molecular topologies of pentapolyphosphate were generated using PRODRG (37).

Structure Determination and Quality of Refined Model-
Recombinant-GMK purified to homogeneity showed the same molecular structure (30 kDa, monomer) and N-terminal sequence and almost the same kinetic parameters as those of native-GMK (3). GMK was crystallized in the presence of glucose, phosphate, and sulfate, and prismatic colorless crystals of the GMK grew to a maximum 0.1 mm in 2 weeks at 20°C. The GMK crystal belongs to space group P2 1 2 1 2 1 with unit cell dimensions of a ϭ 65.21, b ϭ 82.52, and c ϭ 102.06 Å, and the solvent content was 49%, assuming 2 molecules in an asymmetric unit (25). The structure of GMK was determined by   Tables I and II. The electron densities of the main and side chains were generally well defined on the 2 F o Ϫ F c map. The final Rfactor was 19.1% (free R-factor ϭ 22.0%) for 49,926 reflections with F Ͼ 2.0 (F) at 50 -1.8 Å of resolution. Interpretable electron density for GMK begins at residue 11 and ends at residue 263 for 1 molecule of GMK. N-terminal sequence analysis of recombinant-GMK showed that the first residue is missing, and the DNA sequence of the GMK gene exhibits a FIG. 2. Substrate-binding sites. a, electron density map of glucose bound to GMK. Glucose oxygens are numbered. b, binding sites of GMK with glucose (red), phosphate A (orange), and phosphate B (yellow). Hydrogen bonds are indicated by dotted lines. c, binding sites of C-HKI with glucose (red) and ADP (orange) (PDB accession code 1DGK) (18). Interactions of residues with the ␤-phosphoryl group are not indicated since this group was not located at the proper site (18). polypeptide of 267 residues of GMK. Electron density is, thus, present for all but the first 10 and last 4 residues of the polypeptide chain. Based on theoretical curves in the plot calculated by Luzzati (38), absolute positional error was estimated to be close to 0.19 Å for the structure refined with data between 5.0 and 1.8 Å of resolution. The result of Ramachandran plot analysis (39) showed that most of the nonglycine residues (93.9%) lie within most favored regions, and other residues (6.1%) fell within additional and generously allowed regions. There was one cis-peptide between Gln 47 and Pro 48 residues.
Overall Structure of GMK Complexed with Glucose and Phosphates-The refined model consisted of 506 residues, 373 water, 2 glucose, and 4 phosphate molecules per 2 molecules of GMK (GMK-A/-B) in an asymmetric unit. Note that it was unclear if phosphate or sulfate was contained in GMK due to the similar molecular structures of the 2 anions and due to the presence of the 2 anions in the crystallization solution. To avoid confusion, we regard anions as phosphate A and B in this article.
A ribbon model of the overall crystal structure of GMK together with one bound glucose and two bound phosphates (phosphate A and B) in the interdomain cleft is shown in Fig.  1a, and the topology of secondary structure elements of GMK is shown in Fig. 1b. One molecule of GMK, which is active as a monomer (3), consisted of two globular domains. The two domains are called N and C domains and are connected by two segments between SA5 and H3 and between SB5 and H9 (Fig.  1, a and b). Each domain consisted of three layers (␣/␤/␣). The N domain consisted of H1:H2/SA3:SA2:SA1:SA4:SA5/H9, and the C domain of H3/SB3:SB2:SB1:SB4:SB5/H4:H5:H6:H7:H8. The 5-stranded mixed ␤-sheets in N and C domains showed order 32145 with strand 2 antiparallel to the rest of the sheet. The overall topology of the N domain was similar to that of the C domain (Fig. 1b). The primary structure of the N domain (residues 11-122) also showed homology with that of the C domain (residues 123-263), 18% identity, and 48% similarity over 116 residues, as calculated by ClustalW. The superimposition of the N domain on the C domain gave the root mean square deviation of 1.1 Å with 41 C␣ within 2.0 Å (Fig. 1c).
Taken together, we propose that the two domains stem from the duplication of one primordial domain.
From overall crystal structural features, GMK was classified into the "ASKHA (acetate and sugar kinase/hsc70/actin)" superfamily on classification by Cheek et al. (40) or the "actin-like ATPase domain" superfamily on SCOP (36). Both superfamilies are characterized by duplicate domains of a "ribonuclease Hlike"-fold, where the fold consists of 3 layers (␣/␤/␣) and the ␤-sheet has order 32145 as described above. Both superfamilies also contain eucaryotic HKs (40). Accordingly, homology analysis using the crystal structure as a query on DALI showed that GMK resembled eucaryotic HKs such as human HK I (PDB accession code 1QHA) and yeast HK PII (PDB accession code 2YHX).
