Molecular insights into the unusually promiscuous and catalytically versatile Fe(II)/α-ketoglutarate-dependent oxygenase SptF

Non-heme iron and α-ketoglutarate-dependent (Fe/αKG) oxygenases catalyze various oxidative biotransformations. Due to their catalytic flexibility and high efficiency, Fe/αKG oxygenases have attracted keen attention for their application as biocatalysts. Here, we report the biochemical and structural characterizations of the unusually promiscuous and catalytically versatile Fe/αKG oxygenase SptF, involved in the biosynthesis of fungal meroterpenoid emervaridones. The in vitro analysis revealed that SptF catalyzes several continuous oxidation reactions, including hydroxylation, desaturation, epoxidation, and skeletal rearrangement. SptF exhibits extremely broad substrate specificity toward various meroterpenoids, and efficiently produced unique cyclopropane-ring-fused 5/3/5/5/6/6 and 5/3/6/6/6 scaffolds from terretonins. Moreover, SptF also hydroxylates steroids, including androsterone, testosterone, and progesterone, with different regiospecificities. Crystallographic and structure-based mutagenesis studies of SptF revealed the molecular basis of the enzyme reactions, and suggested that the malleability of the loop region contributes to the remarkable substrate promiscuity. SptF exhibits great potential as a promising biocatalyst for oxidation reactions. Non-heme iron and α-ketoglutarate-dependent (Fe/αKG) oxygenases have attracted attention for their application as biocatalysts due to their flexibility and high efficiency. Here, the authors show the biochemical and structural characterizations of the versatile Fe/αKG oxygenase SptF, involved in the biosynthesis of fungal meroterpenoid emervaridones.

O xygenases are widely distributed in nature and catalyze the incorporation of oxygen into various natural products. One of the largest oxygenase groups is the nonheme iron and α-ketoglutarate-dependent (Fe/αKG) oxygenase superfamily, which is involved in both primary and secondary metabolism 1 . In animals, Fe/αKG oxygenases play important roles in physiological processes via hydroxylation or N-demethylation reactions, with substrates including proteins, nucleic acids, and lipids 1 . In plants and microorganisms, in addition to the typical hydroxylation reactions, Fe/αKG oxygenases catalyze a wide range of chemical transformations in the biosynthesis of natural products  , including complex skeletal rearrangement 11,12 , ringexpansion 15 , and C-C bond formation [21][22][23] . Given their intriguing chemical reactions, detailed biochemical, structural, and calculation studies of Fe/αKG oxygenases have been conducted over the past few decades 5,6,8,[28][29][30][31][32] .
Fe/αKG oxygenases utilize αKG as a co-substrate and Fe(II) as a cofactor [33][34][35] . Through the oxidative decarboxylation of αKG, a reactive Fe(IV)-oxo intermediate is generated to activate a selective aliphatic C-H bond, which is normally difficult to cleave due to high bond-dissociation energy, and then oxidative products are generated following radical recombination. Unlike other oxidases, including P450 monooxygenases, Fe/αKG oxygenases do not require any reductase partners or expensive cofactors such as NAD(P)H and FAD/FMN 36 . Furthermore, these enzymes exhibit high catalytic efficiencies, and typically have large turnover numbers 11,14,37,38 . These features of Fe/αKG oxygenases, along with their catalytic versatility, have attracted considerable attention as a source of biocatalysts 36,39 .
SptF is an Fe/αKG oxygenase first discovered in the biosynthesis of the fungal meroterpenoid emervaridones 40 . It catalyzes two oxidation steps to form emervaridone C (3) from andiconin D (1) (Fig. 1, Supplementary Fig. 1). The proposed reaction scheme involves the initiation by a C11 hydroxylation reaction, followed by water elimination, carbon skeletal rearrangement, and deprotonation reactions to yield 2. Compound 2 is then accepted by SptF again and undergoes an epoxidation reaction to generate 3. Thus, SptF accepts two skeletally distinct molecules as substrates and plays a key role in the structural complexification of andiconin-derived meroterpenoids. Although the function of SptF has been investigated by in vitro enzymatic reactions, the detailed mechanisms of the dynamic skeletal rearrangement and the accommodation of structurally distinct substrates remain to be elucidated.
In this study, we investigated SptF further by in vitro enzyme reactions, crystallization, and structure-based mutagenesis. To our surprise, SptF is an unusually promiscuous and catalytically versatile enzyme that performs up to four rounds of oxidative reactions with its natural substrate 1, including hydroxylation, desaturation, epoxidation, and skeletal rearrangements. SptF also catalyzed the formation of cyclopropane-ring-fused, highly congested 5/3/5/5/6/6 and 5/3/6/6/6 cyclic skeletons from structurally distinct meroterpenoid terretonins. Moreover, SptF hydroxylated steroids, including androsterone, testosterone, and progesterone, with different regiospecificities. Our crystallographic analyses of SptF in complex with (un)natural substrates, along with our structure-based mutagenesis studies, have revealed the molecular basis for the substrate recognition and suggested that the malleability of the loop region contributes to the remarkable substrate promiscuity.

