Human OLA1 Defines an ATPase Subfamily in the Obg Family of GTP-binding Proteins*

Purine nucleotide-binding proteins build the large family of P-loop GTPases and related ATPases, which perform essential functions in all kingdoms of life. The Obg family comprises a group of ancient GTPases belonging to the TRAFAC (for translation factors) class and can be subdivided into several distinct protein subfamilies. The founding member of one of these subfamilies is the bacterial P-loop NTPase YchF, which had so far been assumed to act as GTPase. We have biochemically characterized the human homologue of YchF and found that it binds and hydrolyzes ATP more efficiently than GTP. For this reason, we have termed the protein hOLA1, for human Obg-like ATPase 1. Further biochemical characterization of YchF proteins from different species revealed that ATPase activity is a general but previously missed feature of the YchF subfamily of Obg-like GTPases. To explain ATP specificity of hOLA1, we have solved the x-ray structure of hOLA1 bound to the nonhydrolyzable ATP analogue AMPPCP. Our structural data help to explain the altered nucleotide specificity of YchF homologues and identify the Ola1/YchF subfamily of the Obg-related NTPases as an exceptional example of a single protein subfamily, which has evolved altered nucleotide specificity within a distinct protein family of GTPases.

In all organisms, P-loop NTPases play a pivotal role in the regulation of diverse cellular processes such as protein translation, intracellular transport, signal transduction, cell motility, cell division, growth, and others. Most P-loop GTPases act as molecular switches such that the GTP-bound form constitutes the active state and triggers the biological output whereas the GDP-bound form is inactive (for review see Refs. [1][2][3]. The guanine nucleotide-binding domain (G domain) is about 20 kDa in size and structurally conserved. The G domain adopts an ␣,␤ fold, typical for many nucleotide-binding proteins, formed by six central ␤-strands surrounded by five ␣-helices. At the sequence level, the G domain contains five characteristic sequence motifs, termed G1-G5, involved in nucleotide binding and hydrolysis (2). The G1/Walker A motif (GXXXXGK(S/ T)), also referred to as P-loop (4), helps to position the triphosphate moiety of the bound nucleotide. The G2 (X(T/S)X) and G3/Walker B (hhhDXXG) motifs are involved in the coordination of a Mg 2ϩ ion that is required for nucleotide binding and hydrolysis. Specificity in nucleotide binding is conferred by the G4 motif, which has a (N/T)KXD signature in guanine nucleotide binding P-loop NTPases. The G5 motif ((T/G)(C/S)A) supports guanine base recognition (2).
The conformational switch between the GDP-and GTPbound forms of P-loop GTPases manifests itself in two peptide segments defined as switch I and switch II. The switch I region, a loop preceding strand ␤2, has also been termed the effector loop because it exerts downstream functions in many GTPases by binding to effector molecules. The switch I and II regions interact with the ␥-phosphate of the bound GTP, resulting in a protein conformation that is highly responsive to GTP hydrolysis and loss of the ␥-phosphate. Upon GTP hydrolysis, the molecule undergoes a conformational change, and the switch domains move into a different position and conformation, a mechanism that has been compared with the relaxation of a loaded spring (3). P-loop GTPases can be classified into TRAFAC (translation factor-related) and SIMIBI (signal recognition particle, MinD, and BioD) NTPases (5). Based on conserved sequence motifs and structural features, these two classes are further split into more than 20 distinct families and about 60 subfamilies (5).
YchF is one of the highly conserved GTPases (Escherichia coli to Homo sapiens: 45% identity, 62% similarity) belonging to the Obg-related GTPase family. The Obg protein family is thought to comprise five ancient subfamilies, namely Obg/ CgtA, YchF/YyaF, Drg/Rbg, Nog1, and Ygr210 (5). Functionally, the Obg-like GTPases are poorly characterized. The beststudied member is the essential Obg/CgtA (SpoOB-associated GTP-binding protein) from Bacillus subtilis (11). Obg and its homologues have been implicated to function in cellular processes as diverse as sporulation, stress response, control of DNA replication, and ribosome assembly (for review see 12).
In contrast to Obg, YchF/YyaF has been hardly investigated, and functional information on YchF homologues is scarce. In Brucella melitensis, the YchF homologue DugA has been suggested to be involved in iron metabolism and may be a regulator of the Ton system (13). Structural analysis of yeast and bacterial YchF revealed a modular organization formed by three domains, a central G domain, flanked by a coiled-coil domain and a TGS (ThrRS, GTPase, SpoT) domain of unknown function (14).
