Structural and kinetic studies on RosA, the enzyme catalysing the methylation of 8‐demethyl‐8‐amino‐d‐riboflavin to the antibiotic roseoflavin

N,N‐8‐demethyl‐8‐amino‐d‐riboflavin dimethyltransferase (RosA) catalyses the final dimethylation of 8‐demethyl‐8‐amino‐d‐riboflavin (AF) to the antibiotic roseoflavin (RoF) in Streptomyces davawensis. In the present study, we solved the X‐ray structure of RosA, and determined the binding properties of substrates and products. Moreover, we used steady‐state and rapid reaction kinetic studies to obtain detailed information on the reaction mechanism. The structure of RosA was found to be similar to that of previously described S‐adenosylmethionine (SAM)‐dependent methyltransferases, featuring two domains: a mainly α‐helical ‘orthogonal bundle’ and a Rossmann‐like domain (α/β twisted open sheet). Bioinformatics studies and molecular modelling enabled us to predict the potential SAM and AF binding sites in RosA, suggesting that both substrates, AF and SAM, bind independently to their respective binding pocket. This finding was confirmed by kinetic experiments that demonstrated a random‐order ‘bi‐bi’ reaction mechanism. Furthermore, we determined the dissociation constants for substrates and products by either isothermal titration calorimetry or UV/Vis absorption spectroscopy, revealing that both products, RoF and S‐adenosylhomocysteine (SAH), bind more tightly to RosA compared with the substrates, AF and SAM. This suggests that RosA may contribute to roseoflavin resistance in S. davawensis. The tighter binding of products is also reflected by the results of inhibition experiments, in which RoF and SAH behave as competitive inhibitors for AF and SAM, respectively. We also showed that formation of a ternary complex of RosA, RoF and SAH (or SAM) leads to drastic spectral changes that are indicative of a hydrophobic environment. Database Structural data are available in the Protein Data Bank under accession number 4D7K.

N,N-8-demethyl-8-amino-D-riboflavin dimethyltransferase (RosA) catalyses the final dimethylation of 8-demethyl-8-amino-D-riboflavin (AF) to the antibiotic roseoflavin (RoF) in Streptomyces davawensis. In the present study, we solved the X-ray structure of RosA, and determined the binding properties of substrates and products. Moreover, we used steady-state and rapid reaction kinetic studies to obtain detailed information on the reaction mechanism. The structure of RosA was found to be similar to that of previously described S-adenosylmethionine (SAM)-dependent methyltransferases, featuring two domains: a mainly a-helical 'orthogonal bundle' and a Rossmann-like domain (a/b twisted open sheet). Bioinformatics studies and molecular modelling enabled us to predict the potential SAM and AF binding sites in RosA, suggesting that both substrates, AF and SAM, bind independently to their respective binding pocket. This finding was confirmed by kinetic experiments that demonstrated a random-order 'bi-bi' reaction mechanism. Furthermore, we determined the dissociation constants for substrates and products by either isothermal titration calorimetry or UV/Vis absorption spectroscopy, revealing that both products, RoF and S-adenosylhomocysteine (SAH), bind more tightly to RosA compared with the substrates, AF and SAM. This suggests that RosA may contribute to roseoflavin resistance in S. davawensis. The tighter binding of products is also reflected by the results of inhibition experiments, in which RoF and SAH behave as competitive inhibitors for AF and SAM, respectively. We also showed that formation of a ternary complex of RosA, RoF and SAH (or SAM) leads to drastic spectral changes that are indicative of a hydrophobic environment.