Results
Substrate promiscuity and catalytic versatility. Emervaridone C (3) was previously reported as the final product of SptF from andiconin D (1) 40 . However, our careful reexamination of the enzyme reaction revealed that SptF actually produced two more products 4 and 5, in addition to 2, 3, and previously reported emeridone F 40 (Fig. 1a and Supplementary Fig. 1). Structural analyses established that 4 contains an additional C9 hydroxyl group and C1-C2 epoxide (Fig. 1c, Supplementary Figs. Table 8). In contrast, 5 has a new ether bridge between C4' and C2', along with a C5' hydroxyl group (Fig. 1c, Supplementary Figs. 25-30 and Supplementary Table 9). Notably, SptF, in the emervaridone biosynthetic pathway, shares 94% amino acid sequence identity with AndF, which is responsible for the generation of anditomin (7) from andilesin C (6), in the closely related anditomin pathway (Fig. 1c) 37 . As expected, SptF also accepted 6 as a substrate to produce 8 and 9 in addition to 7 (Fig. 1b, Tables 10 and 11). Interestingly, 8 and 9 share the same skeleton, but have different stereochemistry at the C3'-C9' epoxide ring.

2, 19-24 and Supplementary
To determine the reaction sequence, putative intermediates, including 2, 3, and emeridone F were used as substrates of the SptF enzyme reactions. As a result, both 2 and 3 were efficiently converted into 4 and 5, which clearly indicated that SptF converts 1 into 3 via 2, and then 4 and 5 are generated from 3 (Supplementary Fig. 11c, d). Similarly, 8 and 9 were generated from 6 via 7 ( Supplementary Fig. 11e). In contrast, emeridon F with the C9β hydroxyl group was not accepted by SptF, indicating that it is a shunt product. It is remarkable that SptF, a single enzyme, accepts structurally distinct meroterpenoids as substrates and catalyzes sequential oxidation reactions, including desaturation, epoxidation, hydroxylation, ring-opening, and carbon skeletal rearrangements, to generate skeletally diverse products.
To further evaluate the utility of SptF as a biocatalyst, we also tested various steroids, including androsterone (21), testosterone (24), and progesterone (28), as substrates. To our surprise, SptF accepted 21, 24, 28 of them with high conversion rates of 34%-53% over 24 h (Supplementary Table 2). The structures of the products were determined by NMR and the crystalline-sponge-based X-ray diffraction method (Fig. 3 [15][16][17][18]. SptF thus accepted 21 and installed one hydroxyl group at C5 to yield 22, and the second hydroxyl group at C9 to form the steroid 23. In contrast, 24 underwent hydroxylation at C9 to yield 25. Moreover, the hydroxylation and following oxidation at C6 of 24 generated 26, which is interchangeable with 27 according to the NMR data. Although the structures of products 29-31 from 28 have not been determined yet, due to separation and stability problems, they are also predicted to be hydroxylated and/or oxidized products as judged from their molecular weights ( Supplementary Fig. 2). The substrate promiscuity and catalytic versatility of SptF are thus quite exceptional. In contrast, ergosterol and lanosterol, produced by the natural host, were not accepted as a substrate by the enzyme.
In the active site, 1 or 6 is located on the opposite side of the facial triad in an almost identical fashion (Fig. 4a, b and Fig. 5a, b), indicating that the size of the A-ring does not affect the binding mode. The distance between the iron and the initial   Fig. 1 Characterization of SptF function in vitro using andiconin D (1) and andilesin C (6) as substrates. a LC-MS EICs of products from the in vitro assay using 1 as the substrate, and b using 6 as the substrate (the + m/z value for each compound is shown). c Biosynthesis and structures of SptF and AndF products, using 1 and 6 as substrates in vitro. Note that the products 4, 5, 8, and 9 were isolated from large scale enzymatic reactions, and their structures were determined by NMR.  Fig. 8) was only observed in complex structures with 1 or 6 ( Fig. 4a, b). This lid-like loop region interacts with the A-ring of 1 or 6 via only one hydrogen bond with Asn65, suggesting loose interactions between the lid-like region and substrate. In contrast, the D/E-rings of 1 and 6 are tightly fixed by several hydrophilic residues located on the bottom surface of the active site cavity: the carbonyl oxygen of the E-ring interacts with Ser114, while the carbonyl oxygen of the D-ring interacts with the main chains of Leu199 and Thr148 via water molecules (Fig. 4a, b). Interestingly, the unnatural substrate 15 shows a different ligand binding mode in the active site. The D-ring of 15 forms a hydrogen-bond network with Thr148 and Leu199 via water molecules, but also forms a hydrogen bond directly with Asn150 (Figs. 4c and 5c). Notably, the lid-like loop region was not observed upon the binding of 15, while the conformations of the other active site residues were almost identical. In this structure, Phe133 and Ile231 are located close to the methyl groups of C8, C10, and C13, and potentially form hydrophobic interactions. The C6 enol group of 15 is close to the iron with a distance of 3.7 Å, suggesting that the enolic hydrogen atom is abstracted to initiate the cyclopropane-ring formation reaction via the proposed path c in Supplementary Fig. 9.
