Crystal Structure of the Citrobacter freundii Dihydroxyacetone Kinase Reveals an Eight-stranded α-Helical Barrel ATP-binding Domain*

Dihydroxyacetone kinases are a sequence-conserved family of enzymes, which utilize two different phosphoryldonors, ATP in animals, plants and some bacteria, and a multiphosphoprotein of the phosphoenolpyruvate carbohydrate phosphotransferase system in bacteria. Here we report the 2.5-Å crystal structure of the homodimeric Citrobacter freundii dihydroxyacetone kinase complex with an ATP analogue and dihydroxyacetone. The N-terminal domain consists of two α/β-folds with a molecule of dihydroxyacetone covalently bound in hemiaminal linkage to the Nϵ2 of His-220. The C-terminal domain consists of a regular eight-helix α-barrel. The eight helices form a deep pocket, which includes a tightly bound phospholipid. Only the lipid headgroup protrudes from the surface. The nucleotide is bound on the top of the barrel across from the entrance to the lipid pocket. The phosphate groups are coordinated by two Mg2+ ions to γ-carboxyl groups of aspartyl residues. The ATP binding site does not contain positively charged or aromatic groups. Paralogues of dihydroxyacetone kinase also occur in association with transcription regulators and proteins of unknown function pointing to biological roles beyond triose metabolism.

Dihydroxyacetone (Dha) 1 kinases can utilize either of two sources for high energy phosphate, ATP or a phosphoprotein of the bacterial phosphoenolpyruvate:sugar phosphotransferase system (PTS) (1). Little is known about the function of this enzyme and the metabolic origin of its substrate, Dha. Dihydroxyacetone phosphate (DhaP) can be formed by aldol cleavage of fructose-1,6-bisphosphate, by isomerization from glyceraldehyde-3-phosphate, and by oxidation of glycerol-3phosphate in the mitochondrial glycerol phosphate shuttle. DhaP is an obligatory precursor of glyceryl ether phospholipid biosynthesis (2). Free Dha plays a pivotal role in methanol assimilation by methylotrophic yeast and plants where it is produced in the transketolase reaction between xylulose-5phosphate and formaldehyde catalyzed by dihydroxyacetone synthase (3)(4)(5)(6). Bacteria produce free Dha by oxidation of glycerol under anaerobic conditions and aldol cleavage of fructose-6-phosphate (9 -13). Dha is a carbon source for bacteria, and if added to the medium, it is also used as gluconeogenic precursor by mammalian tissues (14 -18). Although the pathways utilizing free Dha appear few and limited in scope, genes for Dha kinases and Dha kinase homologues are widely distributed among plants and animals where their biological function is not obvious (for a survey see accession numbers PF02733 and PF02734 at www.sanger.ac.uk/Software/Pfam/index.shtml). Dha like other triose sugars has an increased propensity to react with proteins in Maillard-type reactions (19,20), because unlike hexoses and pentoses, it cannot be deactivated by formation of cyclic hemiacetals. The chemical reactivity of Dha might be the rationale for its use as a therapeutic tanning agent (21,22). It has recently been shown that Dha can be toxic to yeast cells and that detoxification is dependent on a functional Dha kinase (23). Scavenging of reactive Dha might be an additional function of Dha kinases and could explain its occurrence in organisms that do not normally metabolize Dha. Dha might be formed as a byproduct by phosphatases and aldolases of limited specificity or in paracatalytic reactions such as the accidental oxidation of Schiff-base intermediates (24). Two representative examples of Dha kinases are the ATP-dependent Dha kinase of Citrobacter freundii (Swiss Protein Data Bank accession number P45510) (25,26) and the PTS-dependent kinase complex of Escherichia coli (1,(27)(28)(29). The Dha kinase of C. freundii consists of two domains, which we term K-and L-domains. The E. coli kinase consists of two catalytic subunits DhaK and DhaL (Swiss Protein Data Bank accession numbers P76015 and P76014) and a third phosphoryl donor subunit DhaM (P37349) as shown in Equations 1-3. * This work was supported by Grant 3100-063420 from the Swiss National Science Foundation, a grant from the Secretaria de Estado de Educacion y Universidades (to L. F. G.-A.), and the Ciba-Geigy Jubilä umsstiftung. The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked "advertisement" in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.
