Structure of the acetophenone carboxylase core complex: prototype of a new class of ATP-dependent carboxylases/hydrolases

Degradation of the aromatic ketone acetophenone is initiated by its carboxylation to benzoylacetate catalyzed by acetophenone carboxylase (Apc) in a reaction dependent on the hydrolysis of two ATP to ADP and Pi. Apc is a large protein complex which dissociates during purification into a heterooctameric Apc(αα′βγ)2 core complex of 482 kDa and Apcε of 34 kDa. In this report, we present the X-ray structure of the Apc(αα′βγ)2 core complex from Aromatoleum aromaticum at ca. 3 Å resolution which reveals a unique modular architecture and serves as model of a new enzyme family. Apcβ contains a novel domain fold composed of two β-sheets in a barrel-like arrangement running into a bundle of eight short polyproline (type II)-like helical segments. Apcα and Apcα′ possess ATP binding modules of the ASKHA superfamily integrated into their multidomain structures and presumably operate as ATP-dependent kinases for acetophenone and bicarbonate, respectively. Mechanistic aspects of the novel carboxylation reaction requiring massive structural rearrangements are discussed and criteria for specifically annotating the family members Apc, acetone carboxylase and hydantoinase are defined.

. Reactions of Apc, Acx and methyl-hydantoinase. Apc requires the hydrolysis of 2 ATP to 2 ADP + 2 P i (a), whereas most Acx hydrolyses one ATP to AMP + 2 P i per carboxylase reaction (b). Hydantoinase hydrolyses ATP to ADP + P i for imidate phosphorylation (c). In all three reactions the keto/enol equilibrium is shifted towards the enolate/imidate side by phosphorylation.

Results and Discussion
Structure of the Apc core complex. The Apc core complex reveals an impressive novel protein structure with a size of ca. 250 Å × 100 Å × 100 Å (Fig. 2). It is present in the crystals as an Apc(α α ′ β γ ) 2 heterooctamer which confirms the previously reported oligomeric state derived from a gel filtration-based molecular mass determination in solution 10 . The heterooctamer is built up by a crystallographic two-fold axis from an asymmetric unit containing one Apcα α ′ β γ protomer. A relatively small protomer interface is formed exclusively between two Apcβ (1130 Å 2 ; 4% of the surface area 18 ). The Apcα α ′ β γ protomer is mainly held together by large contact areas between Apcα and Apcβ (3540 Å 2 ). Apcα ′ is located at the periphery of the Apc core complex and only attached to Apcα . The small Apcγ partly envelops Apcβ and is itself embraced by the extended C-terminal arm of Apcβ (Fig. 2). One Apcα α ′ β γ protomer is expected to contain one bound Zn 2+ ion as detected by quantitative elemental analysis 10 . Although Zn 2+ was not detected in the X-ray structure, two of the found mercury binding sites are also plausible for binding of Zn 2+ ions. One mercury was found to be coordinated to Asp65, His123, Asp126 and His148 in Apcβ and a second to Cys22, Cys25, Cys75 and Cys78 in Apcγ (sFig. 1). We favour the latter as more probable physiological Zn 2+ -binding site in Apc because of the high conservation of the cysteines in Apc and Acx, its location in a classical Zn finger motif 19 and the absence of Zn 2+ in Hyd, which are lacking orthologs of Apcγ . As both potential Zn 2+ binding sites are located far away from the proposed active site, the role of the Zn 2+ ion is probably more structural and not directly related to the catalytic reaction.
