Structure and Mechanism of the Rebeccamycin Sugar 4′-O-Methyltransferase RebM*

The 2.65-Å crystal structure of the rebeccamycin 4′-O-methyltransferase RebM in complex with S-adenosyl-l-homocysteine revealed RebM to adopt a typical S-adenosylmethionine-binding fold of small molecule O-methyltransferases (O-MTases) and display a weak dimerization domain unique to MTases. Using this structure as a basis, the RebM substrate binding model implicated a predominance of nonspecific hydrophobic interactions consistent with the reported ability of RebM to methylate a wide range of indolocarbazole surrogates. This model also illuminated the three putative RebM catalytic residues (His140/141 and Asp166) subsequently found to be highly conserved among sequence-related natural product O-MTases from GC-rich bacteria. Interrogation of these residues via site-directed mutagenesis in RebM demonstrated His140 and Asp166 to be most important for catalysis. This study reveals RebM to be a member of the general acid/base-dependent O-MTases and, as the first crystal structure for a sugar O-MTase, may also present a template toward the future engineering of natural product MTases for combinatorial applications.

The indolocarbazole alkaloids are typically divided into two major classes dependent upon their structure and mechanism of action (1)(2)(3). Specifically the staurosporine-type indolocarbazoles (Fig. 1A, 1 and 2) possess indole nitrogens bridged by a single glycosyl moiety at C-1Ј and C-5Ј and are potent inhibitors of protein kinases A, C, and K (4), whereas the rebeccamycin-type variants (Fig. 1A, 3 and 4) are distinguished by a ␤-glucoside attachment critical for the potent topoisomerase I poisoning effects and notable anticancer activities of this class (5)(6)(7). Biosynthetically the indolocarbazoles are derived from tryptophan, glucose, methionine, and, in the case of the rebeccamycin-type indolocarbazoles, chloride (5,8). The biosyn-thetic gene clusters encoding 1, 3, and 4 have been reported (9 -11), and a variety of in vitro and in vivo studies have contributed to an understanding of the enzymes responsible for indolocarbazole core biosynthesis (Fig. 1B). As exemplified by the rebeccamycin core reactions these include RebH-catalyzed tryptophan halogenation (12), RebO/RebD-mediated chromopyrrolic acid formation (12)(13)(14), and RebC/RebP-catalyzed ring closure (5,15,16). Similar studies have delineated the catalysts responsible for the key structural features that mechanistically distinguish the indolocarbazoles including the enzymes to form and attach novel sugars (11,17,18) and a culminating series of S-adenosylmethionine (AdoMet) 3 -dependent N-and O-methylations (11,17,18). Among this set of final tailoring reactions, the differential sugar O-alkylation provided by RebM, AtM, and StaMB greatly influences the biological activity of the corresponding indolocarbazoles (17).
The first indolocarbazole sugar O-methyltransferase (O-MTase) to be characterized in vitro, RebM catalyzes glucosyl C-4Ј O-methylation as the final step in rebeccamycin biosynthesis ( Fig. 1B) (17). RebM displays some subtle distinctions from the other sugar O-MTases studied in vitro including oleandomycin OleY (19), tylosin TylE and TylF (20,21), mycinamicin MycF (22), coumermycin CouP (23), and novobiocin NovP (24). In contrast to typical multimeric methyltransferases (19 -21, 25-27), RebM functions as a monomer. Unlike OleY, TylE, and TylF, RebM activity is observed over a broader pH range and cannot be enhanced by divalent metals. However, the most striking unique feature of RebM is its ability to accept a wide range of alternate substrates. Specifically, in addition to the native substrate (Fig. 1B, 13), RebM can tolerate variation on the imide heterocycle, deoxygenation of the sugar moiety, and even indolocarbazole glycoside anomers (17). Furthermore, RebM was the first secondary metabolite-associated MTase demonstrated to utilize non-natural cofactor analogs of AdoMet, such as synthetic N-mustard adenylates, to provide novel adenylated indolocarbazole conjugates (28). Similar to DNA MTases in recent reports (29 -31), RebM can also accomplish differential alkylation (replacing methyl with alkyl) in the presence of the appropriate AdoMet analogs. 4 Thus, RebM, and natural product MTases in general, may present spectacu-lar opportunities for rapid enzymatic diversification of therapeutically important complex natural products.
