Correction: Crystal Structures of the Kinase Domain of the Sulfate-Activating Complex in Mycobacterium tuberculosis

to correct Figs. 2 and 3, which were distorted during the production process. The publisher apologizes for the error. Please download this article again to view the correct version. The originally published, uncorrected article and the re-published, corrected article are provided here for reference. open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.


Introduction
Mycobacterium tuberculosis, the agent that causes tuberculosis, is one of the most devastating human pathogens. Due to its elaborate defense and adaptation strategies the bacterium is capable of withstanding the primary bactericidal responses of the host. During the course of infection, M. tuberculosis is able to down-regulate its metabolic activity and persist for years in the infected lung tissue until it is reactivated and exits the dormant phase [1]. Approximately one third of the world's population is infected by dormant M. tuberculosis [2] which constitutes a threatening reservoir for new infections. Treatment of tuberculosis requires an unusually long chemotherapy, where not the least lack of patient compliance increases the occurrence of multidrug-resistant (MDR) or extremely drug-resistant (XDR) strains of M. tuberculosis. The elucidation of the molecular events and mechanisms that enable the bacterium to survive the conditions thought to prevail in granulomas during dormancy, i.e. nutrient starvation, hypoxia and oxidative and cell wall stress, is crucial to understand the disease and identify new therapeutic targets for chemotherapy.
There is a growing body of evidence that biosynthesis of sulfur-containing molecules is upregulated in various models of the dormant state [3][4][5]. Sulfate assimilation in M. tuberculosis is initiated by active import of sulfate and the metabolic fate of the sulfate is to a large extent determined by the sulfate-activating complex [6,7] (Fig. 1). This complex consists of two polypeptide chains, a sulfurylase (CysD) and a GTPase (CysN). The CysN polypeptide also contains a C-terminal APS (adenosine-5'-phosphosulfate) kinase domain (CysC), which exists as a separate enzyme in other bacteria and plants. The imported sulfate is adenylated by the ATP sulfurylase (CysD) and the necessary energy is provided by GTP hydrolysis catalyzed by the GTPase activity of CysN [7][8][9]. In the reductive branch of sulfur assimilation the resulting product, adenosine 5 0 -phosposulfate (APS), is then reduced to sulfite by APS reductase (CysH) for the biosynthesis of cysteine [10,11]. Alternatively, APS can be phosphorylated at the 3 0 -position by APS kinase (CysC) to generate 3 0 -phosphoadenosine 5 0 -phosphosulfate (PAPS), a substrate for sulfotransferases that catalyze the transfer of the sulfate group onto a variety of metabolites doi:10.1371/journal.pone.0121494.g001 [12]. Collectively, these reactions constitute the sulfation branch of the sulfate assimilation pathway. PAPS is an essential sulfate donor in sulfolipid biosynthesis [13] and the sulfolipid SL-1 has been linked to virulence in M. tuberculosis [14].
The sulfate-activating complex of M. tuberculosis constitutes a metabolic branch point and the regulatory mechanisms that direct APS either to the reductive or the sulfation branch are not completely understood [6][7][8]. The role of mycobacterial APS kinase in converting APS to PAPS at this branch point of sulfur metabolism has triggered interest in this enzyme as a potential drug target [15]. In M. tuberculosis the APS kinase domain (CysC, residues 424-614 of CysN) is fused to the C-terminus of the GTPase domain of CysN and shares sequence identities with homologous APS kinases, for instance 52% identity with the APS kinase domain of human bifunctional PAPS synthetase 1 (PAPSS1), 50% with E. coli CysC and 46% with the APS kinase from P. chrysogenum (Fig. 2).
As part of a study to characterize mycobacterial CysC as a potential drug target we have determined the crystal structure of the enzyme as a binary complex with ADP, and as ternary complexes CysC•ADP•APS and CysC•AMP-PNP•APS. The crystal structures allowed a detailed comparison of the binding sites with that of other APS kinases, in particular the human homologues hPAPSS 1 [16,17] and hPAPSS 2 (Structural Genomics Consortium, 2AX4) to evaluate the possibility of the design of inhibitors specific for mycobacterial CysC. Site-directed mutagenesis revealed a cysteine residue that plays an important role in ligand-dependent closure of the lid domain and is critical for enzymatic activity.