The subunits of Apc are modularly built up of several domains arranged in a new topological manner. Based on the DALI 20 and SCOP servers 21 Apcα reveals a high structural relationship to Apcα ′ reflected in a rmsd of 4.1 Å between the two subunits (96% of the Cα used) suggesting an evolution from a simpler common ancestor by gene duplication (Fig. 3). Their most prominent structural elements (Apcα : 1-90, 250-275; 275-510; Apcα ′ : 1-80, 215-465) belong to the ASKHA (acetate and sugar kinase/heat shock cognate/actin) superfamily 22 . Moreover, an α /β domain (Apcα : 91-249; Apcα ′ : 81-214) bridges the two domains of the ASHKA modules, and an α + β domain (Apcα : 511-630; Apcα ′ : 470-580) follows the ASHKA modules, respectively. In addition, the C-terminal β -barrel-like domains (Apcα : 640-715; Apcα ′ : 590-660) and the second domains of the ASHKA modules are linked via their central sheets (Fig. 3). Apcβ is composed of a Greek-key like β -domain (30-230), a mixed β -sheet domain (255-380), and a domain (385-605) with a novel fold (Fig. 4a,b). The novel domain can be subdivided in a distorted barrel composed of two β -sheets with a hydrophobic core in between and an unusual folding motif at the barrel bottom. Most strands of the two β -sheets converge to a bundle of eight glycine-rich segments (390-393, 422-427, 431-435, 507-510, 582-584, 475-479, 514-517 and 470-473) with polyproline II helical conformations (manuscript in preparation). In Apcγ , a β -hairpin, a β -meander and a four-strand antiparallel β -sheet are fused together. This subunit is tightly associated to Apcβ and does not assemble to a globular fold (Fig. 2). ATP and putative substrate binding sites of Apc. The X-ray structure of the Apc core complex revealed two ASKHA folding motifs as key components of Apcα and Apcα ′ (Fig. 3) which could not be predicted on the basis of the primary structures. Superposition of the pantothenate kinase-ADP complex 23 (pdb-code: 3BF1) or the actin-ATP complex 24 (pdb-code: 4B1U) structures onto those of Apcα or Apcα ′ basically indicated related ATP binding sites despite substantial deviations of the surrounding polypeptide architecture (Fig. 5). The conformations of the known signature motifs for ATP binding in ASHKA domains, ADENOSINE (GxxPGP), PHOSPHATE 1 (DxGGTxDDT) and PHOSPHATE 2 (DVGGT), are well conserved in Apcα and Apcα ′ , as well as in the modelled structures of Acxα and Hydα (see below). The cleft between the two ASKHA domains in the recorded structure of Apcα is open (Fig. 5a) and is expected to close upon ATP/ADP binding, as found in other structurally characterized ASKHA family members 25 . In contrast, the interdomain cleft in Apcα ′ is closed. This is consistent with the presence of an apparently bound ADP, which has tentatively been fitted into residual electron density at the ATP binding site of Apcα ′ (Fig. 5b). The structural data clearly suggest that both Apcα and Apcα ′ are capable of ATP binding and thus presumably contain the active sites for the two ATP-dependent kinase reactions 10 . Previous kinetic studies on Apc demonstrated uncoupled ATP hydrolysis in the presence of ATP and Figure 2. Structure of the Apc(αα′βγ) 2 core structure. The interface subunit Apcβ is drawn in blue, the potential ATP binding subunits Apcα and Apcα ′ in yellow and green and the small Apcγ in red. Apcα ′ has no contact with Apcβ or Apcγ . The two potential ATP binding sites (black circles) are more than 50 Å apart from each other. either acetophenone or HCO 3 − as sole substrates 10 . Independent phosphorylation reactions for each substrate are consistent with two separated ATP-dependent kinase sites as found in Apcα and Apcα ′ . Moreover, phosphorylations of carboxyl or hydroxyl groups are catalysed by the family members acetate or pantothenate kinase suggesting the suitability of the ASKHA modules for the chemically related reactions of Apc on ketones and bicarbonate. In analogy, the kinase reactions of Apc might proceed mechanistically by a nucleophilic attack of acetophenone in its isomeric form of an enol hydroxylate and of bicarbonate, respectively, onto the γ -phosphoryl groups of the  respective ATP (+ Mg 2+ ) cosubstrates, forming phosphoenol-acetophenone and carboxyphosphate (Fig. 1a). This mechanistic scenario corresponds to the previously deduced pathway based on kinetic data and basic chemical principles 10 .