In an effort to further our understanding of the mechanism of RebM catalysis and the key structural elements for AdoMet recognition and activation, we report the crystal structure of RebM at 2.65 Å. This work revealed RebM to crystallize as a dimer and adopt a fairly typical MTase structural fold. The location and nature of the putative RebM dimer interface is unique in comparison with other MTases and, based upon gel filtration studies, is likely not critical for catalysis. Site-directed mutagenesis of key residues in the AdoMet-binding pocket revealed some tolerance to variation with little overall alteration in structure, AdoMet binding, and/or catalysis. Mutagenesis studies also identified key catalytic residues and confirmed RebM to be a member of the acid/base-dependent O-MTases.
As the first crystal structure for a sugar O-MTase, this work may also present a template toward the future engineering of natural product MTases for combinatorial applications.

MATERIALS AND METHODS
Protein Expression and Purification-Lechevalieria aerocolonigenes RebM and all engineered RebM mutants were produced as amino-terminal His 6 fusion proteins using the expression plasmid pCST28aRebM in Escherichia coli BL21(DE3)pLysS (17). The culture was grown (28°C at 250 rpm) to midlog phase (A 600 ϳ 0.6) at which point isopropyl 1-thio-␤-D-galactopyranoside was added to a 0.4 mM final concentration. Growth was continued for an additional 4 -6 h, the cells were collected by centrifugation (  300 mM NaCl and 20 mM imidazole on ice. The cells were lysed via incubation with 1 mg/ml lysozyme (ϳ50,000 units/mg; Sigma) for 30 min on ice followed by sonication (VirSonic 475; Virtis, Gardiner, NY; 100 watts, 4 ϫ 30-s pulses, ϳ1 min between pulses) on ice. Protein was purified with nickel-nitrilotriacetic acid-agarose resin or spin columns (Qiagen, Valencia, CA) using the manufacturer's protocols. The buffer was exchanged with 20 mM potassium phosphate, pH 8, via PD-10 gel filtration columns (GE Healthcare), and the purified enzyme was subsequently concentrated to Ͼ10 mg/ml, drop frozen in liquid nitrogen, and stored at Ϫ80°C. Protein concentrations were determined by Bradford assay (Bio-Rad) using bovine serum albumin as a standard. For the production of selenomethionine-labeled protein, the E. coli methionine auxotroph strain B834 (DE3) was transformed with the plasmid, and autoinduction medium was used (32). Protein Crystallization-RebM was crystallized by the hanging drop vapor diffusion method at 293 K. The reservoir solution contained 8% (w/v) methyl ether polyethylene glycol 5000, 200 mM ammonium sulfate, and 100 mM sodium acetate, pH 5.0. The hanging drop consisted of 2 l of protein solution (12 mg/ml RebM, 50 mM sodium chloride, and 10 mM Tris-HCl, pH 8.0) mixed with 2 l of reservoir solution. RebM crystals (rectangular, ϳ350 ϫ 50 ϫ 30 m in size) required 1 month to achieve full size. The crystals were subsequently soaked in increasing concentrations of ethylene glycol in mother liquor up to a final concentration of 20% (v/v) and flash frozen in a stream of liquid nitrogen.
Data Collection and Reduction-X-ray diffraction data were collected at the Advanced Photon Source on Life Sciences Collaborative Access Team beamline 21-ID-D at a temperature of 93 K. Reflections were indexed, integrated, and scaled using the HKL2000 package (33).
Structure Determination-A total of 12 of 14 potential selenium sites were identified with HySS (34,35). The data were phased via single wavelength anomalous dispersion using autoSHARP with the help of auxiliary programs from the CCP4 suite (36,37). Coot (38) was used to manually build into the density-modified map produced by autoSHARP. Once a sufficient number of residues were assembled within both RebM molecules of the asymmetric unit to accurately determine the noncrystallographic symmetry operator, density modification was carried out with RESOLVE resulting in a higher quality electron density map (39). The structure was then completed through multiple rounds of model building with Coot and refinement with REFMAC (40). TLS (translation/libration/ screw) groups were incorporated during the final stages of refinement. These groups were selected based on the output of the TLSMD web server (41). Relevant crystallographic statistics are summarized in Table 1.