One anomaly in YchF and its homologues is the unusual G4 motif, which strongly diverges from the (N/T)KXD consensus, suggesting that guanine nucleotide binding specificity of YchF homologues might be either governed by other means or altered. We have biochemically characterized the human homologue of YchF and found that it binds and hydrolyzes ATP more efficiently than GTP. For this reason, we have termed the protein hOLA1, for human Obg-like ATPase 1. To explain ATP specificity of hOLA1, we have solved the x-ray structure of hOLA1 bound to the nonhydrolyzable ATP analogue AMP-PCP. 5 Our structural data help to explain the altered nucleotide specificity of YchF homologues. Biochemical characterization of YchF proteins from different species revealed that ATPase activity is a general but previously missed feature of the YchF subfamily of Obg-like GTPase.

EXPERIMENTAL PROCEDURES
Phylogenetic Tree Construction and Sequence Alignment-Phylogenetic distribution and multiple sequence alignment of Obg-like proteins were calculated with ClustalW (www.ebi-.ac.uk/clustalw) and displayed using TreeView (15) or JalView.
Cloning and Site-directed Mutagenesis-For hOLA1 cloning, the human gene PTD004 coding region was amplified by PCR using cDNA from HeLa cells. The PCR product was digested with NcoI and BamHI and inserted into the pQE60 protein expression vector (Qiagen) to allow for the expression of a C-terminally hexahistidine-tagged protein. The coding regions of S. cerevisiae Ybr025c (yOla1p) was amplified by PCR from yeast genomic DNA. The PCR product was digested with NcoI/ BamHI and inserted into pQE60. Point mutations were generated using the QuikChange site-directed mutagenesis kit (Stratagene) according to the  manufacturer's protocol. DNA oligonucleotides used for PCRs  and mutagenesis are listed in supplemental Table S1.
Fluorescence Spectrometry-Nucleotide binding of 2Ј/3Ј-O-(N-methylanthraniloyl)-labeled nucleotides (mant-nucleotides; Jena Bioscience) was examined by fluorescence spectrometry (Cary Eclipse, Varian). All of the assays were performed at 25°C, and mant-nucleotide fluorescence was monitored at an excitation wavelength of 355 nm (slit width, 5 nm) and an emission wavelength of 448 nm (slit width, 10 nm). The experiments were carried out in 50 mM Tris, pH 7.6, 100 mM NaCl, 5 mM MgCl 2 . hOLA1 was titrated to a 0.5 M mantnucleotide solution up to a final concentration of 15 M. Alternatively, nucleotide binding was followed by polarization measurement with a Fluoromax 2 spectrofluorometer (Spex Industries). Mant-nucleotides were used at a concentration of 0.2 M.
Isothermal Titration Calorimetry-Nucleotide binding affinity was determined by isothermal titration calorimeter (MCS-ITC, MicroCal, Inc.). Calorimetric experiments were performed at 25°C in 50 mM Tris, pH 7.6, 200 mM NaCl, 5 mM MgCl 2 , 2 mM dithiothreitol. The concentration of protein in the sample cell was 58 M, and the titrated nucleotide were used at 600 M. The data were analyzed using the manufacturer's software. Curve fitting were done according to single ligand binding equations.
Alternatively, nucleotide hydrolysis was quantified using [␥-32 P]ATP/GTP. In brief, the proteins were incubated with [␥-32 P]ATP/GTP (50 M, 0.05 Ci), and nucleotide hydrolysis reactions were stopped by the addition of 1 ml of a charcoal suspension. After centrifugation at 10,000 ϫ g, release of [ 32 P]phosphate was determined by scintillation counting of the supernatant. Competition was done with 50 M of unlabeled nucleotides.
Crystallization and Data Collection-hOLA1 was incubated with 0.5 mg/ml trypsin to remove flexible parts of the protein and subsequently purified by gel filtration. Rod-like crystals of hOLA1 approximately 0.4 ϫ 0.1 ϫ 0.1 mm in size were obtained at 4°C by vapor diffusion using the hanging drop method and 12% polyethylene glycol 3350, 50 mM Tris/HCl, pH 8, 10 mM MgCl 2 as crystallization buffer. The nonhydrolyzable ATP analogue AMPPCP was added to the protein drop to a final concentration of 1 mM. The crystals were soaked with 30% glycerol as cryo-protectant, mounted using a nylon-fiber loop (Hampton Research), and flash-cooled to 100 K in a nitrogen stream. The data were collected at Beamline X06SA (SLS, Villigen) as consecutive series of 0.5°rotation images on a MAR-MOSAIC 225 detector. The data sets were indexed and integrated with XDS (17), scaled with XSCALE, and converted to CCP4, XPLOR, or SHELX format with XDSCONV. Data set statistics are given in Table 1.