Introduction
The Gram-positive soil bacteria Streptomyces davawensis and Streptomyces cinnabarinus produce roseoflavin (RoF), the only known natural riboflavin (vitamin B 2 ) analogue with antibiotic activity. In contrast to most other antibiotics, RoF has multiple cellular targets and negatively affects flavoproteins as well as FMN riboswitches [1][2][3][4][5][6][7][8]. RoF may be considered a natural antimetabolite, and was postulated to be biosynthesized from riboflavin through 8-demethyl-8-amino-D-riboflavin (AF) and 8-demethyl-8-methylamino-riboflavin (MAF) [9,10]. The occurrence of the intermediates AF and MAF was confirmed by the discovery of the first enzyme of RoF biosynthesis, the S-adenosylmethionine (SAM)-dependent dimethyltransferase N,N-8-demethyl-8-amino-D-riboflavin dimethyltransferase (RosA) [11]. This enzyme synthesizes RoF from AF (via MAF) in two consecutive methylation reactions, as shown in Scheme 1. The corresponding gene (rosA) is present in a cluster comprising ten genes. The remaining genes of this cluster were found not to be involved in RoF synthesis, and it is presently unclear which genes/enzymes are responsible for synthesis of the key intermediate AF [12].
The important role of RosA for completion of RoF biosynthesis prompted us to determine the threedimensional structure of this enzyme and to investigate the mechanism of the consecutive methylation reactions in more detail. We show that RosA has a topology similar to other SAM-dependent methyltransferases, containing a mainly a-helical 'orthogonal bundle domain' and a Rossmann-like fold. Determination of the three-dimensional structure also enabled us to locate the binding sites for the substrates AF and SAM. In addition, we performed steady-state and rapid reaction studies, and determined the dissociation constants for the substrates of RosA as well as for the products. It was found that AF and SAM bind independently to RosA according to a random-order mechanism. The products of the RosA reaction, RoF and SAH, exhibit higher affinity for the protein than the substrates AF and SAM do, leading to competitive product inhibition. RosA is a comparably slow enzyme, with a turnover rate (k cat ) of 0.06 min À1 for the overall dimethylation reaction. As RoF is a more potent antibiotic than AF and MAF, the slow synthesis of RoF and the tight binding of products may contribute to protection of the strains S. davawensis and S. cinnabarinus from the antibiotic that they produce.

Results
Crystal structure of RosA Initial attempts to solve the structure of RosA by molecular replacement were unsuccessful despite a thorough search for appropriate template structures using the online tools PHYRE2 [13], FUGUE [14] and CASPR [15]. Therefore, RosA labelled with selenomethionine (SeMet) (seven methionine residues in 353 amino acids) was produced, and the structure was solved by multiple-wavelength anomalous dispersion (MAD) using a single tetragonal crystal ( Table 1). The structure (one protomer in the asymmetric unit) was partially refined to a resolution of 3. 5 A, and this model was then used to solve the structure of a triclinic crystal form (space group P1) by molecular replacement. This crystal contained six molecules per asymmetric unit, and the structure was refined at a resolution of 2.2 A, yielding final values of R = 20% and R free = 25% ( Table 2).
The RosA protomer comprises two domains: a mainly a-helical 'orthogonal bundle domain' and a Rossmann-like domain (a/b twisted open sheet). The orthogonal bundle (five a-helices and one antiparallel b-sheet) consists of residues 1-98. Residues 99-178 form an inter-domain region (five helices) that connects the Rossmann motif to the orthogonal bundle. The Rossmann motif (residues 179-353) comprises a seven-stranded b-sheet core consisting of five parallel and two antiparallel b-strands, which are connected by two pairs of a-helices (Fig. 1A). To classify the domains, we performed a CATH database (http:// www.cathdb.info/) search using the CATHEDRAL algorithm [16] A PISA analysis [17] predicted that a dimer is the most probable oligomer of RosA in solution. This prediction was confirmed by size exclusion chromatography (Fig. 2). Dimerization occurs in a head-to-head arrangement (Fig. 1B), and is mediated by interactions of adjacent a-helices of the bundle motif (residues 1-36 and residues 60-73), the inter-domain a-helices comprising residues 101-109 and residues 111-126, and one helix from the Rossmann motif (residues 293-305).
The RosA protomers in the asymmetric unit exhibit very similar structures. Pairwise superpositions of the Rossmann domains in the six crystallographically independent protomers yielded RMSD values in the range from 0.2 to 0.9 A (for approximately 160 aligned Ca atoms). The same analysis yielded mean RMSD values of 0.3 and 0. 6 A for the N-terminal and intermediate regions. The values are slightly higher for superpositions of the entire chains (1.2 A) and the three dimers (1.6 A). To identify protein structures resembling RosA in terms of its tertiary structure, we performed a Dali server analysis [18]. The structures with the highest Z scores were mitomycin 7-O-methyltransferase from Streptomyces lavendulae (PDB ID 3GWZ), a probable phenazine-specific methyltransferase from Pseudomonas aeruginosa (PDB ID 2IP2), the CALO1 methyltransferase from Micromonospora echinospora (PDB ID 3LST), and a caffeic acid O-methyltransferase from Sorghum bicolor (PDB ID 4PGH), with Z scores of 33, 32, 30 and 30, respectively. These proteins share 21-35% sequence identity with RosA. Superposition with the most similar structure (3GWZ, 35% sequence identity, 76% query coverage) resulted in an RMSD of 3.5 A for the isolated subunit (317 of 324 aligned Ca atoms) and 6.6 A for the dimers (618 of 650 aligned Ca atoms). Superposition of the Rossmann domains of RosA (residues 179-345) and 3GWZ (residues 179-349) resulted in an RMSD of 0.9 A (118 of 160 aligned Ca atoms). This indicates that, despite an overall low sequence identity, the structure of RosA, especially the Rossmann domain, closely resembles that of other methyltransferases.