The known ATP binding sites in kinases also define the approximate substrate positions in the vicinity of the γ -phosphates of ATP 26 . In Apcα , the potential substrate binding site resembles similarly-sized pockets in pantothenate or acetate kinases 23,27 (Fig. 5a) and is lined up by rather hydrophobic residues such as Leuα 81, Pheα 281, Asnα 282, Proα 285, Ileα 333, Glnα 344 and Thrα 345. The solvent-exposed pocket is located at the bottom of a wide hollow with a depth of 25-30 Å formed by Apcα , Apcβ and Apcγ (Fig. 6a). In Apcα ′ , the potential substrate binding site is localized in front of the β -phosphate of the bound ADP. Compared to Apcα , the predicted binding site is also solvent-exposed, coated by a higher portion of polar residues as Gluα ′ 248, Aspα ′ 274 and Lysα ′ 281 and narrower, perhaps partly due to the closed inter-domain cleft (Fig. 5b). Based on this qualitative analysis of geometric and polar properties, we tentatively assign the binding and activation of acetophenone to Apcα and that of HCO 3 − to Apcα ′ . However, an unambiguous proposal would require structures of Apc-substrates/substrate analog complexes which are not yet established.
Proposed carboxylation mechanism and a putative role for Apcε. Carboxylases are key enzymes in the biosphere because they catalyze the fixation of inorganic carbon (CO 2 ) into organic matter and extend organic molecules by forming new carbon-carbon bonds. Nature developed diverse types of carboxylases. They can be subgrouped into reductive and non-reductive enzymes, which are further subdivided into carboxylases dependent or independent on biotin 28 . Apc and Acx belong to the non-reductive biotin-independent enzymes, which also include e.g. ribulose-1,5-bisphosphate carboxylase/oxygenase (RuBisCO 29 ). Despite fundamental differences in overall architecture and active sites, these carboxylases share the basic strategy of their catalytic mechanisms consisting of the formation of an electron-rich enolate anion and an electrophilic carbon of CO 2 . Enolate intermediates are formed by ATP-independent and ATP-dependent tautomeric keto-enol shifts as catalyzed by RuBisCO 30 and Apc 10 , respectively. The proposed ATP-dependent activation of HCO 3 − in Apc is shared with biotin-dependent carboxylases. The generated carboxyphosphate might be channeled to the appropriate active site and decompose in-situ to the electrophilic CO 2 which then reacts with the substrate either indirectly via the biotin cofactor 31 or directly as in Apc 10 .
However, the apparent active sites for generating phosphoenol-acetophenone (Apcα ) and carboxyphosphate (Apcα ′ ) in the determined Apc core structure are ca. 50 Å apart from each other, which clearly precludes any direct interaction of CO 2 formed from carboxyphosphate with phosphoenol-acetophenone (Fig. 2). Therefore, we expect a large-scale snapping-type displacement of Apcα ′ relative to Apcα to occur during the reaction which forms a more closed conformation of the complex and positions the two activated substrates closer to each other. Such a rearrangement appears to be feasible due to the presence of a fixed interface consisting of the large contact area between the α /β domains of Apcα ′ (α ′ 81-α ′ 214) and Apcα , and the small contact area between the α /β domain and the exposed rest of Apcα ′ (Fig. 3a). The loose fixation of the exposed part of Apcα ′ to the residual Apc core complex (Fig. 2) is also reflected in its extraordinary flexibility (B = 145 Å 2 ). The postulated movement of the exposed Apcα ′ part into the wide hollow formed by Apcα β γ might be induced by the binding and activation of both substrates. However, due to steric hindrances the residual distance between the kinase sites of Apcα and Apcα ′ in a modelled "tightened" complex is still predicted to be around 20 Å (from modelling results treating Apcα and Apcα ′ as rigid-bodies). The apparent need of extensive conformational changes within the Apc complex during the catalytic cycle may implicate a role for the still elusive subunit Apcε . Apcε might be involved in the transitions between the "tightened" and "relaxed" forms of the Apc core complex and/or in the further bridging of the gap between the kinase sites. Consistent with kinetic and structural data we propose a first scenario