Cofactorless oxygenases guide anthraquinone-fused enediyne biosynthesis

Abstract

The anthraquinone-fused enediynes (AFEs) combine an anthraquinone moiety and a ten-membered enediyne core capable of generating a cytotoxic diradical species. AFE cyclization is triggered by opening the F-ring epoxide, which is also the site of the most structural diversity. Previous studies of tiancimycin A, a heavily modified AFE, have revealed a cryptic aldehyde blocking installation of the epoxide, and no unassigned oxidases could be predicted within the tnm biosynthetic gene cluster. Here we identify two consecutively acting cofactorless oxygenases derived from methyltransferase and α/β-hydrolase protein folds, TnmJ and TnmK2, respectively, that are responsible for F-ring tailoring in tiancimycin biosynthesis by comparative genomics. Further biochemical and structural characterizations reveal that the electron-rich AFE anthraquinone moiety assists in catalyzing deformylation, epoxidation and oxidative ring cleavage without exogenous cofactors. These enzymes therefore fill important knowledge gaps for the biosynthesis of this class of molecules and the underappreciated family of cofactorless oxygenases.

Enediynes are natural products that feature a unique nine- or ten-membered 1,5-diyne-3-ene macrocyclic core and diverse peripheral moieties enabling their triggering mechanisms and potent DNA-binding activities¹. The enediynes calicheamicin and neocarzinostatin are payloads for antibody–drug conjugates and polymer conjugates, respectively, for use as clinical antitumor antibiotics^2,3. The anthraquinone-fused enediynes (AFEs, 1–6), which have recently received attention as antibody–drug conjugate payloads, are a subclass of enediynes that contain an anthraquinone moiety responsible for DNA intercalation fused to variably tailored ten-membered enediyne cores^2,4–10 (Fig. 1a).

Fig. 1 |. Selected AFEs and their BGCs.

a, The AFEs 1–5 share related A- to E-rings, but there are notable differences in the F-rings. The cycloaromatized 6 results from the sgd BGC lacking several critical genes found in all other AFE BGCs. b, BGCs for the selected AFEs. Conserved enediyne core genes are shown in purple. Predicted oxidases are shown in orange. Genes encoding TnmJ, TnmK2 and their homologs are shown in red. tnm, tiancimycin; ucm, uncialamycin; ypm, yangpumicin; dyn, dynemicin; sgd, sungeidine.

The most notable variations among AFEs occur at the F-ring, making up part of the enediyne core and containing the epoxide trigger; however, until the isolation of tiancimycin A (TNM A; 1) from Streptomyces sp. CB03234, detailed biosynthetic studies of AFEs had been limited by poor titers and genetic amenability of known producers⁶. Recently, the first AFE intermediates containing an intact F-ring before epoxide installation, TNM H (7) and TNM I (8), were identified¹¹ (Fig. 2a). These intermediates contain a C16 aldehyde, a rare moiety in natural products, and neither the aldehyde nor the intact F-ring are observed in the next known intermediate, TNM B (9)^11,12 (Fig. 2a)¹². The transformations required for conversion from 8 to 9 include an oxidative F-ring opening, deformylation, C-ring aromatization, isomerization and epoxidation; however, the tnm biosynthetic gene cluster (BGC) does not have any unassigned oxidases predicted for these reactions (Fig. 1b).

Fig. 2 |. Steps and intermediates in F-ring modification of AFEs.

a, Late-stage intermediates and shunt metabolites resulting from TnmK1, TnmJ, TnmK2 and DynA4. For the structures of S1 and S2, see Supplementary Fig. 9. b, Metabolite profiles, upon HPLC analysis with detection at 539 nm and 460 nm, of crude extracts from the ΔtnmJ and ΔtnmK2 mutants and their respective complemented strains, with Streptomyces sp. (S. sp.) CB03234 wild-type as a control. Full traces shown in Supplementary Fig. 4. c, UV–Vis profiles of selected compounds highlighting different levels of conjugation.

Most oxygenases utilize transition metals or organic cofactors such as flavins for O₂ activation. Beyond these standard oxygenases, there is a subset of enzymes termed cofactorless oxygenases that utilize highly-conjugated substrates to mimic flavin for substrate-assisted activation of O₂(refs. 13–16). Because cofactorless oxygenases depend on their substrates for their activities, they can utilize diverse enzymatic folds that are difficult to predict bioinformatically, as exemplified by the α/β-hydrolase and methyltransferase folds of the ring-cleaving Hod and the decarboxylating RdmB, respectively^15,17,18. Though cofactorless oxygenases are rare, they have been characterized to act on a diverse set of aromatic substrates, incorporating either zero, one or two oxygens from O₂ and often catalyzing ring cleavage reactions^17–25 (Supplementary Fig. 1).

In this Article, we report two consecutively acting cofactorless oxygenases involved in TNM A (1) biosynthesis that are responsible for deformylation and C-ring aromatization (TnmJ) and epoxidation, F-ring cleavage and isomerization (TnmK2). Complementary in vivo and in vitro experiments reveal the missing intermediate between 8 and 9, and crystal structures and point mutants of both oxygenases support the proposed mechanisms. A comparative study of TnmK2 with its homologs TnmK1 and DynA4 from 1 and dynemicin A (DYN; 4) biosynthesis, respectively, highlights the levels of fine tuning within the same protein fold to obtain varying oxidation levels even with substrate-assisted catalysis. Together, these results provide a basis for the biosynthetic logic leading to AFE structural divergence and valuable additions to the rare cofactorless oxygenases.

