An unusual aromatase/cyclase programs the formation of the phenyldimethylanthrone framework in anthrabenzoxocinones and fasamycin

Significance

Aromatic polyketides’ structural diversity is primarily determined by basic aromatic frameworks. Currently, only about eight basic aromatic frameworks have been identified, formed by eight different types of aromatase/cyclases (ARO/CYCs). Their formation involves the sequential catalysis of 2 to 4 ARO/CYCs, with only one TcmN-type ARO/CYC catalyzing the C9 to C14 first ring cyclization. In this study, a unique type of ARO/CYCs (Abx₍₊₎D/Abx₍₋₎D/FasL) was identified, capable of cyclizing/aromatizing all four rings, including the C9 to C14 first ring and a rare angular benzene ring within the aromatic framework PDA (phenyldimethylanthrone) shared by (+)-ABX (anthrabenzoxocinone), (−)-ABX, and FAS (fasamycin). The molecular basis of Abx₍₊₎D can serve as a foundation for creating additional multicyclic ARO/CYCs to expand the variety of aromatic polyketide frameworks.

Abstract

Aromatic polyketides are renowned for their wide-ranging pharmaceutical activities. Their structural diversity is mainly produced via modification of limited types of basic frameworks. In this study, we characterized the biosynthesis of a unique basic aromatic framework, phenyldimethylanthrone (PDA) found in (+)/(−)-anthrabenzoxocinones (ABXs) and fasamycin (FAS). Its biosynthesis employs a methyltransferase (Abx₍₊₎M/Abx₍₋₎M/FasT) and an unusual TcmI-like aromatase/cyclase (ARO/CYC, Abx₍₊₎D/Abx₍₋₎D/FasL) as well as a nonessential helper ARO/CYC (Abx₍₊₎C/Abx₍₋₎C/FasD) to catalyze the aromatization/cyclization of polyketide chain, leading to the formation of all four aromatic rings of the PDA framework, including the C9 to C14 ring and a rare angular benzene ring. Biochemical and structural analysis of Abx₍₊₎D reveals a unique loop region, giving rise to its distinct acyl carrier protein-dependent specificity compared to other conventional TcmI-type ARO/CYCs, all of which impose on free molecules. Mutagenic analysis discloses critical residues of Abx₍₊₎D for its catalytic activity and indicates that the size and shape of its interior pocket determine the orientation of aromatization/cyclization. This study unveils the tetracyclic and non-TcmN type C9 to C14 ARO/CYC, significantly expanding our cognition of ARO/CYCs and the biosynthesis of aromatic polyketide framework.

Aromatic polyketides constitute a large family of structurally diverse natural products that exhibit a wide range of biological activities (1–3). Their biosynthesis is primarily programmed by type II polyketide synthases (PKSs), typically comprising a ketosynthase/chain length factor heterodimer (KS/CLF), acyl-transferase (AT), and acyl carrier protein (ACP), collectively known as minimal PKS for polyketide chain generation (4, 5). Nascent polyketide chains are then cyclized by aromatase/cyclases (ARO/CYCs), forming basic aromatic frameworks. In a few cases, ketone groups in polyketide chains can be selectively reduced by ketoreductases (KR) either before or after aromatization/cyclization (4–6). Afterward, additional enzymes catalyze post-modification reactions, such as methylation, hydroxylation, and oxidative rearrangement, further enriching the structural diversity by decorating the basic frameworks (4, 5).

As structural templates, basic frameworks play a determinant role in the generation of structural diversity. However, despite the abundance of aromatic polyketides, only about eight basic frameworks, formed by eight different types of ARO/CYCs, have been identified so far (Fig. 1) (5, 7–9). Moreover, these basic frameworks typically consist of successive conjugated aromatic ring scaffolds, necessitating 2 to 4 different types of ARO/CYCs for their formation (5, 7). Interestingly, a remarkable exception to this biosynthetic scheme is observed in anthrabenzoxocinones (ABXs) (10–16) and fasamycin (FAS) (17–30) (Fig. 2), two classes of aromatic polyketides with potent antimicrobial and other biological activities. ABXs, including (+)/(−)-ABXs and FAS, possess an identical phenyldimethylanthrone (PDA) core, which contains a 1,3-dihydroxy-10,10-dimethylanthrone unit (A–B–C ring) connected to an angular benzene ring (D ring) by a pair of different chiral oxygen-bridges and a phenyl moiety (E ring), respectively (Fig. 2).