Prediction of the SAM/SAH binding site
Unfortunately, all efforts to co-crystallize RosA with SAM, SAH, AF or combinations of these compounds and cofactors were unsuccessful. Therefore, we used structural bioinformatics methods to locate putative binding sites for the substrate and cofactor. A cavity analysis using the program CASOX [19] yielded several cavities, but only one of those was large enough to accommodate AF and SAM, and was located at the interface between the Rossmann domain and the Nterminal domain (Fig. 3A).
In addition, the web tools PROSITE [20], 3DLIGANDSITE [21] and COACH [22] were used to predict and identify ligand binding regions in RosA. PROSITE identified a SAM binding region (residue 235-237) in the C-terminal Rossmann domain. This finding is in line with results obtained using 3DLIGANDSITE, COACH and CaSoX. 3DLIGANDSITE identified 25 crystal structures of proteins structurally related to RosA with bound SAM or SAH, and predicted 22 residues that are putatively involved in cofactor binding to RosA. On the other hand, the program COACH (http://zhanglab.ccmb.med.umich.edu/COACH/) found several structures of methyltransferases in complex with various ligands, and clustered them according to a confidence score (C-score between 0 and 1). The best-scored cluster pre-dicted 13 consensus SAH/SAM binding residues with a C-score of 0.54 and a cluster size (total number of templates in a cluster) of 103. Moreover, COACH generated a structure of RosA with SAM bound in the predicted active site. All other ligand-enzyme complexes had significantly lower C-scores. Among those structures with C-scores between 0.5 and 0.05 were those with ligands such as N-acetyl serotonin (C-score 0.10), pisatin (C-score 0.08) and 5-(3,3-dihydroxypropeny)-3methoxy-benzene-1,2-diol (C-score 0.05). Flavin derivatives such as AF did not occur in these clusters identified by COACH. However, a cluster with an even lower C-score of 0.02 contained structures of methyltransferases with bound SAH and 4-methoxy-E-rhodomycin. 4-methoxy-E-rhodomycin resembles AF with regard to the planar ring system. A comparison of the modelled RosA-SAM complex with the structure of carminomycin 4-O-methyltransferase from Streptomyces peucetius (PDB ID 1TW2), a representative of the 4-methoxy-E-rhodomycin cluster, showed that, despite the low sequence identity of 33%, the Rossmann domain and especially the b-sheet core appeared to be structurally conserved. Therefore, we superimposed the b-sheet core residues of RosA and 1TW2 (RMSD 0.805 A; 42 of 42 aligned Ca atoms), and aligned a molecule of AF with 4-methoxy-E-rhodomycin bound to 1TW2 (Fig. 4A). The RosA-AF-SAM complex thus generated was energy-minimized and analysed in terms of the binding interactions of AF and SAM (Figs 3B and 4B).

Determination of dissociation constants
Inspection of the AF and SAM binding sites in the RosA structure suggested that both substrates, AF and SAM, bind independently to their respective pockets. Thus we determined the dissociation constant for AF and SAM by isothermal titration calorimetry (ITC). As shown in Fig. 5A,C, titration with either AF or SAM produced exothermic signals that were fitted to a one-binding-site model, yielding dissociation constants of 10 AE 1 and 22 AE 2 lM, respectively. Surprisingly, both products, i.e. RoF and SAH, bind approximately ten times more tightly than the substrates, displaying dissociation constants of 0.8 AE 0.05 and 2 AE 0.1 lM, respectively (Fig. 5B,D).

