Catalytic Control of Spiroketal Formation in Rubromycin Polyketide Biosynthesis

Abstract The medically important bacterial aromatic polyketide natural products typically feature a planar, polycyclic core structure. An exception is found for the rubromycins, whose backbones are disrupted by a bisbenzannulated [5,6]‐spiroketal pharmacophore that was recently shown to be assembled by flavin‐dependent enzymes. In particular, a flavoprotein monooxygenase proved critical for the drastic oxidative rearrangement of a pentangular precursor and the installment of an intermediate [6,6]‐spiroketal moiety. Here we provide structural and mechanistic insights into the control of catalysis by this spiroketal synthase, which fulfills several important functions as reductase, monooxygenase, and presumably oxidase. The enzyme hereby tightly controls the redox state of the substrate to counteract shunt product formation, while also steering the cleavage of three carbon‐carbon bonds. Our work illustrates an exceptional strategy for the biosynthesis of stable chroman spiroketals.


Recombinant production and purification of RubL
In order to allow for the recombinant production of RubL, the respective gene was purchased, codon optimized for Escherichia coli and subcloned into a pET16b-vector (BioCat), in frame with the nucleotide sequence coding for an N-terminal histidine-tag. In addition, a tobacco etch virus (TEV)-cleavage site was inserted 5' to the gene sequence, to enable the selective removal of the affinity tag prior to protein crystallization. Upon arrival, the dry plasmid DNA was dissolved in ddH2O to a final concentration of 100 ng µL -1 and transformed into E. coli BL21(DE3)-cells (Thermo Fisher Scientific) for subsequent gene expression.
For the recombinant production of the protein, TB-medium containing 100 µg mL -1 ampicillin was inoculated with a pre-culture grown in LB-medium (supplemented with the same amount of antibiotic) to an optical density at 600 nm (OD600) of 0.

Cloning and recombinant production of MBP-TEV-GrhO5
Since the initial MBP-GrhO5 construct generated by Frensch et al. [1] did not contain a protease cleavage site allowing for the removal of the solubility tag (MBP-tag) prior to protein crystallization, grhO5 was recloned to the vector pETHis6-MBP-TEV (between SspI and XhoI, site). Having confirmed the proper insertion of the gene by automated sequencing, the plasmid was transformed into E. coli BL21 (DE3)-cells for gene expression. Protein production and affinity purification (incl. removal of the affinity tag) were carried out as described for RubL above.
Like RubL, GrhO5 used for crystallization was applied to a Superdex 200 pg (10/600 HiLoad) column, in this case pre-equilibrated with 20 mM Tris, 50 mM NaCl, pH 7.4, for further purification. Again, yellow fractions corresponding to pure monomeric GrhO5 were combined, concentrated to 200 µM (11 mg mL -1 ), flash frozen in liquid N2 and stored at -80°C until further use.

Determination of the molar extinction coefficient of GrhO5-/RubL-bound FAD
To allow for the determination of the molar extinction coefficients of the protein-bound FAD cofactors, a UV-visible absorption spectrum of both native and denatured GrhO5/RubL was recorded. Based on the assumption that the spectrum of denatured protein equals the one of free FAD (450: 11 300 M -1 cm -1 ), molar extinction coefficients of 10 250 M -1 cm -1 and 11 300 M -1 cm -1 were calculated for GrhO5 and RubL, respectively.

Anaerobic photoreduction of RubL-bound FAD
Photoreduction of RubL-bound FAD was carried out in a UV-1800PC photometer (Shimadzu) placed in an anaerobic chamber. A solution containing 11 µM RubL, 25 mM EDTA and 5 µM methylviologen (all components were rendered anaerobic prior to mixing) was prepared in 50 mM Tris, 300 mM NaCl, 10 % glycerol, pH 7.4 and a UV-visible absorption spectrum was recorded between 300 and 800 nm. Then, the sample was irradiated and changes in the UVvisible absorption characteristics were monitored until full reduction of the protein-bound FAD cofactor was observed (no further spectral changes). Finally, the solution was exposed to air for several minutes and a spectrum of "reoxidized" RubL was recorded.
Finally, catalytic amounts of xanthine oxidase (0.06 U, Roche) were added and changes of the UV-visible absorption characteristics were monitored until full reduction of the protein-bound FAD cofactor and the reporting dye was observed (30 min). By plotting the log(RubLox/RubLred) at 357 nm (isosbest of anthraquinone-2,6-disulfonate reduction) as a function of the log(dyeox/dyered) at 343 nm (isosbest of RubL-bound FAD reduction) a Nernstplot was generated, allowing for the calculation of the reduction potential of RubL from the resulting y-axis intercept [3] (quadruplicate determination).

Steady state kineticsreduction of 3 by RubL under anoxic conditions
Steady-state parameters for the reduction of collinone by RubL wild type were determined using a UV-1800PC photometer (Shimadzu) placed in an anaerobic tent. All assay components were rendered anaerobic and reaction mixtures containing 1 mM NADPH and (rates at each substrate concentration were determined at least in triplicate).