Structure determination and model quality
The structure of CysC from M. tuberculosis was determined in complex with different nucleotide combinations: ADP, ADP/APS, and AMP-PNP/APS. The ADP complex crystallized in space group P2 1 with two molecules in the asymmetric unit. Initial phases were obtained by molecular replacement and the model was refined to 1.47 Å resolution resulting in a final R-factor of 15.8% and R free of 18.1% (Table 1). In both subunits, the 18 N-terminal residues were not defined in the electron density map, indicating disorder in this inter-domain linker region. For both chains in the asymmetric unit the electron density is continuous from Pro440 to the last residue present in the construct (Ser612) except for the two surface residues Ile573 and Thr574. For these residues only weak electron density was found, indicating partial disorder for this peptide stretch. The electron density maps clearly show the presence of one molecule of ADP per subunit, but the maps suggested that the ligand is bound in two slightly different conformations (Fig. 3).
Both ternary CysC complexes (ADP•APS and AMP-PNP•APS) crystallized in space group P2 1 2 1 2 1 with two subunits in the asymmetric unit. The structures of these complexes were determined by molecular replacement and refined to R-/R free factors of 19.6/25.3% (ADP•APS) and 17.4/21.6% (AMP-PNP•APS) to resolutions of 2.1 and 1.7 Å, respectively. As for the ADP structure, the 18 N-terminal residues are disordered in the ternary complexes, whereas continuous electron density for the remaining part of the polypeptide chains was found. The difference electron density for the bound nucleotides in the two ternary complexes allowed an unambiguous modeling of the ligands (Fig. 3). The electron density maps also showed a citrate molecule from the buffer solution bound at the surface of each subunit in both ternary complexes. Data collection and refinement statistics are presented in Table 1.

Structure of the CysC•ADP binary complex
Overall structure of the subunit. The structure of CysC is characterized by the canonical α/β-purine nucleotide binding fold, consisting of a five-stranded parallel β-sheet which is flanked on both sides by two α-helices (Fig. 4). Additional secondary structure elements in CysC are one short α-helix in the loop between β2 and α2 and two helices in the loop between β4 and β5. Structural features typical of APS kinases of this fold family [16][17][18][19][20][21], including the P-loop required for ATP binding (residues 452-458 in CysC), the conserved LDGD motif (residues 477-480) and the lid region (residues 552-581) above the P-loop which closes upon substrate binding, are also found in CysC (Figs. 2 & 4). , the kinase domain of human PAPS synthase 2 (PAPSS2, 2ax4) and APS kinases from A. thaliana (4fxp), P. chrysogenum (1dj6), T. denitrificans (3cr8), A. aeolicus (2gks), and A. pernix (2yvu). Note that the alignment is shown for the APS kinase domains only in case of bifunctional enzymes. First residues included in the alignment are numbered for each sequence. White font with red background highlights identical residues and the black font with yellow background indicates similar residues. Residues Cys514, Cys549 and Cys556 of M. tuberculosis are highlighted by red boxes and marked with red triangles. Secondary structure elements for CysC from M. tuberculosis are shown on top. The figure was made using ESPript [40]. Quaternary structure. The asymmetric unit of the crystals of the CysC•ADP binary complex contains two subunits, which are related by a two-fold molecular symmetry axis. Superimposition of the two polypeptide chains gives a root mean square deviation (r.m.s.d.) between 173 equivalent Cα atoms of 0.3 Å. Manual inspection and analysis by the PISA server [22] suggests a dimeric quaternary structure with a buried surface area of 3300 Å 2 , with the dominant part of the interface formed by interactions of residues from helices α2 and α3 with their Values in parentheses are for the highest-resolution shell.