The x-ray structure of the DhaK subunit of E. coli in complex with its substrate has recently been solved at 1.75 Å (33). Dha is bound in hemiaminal linkage to the imidazole nitrogen of an invariant histidine. It has been proposed that Dha kinases exploit the chemical reactivity of the carbonyl group to discriminate between the potentially toxic Dha and the structurally similar compatible solute glycerol (33). In contrast, glycerol kinases cannot discriminate between the two substrates (10). Here we present the x-ray structure of the full-length Dha kinase of C. freundii in complex with Dha, an ATP analogue and a phospholipid.

EXPERIMENTAL PROCEDURES
Plasmid Construction and Site-directed Mutagenesis-The dhaK gene was PCR-amplified with C. freundii genomic DNA as template and primers encoding NsiI and BglII restriction sites. The PCR fragments were digested with NsiI and BglII and ligated with a plasmid vector obtained by digestion of pTSGH11 (35) with the same enzymes. The recombinant dhaK gene encodes a Dha kinase with a C-terminal His tag. The point mutations D380A, D385A, D387A, and T388H in DhaK of C. freundii and D30A, D35A, D37A, H38T, H38C, H38A in DhaL of E. coli were introduced using the QuikChange method (Stratagene) and primers encoding the mutation and a diagnostic restriction site. Successful mutagenesis was verified by DNA sequencing. Standard procedures were used for plasmid purification, restriction analysis, ligation, and transformation.
Protein Purification and Activity assay-The Dha kinase of C. freundii was overexpressed in E. coli K12 WA2127⌬ HIC (⌬ manXYZ ⌬ ptsHIcrr) (36). Cells (3 liters) were grown at 37°C and induced with 200 M isopropyl-1-thio-␤-D-galactopyranoside at A 600 ϭ 0.8, and incubation continued for 5 h. Cells were collected by centrifugation, and the sediment was resuspended in buffer A (25 ml, 50 mM NaP i , pH 8.0, 10 mM ␤-mercaptoethanol, 500 mM NaCl), broken by two passages in a French pressure cell, and fractionated by differential centrifugation into cell debris (10 min, 3000 ϫ g), membrane fraction (60 min, 150,000 ϫ g), and cytoplasmic fraction (supernatant), which was mixed with 10 ml of Ni 2ϩ -nitrilotriacetic acid resin (Qiagen) and allowed to adsorb for 30 min. The resin was transferred to a column washed with 100 ml buffer A containing 0 and 50 mM imidazole. The His-tagged protein was eluted with 200 mM imidazole in the same buffer, concentrated, and purified by gel filtration over Superdex 200 (Amersham Biosciences) with a buffer containing 10 mM HEPES, pH 7.5, 400 mM NaCl, and 2 mM dithiothreitol. Seleno-Met-Dha kinase was prepared using the methionine pathway inhibition method (37) and purified like the native protein. The presence of seleno-Met was confirmed by electrospray ionization mass spectrometry (results not shown). Dha kinase was concentrated to 30 -50 mg ml Ϫ1 in 5 mM HEPES, pH 7.5, 2 mM dithiothreitol. Kinase activity was measured in a coupled assay by reduction of dihydroxyacetone phosphate with glycerol-3-phosphate dehydrogenase as described previously (1) with ATP instead of phosphoenolpyruvate plus PTS proteins as phosphoryl donor. The disappearance of NADH was monitored continuously in a Spectramax-250 plate reader at 30°C.