Reduction of 3 by RubL-wild type and -variants under anoxic conditions
To assay the effect to several amino acid replacements in the active site of RubL on 3 reduction, steady-state assays were performed under anoxic conditions. Reaction mixtures

NADPH-concentration dependence of collinone reduction
To study the effect of the NADPH-concentration on the rate of collinone reduction, again a photometric steady-state assay was performed under anoxic conditions. Reaction mixtures were extracted and plotted as a function the respective NADPH-concentration, revealing a liner correlation between the two parameters (rates at each NADPH-concentration were determined in triplicate).

Site-directed mutagenesis
In order to investigate the role of several active site residues in catalysis, different RubLvariants were generated using PCR-based mutagenesis. In all cases, the mutations were introduced to the pET16b-RubL wild type expression plasmid using both forward and reverse primers carrying the desired nucleotide replacements. production and purification were carried out as described for RubL wild type (see above).

Short-time conversion assays (HPLC-analysis based)
To study the effect of several amino acid replacements on the efficiency of collinone reduction, apart from the steady-state assay conducted under anoxic conditions, also a short turnover

10 min conversion assays (HPLC-analysis based)
To investigate the role of the key active site residues on the efficiency of collinone conversion/lenticulone formation, a 10 min turnover assay was performed. Reaction mixtures

Structure-based sequence alignment (group A monooxygenases)
Sequences for the alignment were selected based on a blastp of the RubL sequence against the sequences/structures deposited in the protein database (PDB)duplicates and sequences with no significant similarity to RubL were omitted. Instead, the sequences of the known RubLhomologs GrhO5, HlqO5, and HyalO5 were added. Then, a structure-based alignment was generated using the online program T-coffee [4] and visualized in Jalview [5] . Upon comparison of the sequence alignment with the overlaid 3D-structures, errors in the alignment of phenol hydroxylase and m-hydroxybenzoate hydroxylase to the other sequences were observed, requiring manual editing of the alignment. Finally, the alignment was visualized using SeaView [6] and a bootstrapped distance tree (1000 repetitions) was generated in the same program. The rooted tree was exported in Newick format and visualized and color-edited using the online program iTOL [7] .

Analytical gelfiltration
Analytical gelfiltration was carried out to determine the biologically active oligomer of RubL wild type and its variants as well as of GrhO5 in solution. Protein samples were applied to a Superdex 200 10/300 GL column, pre-equilibrated with 20 mM Tris, 50 mM NaCl, pH 7.4 and the molecular weight of the protein eluting in the main peak fraction was estimated based on a previously generated calibration curve.

HPLC detection
Turnover assays requiring offline detection, were analyzed by HPLC using an Agilent 1100 chromatographic system (Technologies), equipped with a VP NUCLEODUR 100-5 C18ec were selected to be monitored by the diode array detector connected to the chromatographic system.

Crystallization -GrhO5
GrhO5 was crystallized using sitting GrhO5 + 3 complex crystals were prepared by washing a dislodged single GrhO5 crystal in reservoir solution (2 times) and by subsequently soaking in reservoir solution supplemented with 4 mM 3 in 10 % DMSO for 3 hours prior to cryoprotection and freezing.