Glucose-and Phosphates-binding Sites of GMK-Ligands (glucose and phosphates) were refined with fractional occupancies of 1.0. The average B-factor for glucose was 13.4 Å 2 and lower than the average for all atoms (24.6 Å 2 ), but B-factors for phosphate-A (46.3 Å 2 ), especially for phosphate-B (64.3 Å 2 ), were higher than the average for all atoms. The electron density of the glucose O1 atom was observed for the ␤-anomer configuration (Fig. 2a). Selected contacts between GMK and ligands are listed in Table III. The O1 atom of phosphate B interacted with the Arg 42 NH 2 atom in GMK-A but with water 466 in GMK-B due to different directions in the side chain of Arg 42 between GMK-A and GMK-B (data not shown). The binding sites of GMK with bound glucose, phosphate A, and phosphate B and their interactions with GMK are shown in Fig. 2b. DISCUSSION The crystal structure of GMK presented here is the first among bacterial GKs and poly(P)-dependent kinases, enabling us to understand the structural relationship of bacterial GK to eucaryotic HK and ADP-GK by comparing overall structures of enzymes and their substrate-binding sites. These comparisons also led us to propose putative binding sites of GMK for phosphoryl groups of ATP and for poly(P) and, finally, to obtain insights into the evolution of kinase as described below. Note that bacterial GK consists of Gram-positive and -negative bacterial GKs (6), where GMK belongs to the former (3), but no crystal structure of Gram-negative bacterial GK is known. Hence, we regard GMK as representative of bacterial GK.

Overall Structures of GMK, Eucaryotic HK, and ADP-GK-
The structural relationship of bacterial GK to eucaryotic HK and ADP-GK has not been understood. As mentioned above, however, the crystal structure of GMK appears to be related, at least to those of eucaryotic HK. We, therefore, attempted to compare GMK and eucaryotic HK in detail.
Eucaryotic HKs consist of isozymes (e.g. mammalian (human) HK I-VI) (7,8). HK I (100 kDa) consists of N-and C-terminal halves (50 kDa), which have primary and tertiary structures, and substrate-binding sites, resembling each other and those of other 50 kDa eucaryotic HKs (human HK IV, also called GK, S. mansoni HK, and yeast HK PI and PII) (9 -20). The C-terminal half of human HK I (C-HKI) has catalytic activity, but the N-terminal half is inactive (14 -18). The tertiary structure of C-HKI (residues 525-913) complexed with glucose and ADP (18) was chosen as representative of eucaryotic HK for comparison with GMK. The tertiary structure of C-HKI consists of two domains, called large and small, between which glucose-and ADP-binding sites are formed, and shows "closed conformation" observed in eucaryotic HKs complexed with glucose (9 -20) (Fig. 3a).
The secondary structure of GMK was homologous with that of C-HKI ( Fig. 1b and Fig. 3b). The tertiary structure of GMK also appeared to be similar to that of C-HKI ( Fig. 1a and Fig.  3a) and, accordingly, could be superimposed on that of C-HKI, giving the root mean square deviation of 1.35 Å with 132 C␣ within 2.0 Å, especially the N domain of GMK on the small domain of C-HKI (Fig. 4). The primary structure of GMK corresponds to that of C-HKI (Fig. 5) when the two sequences are aligned based on superimposition of their substrate-binding sites (presented below), not as aligned by ClustalW. The primary structure of GMK is similar to that of C-HKI in this alignment (53% similarity, 16% identity over 253 residues), but several additional elements were found in C-HKI, suggesting that C-HKI evolved by acquiring elements from a GMK-like ancestor. The additional element corresponding to residues 753-818 in C-HKI was most apparent and corresponded approximately to the region previously designated a "flexible sub- domain" (residues 766 -811) (15). The flexible subdomain contains adenine binding residues (see below) (18), is found in other crystal structures of eucaryotic HKs (9 -20), and appears to be involved in allosteric regulation of HK I by glucose 6-phosphate and phosphate through interaction with the N-terminal half (18,41). Collectively, the close relationship of the overall tertiary structure of GMK to that of C-HKI, namely eucaryotic HK, was demonstrated, but the overall tertiary structure of GMK apparently differs from those of ADP-GKs (21-23) (Fig.  1a and Fig. 6).