Results and discussion

Identification of genes responsible for F-ring modification

Given the differences between 8 and 9, multiple enzymes were expected to be required for deformylation, epoxidation, C-ring aromatization, oxidative F-ring cleavage and olefin isomerization. Notably, comparison of BGCs for AFEs with (1–3) and without an intact F-ring (4 and 6) highlighted the tnmK2 product as a candidate for F-ring cleavage, as the dyn and sgd BGCs encode a single TnmK1/TnmK2 homolog, DynA4, and no homologs, respectively, in contrast to the two homologs in the 1–3 BGCs (Fig. 1b)^8,26. This hypothesis was strengthened by the earlier finding that TnmK1 acts upstream of TnmK2. The fact that the ΔtnmK1 mutant (SB20033) could be partially complemented by tnmK2 supports TnmK1 and TnmK2 acting on similar substrates¹¹. However, TnmK2 was not expected to act directly on 8 given the conserved epoxidation and deformylation required for 1–4. Consequently, tnmJ, located within the same operon as tnmK1–tnmK2, was also prioritized for functional characterization. Like the predicted α/β-hydrolases TnmK1/TnmK2/DynA4, the predicted Rossmann methyltransferase TnmJ had no obvious role in the biosynthetic pathway for 1. Cofactorless oxygenases have been demonstrated to utilize both protein folds in other metabolic systems (Supplementary Fig. 1), potentially accounting for the necessary oxidations in the absence of other alternatives encoded within the tnm BGC¹⁹.

Both tnmJ and tnmK2 were separately inactivated in the wild-type Streptomyces sp. CB03234, yielding the ΔtnmJ mutant strain SB20043 and the ΔtnmK2 mutant strain SB20044 (Supplementary Figs. 2 and 3). SB20043 and SB20044 were fermented, with the wild-type Streptomyces sp. CB03234 as a control, under established conditions¹², and the corresponding crude extracts were analyzed by high-resolution liquid chromatography–mass spectrometry (HR-LC–MS) for their metabolite profiles (Fig. 2b and Supplementary Fig. 4). Both strains abolished production of 1. The accumulation of 8, the product of TnmK1, in SB20043 was confirmed by mass spectrometry, nuclear magnetic resonance (NMR), ultraviolet–visible (UV–Vis) and comparison with an authentic standard, indicating that TnmJ acts directly after TnmK1 (Fig. 2 and Supplementary Fig. 5)¹¹. Low levels of 7 were also accumulated (Supplementary Fig. 6). Complementation of SB20043 by expressing tnmJ under the constitutive kasO* promoter (pBS20050) yielded SB20045 and restored production of 1 and 9 (Fig. 2b), excluding the potential polar effect of the ΔtnmJ mutation.

By contrast, examination of the SB20044 extract revealed a single previously identified cycloaromatized metabolite 10, as confirmed by comparison with a standard¹² (Supplementary Fig. 7), and several minor compounds. Five minor compounds, 7, 11, 12, S1 and S2, were isolated and characterized by MS and NMR analyses (Fig. 2 and Supplementary Figs. 6 and 8–13, and Tables 7 and 8). Compound 11, the enediyne form of 10, had been hypothesized to exist with the epoxide intact, but it typically degrades to 10 in vivo¹². While S1 is a cycloaromatized shunt product with a rearranged carbon skeleton, S2 differs from 9 only by its (Z)-configuration of the C26–C28 olefin (Supplementary Fig. 9). Though never observed as a natural product, S2 was previously obtained during synthesis of TNM analogs²⁷. A third shunt product, the cycloaromatized spiro compound 12, resembles sealutomicins (SEA) B and D (ref. 10). Analogous to SB20045, tnmK2, under the control of kasO* (pBS20037), was expressed in SB20044, yielding SB20046 with a similar fermentation profile as SB20045 (Fig. 2b)¹¹. The wide range of metabolites isolated from the ΔtnmK2 strain SB20044 indicated a potentially unstable TnmK2 substrate; however, the accumulation of multiple metabolites with an intact F-ring confirmed TnmK2 acts upstream of 9.

TnmJ is a cofactorless aldehyde decarbonylase

Despite TnmJ acting immediately downstream of TnmK1, no obvious role for a methyltransferase could be predicted. TnmJ was produced in Escherichia coli, and the purified enzyme was assayed with 8 in vitro (Fig. 3a and Supplementary Fig. 14). This assay resulted in a new compound, 13, which was not observed in vivo. However, during the isolation and purification process from a large-scale reaction, 13, designated TNM J, was unstable, spontaneously decomposing into several cycloaromatized compounds including 10 and 12 (Supplementary Figs. 15–17). A [M + H]⁺ ion at m/z 446.1385 was detected for 13 for a molecular formula of C₂₉H₂₀NO₄ (calculated [M + H]⁺ ion at m/z 446.1387), consistent with the loss of an aldehyde and two protons from 8 (Supplementary Fig. 15 and Supplementary Table 5). Furthermore, 13 exhibits a similar UV–Vis spectrum to 9, indicating aromatization of the C-ring (Fig. 2c). Rapid processing of trace amounts of pure 13 only allowed collection of ¹H and ¹H-¹H correlation spectroscopy (COSY) NMR spectroscopy data, revealing the aromatic C-ring (δ_H 13.48, s; δ_H 10.20, d (3.4); δ_H 7.47, d (1.7)) and a new methine signal (δ_H 4.04, m), which was assigned to H-16 based on the COSY correlation of H-16 with H-17, confirming the loss of the aldehyde in 13 relative to 8 (Supplementary Fig. 15). Critically, the major decomposition products 10 and 12, which are comparatively stable, were isolated and structurally characterized (Supplementary Figs. 7, 11, 16 and 17), further supporting the assigned structure of 13 with the (R)-configuration at C16. To further confirm the TnmJ-catalyzed conversion of 8 to 13, kinetic measurements of TnmJ with 8 were attempted; however, the instability of the product prevented accurate measurements. Instead, a time course revealed the conversion of nearly all of 100 μM 8 by 0.2 μM TnmJ within 2 h, suggesting a k_cat well within the normal range for secondary metabolism enzymes²⁸ (Fig. 3a). Further support for 8 as the native substrate of TnmJ was provided through the determination of a dissociation constant (K_d) of 3.4 μM by isothermal titration calorimetry (Supplementary Fig. 18).