Ternary complex formation with substrates and products
The catalytic methylation of AF by RosA was investigated by spectrophotometry using a stopped-flow device and a conventional spectrophotometer, exploiting the spectral changes that occur during methylation [k max (AF) = 478 nm; k max (RoF) = 505 nm]. At low concentrations of RosA, the final absorption spectrum features an absorption maximum at 505 nm, which is characteristic of complete formation of RoF (Fig. 6A). However, at high concentrations of RosA the maximum at 505 nm is less pronounced (hypsochromic effect), and additional maxima at a shorter wavelength are observed (Fig. 6B,C). Analysis by HPLC revealed that RoF is the main pro-duct in both cases, suggesting that the observed spectral changes are due to formation of a complex with RosA and either SAM or SAH. The observed spectral differences suggest considerable changes in the flavin binding site upon formation of a ternary complex, and the UV/Vis absorption spectra of RoF in various solvents suggest that an apolar environment induces similar spectral changes to those observed in our experiments (Fig. 6D). In addition, it is conceivable that steric restriction induced in the ternary complex  prevents de-localization of the electron lone pair into the aromatic system of the isoalloxazine ring, resulting in the observed hypso-and hypochromic effect. Therefore, we assume that formation of the ternary complex generates a more apolar and sterically constrained environment in the RoF binding pocket. In order to reveal the nature of the ternary complex, we also determined the affinity of RosA for SAM and SAH in the presence of RoF. As shown in Fig. 7, the presence of RoF greatly decreases the affinity for SAM, yielding a dissociation constant of 180 AE 20 lM, whereas the affinity of SAH is approximately fourfold higher (K d = 0.4 AE 0.05 lM). Thus the ternary complex with SAH (product complex) is clearly favoured over a complex with SAM (mixed substrate/product complex).
Next, we performed difference absorption spectroscopy to determine the dissociation constants of RoF from the binary complex of RosA:SAH and RosA:SAM, respectively, as well as the affinity of AF for the RosA:SAH complex. The former experiments, shown in Fig. 8A,B, yielded virtually identical dissociation constants of 14 AE 2 and 15 AE 2 lM, demonstrating that SAH and SAM cause similar decreases in the binding affinity of RoF by a factor of 17-18. In contrast to product binding, AF binding to RosA is not affected by SAH, as the determined dissociation constant of 10 AE 1 lM is almost identical to that observed in the absence of SAH (Fig. 9). Table 3 lists the dissociation constants obtained by ITC or spectrophotometric titrations.

Mechanistic studies of the dimethylation reaction catalysed by RosA
With regard to the reaction mechanism, the structure of RosA as well as the independent binding of AF and SAM suggest that the enzyme operates by a randomorder kinetic mechanism, as shown in Scheme 2. To test this hypothesis, we performed a series of presteady-state kinetic measurements in the stopped-flow device. In these experiments, different mixing orders were employed, and the progress of the reaction was monitored at 479 and 530 nm. As shown in Fig. 10A, the rate of reaction was independent of the experimental set-up, i.e. whether RosA was directly mixed with AF and SAM or pre-incubated with either of the substrates. To obtain further insight into the kinetic mechanism, we performed a set of steady-state experiments at various AF and RoF concentrations. As shown in Fig. 10B, a primary reciprocal plot of the reaction velocity against the reciprocal SAM concentration resulted in a set of converging lines in accordance with a random-order mechanism. From a re-plot of the primary data, i.e. slopes and y-axis intercepts versus the reciprocal AF concentration (Fig. 10B, insets), the Michaelis constants of AF and SAM were deduced to be 4 and 70 lM, and the k cat was estimated as 0.06 min À1 (Table 4). As generation of RoF from AF requires two consecutive methylation steps, we also wished to obtain insight into the individual methylation reactions. Towards this aim, we analysed the reaction under single-turnover conditions at two different wavelengths that represent the consumption of the substrate (k = 479 nm) and the generation of the product (k = 530 nm). The reaction of RosA (20 lM) in the presence of AF (20 lM) was investigated by mixing with various concentrations of SAM (80-2000 lM).
The rates of the observed spectral changes at 479 and 530 nm were biphasic, and therefore fitted to two rate equations (Fig. 11A). The first phase consisted of a decrease in absorption at 479 nm and a concurrent small increase at 530 nm (approximately 25%). This first phase was interpreted to represent monomethylation of AF to MAF, and showed a hyperbolic dependence on the SAM concentration, yielding a limiting rate of 0.5 AE 0.05 min À1 and an equilibrium constant of 180 AE 40 lM (Fig. 11B). The second slower phase corresponded to a decrease at 479 nm and a concomitant increase at 530 nm. This phase showed an inverse dependence on the SAM concentration approaching 0.05 AE 0.002 min À1 (Fig. 11C). The latter result suggested that monomethylation of AF generates an inhibitory product that prevents binding of SAM. In fact, our binding studies have revealed that SAH binds much more strongly to RosA than SAM does (Table 3). Furthermore, the presence of RoF further weakens SAM binding but the affinity for SAH is increased (Table 3). Thus we reasoned that SAH, which is produced in the first monomethylation reaction, acts as a competitive inhibitor of SAM. To test this hypothesis, we determined the rate of the methylation reaction as a function of SAM and SAH concentrations. The initial velocities were determined as a function of SAM concentration at various fixed concentrations of SAH to obtain information on the type of inhibition. However, this plot did not allow us to discern the inhibition type, i.e. competitive, non-competitive or uncompetitive. As it was shown that SAH binds more tightly to RosA than SAM does, we used the linearized Henderson equation for data analysis [23]. The resulting Henderson plot, as shown in Fig. 12A, yielded a set of lines that converge on the y axis. A re-plot of the slopes, as shown in the inset to  (Table 3). Thus we conclude that the observed rate decline of the second kinetic phase is due to the competitive effect of SAH, with the limiting value corresponding to k off of SAH.
In a similar set of experiments, it was shown that RoF is a competitive inhibitor of AF, exhibiting a K i value of 27 lM (Fig. 12B), which represents the disso-   Tables 3 and 5).