Structure-based mutagenesis. To investigate the importance of the active site residues, we performed structure-based mutagenesis. Firstly, the hydrophobic residues Ile63, Phe133, and Ile231, lining the active site cavity, were substituted with Ala ( Fig. 6a-c, Supplementary Figs. 10-12). As a result, the I63A variant abolished the formation of 4 and 5, and instead generated the new product 32 (Fig. 6d) in addition to 2. The structure of 32 was determined to be an E-ring opened dicarboxylic acid, which is possibly formed by two rounds of oxidation at the C1' position of 1 (Fig. 6b, e, Supplementary Figs. 10, 83-88, and Supplementary  Table 19). On the other hand, I63A oxidized 2 to 3, and converted 3 to small amounts of 4 and 5 ( Supplementary Fig. 11a, b), whereas F133A abolished the production of 4 and 5, but accumulated the early-stage intermediates 2 and 3 (Fig. 6b)  k cat /K M value of F133A for 1 decreased more than 1000-fold as compared with the wild-type enzyme (Supplementary Table 4). F133A also converted 1 into the C11-hydroxylated product 33 (Supplementary Figs. 89-94 and Supplementary Table 20). In contrast, the substitution of Phe133 with Tyr retained the generation of both 4 and 5 (Fig. 6b). Moreover, the I231A variant showed a similar LC-MS profile to that of F133Y, while affording 32 as the major product. The loss of the sec-butyl group increases the active site space around the D-ring of 1, which could lead to an alternative substrate binding mode. Notably, similar results were observed in the enzyme reactions with the unnatural substrate 15. The activity for the generation of the cyclopropane ring was abolished in F133A, but retained in F133Y. Moreover, I231A significantly reduced the activity to produce 16 ( Fig. 6c), suggesting the similar role of the hydrophobic residues in the recognition of the unnatural substrate 15. Compound 33, generated by F133A variant, is a proposed biosynthetic intermediate of 2 (Supplementary Fig. 1) 40 . However, in vitro enzyme reaction of 33 with SptF revealed that 33 is not converted to the final products 3-5, and therefore 33 is not an intermediate in the pathway ( Supplementary Fig. 13). Secondly, the hydrophilic residues that interact with substrates, including Asn65, Ser114, Thr148, and Asn150, were substituted with alanine ( Fig. 6a-c). Due to the insolubility of T148A, the T148S variant was used for the enzyme reaction. As a result, N65A significantly reduced the 4 and 5 formation activity to 20% and accumulated 3 (Fig. 6b, e), suggesting that the hydrogenbond interaction with Asn65 is not completely essential but moderately important for the binding of 3. Interestingly, as compared with wild-type SptF, the activities for the formation of 4 and 5 were maintained at 83%, 93%, and 52% in the S114A, T148S, and N150A variants, respectively. The ratio of 4 and 5 was altered to 7:3 in S114A and T148S, as compared with 4:6 in the wild-type, and decreased to 1:9 in N150A (Fig. 6e). Our observations suggested that these three residues determine the position of the hydrogen atom abstraction (C9 or C7) by anchoring 3 within the binding pocket, to generate 4 and 5 via a hydrogen-bond network.
Asn65 in SptF is uniquely substituted with threonine in the closely related AndF ( Supplementary Fig. 12). To clarify the function of Asn65, we constructed the SptF N65T variant, which converted 1 into 5 as the major product, with a slight amount of 4 ( Fig. 6b, e). Notably, in the crystal structure of SptF N65T-Fe/ NOG/1, a hydrogen bond was observed between the C3 ketone at the A-ring and the newly introduced Thr65 (Figs. 5d, 6f, and Supplementary Table 3). These results indicated that a subtle change of the hydrogen-bond interactions affects the position of the hydrogen atom for abstraction and determines the product specificity. Interestingly, only the N150A variant lacked the activity for the conversion of 15 into 16 (Fig. 6b, c), suggesting that the direct hydrogen bonding between Asn150 and 15 is crucial for the substrate binding.