Energy Minimization-The substrate complex model was generated by manual insertion of 13 into the suspected active site. The simulation assumed an S N 2 mechanism and was based upon the closely related mycolic acid synthase-mycolic acid ternary complex (42) (Protein Data Bank code 1KPI). In the superimposed model, 13 was positioned in the cavity above the AdoMet binding site in an equivalent position of mycolic acid, and the 13 4Ј-hydroxyl was linearly aligned with respect to the methyl donor to facilitate the nucleophilic attack. Model refinement relied upon 200 steps of conjugate gradient minimization as implemented in the CNS (Crystallography and NMR System) model minimization procedure (43). Topology and parameter files for 13 and S-adenosylhomocysteine (AdoHcy) were obtained from the HIC-Up server (44).
Site-directed Mutagenesis-RebM mutants were generated with the QuikChange II mutagenesis kit (Stratagene, La Jolla, CA) using the parent expression plasmid pCST28aRebM as template (17). The mutagenic primers are listed in supplemental Table S1. All mutant plasmids were confirmed by DNA sequencing to carry the desired mutations. Synthesis of oligonucleotide primers and sequencing of DNA were performed at the University of Wisconsin-Madison Biotech Center. Plasmids containing the confirmed rebM mutations were then transformed into E. coli BL21(DE3), and the corresponding overproduced recombinant mutant proteins were purified as described for the wild-type enzyme.
CD Spectroscopy-To confirm the influence of the targeted RebM mutations upon the global protein fold, all mutant proteins were analyzed by CD spectroscopy. For CD analysis, the mutant proteins were exchanged at 4°C with buffer (50 mM phosphate, pH 8), and the protein concentration of each sample was subsequently determined by the Bradford assay (Bio-Rad). An AVIV model 202 circular dichroism spectrometer (AVIV Associates, Lakewood, NJ) was used to record the CD spectra using a 1-mm-path length quartz cell containing 300 l of each protein sample (ϳ0.25 mg/ml) maintained at 25°C in a ther- is the intensity of an individual measurement of the reflection and I(h) is the mean intensity of the reflection. b R cryst ϭ ⌺ h ʈF obs ͉ Ϫ ͉F calc ʈ/⌺ h ͉F obs ͉ where F obs and F calc are the observed and calculated structure factor amplitudes, respectively. c R free was calculated as R cryst using 5.0% of the randomly selected unique reflections that were omitted from structure refinement.
moelectrically controlled cell holder. Data were collected every 1 nm with an averaging time of 5 s and are expressed as the mean residue ellipticity in units of degrees⅐cm 2 /dmol (supplemental Fig. S1). Enzymatic Reactions-The enzyme assay was accomplished as reported previously (17). Reactions were performed in phosphate buffer (50 mM, pH 8.0) containing 50 M 13 (Fig. 1B), 20 M purified RebM, and 100 M AdoMet (Sigma) with incubation at 30°C for 4 h in a total volume of 20 l. All reactions were quenched by the addition of 20 l of methanol followed by centrifugation (10 min at 10,000 ϫ g) to remove precipitated protein. Reactions were analyzed using an analytical Varian ProStar high pressure liquid chromatography system (Phenomenex Luna C 18 , 5 m, 250 ϫ 4.6 mm, 0.1% aqueous trifluoroacetic acid with a gradient of 10 -100% CH 3 CN over 20 min at 1.0 ml/min, A 316 ), and reaction products were confirmed as described previously (17).

RESULTS AND DISCUSSION
Quality of the Maps and the Model-RebM, a 30-kDa protein consisting of 273 amino acids, crystallized as a homodimer, and the structure was refined to a nominal resolution of 2.65 Å. The asymmetric unit contained two protein molecules (labeled A and B), one molecule of S-adenosylhomocysteine per protein, and six ordered water molecules. The first 42 residues of chain A and the first 46 residues of chain B (20 residues of which comprised the amino-terminal His tag in each case) were not modeled because of insufficient electron density. In addition, the surface loop of chain B containing residues 35-40 could not be modeled. The ␣ carbons of the two chains align with a root mean square deviation of 0.514 Å. The final structure was refined to an R cryst and R free of 21.4 and 26.0%, respectively. The slightly high residual values as well as the low ordered solvent content may be related to the extremely strong diffuse scattering observed in the diffraction images. This diffuse scattering was also observed during room temperature data collection suggesting that it was inherent to the crystals and not the result of cryoprotection. The RebM crystals described belong to the space group P4 3 2 1 2 with unit cell parameters a ϭ b ϭ 119.2 Å, c ϭ 84.4 Å.