Discussion
Structure analysis revealed that RosA adopts a topology that is characteristic of SAM-dependent methyltransferases. Although attempts to obtain structural information for complexes with AF, SAM and SAH were unsuccessful, further inspection of the active site and the AF and SAM binding pockets provided useful hints for mechanistic considerations that were tested by binding studies and steady-state as well as presteady-state kinetic experiments. The final biosynthesis of RoF from AF by RosA is unusual in the sense that the enzyme performs two consecutive methylations in the same active site, and hence SAH must dissociate to make room for another SAM while the monomethylated intermediate, MAF, may remain bound to the flavin-binding pocket. This complexity of the overall conversion of AF to RoF was analysed by binding studies as well as kinetic experiments to evaluate the interaction of substrates and products in the RosA-catalysed reaction. Much to our surprise, our binding studies revealed that both products, RoF and SAH, bind approximately ten times more tightly to RosA than AF and SAM do (Table 3). Interestingly, the affinity of RoF decreased approximately 20-fold in the presence of SAM and SAH, whereas the affinities of SAM and SAH respond differently to the presence of RoF, i.e. the dissociation constant for SAM increased approximately eightfold and that for SAH decreased fourfold. These relative binding affinities clearly suggest that the ternary complex observed in spectrophotometric experiments (Fig. 6) consisted of RoF and SAH bound to RosA. The fact that the tightest binary complex formed is that between RosA and RoF also suggested a potential mechanism to control the concentration of free RoF in S. davawensis. It may be assumed that the concentrations of both SAM and SAH are too low to form a ternary complex with RosA:RoF, and thus RoF inhibits binding of AF and further generation of the product. This scenario is also supported by the competitive inhibition of the RosAcatalysed methylation of AF by RoF (Fig. 12). The resistance of S. davawensis to RoF appears to be related to the ability of the ribB FMN riboswitch to discriminate between FMN and roseoflavin-5'-phosphate [7]. However, it is not clear yet whether other additional mechanisms contribute to resistance in S. davawensis. Therefore, it is plausible that the tight binding of RoF to RosA constitutes an additional mechanism to confer resistance towards RoF. Although it is still unknown how RoF is eventually secreted, it is conceivable that RosA also serves as a chaperone to guide RoF to the secretion machinery that is in charge of its translocation from the cell into the environment.
The X-ray crystallographic structure of RosA suggested that both substrates, AF and SAM, bind independently to their respective pockets. This feature was confirmed by kinetic experiments that demonstrated a random-order binding mechanism for the first methylation reaction yielding MAF and SAH ( Fig. 10 and Scheme 2). The rate of monomethylation of AF to MAF showed a hyperbolic dependence on the SAM concentration, yielding a limiting rate of 0.5 AE 0.05 min À1 (Fig. 11B). Surprisingly, the subsequent reaction phase was characterized by an inverse dependence on the SAM concentration, approaching a  limiting rate of 0.05 AE 0.002 min À1 (Fig. 11C). As noted above, SAH binds tightly to RosA, and its affinity is enhanced in the presence of RoF (Table 3). It is likely that the binding affinity of SAH in the presence of MAF is in the same range, i.e. between 2 AE 0.1 and 0.4 AE 0.05 lM, and thus the second methylation reaction depends on the rate of dissociation of SAH (k off ).
This k off appears to be the limiting step for conversion of MAF to RoF, and therefore the intrinsic rate for this reaction step is not accessible in our kinetic experiments. A summary of the reaction steps characterized in this study is shown in Scheme 3.