Glucose-binding Sites of GMK, Eucaryotic HK, and ADP-GK-In C-HKI, residues and their atoms interacting with each of the oxygen atoms of glucose were comparable with those of GMK (17) ( Table IV). The residues of GMK, Asn 122 , Asp 123 , Glu 168 , and Glu 180 and their atoms corresponded well to those of C-HKI, Asn 656 , Asp 657 , Glu 708 , and Glu 742 in tertiary structure (Fig. 2, b and c) and primary structure (Fig. 5). Among   Fig. 1b and Fig. 3b. Similarity was calculated by ClustalW using the primary structure of C-HKI, from which all additional elements in the structure of C-HKI were removed, and specified as dots and plus symbols (lower and higher similarities). Residues specified in the text are boxed in brown. Flexible subdomains (residues 766 -811) in C-HKI are boxed in blue.
these residues in C-HKI, the significance of Asp 657 , Glu 708 , and Glu 742 has been confirmed by site-directed mutagenesis (42,43), and Asp 657 (corresponding to Asp 123 in GMK) is proposed as a catalytic base (42). The glucose-binding site of GMK, thus, resembled that of C-HKI, although glucose bound to GMK as a ␤-anomer (Fig. 2, a and b) and to C-HKI as an ␣-anomer (17) (Fig. 2c). The glucose-binding site of GMK showed no homology to that of P. furiosus ADP-GK (23) except for a similar spatial arrangement of Asp 440 and Glu 88 in ADP-GK to Asp 123 and Glu 180 in GMK (data not shown). Collectively, the glucosebinding sites of eucaryotic HKs resembled that of GMK, but not that of ADP-GK, except for Asp 440 (putative catalytic base) and Glu 88 in ADP-GK (23).
Putative Binding Sites of GMK for Poly(P) and for Phosphoryl Groups of ATP-Phosphate A in GMK appeared to occupy a site near the ␣-phosphoryl group of ADP in a tertiary structural comparison (Fig. 2, b and c), whereas phosphate B is located slightly apart from phosphate A (Fig. 2b). We expected that phosphate A and B binding gave us information concerning the binding sites for poly(P) and ATP. Although the possibility remains that the phosphates may be sulfates, we assume that sulfate binding can also represent phosphate binding, since sulfate can bind to the phosphate-binding site of yeast HK PII (13). Two phosphoryl groups in pentapolyphosphate consisting of five phosphates could be superimposed on phosphate A and B such that the distance between the phosphorus atom in the terminal phosphoryl group of pentapolyphosphate and the oxygen atom of the 6-hydroxyl group of glucose was 3.0 Å (Fig.  7a), which is a suitable distance for phosphoryl transfer (18). Thus, we tentatively propose that phosphate A-and B-binding sites in GMK represent a pentapolyphosphate-binding site. Accordingly, poly(P), which consists of up to thousands of phosphates (5), could enter into the putative pentapolyphosphatebinding site, since this site is located at the side of the interdomain cleft (Fig. 7b).
Our pentapolyphosphate binding model also suggested that phosphate A is located at the binding site for the ␤-phosphoryl group of ATP (Fig. 7a). Note the ␤-phosphoryl group in C-HKI complexed with ADP is not located at the proper site (18) (Fig.  2c). In C-HKI the ␤-phosphoryl group is proposed to interact with N and OG1 atoms of Thr 680 (18). Thr 680 corresponded to Thr 151 in GMK in primary and tertiary structures ( Fig. 5 and Fig. 2, b and c), and N and OG1 atoms of Thr 151 in GMK interacted with phosphate A (Table III). Furthermore, Asp 532 in C-HKI is proposed to interact with ␤and ␥-phosphoryl groups via Mg 2ϩ and water (18) and also corresponds to Asp 18 in GMK (Fig. 5 and Fig. 2, b and c). Taken together, these data suggest that phosphate A is located at the binding site for the ␤-phosphoryl group of ATP, and that the phosphoryl groupbinding site of GMK is homologous with that of C-HKI. Our model of binding for phosphoryl groups of poly(P) and ATP  further indicates that phosphorylation sites for ATP and poly(P) are shared, in agreement with previous results for a competition plot for GMK (3). Residues in GMK, Thr 22 and Lys 25 , interacting with phosphate B (Fig. 2b) were conserved in the primary structures of C-HKI as Thr 536 and Arg 539 (Fig. 5) and in other eucaryotic HKs (13), Gram-negative and -positive bacterial GKs including poly(P)-and poly(P)/ATP-GKs (1,3,6), and even some proteins in the ASKHA superfamily (40,44). The sequence around residues highly conserved in these proteins is called a "phosphate-1" motif (44). The phosphate-1 motif is regarded as conserved in ATP-utilizing proteins in the ASKHA superfamily such as acetate and sugar kinases, heat shock protein, and actin, and plays a significant role in interaction with ATP (40,44). Thr 536 , Arg 539 , and Asp 532 in this motif of C-HKI are proposed to interact with phosphoryl groups of ATP via Mg 2ϩ and water (18). Note that all crystal structure-solved proteins containing this motif except for GMK are ATP-specific, and we could find no crystal structures of these ATP-specific proteins showing the direct interaction of this motif with phosphate, suggesting that some structural elements in these ATP-specific proteins, not existing in GMK, possibly prevent this motif from binding to poly(P).