Crystallisation and Data Collection-(Seleno-Met)-Dha kinase was crystallized in buffer B (100 mM HEPES, pH 7.3, 50 mM NaCl, 9% (w/v) polyethylene glycol 3000, and 25% (v/v) glycerol) using sitting dropvapor diffusion. Crystals belong to the space group C222 1 with cell dimensions of a ϭ 100.4 Å, b ϭ 124.6 Å, and c ϭ 236.4 Å and contain one dimer per asymmetric unit. For ANP and Dha complexed Dha kinase, native protein crystals were soaked with 5 mM ANP, 10 mM MgCl 2 or MnCl 2 , and 2 mM Dha for 3 h at 20°C in buffer B. The best crystals of the native protein diffracted to 2.5 Å after flash-freezing at 105 K in buffer B. Native diffraction data were collected at the Swiss Norwegian Beamline at European Synchrotron Radiation Facility (Grenoble, France) employing a MAR345 image plate detector (Xrayresearch, Hamburg, Germany). The seleno-Met x-ray data sets were measured at beamline BW7A at the EMBL Hamburg Outstation at the DESY using a MAR CCD-165 detector. Because of the smaller area of the detector, which caused reflection overlap, and also because of the inferior quality of the selenomethionine crystals, data were only collected to a resolution of 3.5 Å. Three data sets were collected at peak (0.9743 Å), remote (0.9252 Å), and inflection point (0.9750 Å). The ANP/Dha-associated Dha kinase crystals were measured at the SLS beamline X06SA at the PSI Villigen using again a MAR CCD-165 detector. All of the datasets were collected employing a MAR CCD-165 detector (Xrayresearch, Hamburg, Germany) and were processed using the HKL program package (38).
Structure Solution and Refinement-The Dha kinase structure was determined by MAD methods. Of the 28 methionines expected in the asymmetric unit, the positions of 20 were readily determined using Shake-and-Bake (39). Phases were computed using SOLVE (40) and density modification, and phase extension to 2.5-Å resolution using the native data set was effected by the program RESOLVE (40). The resulting map was of high quality and allowed the tracing of the whole polypeptide chain. An initial protein model was built into the electron density using the program O (42). Non-crystallographic symmetry restraints were used to refine the model to a 2.5-Å resolution using the programs CNS (43) and REFMAC (44). All of the data within the resolution range were included during refinement. Data collection and is the intensity of an individual measurement and ͗I(hkl)͘ is the average intensity from multiple observations. c R factor ϭ ⌺ hkl ʈF obs ͉ Ϫ k͉F calc ʈ/⌺ hkl ͉F obs ͉. d R free equals the R-factor against 5% of the data removed prior to refinement. e r.m.s.d. is the root-mean-square deviation from ideal geometry. f Friedel partners are accounted as different reflections.
refinement statistics are given in Table I. A Ramachandran plot as defined by Kleywegt and Jones (45) showed no residues in disallowed regions but showed nine outliers in generously allowed regions (1.0%).
The last four C-terminal residues and the additional 8-residue C-terminal histidine tag are not visible in the structure, and the protein model ends with the glutamate 550. Lipid Analysis-Phospholipids were separated by thin layer chromatography on silicic acid. 10 l of concentrated DhaK (15 mg/ml) were applied, the chromatogram was developed with CHCl 3 :MeOH:H 2 O 76: 21:3, and lipids stained with 40% sulfuric acid, ninhydrin, or phosphate stain (46).
Sequence Analyses-To retrieve DhaL sequences from the non-redundant protein data base at NCBI (www.ncbi.nlm.nih.gov/entrez/), the combined DhaK and DhaL sequences of E. coli and C. freundii were used as queries (Swiss Protein Data Base accession numbers P45510, P76015, and P76014). The alignments were performed with MultAlin (47) using the default parameters. Phylogenetic trees were constructed with Gene-Bee (48) using the default parameters.
Coordinates-Crystallographic coordinates and structure factor amplitudes have been deposited in the Protein Data Bank (accession codes 1UN8 (apo form); 1UN9 (Dha/ANP form)).