Fig. 3 |. TnmJ catalyzes the SAM-independent cofactorless cryptic oxidation of TNM I (8).

a, HPLC analysis, with detection at 539 nm and 460 nm, of the in vitro TnmJ-catalyzed conversion of TNM I (8) to TNM J (13). std, standard. b, The homodimeric apo-structure of TnmJ has a characteristic α,β-Rossmann methyltransferase fold, with one monomer depicted in red. c, The active site of TnmJ with 8 docked based on results from a 100-ns molecular dynamics simulation depicts key residues implicated in binding and catalysis.

Aldehyde decarbonylases are poorly studied and considered rare in nature. They have been identified in long-chain alkane biosynthesis, with different taxa using varied mechanisms and protein folds²⁹. Aldehyde decarbonylases utilize iron, O₂ and external reducing equivalents to eliminate the aldehyde as either formate (cyanobacteria) or CO₂ (insects), while plants are proposed to release CO (refs. 29–33). The human P450 aromatase also demonstrates formate-releasing aldehyde decarbonylase activity during its third catalytic step³⁴. To evaluate whether TnmJ is a cryptic oxygenase, large-scale reactions of TnmJ with 8 were analyzed by gas chromatography–mass spectrometry (GC–MS), and a mass consistent with CO₂ was observed only with functional TnmJ (Fig. 4a and Supplementary Fig. 19). By contrast, an in situ click reaction with 2,4-dinitrophenyl-hydrazine designed to capture formate resulted in only trace product (<1% of 13 produced)^35,36 (Supplementary Fig. 20). TnmJ was further assayed in the presence of H₂¹⁸O or ¹⁸O₂, and CO₂ labeled with a single ¹⁸O was observed only in the presence ¹⁸O₂ (Supplementary Fig. 21).

Fig. 4 |. TnmJ deformylates TNM I (8) to yield TNM J (13), H₂O and CO₂.

a, GC–MS results from in vitro reactions of TnmJ with TNM I (8) releases CO₂ as a byproduct. std, standard. b, The proposed mechanism for conversion of 8 to 13 by TnmJ is initiated by deprotonation of the C11-OH and a subsequent electron transfer to O₂ (i), followed by formation of a peroxide intermediate at C14 (ii) and a six-membered ring with the C16 aldehyde (iii). Collapse of the ring yields a C14-OH and C16-COO⁻ (iv), which are lost as water and CO₂, respectively, resulting in 13 (v). For a more detailed mechanism, see Supplementary Fig. 28. O₂-derived oxygens incorporated by TnmJ are highlighted in red. c, The synthetic carboxylic acid 14 was assayed with 20 μM TnmJ in vitro and compared with 8 and 1 μM TnmJ. Data are mean and s.d. (n = 3 independent experiments).

Beyond releasing the aldehyde as CO₂, TnmJ exhibits several differences from characterized aldehyde decarbonylases. Most prominently, conversion of 8 to 13 by TnmJ in the presence of ethylenediaminetetraacetic acid (EDTA) revealed no metal dependence for this reaction (Extended Data Fig. 1a). Likewise, no exogenous reducing agents were added to the reaction, and no rate enhancement was observed upon addition of common reducing agents (Extended Data Fig. 1b). Lastly, the Rossman-fold methyltransferase fold of TnmJ stands in contrast to the integral membrane aldehyde decarbonylases of plants, the membrane-associated P450s of insects and humans, and the cytosolic nonheme diiron oxygenases of cyanobacteria²⁹. Together, these findings establish TnmJ as a distinct cofactorless AD.

TnmJ structure and mutagenesis reveal cofactorless mechanism

As TnmJ lacks homology to any known AD, a structural study of TnmJ was pursued. The apo-TnmJ protein was crystallized, and its structure was solved to a 1.5 Å resolution (Protein Data Bank (PDB) accession 8G5S; Supplementary Table 4). TnmJ crystallized as a 69-kDa homodimer, consistent with its solution state as determined by size-exclusion chromatography (Supplementary Fig. 14). The overall structure of TnmJ exhibits a classical C-terminal α,β-Rossmann methyltransferase fold containing a parallel five-stranded β-sheet, a central α-helical domain and an N-terminal helical dimerization domain (Fig. 3b and Supplementary Fig. 22). TnmJ only shows moderate similarity to the S-adenosyl-l-methionine (SAM)-dependent cofactorless oxygenase RdmB (PDB 1XDS; Cα root mean square deviation: 9.91 Å) and lacks the critical arginine (or any related residue) responsible for decarboxylation in RdmB¹⁸ (Supplementary Figs. 23 and 24 and Supplementary Table 9).