doi:10.1371/journal.pone.0121494.t001 symmetry mates (Fig. 4). The same quaternary structure is also found in the orthorhombic space group of the ternary complexes (see below). Analytical gel filtration experiments (S1 Fig.) show that CysC also forms a dimer in solution, consistent with the crystallographic analysis. Relation to structural homologues. A search of the PDB using the coordinates of CysC reveals the APS kinases from Penicillum chrysogenum [18], Aeropyrum pernix (RIKEN Structural Genomics/Proteomics Initiative, 2YVU), Thiobacillus denitrificans [21], Aquifex aeolicus [19], Arabidopisis thaliana [20] and the human PAPPS1 [16,17] as the closest structural homologues. Superimposition of these structures with the coordinates of CysC using the SSM option [23] (Krissinel & Henrick, 2004) in COOT [24] resulted in rmsd values in the range of 0.9-1.5 Å ( Table 2).  Structural superimpositions were carried out using the SSM option [23] in Coot [24]. doi:10.1371/journal.pone.0121494.t002 The ADP binding site in CysC. The ADP binding site in CysC is located in a crevice formed by the loop between β1 and α1 (residues 449-456), the loop 551-562 including helix α5 and the loop region 593-599 (Fig. 4). In each of the subunits the bound nucleotide was modeled in two different conformations that accounted best for the electron density map (Fig. 3a). The differences between both conformations are small; the adenine rings are slightly shifted with respect to each other due to a rotation of the glycosidic bond. The largest difference is observed for the α-phosphate which is shifted in the two conformations by about 1 Å.
ADP is bound to CysC in the anti conformation, similar to ADP binding in APS kinase from P. chrysogenum [18], but different from the situation in human PAPSS1 where both the syn and anti conformation were observed [16]. The adenine ring of ADP is sandwiched between the guanidinium group of Arg559 and the side chain of Ile599 (Fig. 3b). The π-stacking interaction with the guanidinium group of a conserved arginine residue is observed in other complexes of APS kinases [16,17,19,20,25]. The NH 2 group of adenine forms hydrogen bonds to the backbone carbonyl group of Arg597 and the side chain of Gln602. The ring nitrogen atom N7 also forms an indirect hydrogen bond to the side chain of Gln602 via a water molecule. The remaining parts of the ring system are accessible to solvent and the N1 and N3 nitrogen atoms interact with ordered water molecules. The ribose moiety is also accessible from the solvent region and forms hydrogen bonds to solvent molecules.
The diphosphate group is tightly anchored to the protein via the conserved P-loop at the Nterminus of helix α1 and the β phosphate interacts with main chain nitrogen atoms of residues 453-458. In addition the side chains of Lys456 and Ser458 form hydrogen bonds to oxygen atoms of the α and β-phosphate group, respectively.

Structure of the CysC•ADP•APS ternary complex
Steady-state kinetics have shown that APS kinases follow a sequential ordered mechanism in which MgATP binds before APS, and PAPS leaves before MgADP [26]. Excess of ATP and APS leads to the formation of a catalytically inactive ternary E•MgADP•APS complex due to binding of APS right after PAPS is released and before MgADP leaves the enzyme active site. To obtain more insights into the active site of CysC from M. tuberculosis and modes of ligand binding, we co-crystallized CysC in the presence of ADP, APS and Mg 2+ and determined the structure of this complex.
The overall structure of the CysC protein core and the subunit-subunit interface in the ternary complex CysC•ADP•APS is similar to the structure of the binary complex CysC•ADP. Superimposition of the individual subunits resulted in an rmsd of 0.9 Å. This relatively high value is, however, due to the shift of one loop in response to APS binding. This loop, comprising residues 477-499, moves towards the core of the enzyme and closes off the APS binding site (Fig. 5).
In the ternary complex ADP is bound in a well-defined conformation in the canonical nucleotide binding site in the vicinity of the P-loop and the enzyme-ADP interactions are well preserved compared to the binary complex ( Fig. 3c and d). An additional hydrogen bond from the N1 atom of the adenine ring to the backbone carbonyl group of Arg597 is found in this complex. We note, however, that in spite of the presence of 2 mM Mg 2+ during crystallization no electron density indicating a bound Mg 2+ ion is observed in the vicinity of the diphosphate group.