Fig. 1.

ARO/CYCs and basic aromatic frameworks in the biosynthesis of type II aromatic polyketides.

Fig. 2.

The BGC and proposed biosynthetic pathway of (+)/(−)-ABXs and FAS. (A) The three BGCs, with identical genes indicated by dashed lines. (B) Proposed biosynthetic pathway of the PDA framework (1) in (+)/(−)-ABXs and FAS, along with structures of three typical aromatic polyketides with gem-dimethyl groups.

The unique structure of the (+)/(−)-ABXs and FAS, particularly the discontinuously conjugated angular ring scaffolds, represents an unprecedented aromatization/cyclization pattern, suggesting the existence of an unusual aromatic polyketides biosynthetic strategy. To elucidate its biosynthetic mechanisms, we conducted in vivo and in vitro experiments to characterize the formation of PDA frameworks in (+)-ABX, (−)-ABX, and FAS. Our results revealed a common biosynthetic logic for the PDA framework among these three products, involving an unusual ARO/CYC (Abx₍₊₎D/Abx₍₋₎D/FasL) and a methyltransferase (MT) (Abx₍₊₎M, Abx₍₋₎M, and FasT), as well as a helper ARO/CYC (Abx₍₊₎C, Abx₍₋₎C, and FasD), catalyzing the formation of the tetracyclic aromatic system.

Results

Reconstituting the Biosynthesis of PDA Framework In Vitro.

The biosynthetic gene clusters (BGCs) of (+)-ABX, (−)-ABX, and FAS were recently cloned by us and other groups (26, 30, 31). Although some post-modifications have been elucidated, the biosynthesis of their polycyclic aromatic frameworks remains elusive. (+)/(−)-ABXs and FAS are phylogenetically categorized as C24 to C30 pentangular polyphenols within the long-chain polyketide family (5, 32). The biosynthesis of these polyketides involves three classes of ARO/CYCs: TcmN-like, TcmI-like, and TcmJ-like ARO/CYCs, which are responsible for the formation of the pentangular polyphenols scaffold (Fig. 1) (5, 32). (+)/(−)-ABXs and FAS BGCs have these three ARO/CYCs as well, including TcmN-like (Abx₍₊₎C, Abx₍₋₎C, and FasD), TcmI-like (Abx₍₊₎D, Abx₍₋₎D, and FasL), and TcmJ-like ARO/CYC (Abx₍₊₎R, Abx₍₋₎R, and FasR). However, unlike the continuously conjugated frameworks in other pentangular polyphenols, (+)/(−)-ABXs and FAS have very different scaffolds, suggesting that at least part of their ARO/CYCs may have a distinct cyclization mechanism. Apart from the E-ring, the folding patterns and carbon numbering of the A-, B-, C-, and D-rings in (+)/(−)-ABXs and FAS are identical. In addition, the three pathways share considerable homologous enzymes (Fig. 2A) for polyketide chain synthesis and aromatization/cyclization. These clues suggest that the three pathways may share a common biosynthetic logic for the putative PDA framework. This PDA framework is then assumed to be catalyzed by unidentified enzymes to give rise to the structurally distinct E ring in the (+)/(−)-ABXs and FAS (26, 30, 31).