Lid-like loop region and substrate recognition. The mutagenesis studies revealed that the hydrogen-bond and hydrophobic interactions in the active site cavity are important for the substrate recognition. In contrast, while Ile63 and Asn65 on the lidlike loop region are important for the formation of 4 and 5, these residues are not essential for the reaction with the unnatural substrate 15, suggesting that 15 is mainly recognized by active site residues but not via residues on the lid-like loop. Indeed, the kinetic analysis of I63A and N65A variants toward 1 and 15 revealed that the k cat /K M values for these variants towarding 15 were comparable to that of wild-type SptF due to the decrease of testosterone (24) Table 4).
To further investigate the importance of the flexible lid-like loop region, with a length of 19 amino acid residues, we tested truncated loops. Based on the sequence comparison as well as the structures of SptF in complex with 1, loop-truncated variants of 6 residues (His61-Val66), 9 residues (Lys58-Val66), 16 residues (Asn54-Lys69), and 19 residues (Trp53-Lys71) were constructed ( Supplementary Figs. 14 and 15). Since the Δ19 variant was expressed in inclusion bodies, the Δ6, Δ9, and Δ16 variants were tested for their enzyme reaction activities with 1, 2, and 3. Interestingly, although the activities of the truncated variants were dramatically decreased as compared to the wild-type, all the loop-truncated variants still consumed 60-94% and 17-77% of 1 and 2 over 24 h, respectively, and generated many minor products, including the E-ring-opened product 32, but not 3-5 (Fig. 6b, Supplementary Fig. 16 and Supplementary Table 6). The differences in the activities among Δ6, Δ9, and Δ16 should be due to the folding of the enzymes. On the other hand, these variants also showed 26-56% activities toward 3 and generated small amounts of 4 and 5, as well as other minor compounds (Supplementary Fig. 16 and Supplementary Table 6). Although we could not determine the structures of the minor products due to low yields, these results suggested that the lid-like region is not essential for the reactions with 1-3, but important for the reaction efficiency and product selectivity. We also tested the activities of the truncated variants with unnatural substrates, including the fungal meroterpenoids 10 and 13, and steroids 21, 24, and 28. As a result, all the variants still maintained 24-40% and 56-84% activities toward 10 and 13, but dramatically decreased activities toward 21, 24, and 28 (<20%) (Fig. 6c, Supplementary Fig. 17, and Supplementary Table 6). These observations indicated that the lid-like loop region is crucial for the steroid substrate recognition, but not essential for the meroterpenoid substrates, which is consistent with the absence of the electron density of the loop in the complex structure with 15. The kinetics analysis of the loop-truncation variant Δ9 revealed that the variants also maintained comparable activity toward 15 to that of wild-type (Supplementary Table 4).
Thus, the mutagenesis and kinetic assay results strongly suggest that the malleability of the lid-like loop region contributes to the remarkable substrate promiscuity and catalytic versatility of SptF.

Discussions
Substrate promiscuity plays a key role in both biomedical and industrial applications, as it provides a starting point for protein engineering and drug design, enzyme evolution with different reaction/substrate specificities, and chemoenzymatic total synthesis 36,39,[43][44][45][46] . Therefore, the identification and investigation of promiscuous enzymes enrich our understanding of how the enzymes obtain such a wide range of diverse functions, and will further guide the rational engineering of these enzymes as biocatalysts for the production of useful molecules with improved biological activities.
Hydroxylation of steroids has been previously reported for cytochrome P450 enzymes 47 Fig. 6 Mutagenesis studies. a Active site cavity of SptF with 1 (shown as cyan sticks). Hydrophobic residues (shown as orange sticks), Ile63, Phe133, Ile231, and V264', form a hydrophobic surface to interact with the hydrophobic part of the substrate, whereas hydrophilic residues (shown as salmon sticks), Asn65, Ser114, Thr148, and Asn150, interact with the hydrophilic part of the substrate. b, c LC-MS EICs of products from in vitro assays with mutants using 1 or 15 as the substrate. The + m/z value used for each compound is shown. d Structures of new products generated by mutants. Compounds 32 and 33 were isolated from large scale enzymatic reactions using the I231A and F133A mutants, respectively. Structures were determined by NMR. e Ratios of 4 and 5 from mutants compared with those from wild-type SptF. All reactions were performed in triplicates. Data are presented as mean values and the error bars indicate standard deviations (SD). f Active site residues interacting with 1 (shown as cyan sticks), including hydrogen bonding directly with Ser114 and the newly introduced Thr65, and indirectly with Thr148 and Leu199 via waters in N65T (shown in purple). Note that the coloring scheme for ligands is the same as in Fig. 4.