APS binding site. The APS binding site is located opposite to the ADP binding site (Fig. 4). The bound APS in the ternary complex is well defined (Fig. 3c and d). The ribose moiety is observed in the 3'-endo conformation and located in the active site pocket. The adenine ring, in syn conformation, points into a pocket formed by loops 490-495 and 562-582. The APS phosphate group rides on the C-terminal end of strands β2 and β3 of the β-sheet.
The APS adenine ring is sandwiched between Phe492 and Phe576 in a π-stacking arrangement, an interaction conserved in other complexes of APS kinase [16,20,25]. Phe492 is part of the loop that shifts upon APS binding and shows the largest main chain displacement of 2.8 Å. The adenine ring forms several hydrogen bonds with protein residues: the amino group interacts with the carbonyl oxygen atom of His575 and the N1 ring nitrogen atom with the side chain of Arg497 ( Fig. 3c and d). The ribose moiety is anchored to the protein through hydrogen bonds. The 2'OH group interacts with the side chains of the conserved residues Asp480 and Lys562, whereas the 3'OH group, which is phosphorylated during catalysis, is linked via a water molecule to the side chain of Lys456.
The phosphosulfate group of APS is tightly anchored by several hydrogen bonds and salt bridges. The phosphate group interacts with the side chain of Arg483 and the main chain nitrogen atom of Ile523 and the sulfate group forms hydrogen bonds to the side chains of Arg483 and Arg497 and the main chain nitrogen atom of Ser524. The two arginine residues are conserved (Fig. 2) and interact with the phosphosulfate moiety of APS also in the ligand complexes of other APS kinases [16,20,25].

A mimic of the Michaelis complex-the structure of CysC•AMP-PNP•APS
The structure of CysC with the non-hydrolyzable ATP analogue AMP-PNP, Mg 2+ , and APS bound in the active site provides a view of the Michaelis complex in the reaction sequence. A superimposition of the subunit structure of this complex with that of the binary complex CysC•ADP, rmsd 0.9 Å, shows the same conformational change as observed in the ternary CysC•ADP•APS complex (Fig. 5). The structure of the enzyme in the two ternary complexes is thus essentially identical, which is reflected in the r.m.s.d. value obtained upon superposition of 0.14 Å. The bound ligands AMP-PNP and APS are well defined in electron density (Fig. 3e) and their interactions with CysC are, with the exception of the γ-phosphate group of AMP-PNP, essentially identical to those observed in the CysC•ADP•APS complex (Fig. 3f). The γphosphate extends from the ADP moiety towards the ribose group of APS, and the phosphorous atom is about 3.4 Å away from the 3' hydroxyl oxygen, the acceptor of the phosphoryl group. This moiety interacts with the side chain of Lys456 from the P-loop and Lys562 from the lid region. Lys456 also forms a hydrogen bond to the neighboring β-phosphate. An important contribution to the binding of AMP-PNP is the Mg 2+ ion that is bound in octahedral coordination geometry in the active site. The ligand sphere of this ion consists of one oxygen atom each from the βand γ-phosphate groups, the side chain oxygen of Ser457 and three water molecules (metal-ligand distances 2.0-2.2 Å). Finally, the bridging nitrogen atom of the AMP-PNP forms a hydrogen bond with the main chain nitrogen atom of Gly453 (Fig. 3f).