The PDA framework could be synthesized by KS_α/CLF, ACP, MT, and three ARO/CYCs. To test this assumption, we cloned and overexpressed each enzyme, including KS_α (Abx₍₊₎P, Abx₍₋₎P, and FasA), CLF (Abx₍₊₎K, Abx₍₋₎K, and FasB), ACP (Abx₍₊₎S, Abx₍₋₎S, and FasC), MT (Abx₍₊₎M, Abx₍₋₎M, and FasT), TcmN-like ARO/CYC (Abx₍₊₎C, Abx₍₋₎C, and FasD), TcmI-like ARO/CYC (Abx₍₊₎D, Abx₍₋₎D, and FasL), and TcmJ-like ARO/CYC (Abx₍₊₎R, Abx₍₋₎R, and FasR) from the three pathways (30, 31) in Escherichia coli or Streptomyces lividans TK24. We also included malonyl-CoA synthetase (MatB) (33) from Streptomyces coelicolor M1154 and malonyl-CoA transferase (FabD) (34) from Streptomyces albidoflavus J1074 to supply and prime the extender unit malonyl-CoA. Although most enzymes can be obtained as N-terminal 6×His-tagged proteins from E. coli, attempts to soluble expression of the KS_α/CLF heterodimers Abx₍₊₎PK and Abx₍₋₎PK in E. coli and S. lividans TK24 were unsuccessful. Fortunately, the KS_α/CLF FasAB from the FAS pathway can be effectively expressed and purified from S. lividans TK24 (SI Appendix, Fig. S1).

FasAB is homologous to Abx₍₊₎PK and Abx₍₋₎PK (56.83% and 57.06% identity, respectively), thus, they should be interchangeable. Therefore, we set up three enzymatic reactions, each containing FasAB, FabD, MatB, and the respective enzymes of ACP, MT, and ARO/CYCs from the (+)-ABX, (−)-ABX, and FAS pathways. To our delight, all the three reactions produced the same compound (1) (Fig. 3A, trace I-III). To exclude that the production of 1 in ABX reaction is derived from the misinteraction between FasAB and the ABX enzymes, we performed in vivo studies. The (+)-ABX BGC contains the genes (Abx₍₊₎P, K, S, C, D, M, and R) encoding 1 and lacks the ketoreductase gene (abxE), which is assumed to be involved in the biosynthesis of the E ring (31). We constructed a cosmid (pJTU103) containing the (+)-ABX BGC as well as a few other nonrelated flanking genes (SI Appendix, Table S1) from the (+)-ABX-producing strain Actinomycetes sp. MA7150. Heterologous expression of pJTU103 in S. coelicolor M1154 yielded compound 1 (Fig. 3B, trace I), demonstrating that 1 can indeed be produced by the (+)-ABX pathway. By scaling up the fermentation, 1 was further isolated and structurally characterized by NMR, confirming the presence of the anticipated PDA framework (Fig. 2B). Interestingly, 1 has 25 carbons in the main chain, suggesting that the terminal carboxyl group of the polyketide chain was decarboxylated after being released from ACP.

Fig. 3.

UPLC analysis of the production of 1 (λ = 365 nm). (A) In vitro enzymatic reactions for synthesis of 1. Three reaction systems with minimal PKS containing FabD, FasAB, ACP (Abx₍₊₎S, Abx₍₋₎S, or FasC), MatB, and cofactor SAM, along with CYC/AROs and MTs from each pathway (I-III). (B) In vivo production of 1 in S. coelicolor M1154. I) with pJTU103; II) without pJTU103 (control); III) with abx₍₊₎C-deleted pJTU103; IV) with abx₍₊₎D-deleted pJTU103; V) with abx₍₊₎M-deleted pJTU103.

To further exclude the possibility of 1 being a shunt product due to missing co-factors/enzymes substrate, feeding experiments were performed. We prepared a mutant incapable of producing (−)-ABX by deleting the ARO/CYC abx₍₋₎D, as well as the halogenase gene abx₍₋₎H from the cosmid 4E10 containing (−)-ABX BGC and heterologously expressed it into S. coelicolor M1154 (31). The production of (−)-ABX was restored upon feeding 1, unequivocally establishing that 1 is a real cyclized product and can serve as a substrate for downstream enzymatic reactions (SI Appendix, Fig. S2). Thus, these findings affirm that the mechanism of A-B-C-D ring cyclization is the same for ABXs and FAS, with variations occurring only in the E-ring closure steps, and the minimal PKS, MT, and three ARO/CYCs are sufficient for the tetracyclic cyclization of 1.