for the mechanism of Apc (Fig. 6b): The reaction starts by binding acetophenone, HCO 3 − and the respective ATP molecules which induce a conformational change of the exposed part of Apcα ′ to move towards the wide hollow (Fig. 6b, 1). Subsequently, Apcε binds to the Apc core complex in the closed "tightened" state and triggers the kinase reactions (Fig. 6b, 2). As proposed for biotin-dependent carboxylases 32,33 , carboxyphosphate might be hydrolyzed on-site to the highly reactive CO 2 . For bridging the remaining gap between the active sites of Apcα and Apcα ′ , we speculate that the closed Apc complex with bound Apcε either forms a channel for carboxyphosphate or CO 2 diffusion to its reaction partner (Fig. 6b, 3), or induces additional conformational changes in the complex to provide a closer approach between the kinase sites of Apcα and Apcα ′. Now, the electrophilic CO 2 reacts with phosphoenol-acetophenone and the generated intermediate is hydrolytically dephosphorylated to the product benzoylacetate (Fig. 1) 10 . Finally, the Apc complex dissociates, releases Apcε , benzoylacetate, ADP and P i (Fig. 6b, 4), and leaves behind the "relaxed" open form of the Apc core complex (Fig. 6a). This postulated reaction cycle involves all five subunits of Apc, including Apcε which is apparently not attached to the core complex, but essential for the reaction. The required rearrangement of the core complex and the attachment/detachment cycle of Apcε may also explain the observed extreme tardiness of the reaction even with the purified and reconstituted enzyme 10 .
Compared to other carboxylases, Apc and Acx consume (at least; see below) two ATP equivalents per carboxylation event instead of one (as in biotin-dependent carboxylases) or none (as in RuBisCO). While activation of HCO 3 − to carboxyphosphate is shared by other carboxylases, activation of the respective ketones to reactive enolate states with ATP appears to be essential because of the low acidity of the methyl protons (pK a acetone : 19.2; acetophenone: 15.4) or methylene protons in case of longer substituents (pK a butanone: 14.7; propiophenone: 17.6). We assume that this expensive activation strategy was specifically developed for carboxylating such chemically inert (alkyl)ketones. In other carboxylases substrate activation is achieved by biotin/ATP, by using electron-rich substrates such as phosphoenolpyruvate 34 , by metal ion ligation, and exceptionally by a carbamoylated lysine in RuBisCO 30 . The hydantoinase/ketone carboxylase family. Microbial genome analysis reveals the existence of a large new family of proteins containing highly related α and β subunits which are annotated as putative Hyd, Acx, and Apc complexes. A. aromaticum seems to be a special case of an organism that contains one member of all three branches of this enzyme family, namely Apc 10 , Acx 12 , and a γ -lactamase involved in anaerobic indoleactate metabolism 35 . Based on sequence comparisons and the structural data on Apc, we tried to extract features which are shared or distinct between the members of the three main branches of the enzyme family.
Firstly, we investigated the organization of the encoding genes, which appears to be conserved even if more than one enzyme of the family is encoded in the genome of a bacterial strain (sFig. 3). The gene clusters encoding hydantoinases/amidohydrolases consist of only two genes for the α (HydA) and β (HydB) subunits with a conserved gene order of hydAB. Gene clusters encoding acetone carboxylases exhibit the same order of the genes for α and β subunits (acxAB), but additionally contain a third gene for the γ subunit (acxC) downstream of acxB. Finally, gene clusters encoding acetophenone carboxylase or similar enzymes contain four or five genes in the order apcABCD(E), which code for the subunits Apcα ′ , Apcγ , Apcα and Apcβ of the core complex, followed by the gene encoding the additional subunit Apcε . Note that the fifth subunit is not present in every predicted operon for an acetophenone carboxylase-like enzyme, and that the sequences of these subunits are much more divergent than any of the other subunits. Close orthologues to Apcε from A. aromaticum are only found in the gene clusters of Nevskia ramosa (43% identity) and Rubrobacter xylanophilus (26% identity).