the hydroxylation of testosterone, a 3-keto-Δ 4 -steroid, at the βface of the C3 and C15 positions 48 . However, in spite of intensive engineering efforts, P450 BM3 does not accept 3-hydroxysteroids 49 . Thus, SptF exhibits great potential as a promising biocatalyst for the oxidation of various natural products with important pharmacological activities. Based on our structure-function analyses, we propose enzyme reaction mechanisms for the consecutive oxidation reactions in the emervaridone biosynthetic pathways, as follows (Fig. 7). Hydrogen abstraction at C11 of 1 by the ferryl species first generates a radical, which undergoes recombination to cleave the C-C bond between C4' and C3' and bridge C4' and C11, but not through the hydroxylated intermediate. Subsequently, a double bond is formed to yield 2, which undergoes epoxidation to generate 3. Compound 3 then serves as the branch point intermediate to produce 4 and 5.
In path a, epoxidation at A-ring of 3 generates IM1, which then undergoes hydroxylation at C9 to generate 4. It should be noted that the C9 hydroxylation occurs with attack from the front side of the molecule, whereas all the other attacks by SptF can be explained from the back side, which would suggest that SptF can accept substrates with different binding mode. In path b, hydrogen abstraction at C7 induces the radical recombination, which cleaves the C8-C2' bond to construct a double bond between C7 and C8 on the B-ring. The hydroxyl rebound at C2' forms IM2, which is followed by lactone formation and C4'-C5' bond cleavage to yield IM3, in a similar manner to the formation of emeridone A 40 . In the last step, hydrogen abstraction at C7' and relocation of the radical to C5' complete the formation of the final product 5. The LC-MS analysis of the enzyme reaction mixture suggested the presence of the intermediates IM1-3; however, their structures could not be determined because of instability and low yield ( Supplementary Fig. 11).
The multifunctional catalytic activities largely stem from the unusual substrate promiscuity of SptF, which can accommodate a series of structurally different molecules in the active site for further oxidation reactions. The crystallographic and mutagenesis investigations have provided the structural basis for the promiscuity of SptF. In particular, SptF is unique since the lid-like loop region interacts with the natural substrate 1 via only one hydrogen bond with Asn65, which is clearly distinct from other fungal meroterpenoid Fe/αKG oxygenases, including AndA and PrhA 11,12 . In AndA and PrhA, the residues on the lid-like loop tightly recognize the D/E-rings of substrates via a hydrogen-bond network, while the A/B-rings loosely interact with the active site ( Supplementary Fig. 18). These relaxed interactions and high malleability of the lid-like loop region in SptF are thought to contribute to its remarkable substrate promiscuity.
Truncations of the lid-like loop region demonstrated that the loop is not essential for the activity toward the natural and unnatural meroterpenoid substrates, but important for binding the steroid substrates and the product selectivity from natural substrates. The substrate is thought to be mainly recognized by the active site residues deep inside the cavity, which is then covered by the lid-like loop to further support the substrate loading and ensure the precise and efficient positioning for the C-H bond activation by the ferryl species. This whole process would be rigidly controlled for the natural substrate. In contrast, the flexible loop may interact loosely with the unnatural substrates, but still retains the efficiency and product specificity of the enzyme reactions. In particular, it would be difficult to tightly fix smaller steroids with a few functional groups in the active site. These molecules would probably loosely bind to the enzyme in several different conformations, and the enzymatic C-H bond activation could only take place when the substrate is properly  positioned for the interactions with the lid-like loop region. This would be the reason why we could not obtain the steroid-bound complex structure, despite our soaking experiments. Therefore, understanding how the lid-like loop region interacts with different substrates would provide key information to manipulate the SptF enzyme reaction.
In conclusion, we characterized the exceptionally promiscuous and catalytically versatile Fe/αKG oxygenase SptF, which catalyzes a remarkable repertoire of oxidation reactions of meroterpenoid and steroid molecules. We believe these unique features of SptF will serve as a platform for future efforts to optimize desired specific catalytic functions through directed evolution of the enzyme. SptF thus exhibits great potential as a promising biocatalyst for oxidation reactions.