Elucidating the Bismethylation and Tetracyclic Ring Cyclization of the PDA Framework.

The presence of a gem-dimethyl group in (+)/(−)-ABXs and FAS is an unusual feature not commonly observed in aromatic polyketides, except for a few products like resistomycin, benastatin, and tetarimycin A (Fig. 2B) (7, 35, 36). Previous studies on the biosynthesis of resistomycin and benastatin have confirmed that the geminal bismethylation in these polyphenols occurs after aromatization/cyclization. Inactivation of the corresponding MTs in their BGCs resulted in the production of anthrone and oxidized anthraquinone (36–38). Therefore, we initially speculated that the geminal bismethylation in the (+)/(−)-ABXs and FAS may follow the same strategy. To verify this assumption, we removed Abx₍₊₎M from the 1-producing reaction. However, inconsistent with previous reports, we did not observe any products, including the anticipated C8-demethylated anthrone-like or oxidized anthraquinone-like compounds (Fig. 4A, trace III). This observation suggests that without Abx₍₊₎M, neither the minimal PKS alone nor any combination of the minimal PKS with the three (+)-ABX ARO/CYCs can catalyze cyclization. Similarly, omitting the Abx₍₋₎M and FasT from the corresponding (−)-ABX and FAS reactions (SI Appendix, Fig. S3) or inactivating abx₍₊₎M in vivo gave rise to the same result (Fig. 3B, trace V). Hence, unlike the previous model, bismethylation is essential for the formation of the aromatic framework in the three pathways.

Fig. 4.

UPLC and HR-MS analysis of the in vitro enzymatic reactions (λ = 365 nm). (A) UPLC profiles of enzymatic reactions. Each reaction contains a minimal PKS (FabD, FasAB, and Abx₍₊₎S) and MatB, while CYC/AROs and MTs of (+)-ABX are selectively added according to the indication in each trace (I-IX). Reactions containing Abx₍₊₎M are all added with SAM, except for reaction VII. (B) HR-MS analysis of compound 2, produced by the reaction containing minimal PKS and Abx₍₊₎M (trace VI) using ¹²C-SAM, ¹³C-SAM, ¹²C-malonyl-CoA (MCoA), or ¹³C-MCoA as cofactors and substrates.

We next set up a reaction containing only the minimal PKS (FasAB, Abx₍₊₎S, FabD) and MatB. UPLC (ultraperformance liquid chromatography) analysis revealed that the minimal PKS didn’t yield any products, as expected above (Fig. 4A, trace IX). In contrast to earlier findings (39), which indicated that most KS-CLFs induced C7-C12 cyclization in the presence of only the minimal PKS, our experiments did not yield any detectable products at 254 nm, 365 nm, or other specified wavelengths. However, a small amount of unknown compound (2) was produced upon the addition of both the Abx₍₊₎M and the cofactor S-adenosyl-l-methionine (SAM, Fig. 4A, trace VI and VII). Similar results were achieved by using (−)-ABX and FAS enzymes (SI Appendix, Fig. S3). Inactivation of abx₍₊₎M did not result in the observation of 2, likely owing to the complex enzymatic reactions in vivo, which potentially leads to the conversion or degradation of trace amounts of 2 (Fig. 3B, trace V).