TnmJ could not be co-crystallized with SAM; therefore, to establish whether SAM is necessary, TnmJ was treated with charcoal to extract any potential co-purified SAM and then evaluated for conversion of 8 to 13 with and without SAM or S-adenosyl-l-homocysteine (SAH). No change in conversion was observed, indicating a SAM-independent reaction (Extended Data Fig. 1c). Furthermore, isothermal titration calorimetry data supported that TnmJ was unable to bind SAM (Supplementary Fig. 18). To our knowledge, this is only the second example of a member of the α,β-Rossmann methyltransferase superfamily, joining a family of pericyclases, to be characterized as catalyzing a natively SAM-independent reaction^18,37.

To explore how TnmJ enables substrate-assisted catalysis with only O₂, extensive attempts at co-crystallization and crystal soaking of TnmJ with either 8 or the unstable 13 were made, albeit unsuccessfully. Therefore, docking and 100 ns molecular dynamics simulations with 8 were performed (Fig. 3c and Supplementary Fig. 25). The A–B–C ring system of 8 is predicted to interact primarily via hydrogen bonding between the C11-OH and C13 ketone and Asp264 and Asn363, respectively, and hydrophobic interactions with Trp165, Met158 and Met161. The amide backbones of Pro361 and Leu362 can form an oxyanion hole near the C16 aldehyde. From several evaluated docking poses, several residues were prioritized for mutagenesis (Fig. 3c and Extended Data Fig. 2). Asp264 loses nearly all activity when changed to Ala or Glu. Asp264 is positioned to deprotonate the C11-OH, which could serve as the initiating step for O₂ activation by the aromatic system (Fig. 4b). Asp268/His267 and Asn363 would therefore be positioned to assist in deprotonation and to stabilize the resultant delocalized negative charge, respectively. The N363A and N363D mutants lose 90% of wild-type activity, while N363Q performs at near 50% (Extended Data Fig. 2).

The predicted deprotonation of the B-ring would necessitate direct oxidation of the A–B–C ring system rather than oxidation of the C16 aldehyde to a carboxylic acid. To directly probe whether the C16 carboxylic acid is an intermediate, 8 was converted to 14 via a Pinnick oxidation (Supplementary Fig. 26 and Supplementary Table 6). TnmJ-catalyzed conversion of 14 to 13 was minimal relative to the conversion of 8 (Fig. 4c). Addition of the reducing agent dithiothreitol to the reaction with 14 yielded a 50% increase in conversion, consistent with an initial oxidation away from the aldehyde (Supplementary Fig. 27). Accordingly, a mechanism for TnmJ is proposed in which (1) Asp264 deprotonates the C11-OH of 8, leading to O₂ activation via electron transfer, (2) C14 is oxidized to give a peroxide intermediate, (3) a transient six-membered ring forms, linking C14 and the C16-CHO via a peroxy bridge, (4) collapse of the ring yields C14-OH and C16-COO⁻, and (5) the carboxylate leaves as CO₂ and the C-ring aromatizes via elimination of the C14-OH (Fig. 4b and Supplementary Fig. 28).

TnmK2 is a versatile cofactorless oxygenase that acts twice

TNM J (13) may be the last common intermediate shared across the biosynthetic pathways for the TNM-like enediynes (1–3), SEA (5) and DYN (4) due to differences in the F-ring modifications of their final products (Fig. 1a). While the accumulation of metabolites 11, 12, S1 and S2 isolated from the ΔtnmK2 mutant SB20044 suggested that TnmK2 acts downstream of epoxidation and F-ring cleavage (Fig. 2b and Supplementary Fig. 9), the spontaneous oxidative degradation of 13 to 10 and 12 observed during the TnmJ study revealed 13 as a potential substrate of TnmK2 (Supplementary Fig. 17). To confirm the true substrate, TnmK2, insoluble in E. coli, was overproduced in Streptomyces lividans, purified and assayed (Supplementary Fig. 29). No reaction was observed for TnmK2 with 11 or S1 (Supplementary Fig. 30); with 13, generated in situ by TnmJ with 8 due to its instability, TnmK2 produced 9 as the major product, the identity of which was confirmed by comparison with an authentic standard (Fig. 5a). Remarkably, this transformation requires TnmK2 to catalyze epoxidation at C16–C25, oxidative ring cleavage between C17 and C29, and isomerization of the C26–C28 olefin. To definitively establish that TnmK2 could account for each reaction, freshly prepared 13 from an in vitro TnmJ reaction was rapidly purified and assayed with TnmK2, resulting primarily in 9, with substantial degradation also observed (Fig. 5a). With boiled TnmK2, only the degradation products were observed. Owing to this degradation, as with TnmJ, accurate kinetic parameters could not be calculated for TnmK2, though TnmK2 appears to catalyze its reaction at a rate within the same order of magnitude as the TnmJ reaction.

Fig. 5 |. TnmK2 and homologs TnmK1 and DynA4 catalyze different reactions.

a, HPLC analysis, with detection at 460 nm and 539 nm, of in vitro reactions with 8, TnmJ, and either TnmK2, TnmK1 or DynA4. std, standard. b, The proposed intermediates in the reactions of 13 with TnmK2 (i–v, blue) and DynA4 (i and vi, magenta). The reactions are initiated by deprotonation at C16, electron transfer to O₂, and formation of a peroxide intermediate (i), followed by oxidative ring cleavage by TnmK2 (ii) or epoxidation and release of methanol by DynA4 (vi). Deprotonation of the amine and activation of a second O₂ molecule lead to formation of a second peroxide intermediate (iii). Subsequent epoxidation and isomerization of the C26–C28 olefin result in an oxidized C-ring (iv) before a spontaneous non-enzymatic reduction to 9 (v). For more detailed mechanisms, see Supplementary Figs. 34 and 42.