In the structure of the CysC•AMP-PNP•APS complex, helix α5 is shifted by about 2.2 Å closer to the nucleotide binding site compared to the structure of the CysC•ADP complex (Fig. 5). At the N-terminus of this helix a stretch of invariant residues, Leu-Asp-Gly-Asp (residues 477-480) is found that form part of the active site. The side chain of Asp478 forms a hydrogen bond to a water molecule that is ligand of the Mg 2+ ion. The second conserved aspartate residue, Asp480, is located in the vicinity (3.6 Å) of the O3' hydroxyl group of APS. A similar rearrangement of this conserved loop region has been observed in the APS kinase domain of hPAPSS1 [17] and the APS kinase from A. thaliana [20]. It had been suggested that this conformational change occurs in response to binding of the magnesium ion [17,18]. However, we observe this shift in position of helix α5 already in the CysC•ADP•APS complex, i.e. in the absence of Mg 2+ , implying that APS binding rather than binding of the metal ion triggers this structural re-arrangement in CysC.

Implications for catalysis by CysC
As a mimic of the CysC Michaelis complex the structure of the enzyme with bound AMP-PNP and APS provides a close view of the catalytic machinery and its interactions with the substrate (S2 Fig.). The γ-phosphate group of the ATP analogue and the 3' hydroxyl group of APS accepting the phosphate are well aligned with a distance of 3.4 Å between the phosphorous and the hydroxyl oxygen atom. The P-loop and the metal ion stabilize the negative charges of the triphosphate moiety of ATP and orient the substrates for phosphoryl transfer. In the transition state an increase of negative charge occurs at the γ-phosphate and the side chain of the conserved residue Lys562 is perfectly positioned to facilitate catalysis by stabilizing this charge as it is located between the γ-phosphate and the acceptor oxygen atom of the 3' hydroxyl group of APS. In addition the conserved side chain of Asp480, close to the 3' hydroxyl group of the acceptor substrate APS, could act as a catalytic base in deprotonation of the O3' hydroxyl group (S2 Fig.). Site-directed mutagenesis of the corresponding residue in mouse APS kinase demonstrated its essentiality for catalytic activity [27]. This scenario is in line with previous mechanistic proposals for APS kinases from fungi [25], plants [20] and humans [17] suggesting that the mechanism of this enzyme is preserved also in bacteria.

Comparison of CysC to human PAPS synthase-is there a basis for inhibitor selectivity?
Inhibition of CysC would prevent formation of PAPS, leading to a decrease in the biosynthesis of sulfolipids in M. tuberculosis [6]. However, the related human PAPS synthases contain an APS kinase domain that has relatively high sequence identity (52%) to CysC. Mutations in the PAPSS2 gene have been linked to diseases in both mice and humans [28,29] suggesting that in mammalian cells, functional PAPSS is essential. The question thus arises whether or not the ligand binding sites in the two enzymes are sufficiently distinct to provide a basis for the design of selective inhibitors.
In humans two isoforms of PAPS synthase are found that share 78% sequence identity [30]. In the following analysis the ternary structures of the two isoenzymes, hPAPSS 1 with bound ADP and PAPS (2OFX) [16] and hPAPSS 2 with bound ADP (Structural Genomics Consortium, 2AX4), were compared to the CysC•AMP-PNP•APS complex. The composition of the APS/PAPS and ADP/ATP binding clefts in these enzymes was compared as these sites provide obvious targets for inhibitors of enzyme activity (Fig. 6). The APS binding sites in all three enzymes are highly similar. Of the 15 amino acids that line this cleft and interact with the bound ligand (3.8 Å cutoff), 12 are conserved in the human and mycobacterial enzymes. Two differences are found in a loop closing off the adenosine binding site where Thr574 and His575 of CysC are replaced by Lys183 and Gly184 in hPAPS synthase. Only the backbone carbonyl oxygen of Lys183 is directly involved in the interaction with bound APS, otherwise the side chains are exposed to solvent, away from the bound ligand. Even though there is a large size change from Ala522 in CysC to Phe131 in hPAPSS, this results only in a marginal decrease of active site pocket volume. Overall the high degree of sequence identity and active site topology make it unlikely that specific inhibitors targeting the APS binding site in CysC can be designed.