Since 2 is produced in very low amounts, it is not feasible to isolate a sufficient quantity for structural characterization. Thus, ¹³C-labeled SAM prepared in situ from ¹³C-labeled L-Met and ATP by S-adenosyl-_L-methionine synthase was added to the reaction as a tracer for methyl transfer. High Resolution Mass Spectrometry (HRMS) analysis confirmed a 2.0065+ mass shift of the product, indicating the addition of two ¹³C-methyl groups. Replacing the malonyl-CoA with ¹³C-labeled malonyl-CoA (malonic acid-1, 2, 3-¹³C3) shows an identical mass shift in the product from the ¹³C-labeled SAM, comparing to the product from unlabeled SAM (Fig. 4B). Therefore, 2 is clearly confirmed as a bismethylated product. Since the chemical structure of 2 is not able to be characterized, the exact geminal bismethylation step within the biosynthetic process remains undetermined. However, both in vitro experiments and isotope labeling investigations have revealed a distinct pattern from the prior instances where methylation typically occurs after the final cyclization. In ABXs and FAS, bismethylation is essential for the PDA cyclization and likely takes place before the aromatization/cyclization process.

Next, to delineate the catalytic role of the three putative ARO/CYCs, we removed them sequentially from the 1-producing reaction. Interestingly, the removal of Abx₍₊₎R, a TcmJ-like ARO/CYC, had no impact on the production of 1 (Fig. 4A, trace II). This finding suggests that Abx₍₊₎R may play a role similar to TcmJ in tetracenomycin biosynthesis, acting as an assistant protein but not being necessary for aromatization/cyclization (40). Next, the removal of Abx₍₊₎C, a TcmN-like ARO/CYC, resulted in decreased yield (23.60%) and accumulation of 2, but compound 1 is still produced (Fig. 4A, trace IV). This finding is in stark contrast to the previous understanding that TcmN-like ARO/CYC is essential for the cyclization of C9 to C14 in dodeca/tridecaketide aromatic polyketides (5, 32). To verify this result, abx₍₊₎C was deleted from the cosmid pJTU103, and the resulting cosmid was heterologously expressed in S. coelicolor M1154. Similar to the in vitro enzymatic reaction, the recombinant strain can still produce compound 1, albeit in low yield (Fig. 3B, trace III). Thus, these results establish that Abx₍₊₎C is not an indispensable protein but may serve as a helper or structural protein that stabilizes the binding of poly-β-keto intermediates with the catalytic ARO/CYC.

The last ARO/CYC remaining in the reaction is Abx₍₊₎D. As expected, further removal of Abx₍₊₎D from the reaction leads to the disappearance of 1, and no other products other than 2 is produced (Fig. 4A, trace VI), establishing its essential role in the catalysis of aromatization/cyclization. Additionally, replacing Abx₍₊₎D with Abx₍₊₎C does not restore the production of 1, further confirming the helper or structural function rather than its cyclization function of Abx₍₊₎C (Fig. 4A, trace V). These results are consistent with the in vitro experiments using Abx₍₋₎D and FasL enzymes (SI Appendix, Fig. S3), as well as the in vivo inactivation experiments of Abx₍₊₎D (Fig. 3B, trace IV). Therefore, it is established that the ARO/CYC Abx₍₊₎D/Abx₍₋₎D/FasL catalyzes the cyclization of a geminal bismethylated polyketide chain to form the PDA framework. Notably, the tetracyclic cyclization capability of Abx₍₊₎D/Abx₍₋₎D/FasL, particularly for the formation of discontinuously conjugated angular benzene ring, is unprecedented. Previously, only the TcmN from tetracenomycin demonstrated the ability to catalyze two successive conjugated aromatic rings (41). More so, unlike the TcmN origin, Abx₍₊₎D/Abx₍₋₎D is evolutionarily linked to the putative fourth ring ARO/CYC TcmI of tetracenomycin (42) (Abx₍₊₎D with 39.60% identity) and the putative 4th/5th ring ARO/CYC BenE of pentangular polyphenol (36, 38) (Abx₍₊₎D with 43.81% identity) (SI Appendix, Fig. S4). Thus, Abx₍₊₎D/Abx₍₋₎D/FasL also represents the first non-TcmN type of ARO/CYC capable of catalyzing the C9 to C14 first ring cyclization.