Given the substrate-assisted oxygenation catalyzed by TnmJ and the heavily oxidized degradation products of 13, TnmK2 was evaluated as a putative cofactorless oxygenase. When supplemented with EDTA and reducing agents, the coupled TnmJ–TnmK2 reaction with 8 resulted in no change and reduced conversion, respectively (Supplementary Fig. 31). The reaction run with ¹⁸O₂ revealed the incorporation of three ¹⁸O atoms into 9 from two O₂ molecules, indicating that TnmK2, like TnmJ, acts as a cofactorless oxygenase (Supplementary Fig. 32a–c). As a control, the reaction run in H₂¹⁸O did not show notable isotopic labeling (Supplementary Fig. 32d,e). In 13, only deprotonation at C16 is necessary to extend the conjugated system across the A-, B-, C-, D- and F-rings (Fig. 5b). Once deprotonation and O₂ activation occur, TnmK2 initiates the oxidative ring cleavage via a peroxide intermediate at C29. TnmK2 then activates a second molecule of O₂, leading to an intermediate with a C25-peroxide anion and subsequent isomerization of the C26-C28 olefin and epoxidation at C16–C25. A final spontaneous reduction of the C-ring is required to yield 9, though the reductant is unclear given strong reducing agents may prematurely reduce peroxide intermediates and the reaction can run to completion, indicating excess substrate is not the source of electrons (Fig. 5b and Supplementary Figs. 31, 33 and 34).

TnmK2 chemistry diverges from structural homologs

TnmK2 was predicted as a member of the α/β-hydrolase fold superfamily, a fold shared with other cofactorless oxygenases and the D-ring cyclases TnmK1 and DynA4 (Supplementary Table 10 and Supplementary Fig. 35)¹¹. Therefore, all three enzymes were evaluated in vitro for cyclization and oxidation activities (Fig. 5a). Notably, while TnmK1 does not act as an oxygenase when coupled with TnmJ and 8, TnmK2 can slowly convert the TnmK1 substrate 7 to 9 in the absence of TnmJ, indicative of its catalytic flexibility (Supplementary Fig. 36). DynA4, the only α/β-hydrolase homolog encoded in the 4 BGC, with 13, generated in situ by TnmJ with 8, produces the epoxidized product 11, consistent with the intact F-ring in 4 (Figs. 2a and 5a). Using ¹⁸O labeling, DynA4 was shown to incorporate the epoxide from O₂, while the C29 ketone is derived from H₂O (Supplementary Fig. 37). As DynA4 was previously demonstrated to catalyze the D-ring cyclization with similar efficiency to TnmK1, it serves as one of the rare examples of an enzyme that can act across multiple, nonconsecutive biosynthetic steps^11,38.

To rationalize these mechanistic differences, TnmK2 was crystallized (1.85 Å resolution; PDB 8G5T; Fig. 6a and Supplementary Table 4). There is only a 1.61 Å root mean square deviation between the TnmK2 and TnmK1 (PDB 8E18; Supplementary Fig. 38)¹¹. A central active site cavity is surrounded by an α-domain and the α/β-hydrolase fold and contains the canonical catalytic residues: Ser193, His448 and Asp418 (Fig. 6b). A structure with TnmK2 and 9 (1.81 Å) was obtained (PDB 8G5U; Fig. 6, Supplementary Fig. 39 and Supplementary Table 4), showing a conserved binding mode for the enediyne cores of 7 and 9 by TnmK1 and TnmK2, respectively, but the orientations of the anthraquinone moieties are offset by 143° (Supplementary Fig. 40a)¹¹. Three key residues can be implicated in anthraquinone moiety binding: Ser420 (Phe409_TnmK1 and Phe443_DynA4), Gly298 (Val287_TnmK1 and Val322_DynA4), and Asn301 (Ser290_TnmK1 and Thr325_DynA4; Fig. 6b and Supplementary Fig. 40b). An additional stabilizing interaction between the anthraquinone A-ring of 9 and Met333 was observed to rigidify the region of TnmK2 from Pro324 to Glu339 (Supplementary Fig. 41).

Fig. 6 |. TnmK2 uses a repurposed α/β-hydrolase fold for multiple cofactorless oxidations.

a, The apo-structure (gray) is overlaid on the holo-structure (blue) of TnmK2 with the co-crystallized product TNM B (9; coral) in the active site. TnmK2 shares its α/β-hydrolase fold with TnmK1. b, The active site of TnmK2 with 9 shows the canonical triad far from the substrate, while Lys338 is near C13 and C11 and Asp82 projects close to the amine and C16 (*) of 9. c, The loop containing Asp82_TnmK2 is overlaid with the corresponding loops containing Asp72_TnmK1 (PDB: 8E19) and Asp105_DynA4 (AlphaFold model), highlighting how only Asp82_TnmK2 can access protons from the amine and C16 (*) of 9 (refs. 11,39).

For substrate-assisted oxidation to occur twice, a catalytic base must be positioned for two deprotonations. In TnmK2 and its nearest homologs, a conserved loop with a unique PGDDLG motif positions the second aspartate, Asp82, near C16 and the amine of 9 (and therefore of the substrate 13; Figs. 5b and 6c). The basicity of Asp82 may be enhanced by its interactions with the α/β-hydrolase catalytic triad, though mutagenesis indicates no direct catalytic roles (Extended Data Fig. 3). In contrast, in TnmK1 and DynA4, the loop contains a smaller and more rigid PDDPG motif (Supplementary Fig. 35). In TnmK1, the corresponding Asp72_TnmK1 coordinates with 7 via an extensive water network but cannot deprotonate 7 directly¹¹; however, in an AlphaFold-generated model of DynA4, Asp105_DynA4 is positioned to deprotonate only C16 but not the amine, consistent with initiating a single oxidation event (Fig. 6c and Supplementary Fig. 42). In support, the TnmK2 D82A mutant loses all activity (Extended Data Fig. 3). By contrast, the TnmK1 D72A mutation did not lose its cyclase activity¹¹. Further, the amide backbone of one or more residues in the PGDDLG motif, especially Leu83_TnmK2, is predicted to position O₂ for catalysis, but in DynA4, the smaller loop positions O₂ only for epoxidation instead of F-ring cleavage.