Structure Determination and Overall
Architecture-Dha kinase was purified by nickel-nitrilotriacetic acid affinity and size-exclusion chromatography from where it eluted as a symmetrical peak with the retention time expected of a 2 ϫ 60-kDa homodimer. It was crystallized with polyethylene glycol 3000 as precipitant, and the structures of the apoprotein and the complex with Dha and ANP were solved. Phases were determined using selenomethionine-substituted protein and multiwavelength anomalous dispersion at 3.5-Å resolution. Refinement of the native protein structure against the 2.5-Å data resulted in an R-factor of 19.3% and a R free of 24.5% with reasonable stereochemistry (Table I). The asymmetric unit contains one dimer. The Dha kinase monomer is well ordered (residues 1-550) with the exception of the loop 518 -528.
The K-domain (residues 1-330) and the L-domain (residues 350 -550) are separated by an extended linker (residues 331- 349) and swapped such that the L-domain of one kinase subunit is bound to the K-domain of the other (Fig. 1). The linker is well ordered, and it contains a short ␤-strand, which is hydrogen-bonded to the edge strand of the N-terminal ␤-sheet of the K-domain.
The K-domain of C. freundii has the same structure as the DhaK subunit of E. coli kinase (33). It consists of two ␣/␤-folds, each containing a six-stranded mixed ␤-sheet surrounded by six and three helices, respectively. The core structure of the first ␣/␤-fold (colored blue in Fig. 1) has the same topology as the IIA Man domain of the PTS transporter for mannose (Protein Data Bank code 1PDO) (30) and the N-terminal domain of DhaM (1). The inter-subunit interaction is mediated mainly by the surfaces buried between helices 1 and 8, which intersect at an angle of 70°.
The L-domain displays a novel all-␣-fold. It consists of eight antiparallel ␣-helices arranged in an up-and-down geometry and aligned on a circle. This results in the formation of a helix barrel enclosing a deep pocket. The helices are between 3.5 and 6 turns long with their axes tilted at an angle of ϳ15°with respect to the barrel axis. The first and longest helix (amino acids 355-386) is kinked at an angle of 45°between Leu-372 and Glu-373. The helices are amphipathic with the hydrophobic side chains directed into the pocket of the barrel and with the polar residues surface exposed. All of the loops are well defined with the exception of the longest one (amino acids 513-533) between helix 7 and 8 where residues 518 -527 are not visible in the electron density map. The four loops at the top of the barrel contain highly conserved residues (Fig. 2), whereas the loop sequences on the bottom side (defined by the N and C termini of the chain) are not conserved.
Lipid-binding Pocket-Strong electron density remained in the native protein model that could not be explained by protein atoms and became the predominant feature in F o Ϫ F c maps after sequence fitting and initial model refinement (Fig. 3A). This density has the shape of a 2-myristic(C14)-3-palmitic(C16)-phospholipid. The two acyl chains extend ϳ 15 Å into the pocket where they are surrounded by apolar side chains. Residues in a distance of Ͻ4 Å to the acyl chains are invariant or conservatively exchanged in the L-domains and DhaL subunits of Dha kinases (Fig. 2, open circles). The lipid binding pocket has an ellipsoidal shape (5-11-Å wide) and a solventaccessible surface area of 496 Å 2 . The lipid head group lies solvent-exposed at the entrance of the pocket (Fig. 3B). The lipid was identified as a 1:1 mixture of phosphatidic acid and phosphatidylethanolamine by thin layer chromatography (Fig.  4A). Staining revealed two closely spaced spots of which the second disappeared after pretreatment of the sample with phospholipase D. Both spots were stained phosphate-positive, and the phospholipase-sensitive spot also was ninhydrin-positive (results not shown). The lipids can be removed by incubation with the detergent ␤-octylglucoside and gel-filtration in a detergent-free buffer. The lipid-free kinase (Fig. 4A) retains at least 80% of its original activity. It cannot be stimulated with E. coli liposomes, indicating that the lipid is not essential for kinase activity (results not shown). Given the hydrophobic lining of the pocket, the phospholipid is probably replaced by one or two detergent molecules.