Resolving the Crystal Structure of Abx₍₊₎D.

To elucidate the structural basis for this unique ARO/CYC activity of Abx₍₊₎D, we resolved its high-resolution crystal structure at 1.32 Å (Fig. 5A, PDB: 8IS2) using X-ray crystallography (SI Appendix, Table S2). Abx₍₊₎D forms a tight dimer, with each subunit containing three α-helices (α1, α2, and α3) and five β-strands (β1, β2, β3, β4, and β5). The active site is situated within the cavity of each subunit, accessible through a wide entrance formed by the β1−3 sheet and the α1−3 helix (Fig. 5 B and C). To compare the structure of Abx₍₊₎D with other ARO/CYCs, we performed a structure-based search against the PDB database using the DALI server (43). Consistent with the phylogenetic analysis, TcmI, a fourth ring ARO/CYC of tetracenomycin (PDB: 1TUW) (42) emerged as the most homologous protein with a sequence identity of 39.60% and an RMSD of 1.191 Å to Abx₍₊₎D. These two enzymes share an almost identical 3D architecture, except for a region between β3 and α3, where Abx₍₊₎D has a loop and TcmI has a helix (α3.5) (Fig. 5D). Notably, TcmI catalyzes the last ring closure of the free molecule TcmF2, exhibiting a very different reaction mode compared to Abx₍₊₎D-mediated tetracyclic ring closure of an ACP-tethered polyketide chain.

Fig. 5.

Crystal structure of Abx₍₊₎D and its substrate binding and catalytic mechanism. (A) The apo structure of Abx₍₊₎D (PDB 8IS2). (B) Abx₍₊₎D docking with substrate 3 (cyan, chemical structure see SI Appendix, Fig. S5) and optimized by MD simulation. Hydrogen bonds between residues and hydroxyl groups of the substrate are indicated by dark dashed lines. The key water molecule is shown in red, and its hydrogen bond between the substrate and D27 is indicated by green dashed lines. Potential hydrogen bonds are represented as dashed lines within 3.5 Å. (C) Abx₍₊₎D docking with substrate 4 (cyan, chemical structure see SI Appendix, Fig. S5) and optimized by MD simulation. Interaction details are shown similarly as above. (D) Structural superimposition highlights the distinctions between Abx₍₊₎D (bright yellow) and TcmI (light blue). (E) The relative activity of Abx₍₊₎D mutants.

Given the instability of poly-β-keto intermediates and the challenge in obtaining the enzyme complex structure, we conducted docking simulations of the intermediates into the cavity of Abx₍₊₎D (Fig. 5 B and C and SI Appendix, Fig. S5). The resulting complex structures with substrates 3 and 4 were optimized by molecular dynamics (MD) simulations, revealing minimal conformational changes with an RMSD of 1.447 Å between the two protein structures. Similar to the C9 to C14 ARO/CYC TcmN, which utilizes S67 and R69 to anchor the C11 and C13 hydroxyl groups of the substrate (41), Abx₍₊₎D employs a comparable binding manner. It utilizes the side chain of S26 and R40 at the bottom of the cavity to anchor the first ring of the polyketide intermediate by forming hydrogen bonds with the C11 and C13 hydroxyl groups of the substrates. The C1 to C8 and C15 to 26 segments of the aromatic framework are further stabilized by interactions with the residues Y49, H51, and S76 of Abx₍₊₎D. This binding conformation positions the C1 thiol group of the substrates at the entrance to the cavity, where the ACP is covalently bound to C1 via a thioester bond. As the loop is away from the top of the cavity, the C1 thioester and gem-dimethyl side of the substrates are exposed to the solvent. This “half-open” conformation of Abx₍₊₎D facilitates the docking of the ACP, allowing for the delivery of growing polyketide chains to the active cavity for cyclization and aromatization.