The AFEs are valuable natural products for drug development; however, the biosynthesis of this family, especially for early steps, remains poorly characterized. This study has (1) closed the gap in AFE scaffold biosynthesis and (2) highlighted the diverse chemistry catalyzed by two cofactorless oxygenases that consecutively utilize electron-rich substrates to activate O₂. TnmJ reveals the fate of the 15th carbon of the enediyne core observed in early intermediates. TnmK2 cleaves the F-ring and installs the epoxide responsible for triggering cyclization of the enediyne core. The structural similarity of TnmK2 to TnmK1 and DynA4, which utilize similar protein folds for alternative chemistries, highlights the impact of substrate-assisted catalysis. Given the number of undetermined oxidative events expected for AFE biosynthesis and the relatively low number of annotated AFE-related oxidases (Fig. 1b), it is expected that more cofactorless oxygenases are involved in AFE biosynthesis—and in the biosynthesis of other electron-rich natural products. This study expands the scope of known cofactorless oxygenases by structurally and biochemically describing enzymes that catalyze a cryptic oxidation (TnmJ) and catalyze multiple oxygenations (TnmK2), further opening the door to the discovery, exploration and exploitation of an understudied class of enzymes.

Online content

Any methods, additional references, Nature Portfolio reporting summaries, source data, extended data, supplementary information, acknowledgements, peer review information; details of author contributions and competing interests; and statements of data and code availability are available at https://doi.org/10.1038/s41589-023-01476-2.

Methods

Bacterial strains, plasmids and chemicals

Plasmids and bacterial strains used in this study are listed in Supplementary Tables 1 and 2, respectively. E. coli strains were grown on Luria–Bertani medium at 28 °C or 37 °C; appropriate antibiotics were added at a final concentration of: 100 μg ml⁻¹ carbenicillin, 50 μg ml⁻¹ apramycin (Apr), 50 μg ml⁻¹ kanamycin (Kan), 25 μg ml⁻¹ chloramphenicol, 100 μg ml⁻¹ thiostrepton and 20 μg ml⁻¹ nalidixic acid when necessary. Streptomyces sp. CB03234 and mutant strains were maintained on solid International Streptomyces Project medium 4 (ISP4) plates at 28 °C. Tryptic soy broth was adopted for the spore suspension of Streptomyces sp. CB03234. Production medium (soluble starch 40 g l⁻¹, pharmamedia 20 g l⁻¹, CuSO₄•5H₂O 0.2 g l⁻¹, NaI 20 mg l⁻¹, CaCO₃ 4 g l⁻¹ and 1.5% Diaion HP2MGL resin, adjusted to pH 7.2) was used for fermentation of Streptomyces sp. CB03234 and mutant strains.

Polymerase chain reaction (PCR) primers were synthesized by Sigma-Aldrich or Integrated DNA Technologies. High-fidelity DNA polymerase (Q5), restriction endonucleases and T4 DNA ligase were purchased from NEB. Pfu DNA polymerase or Q5 mutagenesis kit were used for mutagenesis and were purchased from Agilent or NEB, respectively. Plasmid, gel extraction and cycle-pure kits were acquired from Omega Bio-tek. For Southern hybridization, the DIG-High Prime DNA Labeling and Detection Starter Kits I and II were used (Roche Diagnostics), and the Southern hybridization was performed by following the protocols suggested by the manufacturer. SAM and SAH were purchased from Sigma-Aldrich. 2,4-Dinitrophenyl-hydrazine was purchased from Spectrum Chemical Mfg Corp. ¹⁸O₂ and H₂¹⁸O were purchased from Sigma-Aldrich. Other chemical solvents (analytical grade) were all purchased from standard commercial sources. Crystallization kits were purchased from Hampton Research and Anatrace.

General experimental procedures

¹H, ¹³C and 2D (COSY, heteronuclear single-quantum correlation spectroscopy (HSQC), heteronuclear multiple-bond correlation spectroscopy (HMBC) and nuclear Overhauser effect spectroscopy (NOESY)) NMR spectra were recorded at 25 °C with an Avance 600 MHz spectrometer instrument (Bruker). Ultrahigh-performance liquid chromatography–electrospray ionization–high-resolution mass spectrometry (UHPLC–ESI–HRMS) measurements were carried out on a Vanquish UHPLC system (Thermo Scientific) combined with an Orbitrap Exploris 120 mass spectrometer (Thermo Scientific) equipped with an ESI source and an Accucore C18 column (100 mm × 2.1 mm, 2.6 μm). LC–MS was performed on an Agilent 1260 Infinity LC system coupled to an Agilent 6230 TOF (HR-ESI) equipped with a Poroshell 120 EC-C18 column (Agilent, 50 mm × 4.6 mm, 2.7 μm). UV–Vis spectra were measured with a NanoDrop 2000C spectrophotometer (Thermo Scientific). Optical rotations were obtained using an AUTOPOL IV automatic polarimeter (Rudolph Research Analytical). Preparative HPLC was carried out on an Agilent 1260 Infinity LC equipped with an Agilent Eclipse XDB-C18 column (250 mm × 21.2 mm, 7 μm). Semi-preparative HPLC was performed on a Varian liquid chromatography system equipped with a phenol-hexyl column (250 mm × 10 mm, 5 μm). Sephadex column chromatography was conducted using Sephadex LH-20 (GE Healthcare). Electrophoresis was carried out using a Bio-Rad PowerPac 300. All error bars were determined using the standard deviation of at least three replicates, with points indicating values of each replicate.