Molecular Replacement, Model Building, and Refinement -Molecular replacement was performed with the programs MOLREP (18) and PHASER (19) and homologous YchF proteins (Protein Data Bank codes 1JAL and 1NI3) as search models. The search models were modified with CHAINSAW (20) after sequence alignment with the hOLA1 target sequence. Both MOLREP and PHASER gave the same clear molecular replacement solutions for both search models. The initial phases were improved by non-crystallographic symmetry averaging and solvent flattening with DM (21), resulting in an interpretable electron density map with reduced model bias. The correct side chains were modeled into the electron density map followed by real space refinement using the computer graphics programs COOT (22) and MOLOC (23). Modeling of the cofactor AMPPCP was performed by superposition of the putative nucleotide-binding site of hOLA1 with that of GTP-bound p21-H-Ras (Protein Data Bank code 121P; (24)), using the position and conformation of GTP as a starting point. The model was improved by iterative rounds of model building using COOT and MOLOC and of restrained maximum-likelihood refinement using REFMAC5. 2 including TLS refinement (25). The quality of the final model was checked with MOLPROBITY (molprobity.biochem.duke. edu/) (26,27). The refinement statistics are given in Table 1

RESULTS
hOLA1 is closely related to the bacterial protein YchF, a member of the TRAFAC class of P-loop NTPases (5) that constitutes one of four evolutionary conserved subfamilies belonging to the family of Obg-related GTPases (Fig. 1). The other three subfamilies comprise the Obg-, Nog1-, and Drg-like GTPases. Previously, the Obg family had been divided into five subfamilies, with the group of Ygr210-like proteins defining an independent fifth subfamily. Our phylogenetic comparison revealed that the Ygr210 branch and the Ola1/YchF branch are closely related and group together into one subfamily ( Fig. 1A and supplemental Fig. S1).
In general, Obg-related proteins share two conserved sequence features, namely an YXF(T/C)TXXXXXG segment in the switch I region and a glycine-rich motif in the switch II region (GAXXGXGXGXXX(I/L/V)). A distinguishing feature of the Ola1/YchF subfamily is a noncanonical G4 motif, distinct from the (N/T)KXD consensus (Fig. 1B). The G4 motif in the Ola1/YchF subfamily is as variable as NVNE in bacteria (YchF), NMSE in yeast (Ybr025cp/ yOla1p), and NLSE in human (hOLA1). Both, the conserved lysine and aspartate are replaced. In contrast, the G1, G2, G3, and G5 motifs match the signature found in other P-loop GT-Pases (Fig. 1B).
The variation of the G4 motif may lead to an altered nucleotide binding specificity. Indeed, when we analyzed nucleotide binding of recombinant, highly purified hOLA1 using fluorescent, mant-labeled nucleotides by polarization analysis, we noticed that hOLA1 bound ATP. In contrast, we hardly observed GTP binding at a micromolar concentration of hOLA1 ( Fig. 2A). Even binding of a nonhydrolyzable GTP analogue such as mant-GMPPNP was not saturable as shown by the linear dependence on hOLA1 concentration and failed to reach the level of ATP binding (see Fig. 6B). Further evidence that ATP is the preferred nucleotide-binding partner of hOLA1 stems from the observation that only ATP but not GTP could efficiently displace mant-ATP from hOLA1 (Fig. 2B). Together, this analysis shows that purified hOLA1 is an ATP-binding protein rather than a GTP-binding protein.

hOLA1, an Obg-like ATPase
Isothermal titration calorimetry was used to determine the relative binding affinities for ATP and GTP. The dissociation constant (K D ) for ATP was calculated to be 8 M (Fig.  2C), whereas the K D for GTP could not be determined exactly because of its low affinity for hOLA1 (data not shown). Curiously, when the G4 motif of hOLA1 was reverted to the G4 consensus NKXD by site-directed mutagenesis, the hOLA1-NKXD mutant retained ATPbinding specificity (Fig. 2D), suggesting that the overall fold of the nucleotide-binding pocket in hOLA1 might define nucleotide specificity (see below).