Expression and purification
Heterologous production and purification of recombinant hexahistidine-tagged RosA from S. davawensis was per-formed as described previously [11]. E. coli Rosetta 2 (DE3) was transformed using pET24a(+)rosA-His 6 (pFJ02) [11]. One of the transformant strains was used to inoculate a pre-culture that was aerobically incubated at 37°C for 16 h (200 rpm) in lysogeny broth containing 50 lgÁmL À1 kanamycin. The cells were harvested by centrifugation (5000 g at 4°C for 30 min), and washed three times using PBS (10 min per wash at 4°C). The cells were then suspended in 30 mL PBS and used as an inoculum for a 15-L bioreactor (Bioengineering AG, Wald, Switzerland). Cell growth was allowed to proceed for 7 h at 750 rpm, 37°C, with an aeration of 1 L of oxygen per L of culture per minute in a total volume of 10 L until the cells reached an absorption at 600 nm of 0.8. Synthesis of the recombinant protein was induced by addition of 0.5 mM isopropyl-b-Dthiogalactopyranoside. The culture was further incubated for 16 h at 30°C until an attenuance at 600 nm of 1.2 was reached. Cells were harvested by centrifugation at 5000 g (4°C).
To produce selenomethionine-containing hexahistidinetagged RosA, the expression protocol for native RosA was slightly altered. E. coli BL834 (DE3) was transformed with pFJ02, and incubated for 7 h at 37°C and 200 rpm in selenomethionine minimal medium [26] supplemented with 50 lgÁmL À1 kanamycin. The pre-culture was harvested by centrifugation at 5000 g (4°C) and used as an inoculum for a 15-L bioreactor (Bioengineering AG). Synthesis of the recombinant protein was induced at when the cells reached an attenuance at 600 nm of 0.8 by adding 0.5 mM isopropyl-b-D-thiogalactopyranoside. Cells were harvested when they reached an attenuance at 600 nm of 1.2, by centrifugation at 5000 g (4°C). All cell pellets were frozen immediately after harvesting and stored at À80°C.
The enzyme was purified using an € AKTApurifier TM system (GE Healthcare). Frozen cell pellets of E. coli Rosetta 2 (DE3) and BL834 (DE3) over-producing hexahistidine-tagged RosA and SeMet-RosA, respectively, was resuspended in 50 mL HisTrap binding buffer. This buffer contained 20 mM Na 2 HPO 4 , pH 7.4, 500 mM NaCl and 20 mM imidazole, and was supplemented with protease inhibitors (Roche cOmplete TM EDTA-free protease inhibitor cocktail, Roche, Basel, Switzerland). For purification of SeMet-RosA, the HisTrap buffer also contained 2 mM bmercaptoethanol. The cells were disrupted using a French press (three cycles, 2 9 10 8 Pa, 10°C). The cell-free extract was cleared by centrifugation at 10 000 g for 30 min at 4°C. The supernatant was again centrifuged at 100 000 g for 30 min at 4°C to remove cell debris and unbroken cells. The cleared lysate was filtered (0.45 lm), equilibrated with loading buffer (20 mM Na 2 HPO 4 , pH 7.4, 500 mM NaCl, 20 mM imidazole, plus 2 mM b-mercaptoethanol for SeMET-RosA) and applied to a 5 mL HisTrap column (HisTrap TM HP; GE Healthcare). Gradient elution of the hexahistidine-tagged protein was performed using 0 to 50% elution buffer (20 mM Na 2 HPO 4 , pH 7.4, 500 mM NaCl, 500 mM imidazole, plus 2 mM b-mercaptoethanol for SeMET-RosA). Eluted enzyme fractions were pooled and desalted using a HiTrap TM desalting column (GE Healthcare), and equilibrated in a buffer containing 20 mM Tris/ HCl, pH 8.0, with 2 mM b-mercaptoethanol for SeMET-RosA only. To achieve enzyme purity appropriate for crystallization experiments, an additional ion exchange chromatography step was performed. The enzyme fraction was applied to a MonoQ 5/50 GL column (GE Healthcare) and eluted using a linear gradient from 0 to 100% elution buffer comprising 20 mM Tris/HCl, pH 8.0, 500 mM NaCl (plus 2 mM b-mercaptoethanol for SeMET-RosA only). Aliquots of the fractions were analysed by SDS/PAGE with staining using Coomassie Brilliant Blue R-250. The homogeneous fractions were tested directly for RosA activity (see below).
To determine the native molecular mass of RosA gel filtration chromatography at 25°C was performed. The pro-  tein solution was loaded onto a Superdex 200 10/300 GL column attached to a AKTApurifier TM system (GE Healthcare) and then eluted with elution buffer (50 mM Tris-HCl, pH 8.0 containing 100 mM NaCl) at a flow rate of 0.5 mLÁmin À1 . Protein standard markers used in this experiment were ferritin (440 kDa), aldolase (158 kDa), conalbumin (75 kDa), ovalbumin (43 kDa), and ribonuclease a (13.7 kDa).