A similar comparison of the ATP binding sites in the two enzymes reveals larger differences in amino acid composition (Fig. 6). This site is formed by 14 residues that interact with bound ADP and AMP-PNP, respectively. Of these residues six are conserved, five are conservatively substituted and three show a complete change in size and chemical properties. These three residues are located in the adenosine binding pocket and involve changes (human-> CysC) Val170 -> Pro561, Cys/Ser207 -> Arg597 and Cys212 -> Gln602 (Fig. 6). The substitutions result in minor differences in hydrophobicity and size of the adenosine binding pocket and hence might be difficult to exploit for the design of specific inhibitors of the mycobacterial enzyme. Cys556 is essential for enzymatic activity In M. tuberculosis CysC is fused to a GTPase in a single polypeptide (CysN) that is part of the sulfate-activating complex [7,31]. The CysC domain used in this study is catalytically active and shows APS kinase activity (Fig. 7). We confirmed that the observed reaction was not due to an ATPase activity of the enzyme by incubating the enzyme with ATP in the absence of APS. Under these conditions no signal for ADP formation was observed.
In the course of this work we noted that the APS kinase activity of CysC is increased by the addition of reductants such as DDT, glutathione and tris(2-carboxyethyl)phosphine (TCEP). Activation by reducing agents has been reported earlier for the related APS kinases from A. thaliana and A. aeolicus, where disulfide formation has been implicated in enzyme regulation [19,20]. CysC contains three cysteine residues, Cys514, Cys549 and Cys556, and we therefore examined their role in CysC activation. Examination of the electron density maps in the vicinity of these cysteine residues did, however, not show signs of disulfide formation or oxidation of the sulfhydryl groups to their corresponding sulfenic/sulfinic acids.
One of the cysteine residues, Cys514, is located at the subunit interface and is in the vicinity (distance between thiols groups 7 Å) of its symmetry mate from the second subunit. Replacement of Cys514 by alanine results in a mutant enzyme that still displays the activation behaviour as wild-type CysC (Fig. 7). The same observation is made upon replacement of Cys549, located at the C-terminal end of strand β4. The Cys556Ala mutant also displayed activation upon addition of DTT, but to a lesser extent. The redox sensitivity of CysC can thus not be attributed to a single cysteine residue, but might reflect partial oxidation of two or three thiol groups to various extents.
Surprisingly amino acid replacement at position 556 resulted in mutants with severely impaired catalytic activity. The Cys556Ala mutant showed little activity upon treatment with DTT, whereas the conservative replacement of Cys556 by serine resulted in complete loss of enzymatic activity, which could not be rescued by the addition of DTT. The inactive Cys556Ser mutant showed increased sensitivity towards thrombin in the presence of ATP, comparable to the proteolytic susceptibility of the wild-type enzyme in absence of nucleotide (Fig. 8). Overall, the crystal structure of this mutant shows little alteration compared to the wild-type, as indicated by the r.m.s.d. of 0.5 Å over 143 equivalent Cα atoms after superposition with the crystal structure of the binary complex of wild-type enzyme (S3 Fig.). The crystals of the mutant were obtained under the same conditions as those of the wild-type CysC•ADP complex, however no bound ADP was observed in any of the subunits. During structure refinement it became evident that an approximately 30 residue long segment (residues 552-581) in both chains of the asymmetric unit lack electron density and cannot be modeled, possibly due to loop disorder. This peptide stretch corresponds to the lid domain that closes over the active site cleft upon binding of the nucleotide ADP or AMP-PNP, respectively. Loop closure leads to positioning of the side chain of Arg559 close to the adenine ring of the bound nucleotide ADP or ATP, an enzyme-substrate interaction conserved in APS kinases (Fig. 2). The mutagenesis and structural data thus suggest that the loss of enzyme activity in the Cys556Ser mutant is caused by deficiencies in binding of the co-substrate ATP, due to increased flexibility of the lid and accompanying loss of the Arg559-ATP interaction. Inactivation of the kinase domain under oxidative stress conditions in macrophages through oxidation of the cysteine residues might provide a redox switch at this branching point of sulphur metabolism, but this hypothesis would require further experimental evidence.