In contrast, TcmI (PDB: 1TUW) has a cavity with an identical shape and orientation, along with a set of binding residues (H26, D27, R40, Y49, and H51). However, its upper side is closed due to the presence of the helix α3.5 (42). As a result, helix α3.5 in TcmI is very close to the C1 thioester of substrate 3, leading to potential interference with ACP and hindering its accessibility (Fig. 5D). This feature restricts TcmI to catalyzing ACP-free polyketide substrates. Analysis of other typical TcmI-type ARO/CYCs also revealed a similar helix segment (SI Appendix, Fig. S6). Therefore, the replacement of the helix α3.5 by the loop in Abx₍₊₎D appears to be a decisive factor enabling its unique catalytic mode. This difference can be utilized as a criterion to differentiate the catalytic function of Abx₍₊₎D/Abx₍₋₎D/FasL-type ARO/CYC from the conventional TcmI-like ARO/CYCs.

Investigating the Structural Basis of Abx₍₊₎D Activity.

To explore the function of residues (S26, D27, R40, Y49, H51, and S76) in catalysis, we conducted mutagenesis studies. We initially mutated these residues into alanine, but all of the mutants, except for S26A and S76A, resulted in the formation of inclusion bodies. Thus, saturated mutagenesis of R40, Y49, and H51 was employed, and a total of five soluble mutants, including R40N, R40H, Y49L, H51E, and H51Q were obtained. Among these mutants, S26A and R40H significantly reduced the cyclization activity to 15.56% and 2%, respectively, while R40N completely abolished the catalytic activity (Fig. 5E and SI Appendix, Fig. S7). The inactivity of R40N indicates that residue 40 may require more than just a polar functionality capable of anchoring the substrate through hydrogen bonding. Rather, a general acid appears necessary at this position, akin to R69 in TcmN (41). In TcmN, another residue, Y35 functions as a general base, collaborating with R69 to promote the first- and second-ring cyclization. Interestingly, a similar residue doesn’t exist in Abx₍₊₎D. However, a water molecule coordinated by D27 via hydrogen bond in the crystal structure of Abx₍₊₎D has an ideal distance of 3.4 Å to C14 of the substrate (Fig. 5 B and C). We speculate that D27 may polarize the water to act as a general base, abstracting the hydride from C14 and initiating the first ring closure. As anticipated, mutants D27E, D27Y, D27H, and D27K all lost catalytic activity (Fig. 5E and SI Appendix, Fig. S7). Although D27E also has a carboxyl group, its longer side chain could position the water molecule improperly for C14. These results support the crucial role of these three amino acids in controlling and catalyzing the cyclization and aromatization.

H51 is close to the 2nd/3rd ring, as well as the C17 carbonyl/hydroxyl group of substrates (5 to 9 in SI Appendix, Fig. S5). Mutants H51E and H51Q exhibited severe compromises in catalytic activity (9.22% and 11.42%, respectively) (Fig. 5E and SI Appendix, Fig. S7), supporting its role in substrate binding and catalysis. Similar to H109 in ZhuI (44), H51 may participate in acid-base chemistry and charge-transfer activation for 2nd/3rd-ring cyclization through aldol condensation. However, the residual activity of H51E/Q indicates that spontaneous aldol cyclization may also occur for ring closure, as assumed in some ARO/CYCs (39). The C19- and C21-, C23- carbonyl groups of 3 are coordinated by S76 and H51, Y49 via hydrogen bond, respectively. Mutation of S76 to S76A decreases the catalytic activity of Abx₍₊₎D to 54%, supporting its role in substrate binding. Y49 can oscillate between the C21 and C25 ketones, potentially catalyzing the C20 to C25 cyclization of the angular ring by abstracting the hydride of C20 and protonating the nascent C25 hydroxyl group. Its mutation to leucine (Y49L) nearly abolishes the activity of Abx₍₊₎D (only retains 1%), and simultaneous mutation of Y49 and H51 (Y49L-H51E) completely abolishes the activity. These results collectively confirm the critical role of the above residues in substrate binding and catalysis (SI Appendix, Fig. S8). Additionally, none of the above mutants produced cyclization shunt products, suggesting that the Abx₍₊₎D active site not only provides the catalytic moiety but also directs the chemical outcome of cyclization.