Isothermal titration calorimetry (ITC)
All of the experiments to determine dissociation constants (K d ) for AF, RoF, SAM and SAH to RosA were performed at 25°C in 50 mM Tris/HCl buffer, pH 8.0, containing 100 mM NaCl using a VP Steady-state kinetics of RosA RosA catalyses the N,N-dimethylation of AF to yield RoF as the final product. Steady-state kinetic measurements were Product inhibition of RosA Enzymatic activity of RosA in the presence of the reaction products RoF and SAH was studied by measuring the initial rate of the reaction at 25°C using a spectrophotometer. The assay reaction The inhibition by SAH was analysed at various concentrations of SAH and SAM using a fixed concentration of AF (200 lM). Initial rates were obtained from linear data fitting using KINETIC STUDIO software (TgK Scientific Ltd). A Henderson plot of [I]/(1v i /v 0 ) and the reciprocal of the fractional initial velocity (v 0 /v i ) (Eqns 3 and 4) was used to obtain inhibition constants and information on the inhibition mechanism [23]: and for a competitive inhibitor: In Eqns 3 and 4, [I] is the concentration of inhibitor (i.e. RoF and SAH), v 0 and v i are the initial velocity of a reaction in the absence and presence of a inhibitor, respectively, [E] is the total enzyme concentration, [S] is the variable substrate concentration, K m is the Michaelis constant, K i represents the inhibition constant of a tight binding inhibitor and K app i represents the apparent inhibition constant.

Pre-steady-state reaction studies
The methylation of AF bound to RosA was investigated at various concentrations of SAM at 25°C using a spectrophotometer. Briefly, a solution of RosA (20 lM) and AF (20 lM) in 50 mM Tris/HCl buffer, pH 8.0, containing 100 mM NaCl was pre-incubated at 25°C for 3 min, and then the reaction was initiated by adding SAM (final concentrations of 80, 200, 400, 1000 and 2000 lM). Progress of the reaction due to consumption of AF and formation of the final product (RoF) was monitored at 479 and 530 nm, respectively. Data analysis of the kinetic traces was performed using the exponential equations from KINETIC STU-DIO software (TgK Scientific Ltd) to obtain the observed rates (k obs ). Rate constants were determined from plots of k obs as a function of the SAM concentration using the Levenberg-Marquardt algorithm implemented in the KALEI-DAGRAPH software (Synergy Software).

Stopped-flow spectrophotometric studies
All experiments were performed at 25°C using an SF-61SX2 stopped-flow spectrophotometer (TgK Scientific Ltd, Bradford-on-Avon, UK) with a 1 cm path length. Prior to all experiments, the stopped-flow instrument was flushed with 50 mM Tris/HCl buffer, pH 8, containing 100 mM NaCl several times. To analyse the effect of substrate binding order on the reaction kinetics, experiments with different mixing orders of AF and SAM were performed. Various pre-mixed solutions of RosA, RosA:AF or RosA:SAM complexes were prepared, and then mixed with solutions containing AF and SAM, SAM or AF, respectively. The progress of the reactions was monitored at 479 and 530 nm using a KinetaScan diode array detector (TgK Scientific Ltd).

Product analysis by HPLC
HPLC analysis of AF, MAF and RoF was performed using an Atlantis Ò dC18 reversed-phase column (5 lm, 4.6 9 250 mm; Waters, Milford, MA, USA) on an Ultamate 3000 HPLC instrument (Dionex, Sunnyvale, CA, USA) at 25°C. All samples prepared from RosA reactions were loaded onto the column and eluted using a multi-step gradient of 100 mM formic acid and 100 mM ammonium formate (pH 3.7)/methanol (0-3 min, 30% methanol; 3-20 min, 30-75% methanol; 20-22 min, 75% methanol) at a flow rate of 0.8 mLÁmin À1 . Detection of AF, MAF and RoF was performed by measuring the absorption at 479, 490 and 509 nm, respectively.  Crystals of native RosA grew within 5 days after mixing the protein solution (in a 1 : 1 ratio) with a solution containing 200 mM sodium chloride, 100 mM Tris/HCl, pH 5.5, and 25% w/v PEG-3350. The same approach was used to grow crystals of the selenomethionine-labelled derivative. Equal amounts (0.5 lL) of the variant at a concentration of 20 mgÁmL À1 in 20 mM Tris/HCl, pH 8.0, 500 mM NaCl and 2 mM bmercaptoethanol were mixed with precipitant containing 100 mM MES/imidazole, pH 6.5, 20 mM D-glucose, D-mannose, D-galactose, L-fructose, D-xylose and N-acetyl-D-glucosamine, 10% w/v PEG-20000 and 20% v/v PEG-MME 550 [28]. SeMet-RosA crystals grew within 7 days after crystallization set-up. For diffraction data collection, crystals were harvested from their mother liquor using Cryo-Loops TM (Hampton Research, Aliso Viejo, CA, USA), and flash-cooled in liquid nitrogen without any additional cryoprotection.