Methods
General. Oligonucleotide primers, listed in Supplementary Table 7, and DNA sequencing were ordered from Eurofins Genomics. The restriction enzymes NdeI and HindIII, and PrimeSTAR GXL DNA polymerase were purchased from Takara Bio Inc. Solvents and chemicals were purchased from Wako Chemicals, Ltd. (Tokyo, Japan), Merck KGaA Ltd. (Darmstadt, Germany), and Hampton Research (CA, USA), unless noted otherwise. PCR was performed using a TaKaRa PCR Thermal Cycler Dice ® Gradient (Takara Bio Inc.). The NMR spectra of compounds were recorded on Bruker AVANCE III HD 800 MHz, AVANCE III HD 900 MHz (Bruker MA, USA), ECX-500 MHz, and EXZ-500 MHz (JEOL Ltd., Tokyo, Japan) spectrometers.

Construction of SptF expression plasmid.
To obtain stable and pure proteins for in vitro enzyme reactions and crystallization experiments, expression plasmids for SptF lacking three residues, predicted to be flexible by XtalPred server 50 , were prepared as follows. The primer pair SptF-N3-F/SptF-N3-R was used for amplification of the SptF DNA fragment from the previously reported pET-28a(+)-sptF 40 plasmid (see Supplementary Table 7 for primer sequence information). SptF PCR products were purified and ligated into the NdeI/HindIII-digested pET-22b (+) vector, using an In-Fusion® HD Cloning Kit (TaKaRa Clontech). The identities of the resulting vectors were confirmed by DNA sequencing. Escherichia coli DH5α was used for plasmid construction, and E. coli Rosetta TM 2(DE3)pLysS (Novagen) was used for the preparation of recombinant proteins.
Construction of plasmids for expression of SptF mutants. The primers used for the construction of plasmids for site-directed mutagenesis studies are listed in Supplementary Table 7. The plasmid for the expression of wild-type SptF was used as the template for PCR-based site-directed mutagenesis, which was performed with a QuikChange Site-Directed Mutagenesis Kit (Stratagene) according to the manufacturer's protocol. For the construction of loop-truncation mutants, the primer pairs listed in Supplementary Table 7 were used for the amplification of SptF DNA fragments from pET-28a(+)-sptF 40 plasmids. SptF PCR products were purified and ligated into the NdeI/HindIII-digested pET-22b (+) vector, using an In-Fusion® HD Cloning Kit (TaKaRa Clontech). The identities of the resulting plasmids were confirmed by DNA sequencing.
Production and purification of SptF and its mutants. In this study, the recombinant proteins used for in vitro assays were prepared as follows: Plasmids expressing SptF or its mutants were transformed into E. coli Rosetta TM 2(DE3) pLysS (Novagen). The resulting strains were cultured in LB medium supplemented with 34 mg/L chloramphenicol and 50 mg/L ampicillin sodium, with shaking at 37°C. When the OD 600 reached 0.6, the cell cultures were cooled down on ice for 30 min, and then IPTG (0.2 mM) was added to induce the target protein expression at 16°C. After 18 h of post-induction incubation, cells were harvested by centrifugation at 3300 × g for 15 min and suspended in lysis buffer, containing 50 mM Tris-HCl (pH 8.0), 300 mM NaCl, 10 mM imidazole, and 5% glycerol. The cell suspension was sonicated for 4 min on ice. After removal of the cell debris by centrifugation at 20,400 × g for 30 min, the supernatant was mixed with 1 mL Ni-NTA agarose resin and loaded onto a gravity flow column. Unbound proteins were removed with 50 mL lysis buffer containing 25 mM imidazole, and then the Histagged protein was eluted with lysis buffer containing 300 mM imidazole. For the in vitro assay, the protein eluate was concentrated to 10 mg/mL after the removal of imidazole, using a 30 kDa Amicon® Ultra-15 filtration unit (Millipore). For crystallization, the eluate from the Ni-NTA agarose was applied to a HiLoad 16/60 Superdex 200 prepacked gel filtration column (4°C, GE Healthcare), and eluted with a solution containing 20 mM Tris-HCl (pH 8.0), 100 mM NaCl, and 1 mM dithiothreitol. The resulting eluate was concentrated to 10 mg/mL, using an Amicon Ultra-4 (MWCO: 30 kDa) filter at 4°C. The purity of the purified proteins was monitored by SDS-PAGE ( Supplementary Fig. 15), and the protein concentrations were determined with a SimpliNano microvolume spectrophotometer.