Generation of the complementation mutant strains

To construct the complemented strains, the tnmJ gene was amplified using primers tnmJ-com-KasO-F and tnmJ-com-KasO-R, digested with SpeI and EcoRI, and ligated into digested pBS21003 to afford pBS20050, in which the expression of tnmJ is under the control of kasO* promoter. The resulting construct was confirmed by PCR and sequencing. The equivalent construct with the tnmK2 gene was previously constructed as pBS20037 (ref. 11). pBS20050 and pBS20037 were introduced into SB20043 and SB20044, respectively, by intergenic conjugation to afford the strains SB20045 and SB20046, respectively.

Analysis of production profiles of the Streptomyces sp. CB03234 wild-type and mutant strains

Unless otherwise stated, the solvent system of analytical HPLC consisted of solvent A (0.1% formic acid in Millipore H₂O) and solvent B (0.1% formic acid in acetonitrile (ACN)). To analyze the metabolite profiles of Streptomyces sp. CB03234 and its mutants, analytical HPLC was performed on LC–MS with detection at 254 nm, 460 nm and 539 nm. The samples were eluted with a linear gradient of ACN in Millipore H₂O with 0.1% formic acid (0–18 min, 5–100% B; 18–25 min, 100% B; 25–25.01 min, 100–5% B) at a flow rate of 0.4 ml min⁻¹ (Method I). A shorter elution method was also used with a linear solvent gradient (0–7 min, 20–100% B; 7–10 min, 100% B; 10–11 min, 100–20% B; 11–12 min, 20% B) at a flow rate of 0.4 ml min⁻¹ (Method II).

For compound purification, the solvent system used for preparative HPLC or semi-preparative HPLC consisted of solvent A (100% Millipore H₂O) and solvent B (100% ACN).

Fermentation and isolation of compound 8 from the ΔtnmJ mutant (SB20043)

To characterize intermediates and related compounds from SB20043, a two-step large-scale (30 liter) fermentation strategy was used. A general procedure is described herein¹¹. First, a suitable portion of spores from an ISP4 plate was used to inoculate 50 ml tryptic soy broth medium in a 250-ml flask as a seed culture and incubated at 28 °C and 148g for 36–48 h. Then, the seed culture was transferred to 400 ml of production medium in a 2,000-ml flask and incubated at 28 °C and 148g for an additional 7 days. After incubation, the resin and the cell pellets were collected by centrifugation, washed by Millipore H₂O and extracted three times with ethyl acetate. The ethyl acetate extract was concentrated under reduced pressure to yield an oil crude extract. The residue was then subjected to medium-pressure LC chromatography, eluted with gradient ACN/H₂O system (0–90 min, 0–100% B; 90–120 min, 100% B) at a flow rate of 25 ml min⁻¹. The fraction containing the compound of interest was collected, evaporated, and redissolved in acetone, followed by semi-preparative HPLC purification (flow rate 3 ml min⁻¹ with an isocratic gradient of 85% solvent B for 20 min, followed by holding at 100% B for 5 min, and then elution with 85% solvent B for 3 min) to yield compound 8 (19 mg, t_R 15.4 min).

Fermentation and isolation of compounds 10–12, S1 and S2 from the ΔtnmK2 mutant (SB20044)

Using the same fermentation protocol as described for SB20043, a 10-liter fermentation of SB20044 was carried out. The resulting crude extract was first subjected to Sephadex LH-20 Sephadex gel chromatography and eluted with acetone to yield 20 fractions (Fr. A1–A20). Fractions containing the targeted compounds with red or purple coloration were collected, evaporated and redissolved in acetone, followed by preparative HPLC purification to give Fr. B1–B10. Further purification was conducted by semi-preparative HPLC to yield pure compounds 10 (6 mg), 11 (0.4 mg), 12 (0.5 mg), S1 (0.5 mg) and S2 (0.6 mg).

Gene inactivations of tnmJ and tnmK2 in Streptomyces sp. CB03234

The in vivo gene inactivations were performed according to the method described previously^12,40. Primers designed for inactivation are listed in Supplementary Table 3. Specifically, cosmid pBS20002, which contains the partial tnm BGC, was transformed into E. coli BW25113/pIJ790 to produce E. coli BW25113/pIJ790/pBS20002. Then, the PCR product of the gene disruption cassette encoding Kan resistance from pJTU4659 was then transformed into E. coli BW25113/pIJ790/pBS20002 competent cells for λ-RED-mediated recombination to yield the recombinant strain SB20066, and the PCR product of the gene disruption cassette encoding spectinomycin resistance from pIJ778 was then transformed into E. coli BW25113/pIJ790/pBS20002 competent cells for λ-RED-mediated recombination to yield the recombinant strain SB20067. The recombinant cosmids pBS20069 (ΔtnmJ::neo) and pBS20070 (ΔtnmK2::aadA) were transformed into E. coli ET12567/pUZ8002 and conjugated into Streptomyces sp. CB03234 wild-type strain. Double crossover mutants for ΔtnmJ (SB20043) and ΔtnmK2 (SB20044) were selected on the basis of the Apr^S Kan^R and Kan^S Spt^R phenotypes, respectively, and then further confirmed by PCR and Southern blot analysis (Supplementary Figs. 2 and 3) using primers listed in Supplementary Table 3. Finally, the mutant strains of Streptomyces sp. CB03234 were generated (Supplementary Table 2).