Overview of the Structure-In the RebM crystal structure, the two monomers in the homodimer are related by 2-fold crystallographic symmetry (Fig. 2A). The dimer interface in previously reported small molecule MTases is typically found above the substrate binding site and often contains an extra amino-terminal dimerization domain also important for catalysis. In contrast, the RebM dimer interface is composed predominantly of hydrogen bonding interactions encompassing residues Thr 211 -Asp 215 of ␤-strand-6 aligned in an antiparallel orientation ( Fig.  2A). Four backbone-backbone amide-carbonyl hydrogen bonds are involved in the dimerization interface ( Fig. 2A), the span of which is quite small overall, comprising only 380 Å 2 of the buried surface area (ϳ3% of the total surface area). Consistent with gel filtration studies that revealed RebM to be a monomer in solution (17), the relatively small size of the unique RebM dimer interface suggests it to be of low affinity. This is unlike the previously reported small molecule MTases (such as  (46)) wherein the dimer interface is comprised of nearly 30% of the total surface area. Because the dyad-related monomer of RebM does not contribute to the active site of its partner molecule, dimerization does not appear to be necessary for the substrate recognition or transmethylation. In addition to RebM, only two other MTases invoke an atypical ␤-strand-6 dimer interface. Of these, the caffeoyl-coenzyme A 3-O-methyltransferase (47) ␤-strands of the dimer interface adopt a parallel orientation, whereas the analogous ␤-strands in sarcosine dimethylglycine methyltransferase (Protein Data Bank code 2O57) align in a RebM-like antiparallel fashion but, unlike RebM, form four salt bridges. In the context of catalysis, caffeoyl-coenzyme A 3-O-methyltransferase dimerization is not critical for activity, whereas the catalytic influence of sarcosine dimethylglycine methyltransferase dimerization has not been reported.
Structural Homology-Although this work represents the first reported sugar O-MTase crystal structure, based upon the Research Collaboratory for Structural Bioinformatics (RCSB), over 260 functionally diverse MTases have been characterized structurally (49,50). These enzymes show no or very low overall sequence identity to each other, but most share a common AdoMet-dependent MTase fold. Many members also contain additional domains outside the core MTase structure that play a role in substrate recognition or alternative functions.
A DALI search for structures similar to that of RebM returned several hits with Z-scores of Ͼ12 (Table 2) (47), and phenylethanolamine N-methyltransferase (55). This entire panel shares very low overall sequence identity (ϳ7-23%) but high sequence conservation among residues associated with AdoMet binding, specifically the core fold between ␤1 and ␤2 that interacts with the AdoMet homocysteine and ribosyl moieties and an acidic residue in the loop between ␤2 and ␤3 that interacts with the exocyclic N 6 and ring N-1 of the adenine ring of the cofactor. Although most of the RebM struc- tural homologs identified catalyze methylation of heteroatoms, it is interesting to note that the three closest structural homologs (Hma, Cmaa1, and RdmB) are involved in rather unique AdoMet-dependent transformations. Specifically Hma and Cmaa1 are responsible for branching and cyclopropanation in mycolic acid biosynthesis (C-alkylation) (42,51), whereas RdmB is a novel AdoMet-dependent anthracycline hydroxylase (53). Superposition of a monomer of RebM and its closest structural homologs, sarcosine dimethylglycine methyltransferase (Protein Data Bank code 2O57) and the mycolic acid synthase Hma (51), revealed an overlap with root mean square deviation of 1.8 and 2.0 Å for 237 and 241 C ␣ atoms, respectively (Fig. 3, A and B). AdoMet/AdoHcy Binding Site-The electron density for the cofactor AdoHcy (demethyl-AdoMet) is well defined in the RebM structure. The AdoMet/AdoHcy binding site is located in the carboxyl-terminal end of a cleft formed by the ␤-strands highly conserved throughout MTases. AdoHcy is bound through an extensive hydrogen bond network and van der Waals interactions and is partially exposed to solvent (Fig. 2B). Specifically the RebM-AdoHcy interaction mainly engages residues in the loops between ␤1 and ␤2 (L1; binds the amino acid and ribosyl moieties), between ␤2 and ␤3 (L2; contacts ribose and adenine), and after ␤3 (L3; interacts with adenine) and also helix ␣4 (binds the amino acid and adenine). The adenine ring of AdoHcy is situated in a hydrophobic pocket formed by the aliphatic side chains Ile 92 (L2), Ala 120 (L3), and Met 142 (helix ␣4). The exocyclic N 6 and ring N-1 of the adenine ring are hydrogen-bonded to the side chain carboxyl oxygen of Asp 119 (L3). The ribosyl moiety is anchored via hydrogen bonds from the C-2Ј and C-3Ј hydroxyl groups to the side chain of Gln 96 (L2) and Ser 91 (L2). Helix ␣4 presents the closest neighbors of the AdoHcy sulfur, which include the side chain ring nitrogen of His 141 (3.7 Å), main chain carbonyl oxygen of Glu 26 (3.4 Å), and side chain carboxyl oxygen of Glu 137 (5.5 Å). The AdoHcy carboxyl is locked by the side chains of His 29 and Pro 75 (L1), whereas the corresponding AdoHcy amine is anchored to the main chain carbonyl oxygen atoms of Gly 69 (L1) and Leu 136 (helix ␣4) (Fig. 2C). Structural comparison and primary sequence analysis (Fig. 3B) of other AdoMet-dependent MTases revealed that L1, which contains a glycine-rich consensus sequence (LDXGXGXG), and an acidic residue in loop L3 (Asp 119 in RebM) are highly conserved, whereas residues in loop L2 and helix ␣4 are less well conserved.