The ATP-and Dha-binding Sites-Crystals were soaked in a solution containing ANP, MgCl 2 , and dihydroxyacetone, and a complete data set was collected to a 3.1-Å resolution from a single crystal (Table I) and one molecule of Dha per monomer could be fitted unambiguously. After rigid body and positional refinement using the model of the apoenzyme, two magnesium ions complexed to the ANP molecule were also identified (Fig. 3C, peaks at 8 and 6 in the F o Ϫ F c map). To further confirm their position, the crystals were soaked with ANP, Dha, and Mn 2ϩ instead of Mg 2ϩ , and a dataset of a single crystal was collected to a 4.0-Å resolution (Table I). Isomorphous and anomalous difference Fourier maps showed clear peaks at the putative Mg 2ϩ positions, confirming the original interpretation (Fig. 3B). The two magnesium atoms have a 3.99-Å center to center distance. ANP binds to the L-domain at the top of the ␣-barrel (Fig. 3, C and  D). The bottom of the nucleotide-binding site is formed by Met-428, Met-436, Met-477, and Phe-392, which septum-like separate the ANP-binding depression from the lipid-binding pocket. The side walls are formed by the helix termini and the connecting loops. The adenine base is packed between the hydrophobic side chains of Leu-435 (H3-H4), Thr-476, and Met-477 (H5-H6). The ribose moiety is bound by residues Met-477 (H5-H6), Asp-533, and Pro-534, its hydroxyl groups are coordinated by Asp-533 (H7-H8), and the ring oxygen is within hydrogen bonding distance to Ser-432. The ␣-phosphate is coordinated by Ser-432, and the ␤and ␥-phosphates are hydrogenbonded to Ser-431 (H3-H4). The N terminus of helix H4 is directed toward the ␥-phosphate and by its helix dipole might stabilize the phosphate. The phosphate groups are complexed by two Mg 2ϩ ions. One Mg 2ϩ coordinates the ␥-phosphate with Asp-385 and Asp-387. The second Mg 2ϩ connects the ␣and ␤-phosphates with the ␥-carboxyl groups of Asp-380, Asp-385, Asp-387, and the hydroxyl group of Thr-388 (H1-H2) (Fig. 3D). The coordination geometry of the magnesium atoms cannot be determined because water ligands that complement the coordination sphere are not visible at the resolution of the structure. The H1-H2 and H3-H4 loops contain invariant glycines, reminiscent of the mononucleotide-binding motives of ATPbinding proteins and protein kinases (49,50). However, the ATP-binding site of the Dha kinase differs from those sites because (i) the loops are centered between two ␣-helices and not in a ␤-hairpin or ␤-␣-loop, (ii) the loops do not contain the characteristic aromatic residue or C-terminal lysine, respectively, and (iii) the negative charge of the nucleotide is neutralized by two Mg 2ϩ ions rather than one.
The Asp-380, Asp-385, and Asp-387 mutants of the ATP-dependent C. freundii kinase are completely inactive (Fig. 4B), whereas the corresponding mutants of the PTS-dependent E. coli kinase have reduced activity (Fig. 4C). The reduced activity may be caused by protein instability rather than by catalytic inefficiency because the E. coli mutants were unstable and expressed only in minute amounts compared with the wild-type protein. The conserved Thr-388 is of particular interest. Mostly two-domain Dha kinases (presumably ATP-dependent) have a Thr in this position, whereas the two-subunit kinases (presumably PTS-dependent) always have a His (Fig.  2). Whereas Thr-388 is not essential for activity of the C. freundii enzyme, His-38 is essential for the E. coli enzyme (Fig.  4, B and C). This invariant histidine in the first loop of a DhaL subunit might thus be diagnostic for the PTS dependence of a Dha kinase.