To test whether hOLA1 hydrolyzes ATP, we used HPLC analysis to assess the intrinsic ATP hydrolysis rate of hOLA1. Recombinant hOLA1 was incubated with an excess of ATP, and the amount of ADP production was measured at different time points (Fig. 3). The hydrolysis rate of hOLA1 was calculated to 0.050 Ϯ 0.002 min Ϫ1 . The affinity of hOLA1 for ATP and the intrinsic ATP hydrolysis rate are comparable with the GTP binding and hydrolysis properties of other Obg family members (28 -30). hOLA1 might therefore constitute a novel group of ATPases, with biochemical features resembling those of related Obg-like low affinity GTPases (16,28,29).
To obtain structural insight into ATP binding of hOLA1, we determined the crystal structure of hOLA1 complexed with AMPPCP, a nonhydrolyzable ATP analogue, at a resolution of FIGURE 1. hOLA1 belongs to the Obg family of GTPases. A, phylogenetic tree of the Obg family of GTPases. Tree topology of full-length Obg-related proteins was generated by ClustalW and displayed using TreeView. Obg-related proteins assemble into four main branches, namely Obg (red), Nog1 (purple), Drg (blue), and Ola1/YchF (yellow). Note that Ygr210 homologues are only found in fungi and archaea and build a distinct subclass in the Ola1/YchF branch. In converse, hOLA1 homologues are not present in archaea. Interestingly, the so far uncharacterized human protein GTPBP10/LOC85865 and its homologues from Xenopus laevis, Drosophila melanogaster, and Caenorhabditis elegans form a novel, independent branch close to the Obg branch (supplemental Fig. S2). Note that Obg-like proteins of higher eukaryotes were named after their closely related human homologues GTPBP5 and GTPBP10. Proteins from S. pombe were named after the homologous protein from S. cerevisiae. Alternative names or corresponding gene loci are listed in supplemental  Table S2.  20) with one monomer in the asymmetric unit and no obvious formation of higher oligomeric states in the crystal packing, consistent with the observation that the protein is a monomer in solution (static light scattering; data not shown). We could build 327 of 396 amino acids into the electron density map. Missing are the N-terminal residues 1-15, the switch I and switch II regions with residues 45-57 and 96 -106, respectively, some disordered loop regions with residues 127-138, 172-179, 196 -197, and the C-terminal residues 389 -396. The model was completed by building AMP-PCP and 26 water molecules into the electron density maps. The final model has a crystallographic R factor of 0.23 and a free R factor of 0.29 with very good stereochemistry (Table 1). hOLA1 consists of an N-terminal G domain, flanked on either side by an inserted coiled-coil and a C-terminal TGS domain (Fig. 4) S3). The coiledcoil domain is tilted toward helix ␣7 in the latter structure, making it more compact. Helices ␣8b and ␣10, but not the coiled-coil domain, show similar changes in the superposition of the yeast and bacterial YchF structures, indicating a general variability in these structural elements.
The triphosphate moiety appears tightly bound through a network of H-bonds of main and side chain donors to the phosphate oxygens (Fig. 5B). The conserved interaction of the P-loop Lys 35 with the ␤-phosphates (via main chain NH) and ␥-phosphates (via NH ) is also observed in hOLA1. We observe electron density at the potential binding site of the essential magnesium between Ser 36 in the G1 P-loop, Thr 55 in the G2-loop (switch I region), and the ␤ and ␥-phosphates, but the resolution is too low to identify the ion or its position unambiguously. The cofactor triphosphate moiety forms hydrogen bonds to residues from the P-loop, whereas its sugar moiety has no interaction with hOLA1, and its adenine base has only a few weak interactions with hOLA1. N-7 of the pyrimidine ring in the adenine base can form a hydrogen bond with NH ␦ 2 of Asn 230 (distance 3.3 Å). The N-6 of the base can engage into a weak potential hydrogen bond to the main chain carbonyl oxygen of Leu 231 with an N-O distance of 2.7 Å but a suboptimal C-O-N-angle of 103°. An interaction between the side chain carboxylate group of Glu 233 and the nucleotide as proposed for GTP bound to YchF (14) appears not to be possible, because Glu 233 is too far away.