Conclusions
The present study highlights several major features in CysC from M. tuberculosis. The structures of the binary and ternary complexes provide the first insights into the enzyme-ligand interactions of APS kinase from a major human pathogen. The comparison to the human homologues PAPSS1 and PAPSS2 suggest that the ATP and APS binding sites are highly conserved, questioning the possibility to develop selective inhibitors of the mycobacterial enzyme. The mutagenesis results highlighted the role of Cys556 as an essential residue in proper function of the lid region of the enzyme. Lid closure is essential to enzymatic activity and disturbance of the packing interactions of the lid with the protein core, even by a conservative cysteine to serine substitution, leads to the loss of enzymatic activity.

Cloning and mutagenesis
The coding sequence for full length ATP sulfurylase CysDN (Rv1285-Rv1286) (provided by Mahavir Singh, LIONEX GmbH, Braunschweig) was cloned into the pET28a expression vector (Novagen) with upstream NdeI and downstream HindIII sites. This results in pET-His 6 -CysDN that carries a thrombin cleavable hexahistidine tag on the N-terminus of CysD. The coding sequence for the APS kinase domain of CysN (Rv1286, residues Ser424 to Ser612) was amplified by PCR using Pfu-Turbo DNA polymerase (Stratagene), appropriate primers and the pET-His 6 -CysDN plasmid as template, and cloned into pET28a as a NdeI-HindIII fragment. This expression construct of the APS kinase domain, denoted as CysC, carries an N-terminal thrombin cleavable His 6 -tag and lacks the two C-terminal serine residues.
Mutant variants of CysC were produced using the QuikChange approach, Pfu-Turbo DNA polymerase (Stratagene) and appropriate mutagenesis primers to introduce the Cys514Ala, Cys549Ala, Cys556Ala, and Cys556Ser mutations. The triple cysteine mutant CysCΔΔΔ was produced from the single cysteine mutant Cys514Ala using the megaprimer method [32]. All constructs were confirmed by DNA sequencing.

Protein production
Recombinant CysC was produced in E. coli BL21(DE3). An overnight culture was inoculated 1:2000 (v/v) into 0.75 L of LB medium supplemented with kanamycin (30 μg/mL). Gene expression was induced by adding 0.1 mM isopropyl β-thiogalactoside (IPTG) to mid-exponentialphase cultures growing at 20°C. The cells were harvested 24 h later, the pellet suspended in buffer (10 mM Tris, pH 8.0, 300 mM NaCl, 10 mM imidazole) at 4°C and disrupted by sonication in the presence of DNase I (Roche) and a protease inhibitor cocktail (Roche). The suspension was centrifuged for 30 min at 20000g and the supernatant incubated with 1 mL of Ni 2+ -nitrilotriacetate (NTA) resin (Qiagen) on a rocking shaker for 1 h at 4°C. The resin suspension was filled into a column, washed extensively with binding buffer (10 mM Tris, pH 8.0, 300 mM NaCl, 10 mM imidazole) and CysC was eluted with elution buffer (10 mM Tris, pH 8.0, 300 mM NaCl, 200 mM imidazole). Fractions containing CysC were combined and concentrated with a Centriprep Ultracel concentrator with a 10 kDa molecular weight cut-off (Millipore Ldt, Ireland). The solution was loaded on a PD10 desalting column (GE Healthcare) and CysC was eluted with buffer (25 mM Tris, pH 8.0, 150 mM NaCl). For crystallisation removal of the His tag proved to be essential. In the absence of nucleotide ligands thrombin digestion for tag removal led to extensive unspecific cleavage of CysC. Therefore ADP or ATP (final concentration: 5 mM) was added to the protein solution prior to tag removal to reduce nonspecific cleavage (Fig. 8). 5-10 U of thrombin per mg of fusion protein were added and the digestion mix was incubated for 1 h at 20°C. Processed and uncut proteins were separated by a second Ni-NTA chromatography step whereby cleaved CysC was collected in the flow-through. As final purification step, the protein was subjected to gel filtration chromatography using a Superdex 200 column (GE Healthcare) equilibrated with buffer (25 mM Tris, pH 8.0, 150 mM NaCl). Elution fractions containing pure CysC according to SDS-PAGE analysis were combined and concentrated. ESI-MS data confirmed the integrity of the protein preparations with a protein mass of 21259.7 Da, which is in agreement with the theoretical mass derived from the sequence (21261 Da). To remove bound ADP, the protein solution was dialyzed overnight against excess of gel filtration buffer at 4°C (Pierce, 10 kDa cutoff, 3 mL). ADP-free CysC was concentrated to 27.5 mg/mL, flash-frozen in liquid nitrogen and stored at -80°C until further use. The purifications of the mutant enzymes were carried out as described for wild-type CysC. Since the His-tag did not interfere with the enzyme assay, wild-type and mutant samples prepared for this purpose were not treated with thrombin and the purification consisted of affinity chromatography followed by gel filtration.