In vitro assay of SptF and its mutants. Substrates used for in vitro enzyme reactions, including 1, 6, 10, 12, 13, 15, 17, and 19, were prepared according to previous studies 12,37,40,51 , ergosterol and lanosterol were purchased from TCI (Japan). and the proteins were prepared as described above. All reactions in this study were performed following the previous studies unless noted otherwise 40 . Reactions were performed on a 50 μL scale with 50 mM PIPES (pH 7.5), 0.2 mM FeSO 4 , 5 mM α-ketoglutarate, 4 mM ascorbate, 100 μM of substrate, and 15 μM of wild-type or mutant SptF. The reaction mixture without protein was used as the negative control. Enzyme reactions, except for steady-state enzyme kinetics, were incubated at 30°C for 1 h or overnight and quenched by adding 50 µL methanol. After centrifugation and filtration, the reaction mixtures were analyzed by LC-MS. Crystallization and structure determination. Crystals of SptF lacking three residues, as well as the N65T and S114A mutants, were obtained after 1 day at 10°C by using the sitting-drop vapor-diffusion method with the following reservoir solutions: SptF-apo: 0.27 M magnesium chloride, 21% w/v PEG 3350, 1.5% trimethylamine N-oxide dihydrate. SptF N65T: 0.2 M magnesium formate dehydrate, 20% w/v PEG 3350. SptF S114A: 0.27 M magnesium chloride, 20% w/v PEG 3350, 0.2 M sodium chloride, 1.5% trimethylamine N-oxide dihydrate. The protein concentrations used for the crystallizations of SptF, N65T, and S114A were 7.5 mg/L, 10 mg/L, and 7.5 mg/L, respectively. Since S114A can produce 4 and 5, and has better diffraction as compared with wild-type SptF, we employed it for the soaking experiments of 15. The complex structures were prepared by incubating SptF, N65T, or S114A crystals at 10°C for 12 h with 5 mM of compound, 5 mM of αKG/NOG, and 100 µM of Fe in the crystallization drop. When αKG was used, the soaking experiment was performed in an anaerobic chamber.
The crystals were transferred into the cryoprotectant solution (reservoir solution with 25% (v/v) glycerol), and then flash cooled at −173°C in a nitrogengas stream. The X-ray diffraction datasets were collected at BL-1A (Photon Factory, Tsukuba, Japan), using a beam wavelength of 1.1 Å;. The diffraction datasets for SptF were processed and scaled using the XDS program package 53 and Aimless in ccp4 54 . The initial phase of the SptF structure was determined by molecular replacement, using AndA (PDB ID: 5ZM4) as the search model. Molecular replacement was performed with Phaser in PHENIX 55,56 . The initial phase was further calculated with AutoBuild in PHENIX 57 . The SptF structures were modified manually with Coot 58 and refined with PHENIX.refine 59 . The cif parameters of the ligands for the energy minimization calculations were obtained by using the PRODRG server 60 . The ligands were added in the unassigned density at active site with several conformations and compared their 2mFo-DFc and mFo-DFc maps and Real-space fit (R-value, RSR, and corr. coeff., RSCC) values calculated by PHENIX.refine and PDB validation server, respectively ( Supplementary Fig. 7). The final crystal data and intensity statistics are summarized in Supplementary  Table 3.
Steady-state enzyme kinetics. Each assay was performed in a total volume of 50 μL, containing 50 mM PIPES buffer (pH 7.5), 4 mM ascorbate, 200 μM FeSO 4 , and 1 mM αKG, at 30°C, and quenched by the addition of 50 μl methanol.  Table 4. All reactions were analyzed by LC-MS, and substrate consumption or product formation was determined by calculating the total peak areas of the substrate or product in each reactions as compared with negative control (without enzyme). For compound 1 and 6, kinetic parameters were determined based on consumption of substrates, while for 15, formation of product was used instead. Each reaction was performed in triplicate. GraphPad Prism (GraphPad Prism Software Inc., San Diego, CA) was used for statistics data analysis (Supplementary Table 4).
Enzyme reaction products from 1, 6, 10, 13, and 15. To isolate the products from 1, 6, 10, 13, and 15 for structure determination, in vitro enzymatic reactions were performed as described above in a total volume of 100 mL, divided into 0.5 mL portions in 2 mL tubes. After an overnight incubation at 30°C, the reactions were quenched and extracted twice with an equal volume of ethyl acetate. After the ethyl acetate removal, 1 mL methanol was added to dissolve the products for separation by semi-preparative HPLC, which was performed on a Shimadzu LC20-AD HPLC system.
Compounds 4, 5, and 33 were purified from large scale enzymatic reactions as mentioned above, using wild-type SptF and the F133A mutant with 1. Compounds 8 and 9 were prepared using wild-type SptF from 6. These products were purified on an XSelect HSS T3 OBD Prep column (10.0 × 250 mm, 5 μm) using 50% methanol supplemented with 0.1% v/v formic acid at 40°C. The injection volume was 0.1 mL and the flow rate was 2.5 mL/min. Finally, 0.3-0.4 mg of compounds 4, 5, 8, 9, and 33 were obtained.