Dha is bound to the K-domain at the same site as reported for the 1.75-Å resolution structure of the homologous E. coli DhaK subunit (33). The details of bonding are not revealed at the lower resolution of this structure, but given the similarity of sequence (Fig. 2) and structure, an identical binding mechanism can be inferred (Fig. 1C). The binding sites for Dha and ATP are 14 Å apart in the crystal structure presented here (Fig.  1), indicating that the contacts between the K-and L-domains are induced by the crystal lattice. The solvent-accessible surface buried between H3 of the K-domain and H4 and H5 of the L-domain is 1340 Å 2 , which is less than the strongest crystal  contact of 2300 Å 2 that stretches across the K-L interface. Moreover, the contact area does not contain conserved residues as would be expected of a functional interface (helices 4 and 5 in Fig. 2). We envisage the K-and L-domains as mobile in solution and connected only by a long flexible linker as it was shown for the dimeric IIAB Man subunit of the E. coli mannose transporter (51). Attempts to induce alternative conformations by co-crystallization instead of soaking with Dha/ANP and with the heterobifunctional substrate ATP-dihydroxyacetone 2 were unsuccessful.
Amino Acid Sequence and Structure Comparison of the C. freundii and E. coli Dha Kinase Domains-Overlay of the C.
freundii K-domain with the E. coli DhaK subunit using the DALI server (52) yields a root mean square distance of 1.32 Å for 287 C␣ atoms (Fig. 1D, red and blue backbone traces). The only prominent difference is the extended ␤-ribbon that caps one edge of the N-terminal ␤-sheet of E. coli DhaK. This ribbon occurs only in the DhaK subunits of Gram-negative bacteria, which contain large multidomain DhaM subunits and where it might prevent edge-to-edge aggregation between apolar ␤-sheets (53,54). Amino acid sequence comparison reveals that DhaK subunits without this ␤-ribbon always contain a charged residue (Arg, Lys, or Glu) inserted at an invariant position of the edge strand, which is a dominant strategy for blocking edge strand aggregation (53).
The L-domain, an eight-stranded ␣-helical barrel, represents  (33). The lipidprotein complex is reminiscent of lipid transfer proteins of plants (55) and a recently described pollen allergen (41). In lipid transfer proteins, the lipid is enclosed by four amphipathic helices in the pollen allergen by two interlocked EF-hand-folds. However, whereas in these binding proteins the lipids are completely shielded from the medium, the polar headgroup remains solvent-exposed in the L-domain (Fig. 3B). To our knowledge, lipids bound in such a way have so far been observed in membrane proteins only where they are supposed to determine the packing of transmembrane helices (34). The cellular function of the lipid in the L-domain of the Dha kinase is not known, nor how it gets incorporated during protein folding. We favor the hypothesis that the lipid stabilizes the helix barrel and, in comparison with the non-annular membrane lipids, may play a purely structural role.
The amino acid sequence comparison indicates that the ATPbinding site of the L-domain is well conserved in all of the Dha kinases (Fig. 2), regardless of whether they utilize ATP or a phosphoprotein of the PTS as phosphoryl donor. Mutations of the conserved aspartyl residues not only completely abolish ATP-dependent activity of the C. freundii enzyme but also compromise the activity and stability of the PTS-dependent E. coli kinase. Although the function of these invariant aspartyl residues appears obvious from the x-ray structure, namely to provide binding sites for the ATP-magnesium complex, their molecular function in the PTS-dependent kinases is unknown. Here they may participate in the binding of the phosphoryldonor subunit DhaM. In this case, DhaM, the component of the PTS, would have evolved toward binding to the already pre-FIG. 6. Topological diagrams of TM841, K-domain, and DhaK subunit. The IIA Man and FtsZ core-folds are framed. Extra secondary structure elements are filled for TM841 (vertical stripes), for Dha kinase (horizontal stripes), and for E. coli (diagonal stripes). For amino acid sequence alignment of TM841 and YfhG, the common secondary structure elements of the IIA Man core are numbered ␤1-␣5 of the FtsZ core ␤1*-␤6*. Secondary structure elements of the extensions are not numbered. Thr-88 and Ser-121, which coordinate the carboxylic group, are indicated with a star.