In vitro biochemical assay of TnmJ

To assay TnmJ activity with its substrate 8, a preliminary assay of 50 μl containing 20 mM MOPS, pH 7.0, 100 μM 8 and 200 nM TnmJ at 30 °C for 30 min was conducted. The reaction was quenched with the addition of two volumes of cold acetone and then centrifuged. The supernatant was analyzed with an Agilent HPLC using a Phenomenex Prodigy ODS column S5 (150 × 4.60 mm, 5 μm) with detection at 460 nm and 539 nm. Samples were analyzed with Method I.

In vitro biochemical assay of TnmK2

To assay TnmK2 activity, a 50-μl reaction containing 50 mM MOPS, pH 7.5, 50 μM substrate (13, 11 or S2), and 1 μM TnmK2 at 28 °C for 30 min was conducted. The reaction was quenched with the addition of two volumes of cold acetone and flash frozen by liquid nitrogen and stored at −80 °C until HPLC analysis. Protein was precipitated and removed by centrifugation, and the supernatant was analyzed by LC–MS with Method I. For the coupled TnmJ–TnmK2 assay, a 50-μl reaction containing 50 mM MOPS, pH 7.0, 100 μM 8, 20 μM TnmJ and 1–2 μM TnmK2 at 28 °C for 60 min was conducted. To evaluate TnmK2 with 7, the reaction was conducted at 50 mM MOPS, pH 7.5, 100 μM 7 and 5 μM TnmK2 at 28 °C for 48 h. For full conversion of 8 to 13 by TnmK2, a 50-μl reaction was run in 50 mM MOPS, pH 7.0 with 100 μM 8, 20 μM TnmJ and 10 μM TnmK2 at 28 °C for 90 min before following the protocol as described above.

Evaluation of SAM and SAH on the AD activity of TnmJ

A total of 200 μM SAM or 200 μM SAH was added to a 50-μl solution containing 1 μM TnmJ in 50 mM MES, pH 6.0, and the mixture was incubated at room temperature for 15 min. Then 100 μM substrate 8 was added to the reaction mixture to initiate the enzymatic reaction. After 1 h at 28 °C, the reactions were quenched by the addition of 100 μl of cold acetone. All samples were flash frozen by liquid nitrogen and stored at −80 °C until analysis. Protein was precipitated and removed by centrifugation, and the supernatants were analyzed by LC–MS with Method II. The relative activity was calculated as percentage of controls without SAM or SAH. The error bars represent the standard deviation of three independent replicates. Data fitting was performed using GraphPad Prism 9. The results are shown in Extended Data Fig. 1.

Oxygen isotope labeling by TnmJ, DynA4 and TnmK2

For TnmK2, substrate 13 was produced by incubating 8 with TnmJ in situ. The procedure for the reaction is similar to the small-scale TnmJ–TnmK2 coupled assays with slight modifications. The reaction was performed in a 2-ml sealed headspace vial (Agilent). Before starting the reaction, the buffer (50 mM MES, pH 6.0) was degassed by continuous bubbling of argon gas for 30 min. At the same time, the reaction vial was degassed by repeated evacuation and argon flushes (three times). Then, reaction buffer and 8 dissolved in dimethyl sulfoxide were added to the vial using a gas-tight syringe. The reaction vial was then evacuated and ¹⁸O₂ was bubbled into the reaction mixture. TnmJ and TnmK2 were then added to the reaction vial by syringe to initiate the reaction. After 1 h of incubation at 30 °C, the reaction was quenched with cold ACN. LC–MS analysis was performed using Method II. For TnmJ assays with 8, the reaction was performed as above but without TnmK2. For DynA4 assays with 8, the reaction was performed as with TnmK2. Each enzyme was also assayed using H₂¹⁸O in place of water.

Extended Data

Extended Data Fig. 1 |. Evaluation of TnmJ for metal, SAM, and reducing agent dependence.

a, No difference was observed upon addition of 10 mM EDTA to a reaction with TnmJ and 8. b, No difference was observed upon addition of 2 mM reducing agents DTT or TCEP. c, No difference was observed upon addition of 200 μM SAM or SAH to a reaction with TnmJ and 8. Likewise, charcoal-treated TnmJ showed no difference. Bars depict the mean; error bars represent standard deviation (n = 3 independent experiments).

Extended Data Fig. 2 |. Mutagenesis of TnmJ.

Site-directed mutants of TnmJ were evaluated in vitro with 100 μM 8. Results were used to evaluate possible binding modes from molecular dynamics simulations. Bars depict the mean; error bars represent standard deviation (n = 3 independent experiments).

Extended Data Fig. 3 |. Mutagenesis of TnmK2.

TnmK2 mutants were evaluated in vitro with 13 produced in situ by TnmJ from 100 μM 8. N.D. = not detected. Bars depict the mean; error bars represent standard deviation (n = 3 independent experiments).

Data availability

The data that support the findings of this study are available within the main text and its Supplementary Information file. Data are available from the corresponding author upon request. The coordinates of the TnmJ, TnmK2 and TnmK2–9 structures have been deposited to the PDB with the accession codes 8G5S, 8G5T and 8G5U, respectively. Source data are provided with this paper.