In comparison with the G4 motif of the GTPase Ras ( 116 NKCD), the conformation of the hOLA1 G4 motif N 230 LSE differs dramatically (Fig. 5C). Although the G4 motif possesses an extended ␤-strand conformation in hOLA1, it shows a zigzag-like conformation in Ras (31). Consistently, Ras Lys 117 can stack with the cofactor, and Asp 119 interacts with the N-1 and N-2 group of the guanine. In hOLA1, the individual positions of the G4 motif are shifted such that only position 1 (Asn 230 ) matches with position 1 of Ras (Asn 116 ), whereas position 3 of hOLA1 (Ser 232 ) matches with position 4 of Ras (Asp 119 ). The Glu 233 of the hOLA1 G4 motif lies at the beginning of a small helix ␣8a preceding helix ␣8b, and intramolecular H-bonds force the entire motif into an extended conformation. This places Glu 233 too far away to interact with the nucleotide. Apparently, the intramolecular forces that define the position of helices ␣8a and ␣8b do not allow the hOLA1 G4 motif to adopt a Ras-like conformation, explaining why the ATP specificity of hOLA1 could not be switched to GTP via mutations of the G4 motif to NKXD (Fig. 2D).
To validate the role of key amino acids in the nucleotidebinding pocket of hOLA1, we mutated selected residues and tested these mutants for their ability to bind to mant-labeled ATP (Fig. 6A). As expected from the structural analysis, mutation of Asn 230 to alanine (N230A) resulted in loss of ATP binding without perturbing solubility or other biochemical characteristics of the protein. Additionally, we mutated the conserved Phe 127 , which had been proposed to contribute to nucleotide binding in H. influenzae YchF through stacking of the aromatic side chain with the base (14). Indeed, when we replaced Phe 127 with Ala (F127A), the mutant protein no longer bound mantlabeled ATP. Like wild-type hOLA1, both mutants also failed to bind GMPPNP or GTP (Fig. 6B and data not shown). This supports that Asn 230 and Phe 127 are essential for ATP binding. It seemed plausible that the conserved Phe 127 , that we could not FIGURE 3. hOLA1 hydrolyzes ATP. A, separation of AMP, ADP, and ATP nucleotides as standards by reverse phase HPLC. AMP, ADP, and ATP were separated on a hydrophobic C-18 column in the presence of tetrabutylammonium, and eluted nucleotides were quantified by absorbance at 260 nm. B, 5 M hOLA1 were incubated with 125 M ATP at 25°C for different periods of time (0 (orange), 15, 30, 45, 60, and 75 min (green)), and ADP production was analyzed as in A. C, quantitation of ATP and GTP hydrolysis by hOLA1 as described in A and B. ADP production was proportional to the incubation time of hOLA1 with ATP. hOLA1 showed low intrinsic ATPase activity, but no GTPase activity. hOLA1, an Obg-like ATPase JULY 6, 2007 • VOLUME 282 • NUMBER 27 see in the electron density map abolished proper conformation of the G4 motif and, together with the unconventional sequence in G4, would prevent efficient binding of GTP. However, combining the F127A and hOLA1-NKXD mutations did not result in a version of hOLA1 that could bind GTP (Fig.  6B), indicating that the overall fold of the nucleotide-binding pocket in hOLA1 must in addition be determined by other parameters.
Because the unconventional G4 motif is present in all members of the Ola1/YchF subfamily, we next tested whether ATPase activity is a general feature of Ola1-like proteins. Analysis of ATP hydrolysis revealed that indeed two other Ola1/YchF subfamily members, namely YchF from H. infuenzae and Ola1p (Ybr025c) from S. cerevisiae preferentially hydrolyzed ATP when compared with GTP (Fig. 7). In contrast, and as reported before, B. subtilis Obg preferred GTP as a substrate for nucleotide hydrolysis. Together, these data uncover an ATPase subfamily in the family of Obg-like GTPases.

DISCUSSION
Our biochemical characterization of hOLA1 reveals a new subclass of ATPases in the Obg family of GTPases. Hence we have named this subclass Ola1, for Obg-like ATPases 1.