formed binding site of the DhaL ancestor rather than vice versa. Indeed, the DhaL domains display a higher degree of amino acid sequence similarity than the IIA domains. For instance, the DhaL subunit of E. coli has 26 and 30% sequence identity with the C. freundii and Homo sapiens DhaL domain, respectively, whereas the IIA domains of DhaM and the mannose transporter of E. coli only have 15% sequence identity. Thr-388 is another residue of interest in the nucleotide-binding site. The equivalent position in the PTS-dependent kinases invariably is occupied by a histidine. There is indirect evidence from isotope exchange and pull-down experiments 3 that a phospho-DhaL intermediate is formed and that E. coli DhaL is transiently phosphorylated at this histidine (His-38 of the E. coli subunit). This intermediate is unusually labile for a PTS protein and therefore escapes detection by autoradiography on polyacrylamide gels (1). Taken together, this finding suggests that the nucleotide-binding site is important not only for phosphoryl transfer from ATP but also for the transfer from the phosphoproteins of the PTS. A phylogenetic analysis of prokaryotic and eukaryotic L-domains indicates that the ATP-dependent Dha kinases are deeper rooted than the PTS-dependent ones. Consequently, it is tempting to speculate that the PTS-dependent forms evolved from an ATP-dependent ancestor to tap the PTS as source of high energy phosphate.
The ␣ helix barrel fold appears not only as C-terminal domain in Dha kinases but also as N-terminal domain in a family of two-domain proteins with unknown function (Fig. 5). One representative example is YfhG of Lactococcus lactis (Swiss Protein Data Bank code Q9CHY7). Whereas the amino acid sequence similarity of its N-terminal domain with the L-domain of Dha kinases is obvious (Fig. 2), the sequence of its C-terminal domain bears no similarity to the K-domain. However, this domain is predicted by protein-fold recognition (www.bmm.icnet.uk/ϳ3dpssm/) (8) to have a similar fold as TM841, a hypothetical fatty-acid binding protein of Thermotoga maritima (7) (Swiss Protein Data Bank accession number Q9X1H9 and Protein Data Bank code 1MGP). TM841, in turn, has a fold strikingly similar to the K-domain of the Dha kinases (Fig. 6). Taken together, this finding suggests that YfhG is a circularly permuted structural variant of a Dha kinase with the domain order of L3 K instead of K3 L (Fig. 5). The K-domains of Dha kinases and of TM841 (and by inference also the Cterminal domain of YfhG) share the IIA Man and the FtsZ corefolds but differ in how these cores are modified (Fig. 6). (i) The K-domains of Dha kinases contain an ␣-helix (dimerization helix) and two ␤-strands preceeding the IIA Man core, which is not present in TM841. (ii) TM841 has an additional ␣/␤-fold inserted between ␤2 and ␣2 of the IIA Man core (corresponds to ␤4 and ␣3 of the K-domain), which is not present in the Kdomain. The FtsZ-fold is conserved in both proteins. Also conserved is the extension of the FtsZ-fold by the two ␤-strands that carry the active site His of the Dha kinases. (iii) The two unequal substrates are bound in a similar way. The carboxyl group of the fatty acid bound to TM841 and the two hydroxyl groups of Dha bound to the K-domain are both oriented and hydrogen-bonded to residues at the topological switch point of the IIA Man core (Fig. 6). Biochemical studies will be necessary to decide whether the ␣-helix barrel is nucleotide binding also in YfhG, whether the lipidic moieties observed in the L-domain on one hand (Fig. 3) and in TM841 on the other (7) are simultaneously present in YfhG, and whether this points to a possible function of YfhG in lipid metabolism.