Where does adenine-ribose (adenosine) of ATP bind to GMK? Gly 747 in C-HKI, interacting with ribose of ADP, appears to correspond slightly to Ala 185 in GMK in primary structural alignment (Fig. 5). Thr 784 , Lys 785 , Ser 788 , and Thr 863 (interacting with adenine of ADP) in C-HKI were not conserved in GMK, however (Fig. 5). Note that Thr 784 , Lys 785 , and Ser 788 in C-HKI are located in the flexible subdomain not found in GMK. Although the primary structure of poly(P)-GK (1) was highly conserved with those of GMK and poly(P)/ATP-GK (2), poly(P)-GK could not use ATP. Identifying the correct adenosine-binding site of GMK would aid in understanding why poly(P)-GK rejects ATP. To specify adenine-and poly(P)-binding sites in GMK, a crystallographic study of GMK complexed with poly(P) or ATP is currently in progress.
Insights into Kinase Evolution-The GMK crystal structure presented here led us to assume evolutionary relationships of bacterial GK with eucaryotic HK and archaeal ADP-GK. Tertiary structural and evolutionary relationships between bacterial GK and eucaryotic HK are currently matters of debate (7,8), whereas relationships between eucaryotic HK and archaeal ADP-GK are distinctly diverse (7,8,(21)(22)(23). Our study showed the close structural relationship between bacterial GK and eucaryotic HK from the points of overall tertiary structure (Fig.  4), binding sites for substrates (glucose and phosphoryl group) (Fig. 2, b and c), and primary structure (Fig. 5), and hence, we propose that eucaryotic HK and bacterial GK diverged from a common poly(P)-specific primordial protein as described below. The structural relationship between bacterial GK and ADP-GK is apparently diverse (Fig. 1a and Fig. 6), and hence, the evolutionary relationship would also be diverse, agreeing with the accepted idea that the structural and evolutionary relationship between eucaryotic HK and archaeal ADP-GK is distinctly diverse (7,8,(21)(22)(23). It appeared that the common ancestor of bacterial GK and eucaryotic HK evolved into eucaryotic HK by acquiring several elements, including a flexible subdomain (Fig. 5), to satisfy physiological criteria required in eucaryotes (organisms containing mitochondria), whereas ancestors evolved into bacterial GK without such acquisitions. Eucaryotic HK appears to evolve to be suitable for organisms containing mitochondria, since (i) eucaryotic amitochondriate protist (without mitochondria), Trichomonas vaginalis, carries only Gram-negative bacterial GK (6), not eucaryotic HK, and (ii) human HK I interacts with mitochondria, permitting direct exchange of adenine nucleotides between the mitochondrial matrix and the HK active site (45,46).
Crystal structural and evolutionary relationships of poly(P)dependent kinase with ubiquitous ATP-specific kinase are also interesting and significant but remain to be clarified due to the lack of crystal structural information on poly(P)-dependent kinase. Here we present the crystal structure of this type of kinase and propose the putative poly(P)-binding site of GMK, which contains the phosphate-1 motif well conserved in ATPspecific proteins. Taking into account that (i) adenine-interacting residues in C-HKI are located in the flexible subdomain not found in GMK (Fig. 5), (ii) our proposition that eucaryotic HK acquired the flexible subdomain, interacting with adenine, during evolution, and finally (iii), GMK takes a fold similar to those of ATP-specific ASKHA proteins containing the phosphate-1 motif, we propose that ATP-specific proteins containing the phosphate-1 motif may have evolved from a primordial poly(P)-specific GMK-like protein and may acquire the ability to use ATP and lose that to use poly(P) during evolution, which agrees with the suggestion that poly(P) is an energy carrier preceding ATP (5).