It is assumed that P-loop ATPases and GTPases have a common P-loop-containing ancestor. Several ATPases have been derived secondarily form the GTPase superclass through loss of GTPase specificity or activity (5). The kinesinmyosin superfamily of motor ATPases, for instance, may have evolved from the TRAFAC class by loss of the NKXD G4 motif, accompanied by a the gain of ATPase activity. Another example is the subfamily of MinD-like ATPases, which derived within the SIMIBI class of the GTPase superclass (5). The Ola1/YchF subfamily of the Obg family of GTPases, however, presents a unique exception of a sin-hOLA1, an Obg-like ATPase gle protein subfamily, which shows altered nucleotide specificity within a distinct protein family of GTPases, whereas in the aforementioned examples whole protein families of distinct nucleotide specificity evolved. This implies that in the case of the Ola1/YchF subfamily the change in nucleotide specificity occurred relatively late in the evolution of the Obg family.
Biochemically, Ola1-like proteins hOLA1, yeast Ola1p (Ybr025c), and YchF behave similarly to the related GTPase Obg, apart from different nucleotide specificity. Both subfamilies of enzymes display a slow rate of nucleotide hydrolysis in vitro and are characterized by a high rate of nucleotide dissociation (our data and Refs. 16, 28, and 29).
The biochemical characterization of the ATPase activity of hOLA1 was supported by structural analysis of hOLA1 complexed with AMPPCP. The interactions between the bound adenine nucleotide and hOLA1 mainly involve the phosphate groups of the bound AMPPCP. The ␣-phosphate contacts the main chain amide of Thr 37 , the ␤-phosphate the main chain amides of Ser 36 and Val 33 , and the ␥-phosphate forms a hydrogen FIGURE 5. The nucleotide-binding site of hOLA1 bound to AMPPCP. A, the cofactor is bound in between the P-loop (magenta), helices ␣1, ␣9, and the G4 motif bearing the loop following strand ␤6. B, close-up view of interactions between AMPPCP and hOLA1. C, comparison of the cofactor binding site in hOLA1 (white) and p21-H-Ras (orange). Note that the G4 motif of hOLA1 ( 230 NLSE) adopts an extended ␤-strand conformation, whereas the G4 motif of Ras ( 116 NKCD) possesses a zigzag-like conformation. The position of Glu 233 at the beginning of the small helix ␣8a preceding helix ␣8b likely prevents hOLA1 from adopting a Ras-like conformation of the G4 motif (see text). D, model of interactions between hOLA1 and ATP (left) or GTP (right). The purine base is stabilized (blue) by hydrogen bonds to Asn 230 and stacking with the aromatic ring of Phe 127 . Note that the contribution of Phe 127 is based on our biochemical data but not visible in the structure. Nucleotide specificity (red) is conferred by Leu 231 main chain (mc), which forms a hydrogen bond with the exocyclic amino group in adenine, whereas guanine would be repelled.  . ATPase activity is a general feature of the YchF/OLA1 subfamily of Obg-like proteins. A, ATP and GTP hydrolysis activity of YchF, yOla1p, hOLA1, and B. subtilis Obg was analyzed by HPLC. Like hOLA1, the two homologues from bacteria and yeast showed higher ATPase than GTPase activity. In contrast, B. subtilis Obg clearly favored GTP hydrolysis. B, analysis of nucleotide hydrolysis was performed in the presence of both ATP and GTP (100 M each). The nucleotides were separated by anion exchange HPLC in a linear gradient. Although Obg prefers GTP, OLA1-like proteins clearly favor ATP as substrate. The N230A mutation in hOLA1 abolishes ATP hydrolysis. C, specificity of nucleotide hydrolysis activity of YchF, yOla1p, hOLA1, and B. subtilis Obg was examined by competition experiments using ␥-32 P-labeled NTPs in combination with unlabeled ATP or GTP. ATP hydrolysis by OLA1-like proteins was efficiently competed by ATP but not GTP, whereas B. subtilis Obg preferentially hydrolyzed GTP.
bond to the main chain amide of Asn 32 in hOLA1. In our structure, we cannot observe any contacts of the ribose to the protein.
The specificity of nucleotide interactions must be determined by contacts of the protein with the adenine base. The (unprotonated) N-7 of the adenine base forms a hydrogen bond with NH ␦ 2 of Asn 230 . The N-6 group of the base might engage into a hydrogen bond with the main chain CO of Leu 231 , which is part of the unconventional G4 motif of hOLA1. The distance of 2.7 Å falls well within acceptable H-bond range, although the geometry is not optimal in the refined model. Therefore, we cannot exclude an energetically more favorable orientation of the peptide carbonyl group. Our biochemical data, however, support an important contribution of Asn 230 to nucleotide binding, because its mutation to alanine abolished nucleotide binding. Another contact from the protein contributing to base recognition might be established by Phe 127 . Mutating this residue to Ala diminishes ATP binding drastically, indicating an important involvement of this residue in nucleotide binding. The electron density for Phe 127 is too poorly defined for the residue to be modeled. Phe 127 may contribute to the fixation of the nucleotide via stacking on the base, as proposed for the H. influenzae YchF case (14).