To interrogate the AdoMet binding site we selected a representative set of amino acids within a 5-Å radius of the cofactor for mutagenesis. This set included both representative primary shell residues that directly interact with AdoMet (P75S and L136V) as well as a series of secondary shell residues with the potential to indirectly influence cofactor binding (C70A, C70S, W134A, and S138A). Interestingly all but one of the designated   (pK a ϳ 16). Of these two putative catalytic residues, His 140 is also within hydrogen bonding distance (3.01 Å) of Asp 166 . Consistent with a putative acid/base mechanism for RebM, EDTA or divalent metals do not influence RebM activity (17).
Interrogation of the three putative catalytic residues via mutagenesis revealed decreases in the apparent k cat by 20-, 10-, and 4-fold for H140A, D166A, and H141A, respectively (Table  3). Both D166A and H141A also displayed significant increases in the apparent K m for AdoMet (Ն8-fold), whereas H141A also led to a ϳ2-fold reduction in affinity, based upon the apparent K m , for 13. In contrast, a slight decrease in the apparent K m for AdoMet (ϳ3-fold) and a moderate increase in the apparent K m for 13 (ϳ2.5-fold) were observed for mutant H140A. Evaluation of the H140A/H141A double mutant revealed a properly folded (based upon CD), but inactive, protein. These results are consistent with His 140 as the preferred general base potentially augmented via hydrogen bonding to Asp 166 , a residue that, although positioned ϳ7-10 Å from AdoMet in the static structural model, is critical for AdoMet binding based upon kinetic characterization. Interestingly, in isoflavone O-methyltransferase, the corresponding general base (His2 57 ) is constrained and oriented via hydrogen bonding with a specific glutamate (Glu 318 ). Given this precedent and the significant reduction of the apparent k cat in RebM D166A, we postulate that RebM Asp 166 orients the RebM general base His 140 in a similar manner. Consistent with the position of His 141 at the AdoMet-13 interface in the static structural model, mutagenesis of His 141 reduced the apparent K m for both AdoMet and 13. In the context of RebM H140A, His 141 may also weakly compensate as a general base, as the double mutant H140A/H141A is completely inactive. Notably alignment of RebM with natural product O-MTases from GC-rich bacteria (Fig. 4) revealed His 140/ 141 and Asp 166 to be invariant residues among this set of enzymes. Based upon this analysis, we propose these invariant residues to play similar catalytic roles (His 140/141 equivalents as the general base potentially constrained by interactions with Asp 166 homologs) in each of these respective MTases.
Summary-The determination of the crystal structure of RebM in complex with S-adenosyl-L-homocysteine enabled the illumination of key catalytic residues (His 140/141 and Asp 166 ) that, based upon this study, were found to be invariant among a wide array of natural product O-MTases from GC-rich bacteria. This study revealed RebM to be a member of the general acid/base-dependent O-MTases and to adopt a typical AdoMet-binding fold. Consistent with the reported ability of RebM to methylate a wide range of indolocarbazole surrogates, the model of the RebM-indolocarbazole complex implicated a predominance of nonspecific hydrophobic interactions. As the first crystal structure for a sugar O-MTase, this study may also present a template toward the future engineering of natural product MTases for combinatorial applications.