Crystallization and Structure Determination
Crystals of the CysC•ADP complex were obtained at 20°C using vapour diffusion. Droplets of an enzyme solution (27.5 mg/mL, pre-incubated with 5 mM ADP) were mixed in a 1:1 (v:v) ratio with crystallization buffer (0.2 M NH 4 NO 3 , 20% (w/v) PEG3350) and equilibrated against the same buffer solution. For data collection, crystals were transferred to cryoprotectant (mother liquor containing ADP and 25% (v/v) ethylene glycol) and then frozen in liquid nitrogen. Crystals of the CysC•ADP•APS complex were obtained by pre-incubating CysC at 27.5 mg/mL with 2.3 mM ADP, 1.9 mM APS and 1.8 mM MgCl 2 . The enzyme solution was mixed in a 2:1 (v:v) ratio with buffer containing 0.1 mM citric acid, pH 5.5, 20% (w/v) PEG 4000 and 10% (v/ v) 2-propanol and the droplets were equilibrated against 500 μL of the same buffer. Crystals of the CysC•AMP-PNP•APS complex were obtained under the same conditions, except that ADP was replaced by 2.3 mM AMP-PNP. Crystals of the Cys556Ser mutant were obtained at 20°C by mixing 1 μL protein solution (40 mg/mL, 5 mM MgCl 2 and 5 mM ADP) and 1 μL of the well solution (200 mM NH 4 NO 3 , 21% PEG3350). Crystals were cryoprotected by dipping them in a solution of 200 mM NH 4 NO 3 , 21% PEG3350, 5 mM ADP, 5 mM MgCl 2 , 25 mM Tris-HCl, pH 8.0, 150 mM NaCl, 22.5% ethyleneglycol) for 5-10 seconds and subsequent flash-freezing in liquid nitrogen.
The structure of CysC•ADP was determined by molecular replacement with PHASER [35] using chain A of the human PAPS synthetase 1 (PAPSS1) structure (2PEYA) as search model [17]. Crystallographic refinement consisted of iterative rounds of refinement with REFMAC5 [36] (Murshudov et al., 1997) and model building using COOT [24]. Local NCS restraints were used in the refinement of all structures.
The crystal structures for both ternary complexes and the Cys556Ser mutant were determined by molecular replacement using the refined coordinates for the CysC•ADP complex as search model, with the ADP molecule removed from the model. Refinement followed the same procedure as described above, except that for the ternary complexes TLS refinement was included. Structure refinement was carried out by manual adjustments in COOT interspersed by minimization in Refmac5. Water molecules were added using COOT or PHENIX [37]. Structure validation was performed with MolProbity [38]. Structure figures were prepared with PyMOL and CCP4MG [39]. The atomic coordinates and structure factors amplitudes reported (chain A) and grey (chain B). The ligand binding motifs such as the P-loop (orange) and the conserved LDGD motif (purple) are highlighted in chain A. The backbone of the ADP complex of wild-type CysC is shown as ribbon in blue (chain A) and yellow (chain B) and the ADP molecule is shown as stick model. (PDF)