So far, there is no published structure of an Ola1-like protein bound to a nucleotide-triphosphate. Structures of YchF from H. influenzae (14) and S. pombe (Protein Data Bank code 1NI3) have been solved in the nucleotide-free form. The fact that Teplyakov et al. (14) were not able to obtain a YchF-GTP cocrystal structure by soaking or cocrystallization supports our results on the binding preference of hOLA1 for ATP and suggests that this specificity extends to other members of the Ola1/ YchF subfamily. Very recently, a structure of YchF from T. thermophilus bound to GDP has been published in the data base (Protein Data Bank code 2DBY). However, in that structure, there is no interpretable electron density for the ribose and the base. Therefore, it is not clear how the base is recognized in this case. Our biochemical data clearly show that ATP binding to Ola1-like proteins is favored over GTP binding, but that GTP association can be forced to occur at high nucleotide concentration in vitro.
How can the preference of hOLA1 for ATP be explained? The nucleotide binding mode deduced from our structure suggests that the relative specificity of hOLA1 for adenine and discrimination against guanine binding is based on the interaction between the adenine N-6 group and Leu 231 main chain CO in G4. In guanine, with the carbonyl oxygen O-6 present at the 6-position, this hydrogen bond binding mode is not possible. Instead, repulsive forces between the O-6 and Leu 231 CO presumably disfavor guanine binding (Fig. 5C).
The ATP binding specificity of hOLA1 appears even more striking when compared with the founding member of the family, B. subtilis Obg bound to ppGpp (16). The 5Ј phosphates of both nucleotides (AMPPCP and ppGpp) superimpose very closely, with equivalent H-bond patterns (supplemental Fig.  S4). Strikingly, both the ribose and base moieties are tilted by about 40°with respect to one another. The different angle of the adenine base relative to the phosphates in AMPPCP allows for the adenine N-6-Leu 231 O H-bond in the hOLA1 case, whereas a different hydrogen bond is formed between the ppGpp and Obg (N-1-Asp 285 O ␦2 and N-2-Asp 285 O ␦1 ). This latter, bidentate interaction represents in fact the only clear discrimination between adenine and guanine base, because the H-bond between the ppGpp O-6 and Ser 310 O ␥ could also be formed with an exocyclic N-6 of an adenine in a similar position.
The G4 motifs of the GTPase Ras ( 116 NKCD) and the ATPase hOLA1 ( 230 NLSE) show different conformations (Fig. 5B). Although the G4 motif in Ras adopts a zigzag-like conformation (31), the G4 motif in hOLA1 possesses an extended ␤-strand conformation. In the G4 motif of hOLA1, Ser 232 and Glu 233 cannot interact with the nucleotide because of the altered conformation of the G4 loop. Our attempts to convert hOLA1 back into a GTPase by conversion of the G4 motif to NKXD did not result in a version of hOLA1 that could bind GTP but retained ATP specificity (Fig. 2D). This indicates that the overall fold of the nucleotide-binding pocket in hOLA1 is determined by additional parameters.
In Ras-like GTPases, a Gln residue is involved in (GAP-induced) GTP hydrolysis. This conserved Gln residue is replaced by a hydrophobic amino acid in all Obg family members (e.g. Leu 96 in hOLA1), a substitution that inactivates Ras-like GTPases (32). This indicates a different NTP hydrolysis mechanism used by Obg-like NTPases. P-loop NTPases undergo conformational changes upon NTP hydrolysis. The structural differences between AMPPCP-bound hOLA1 and nucleotidefree YchF are minimal, and it remains to be seen which structural differences might exist between ATP and ADP-bound hOLA1.
Ola1-like proteins display a low intrinsic nucleotide hydrolysis and a high nucleotide dissociation rate, as observed for Obg, the founding member of the Obg family. It remains to be seen whether these features are important for hOLA1 function in vivo or whether yet to be identified binding partners of Ola1-like proteins influence their nucleotide binding and hydrolysis properties in the cellular context. This awaits the identification of potential binding partners and their functional characterization.