X-Ray Data Collection and Structure Determination
High-resolution X-ray diffraction datasets were collected at the PXI-X06SA beamline (SLS and scaled using SCALA [9] . Structure determination of RubL and GrhO5 was performed by molecular-replacement with PHASER [10] as implemented in CCP4 suite, using a RubL-or GrhO5-structure model generated by the FFAS server (https://ffas.godziklab.org/ [11] ) or SWISS-model server [12] , respectively, based on the crystal structure of Streptomyces purpurascens RdmE of rhodomycin biosynthesis (PDB ID: 3ihg) as the search model [13] . The structures were modified manually with Coot [14] and refined with REFMAC [15] or PHENIX [16] [10]. For omit map calculations Polder OMIT [17] maps tool was used. Restraints for collinone and 1,2,6-hexanetriol were generated using AM1 QM optimization method in eLBOW [18] as implemented in PHENIX [16] .   Table S2: Overview of all group A FPMOs and their origin that are displayed in the phylogenetic tree of Figure 3. A, The active sites feature unique conserved tryptophan residues for GrhO5-like type I enzymes. While the FAD cofactor in GrhO5 is held in the "OUT"-position by sandwich π-π stacking with W281, this particular tryptophan is missing in the type II enzymes such as MtmOIV and other group A FPMOs such as PHHY. These enzymes, however, exhibit another conserved tryptophan residue, which is missing in the type I enzymes. Note that GrhO6 and the functional homologs of GrhO5 (HlqO5 and HyalO5) required for formation of other rubromycin polyketides have not been structurally characterized and that GrhO6 clusters with type II enzymes rather than the GrhO5-like type I. All shown enzymes are involved in the biosynthesis or modification of natural products apart from phenol hydroxylase (PHHY [19] ), meta-hydroxybenzoate hydroxylase (MHBH [20] ) and 2-hydroxybiphenyl 3monooxygenase (HbpA) [21] from bacterial and fungal primary metabolism. For a comprehensive tree including enzymes from primary metabolism and fungal group A FPMOs, see ref. [22] .  Figure S1: Uncropped structure-based multiple sequence alignment (cf. Figure 3 main text) showing the spiroketal synthases RubL and GrhO5, as well as previously crystallized group A monooxygenases. In addition, GrhO5/RubL homologs from heliquinomycin (HlqO5) and hyaluromycin (HyalO5) biosynthesis and the second group A monooxygenase involved in griseorhodin A biosynthesis (GrhO6) were included. The structurebased sequence alignment was generated using the online tool T-coffee [4] and visualized in SeaView [6] . Final color editing to highlight important amino acid residues (FAD-binding and solvent gate keeping) and structurally conserved regions (blue boxes) was carried out in PowerPoint. For PDB-IDs of the proteins included in the alignment, see Table S2.         While only a clear peak corresponding to monomeric enzyme (15 mL; 60 kDa) is observed in the gel-filtration chromatogram of the RubL wild type, additionally, both a substantial aggregate peak at 10 mL and a peak corresponding to free FAD (20 mL) are found in the elution profile of RubL-W289A. Figure S12: Determination of the reduction potential of RubL wild type. A, Selected UV-visible absorption spectra recorded in the course of the xanthine oxidase catalyzed reduction of the RubL-bound FAD cofactor and the reporting dye anthraquinone-2,6-disulfonate (E°: -184 mV; [23] ). B, Nernst-plot generated by plotting the log(ox/red) of the protein-bound FAD (at 357 nm) as a function of the log(ox/red) of the dye (at 343 nm). Using the value of the y-axis intercept, subsequently the midpoint potential of RubL was calculated (-177 ± 1 mV; from 4 determinations). Due to the overlapping spectra, the reduction potential was calculated based on the spectral changes at the isobestic points observed in the course of the separate reduction of RubL (343 nm) and anthraquinone-2,6-disulfonate (357 nm).  [24] ) and anthraquinone-2-sulfonate (B; E°, -225 mV [25] ). A, The spectral changes indicate that only the reporting dye, but not the RubL-bound FAD became reduced, suggesting a more negative reduction potential (than -125 mV) for the enzyme-bound cofactor. B, The spectral changes indicate that first only the protein-bound cofactor and later the reporting dye became reduced, suggesting a more positive reduction potential (than -225 mV) for the enzyme-bound cofactor.    Then, the reactions were stopped with EtOAc:FA (9:1) and the organic layers immediately analyzed by HPLC-DAD. While almost full conversion of 3 into 5 was observed for the wild type enzyme, particularly the arginine-variants (R87K, R374K, and R374M) and the tryptophan-variant (W289A) showed a strongly reduced reduction activity. Note that in the shown WT trace, 3 was almost completely converted to 5 without detectable amounts of the final products 4 and 7 with retention times of 14.6 min and 10.1 min, respectively (compare Figure S18). This suggests that enzyme-bound 5 is most likely displaced by remaining 3 before hydroxylation can occur.

) reduction rates determined under anoxic (column 2 and 3) and aerobic conditions (column 4).
Both in the photometric, as well as in the HPLC-based assays, very similar effects of the amino acid replacements on 3 reduction were observed. While the substitution of E103 or H225 resulted in residual activities of 20-30 %, the replacement of R87, R374 or W289 diminished the reduction activity to 3-10 %. Interestingly though, for some of the variants the relative reduction rates at the lower 3 concentration of 50 µM were slightly higher rather than lower, despite the anticipated role of both arginine residues in substrate binding (however, evaluation of the data at a substrate concentration of 50 µM is challenging, as the absorption at 650 nm (max) is close to the detection limit).  To investigate the influence of several amino acid replacements on the efficiency of 3 turnover and 4/7 formation, RubL wild type and seven variants (10 µM) were separately mixed with a 350 µM 3 solution (in the presence of 1.5 mM NADPH) and incubated at 30°C and 750 rpm for 10 min. Then, the reactions were stopped with EtOAc:FA (9:1) and the organic layers immediately analyzed by HPLC-DAD. While substantial conversion of 3 into 4 and 7 was observed in assays with the wild type protein as well as the glutamine and the histidine variants, turnover was significantly lower in the reactions containing the R374K-variant. No reaction products were found in the assay mixtures containing the W289A-, the R87K-or the R374M variant, suggesting a crucial role of these residues in catalysis. Reactions were quenched before full turnover occurred to allow for comparison of the reaction rates. Shown are relative amounts of residual 3 in the assay mixture (column 1) and of produced compounds 5, 7 and 4 (columns 2, 3 and 4, respectively). n.d. = not detectable.