Compound 32 was purified from large scale enzymatic reactions using I231A, on an XSelect HSS T3 OBD Prep column (10.0 × 250 mm, 5 μm) with H 2 O supplemented with 0.1% v/v formic acid as solvent A and methanol as solvent B, at 40°C. The gradient for separation was 46% B for 40 min, and 56% B for 60 min. The injection volume was 0.1 mL and the flow rate was 2.5 mL/min. The purification yielded 0.3 mg of compound 32.
Compound 11 was purified with an XSelect HSS T3 OBD Prep column (10.0 × 250 mm, 5 μm), using 45% acetonitrile supplemented with 0.1% v/v formic acid, at 40°C. The injection volume was 0.15 mL and the flow rate was 3.0 mL/min, and 0.9 mg of compound 11 was obtained.
Compound 11 was purified with an XSelect HSS T3 OBD Prep column (10.0 × 250 mm, 5 μm), using 45% acetonitrile supplemented with 0.1% v/v formic acid, at 40°C. The injection volume was 0.15 mL and the flow rate was 3.0 mL/min, and 0.9 mg of compound 11 was obtained.
Enzyme reaction products from 21 and 24. Reactions were performed in a total volume of 100 mL, with 100 mM HEPES (pH 7.5), 0.2 mM FeSO 4 , 4 mM α-ketoglutarate, 2 mM ascorbate, 1.6 mM of substrate, 10% glycerol, and 50 μM of SptF, at 4°C for 56 h. Products were separated by normal-phase silica gel chromatography. The elution conditions were as follows: CHCl 3 : MeOH = 100: 0.5 for 540 mL; then CHCl 3 : MeOH = 100: 1 for 90 mL, and CHCl 3 : MeOH = 100: 1.5 for 90 mL. Structure determination of 25 by crystalline-sponge method. A single crystal of [(ZnCl 2 ) 3 (tpt) 2 (cyclohexane) x ] (tpt = 2,4,6-tris-(4-pyridyl)-1,3,5-triazine) 61 was used as the crystalline sponge in this study. The crystalline sponge (typicallỹ 200 × 150 × 50 μm 3 ) was placed in a vial and immersed in a small amount of cyclohexane (50 μL). A 1,2-dichloroethane solution (5 μL) of the compound (1 μg/ μL, 5 μg) was added to the vial. The vials were equipped with a syringe needle for slow solvent evaporation. Guest soaking was carried out at 50°C for 1 d. After confirming solvent evaporation, the X-ray diffraction datasets were collected at BL02B1 (SPring-8, Hyogo, Japan), using a beam wavelength of 0.4115 Å. The sample crystal was cooled to 100 K, using a cold nitrogen stream. The collected data were integrated, corrected, and scaled using the program CrysAlisPro, ver. 1.171.38.46 (Rigaku Oxford Diffraction, 2015). All crystal structures were solved using the program SHELXT ver. 2014/5 62 and refined with the program SHELXL ver. 2018/1 63 . The crystal structure of the 25/crystalline-sponge complex was solved and three independent molecules of 25 per asymmetric unit were clearly observed. All nonhydrogen atoms were refined anisotropically. All hydrogen atoms were generated using the proper HFIX command and refined isotropically using the riding model. All nonhydrogen atoms in the framework were refined without restraints and constraints. Guest molecules were refined with restraints on the atomic displacement parameters (SIMU and RIGU). Three independent guest molecules were characterized as geometrically equal molecules by restraint (SAME). The populations of the three independent guest molecules per asymmetric unit were estimated to be 69%, 56%, and 31%, according to the structure refinement. The solvent cyclohexane molecules suggested in the electron density map were refined using some restraints (DFIX, DANG, SIMU, and ISOR), because the corresponding electron densities were obscured due to their thermal motion and disorder. The absolute configuration of 25 was thus established as 8 S, 9 R, 10 S, Reporting summary. Further information on research design is available in the Nature Research Reporting Summary linked to this article.

Data availability
The data supporting the findings of this study are available within the article and its Supplementary Information Files, source files or from the corresponding authors on request. The crystallographic data that support the findings of this study are available from the Protein Data Bank (http://www.rcsb.org). The coordinates and the structure factor amplitudes for the apo structures of SptF wild-type apo, SptF wild-type in complex with 1 and α-KG, SptF wild-type in complex with 6 and NOG, and SptF N65T variant in complex with 1 and NOG, SptF S114A variant in complex with 15 and NOG, and SptF Δ9 variant in complex with NOG were deposited under accession code 7EYR 64 , 7EYS 65 , 7EYT 66 , 7EYU 67 , 7EYW 68 , and 7FCB 69 , respectively. The crystallographic information file (CIF) for crystal structure of 25 has been deposited to The Cambridge Crystallographic Data Centre (CCDC), under reference number 2090716. Source data are provided with this paper.