Secondly, we constructed phylogenetic trees for the hydantoinase/ketone carboxylase family based on multiple sequence alignments 36 , which revealed a distinct clustering of the three functionally different subfamilies into three branches (sFig. 1). Moreover, several subclusters were identified in the respective branches, allowing to predict the function of so far uncharacterized enzymes. For instance, a number of predicted 2-oxoindoleacetyl-CoA γ -lactamases involved in indoleacetate metabolism 35 form a well-defined cluster within the branch of Hyd variants, and the characterised Acx variants hydrolysing either one or two ATP per acetone to AMP and P i appear to form separate subbranches 12-15 (sFig. 1).
Thirdly, an approximate distinction between Apc, Acx and Hyd is possible by the different lengths of their subunits. In particular, Acxβ consists of ca. 720 amino acids, which is about 50 more than in Apcβ and about 150 more than in Hydβ orthologues. Most of the additional amino acids are located at the N-and C-terminal ends of the subunits. Acxα orthologues are ca. 30 amino acids longer than those of Apcα and more than 100 amino acids longer than Hydα orthologues. Finally, Acxγ orthologues contain ca. 50 more amino acids in their N-terminal regions, compared to Apcγ .
Fourth, a comparative structure-based sequence analysis revealed several signature motifs for the hydantoinase/keton carboxylase family at the interface regions of subunits, at the metal binding site of subunit γ containing four conserved cysteines, at the ATP binding sites of subunits α (and α ′ ) and at the glycine-rich segments characteristic for the new protein fold in subunits β . However, a specific identification of Apc, Acx and Hyd based on structural differences i.e. of the substrate binding sites reflected in a consensus sequence failed. In particular, members of the Apc and Hyd branches are phylogenetically highly diverse. In addition, only few enzymes are biochemically characterized yet (in the case of the Apc branch only one) and an assignment of the function of more distantly related species has to be regarded with caution.
After annotating specific family members as Apc, Acx or Hyd based on items 1-3, preliminary structural models for the AcxABC and HydAB complexes can be calculated on the basis of the established structure of the Apc core complex and sequence identities between the subunits of 20-30% i.e. with the Swissprot server 37 (sFig. 3). In analogy to Apc, the structures are predicted to contain two central Acxβ or Hydβ subunits which form the dimerisation interfaces between the protomers Acxα β γ and Hydα β . The subunit equivalent to Acxα or Hydα in Apc is Apcα and definitively not Apcα ′ because Apcα ′ does not form any contact area to Apcβ . In addition, the interface regions between Apcα and Apcβ are conserved in the Acx and Hyd sequences, whereas the corresponding sequence segments in Apcα ′ deviate substantially. A fingerprint motif distinguishing Apcα and Apcα ′ is found in segment R/KERiDsxG of the α /β domain of Apcα located adjacent to the interface with Apcβ which is not conserved in Apcα ′ . In line with the X-ray structure of Apc, subunit γ is apparently not essential for the integrity of the core complexes. Because of its absence in any Hyd complex this subunit is also not required for the ATP-dependent phosphorylation of amidic carbonyl groups by Hydα β . The large distances between the subunits γ and the putative active sites of Acx or Apc make their direct participation in the carboxylation reaction unlikely either.
While the structure of the Apc core complex provides plausible insights into the functionality of the kinase-related partial reactions of the enzymes of the hydantoinase/ketone carboxylase family, the partial reactions related to carboxylation in Acx and Apc substantially differ as suggested by structural and biochemical data 10 . The structure-inspired mechanistic proposal for acetophenone carboxylation by Apc with Apcα ′ and Apcε as central components cannot be directly applied to Acx due to the absence of subunit α ′ from the core complex and the lack of the extra subunit subunit ε . Therefore, Acx cannot bind and activate the ketone and bicarbonate simultaneously in separate activation sites as Apc and thus the reaction must proceed sequentially. This prediction is consistent with the observed stepwise hydrolysis of ATP to ADP and further to AMP linked to the activation of acetone followed by that of bicarbonate in members of the Xanthobacter Acx subbranch [12][13][14][15] (Fig. 1b). In contrast to Apc, both characterized biochemical variants of Acx exhibit uncoupled ATP hydrolysis only with the ketone, but not with bicarbonate as single substrate 12 which can be correlated with a sequential reaction mechanism on a single ATP-binding subunit. The mechanistic differences of Acx vs. Apc may also correlate with the higher expected stability of the enolphosphate derivatives of aliphatic ketones compared to those of aromatic ketones, because the latter cannot easily isomerise to a conformation with a double bond extending towards the aromatic ring. Further studies are necessary to work out the intricate catalytic cycle for carboxylating these inert ketones and to unravel the astonishing mechanistic differences among ketone carboxylase reactions despite highly related architectural features and reaction types.