Structure determination and refinement
Diffraction data to a resolution of 2.2 A were collected for native RosA from a single triclinic crystal (space group P1) at beamline ID23-1 of the European Synchrotron Radiation Facility (Grenoble, France). MAD data to a resolution of approximately 3.5 A were collected at beamline X06DA-PXIII of the Swiss Light Source at the Paul Scherrer Institute (Villigen, Switzerland). Three MAD datasets were collected from a single tetragonal SeMet-RosA crystal (space group I4 1 22) at various wavelengths (peak, inflection, remote) that were determined from an X-ray fluorescence scan recorded around the selenium absorption edge. Data processing and scaling for the native and SeMet MAD datasets were performed using XDS [29], SCALA and AIMLESS [30].
The calculated Matthews coefficient [30,31] of the MAD datasets indicated a 99.9% probability for the presence of one RosA molecule per asymmetric unit of the crystal. To identify selenium sites, the two MAD pipelines AUTO-RICK-SHAW [32,33] and PHENIX AutoSol [34,35] were used. The combination of these programs yielded first phases and a partial structure. These phases were further used as input phases for the automated chain-tracing/building programs PHENIX AUTOBUILD [36] and BUCCANEER [37] to extend the model and to improve its quality. The combined model of both rebuilding programs produced interpretable electron densities, with preliminary values of R = 37% and  R free = 49%. R free values were computed from a set of randomly chosen reflections (5%), which were not used during refinement [38]. Structure refinement was performed using the refinement programs PHENIX REFINE [34] and REFMAC [39]. Model rebuilding was performed using BUCCANEER [37] and COOT [40] by alternate automated rebuilding and real-space fitting against r A -weighted 2 F o -F c and F o -F c electron density maps together with least-squares optimization. The improved MAD model was used as an initial molecular replacement template for PHASER [41] to determine the structure of native RosA in the triclinic crystal form. Molecular replacement resulted in six molecules in the asymmetric unit. Structure refinement and model building were performed using PHENIX REFINE [34] and COOT [40] by real-space fitting against r A -weighted 2 F o -F c and F o -F c electron density maps and least-squares optimizations. Water molecules were placed into the difference electron density map, and accepted or rejected on the basis of geometry criteria as well as refined B-factors. In later stages of the refinement, four TLS groups per protomer were defined based on an analysis using the TLSMD web server [42]. The final model was refined to R = 20% and R free = 24%. Validation of the structure was performed using MOLPROBITY [43], yielding a Ramachandran plot with 97.20% of the residues in favoured regions, 2.65% in allowed regions and 0.15% in disallowed regions. Data statistics and details of structure refinement are given in Table 1 (MAD data) and Table 2 (native data).

Prediction of the SAM/SAH binding site
In order to locate putative binding sites for the substrate AF and the cofactor SAM, the program CASOX [19] together with the web tools PROSITE (sequence-based prediction) [20], 3DLIGANDSITE (sequence-and structure-based prediction) [21] and COACH (sequence-and structure-based prediction) [22] were used. The results of the cavity analysis using CASOX were compared with consensus ligand binding residues predicted by PROSITE, 3DLIGANDSITE and COACH. COACH additionally generated putative enzyme-ligand model complexes and ranked them according to an intrinsic C-score. This C-score represents a confidence score for the prediction. C-scores range between 0 and 1, with a higher score indicating a more reliable prediction. The highestrated complex, together with other related methyltransferase structures, were superimposed with RosA to identify ligand binding modes. The modelled RosA-ligand complex was further energy-minimized using the program YASARA with the AMBER03 force field, employing the standard optimization protocol [44]. Structure alignments were performed using PYMOL (https://www.pymol.org/).