Scientific RepoRts | 7:39674 | DOI: 10.1038/srep39674 Methods Cultivation and purification. Cultivation and harvesting of ethylbenzene-degrading A. aromaticum cells and purification of the Apc core complex were performed as described previously 10 . Activity of the Apc holoenzyme (purified native Apc core complex substituted with recombinantly overproduced Apcε ) was confirmed by determination of the ATPase activity and carboxylation of acetophenone with [ 14 C]-bicarbonate to [ 14 C]-benzoylacetate 10 . The Apc core complex was concentrated to a solution containing 40 mg ml −1 enzyme in 10 mM Tris buffer, pH 7.5, containing 0.1 M KCl and 5 mM acetophenone and used for crystallization.
Crystallization and structure determination. Crystallization of the Apc core complex was performed by the sitting drop method at 4 °C with diverse screening solutions (JBScreen Classic 1-10, Pentaerythritol Screen, JCSG Core Suite I-IV, Morpheus Screen) using the Cartesian Honey Bee robot system. The optimized conditions are listed in Table 1. Freshly grown crystals normally diffracted to ca. 4 Å resolution but after long-term storage (up to 2 years), the resolution increased to ca 3.5 Å, possibly due to dehydration effects. After soaking with 0.2 mM mersalylic acid a single crystal diffracted to ca. 3 Å resolution. Anomalous data at the HgL III edge with the peak wavelength of 1.008 Å were collected at the beamline PXII of the Swiss-Light-Source in Villigen (Switzerland) and processed with the program XDS 38 (Table 1). The overall data quality was only satisfying to a resolution of 3.3 Å. However, spots were clearly visible at 3.0 Å resolution, although blurred in the quality parameters (Table 1) due to radiation damage and weak anisotropic diffraction properties. The space group is P6 5 22 and the cell axes are 240.1 Å and 336.6 Å, best compatible with an Apc(α α ′ β γ ) 2 heterooctamer in the asymmetric unit. The positions of three anomalously scattering Hg atoms in the asymmetric unit were detected with the programs SHELXC/D 39 . Initial phase determination was performed with SHARP 40 at 4.0 Å resolution by the SAD method and subsequent solvent flattening with SOLOMON 41 at 2.8 Å resolution. Inspection of the resulting electron density only revealed one Apcα α ′ β γ protomer in the asymmetric unit, which corresponds to a solvent content of 79%. Apcα , Apcβ and Apcγ could be partly traced automatically by PHENIX 41 and BUCCANEER 42,43 , while Apcα ′ was entirely traced manually using COOT 44 . The extraordinary high B-factor of 137 Å 2 of Apcα ′ substantially complicated the fitting of the polypeptide into the solvent-flattened electron density map, which was only feasibly because of its structural relationship to Apcα . The particularly low quality of the electron density for Apcα ′ reflects its high mobility compared to Apcα β γ which might even be higher in solution because of the absence of an observed crystal lattice contact to Apcα ′ of a neighboring complex.The resulting model is reminiscent to that of many membrane proteins in exhibiting a high temperature factor of the coordinates (B aver = 109 Å 2 ). Refinement was performed with REFMAC 45 and PHENIX 42 . Rigid body thermal displacement was taken into account by five TLS (translation, libration, screw) groups. The final R/R free factor was 20.2/24.0% in the resolution range 50.0-2.84 Å ( Table 1). The
Data availability. Atomic coordinates and structure factors for the crystal structures are deposited in the protein databank under the accession code 5L9W.