Discussion

The cyclization and aromatization of type II polyketides are pivotal for generating structural diversity and bioactivity. This study has unveiled a unique type of ARO/CYC that programs the aromatization/cyclization of a PDA framework in (+)/(−)-ABXs and FAS. Typically, the production of a polyketide framework involves the action of 2 to 4 ARO/CYCs that sequentially mediate regioselective aldol condensation or form multi-enzyme complex for synergistic catalysis (1 to 5). In very few cases, oxygenases can generate complex polyketide scaffolds (45, 46). The capability of Abx₍₊₎D/Abx₍₋₎D/FasL in catalyzing tetra-ring cyclization is unique among ARO/CYCs. Beyond its diverse cyclization capabilities, Abx₍₊₎D/Abx₍₋₎D/FasL also exhibit a distinctive ability to catalyze the formation of a discontinuously conjugated angular ring and represent the first non-TcmN type of ARO/CYC capable of catalyzing the C9 to C14 first ring cyclization. These combined features are unprecedented among ARO/CYCs, substantially enriching our knowledge of the aromatization/cyclization process in type II polyketides.

The 3D architecture of Abx₍₊₎D closely resembles that of TcmI in tetracenomycin biosynthesis (42), but it contains a unique loop region between β3 and α3, where TcmI has a helix (α3.5). In typical TcmI-like ARO/CYCs, this helix is conserved and positioned right on top of the reaction cavity, partially closing it and impeding the accessibility of ACP. In contrast, Abx₍₊₎D features a loop instead of a helix, creating a cavity that is more exposed to solvent and facilitating the docking of ACP to deliver polyketide chains for cyclization and aromatization. The structural difference confers a distinct catalytic mode to Abx₍₊₎D. Abx₍₊₎D is phylogenetically close to TcmI, and their similar 3D structure and substrate binding mode suggest a possible evolutionary relationship. Engineering of PKS ARO/CYC to alter cyclization pattern is highly challenging and has not been successful, to the best of our best knowledge. This study provides a natural example of how to design ARO/CYCs with distinct functions.

In summary, our study has characterized the biosynthesis of discontinuously conjugated aromatic ring scaffold in (+)/(−)-ABXs and FAS. Through in vitro and in vivo investigations, we unraveled a unique biosynthetic strategy that involves a necessary gem-dimethylation step in aromatization/cyclization process. This process employs an atypical TcmI-like ARO/CYC (Abx₍₊₎D/Abx₍₋₎D/FasL) as well as a nonessential TcmN-like helper protein (Abx₍₊₎C, Abx₍₋₎C and FasD) to program the formation of a discontinuously conjugated tetracyclic aromatic architecture. Our structural analysis of Abx₍₊₎D revealed a distinct loop region that likely to contribute to its unique catalytic mode compared to other conventional TcmI-type ARO/CYCs. Furthermore, mutagenesis analysis identified key catalytic residues and highlighted the conformation of the reaction cavity that directs the chemical outcome of cyclization. Based on these findings, we proposed a biosynthetic mechanism for this unique type of ARO/CYC. Overall, our study significantly expands the understanding of type II aromatic polyketide framework biosynthesis and provides a structural template for the future generation of additional ARO/CYC and aromatic polyketide frameworks.

Materials and Methods

Detailed Materials and Methods are described in SI Appendix.

Gene Expression and Protein Purification.

The genes, templates, plasmids, digestion sites, and expression hosts used for protein expression are documented in SI Appendix, Table S3. Briefly, the amplified DNA fragments from genomic DNA by specific primers and primerstar polymerase were cloned into vectors utilizing the Spark HiFi Seamless Cloning Kit to generate expression plasmids. The resulting plasmids were confirmed by DNA sequencing and then transferred into the host strain for protein expression. All genes, except for fasAB and abx₍₊₎S, were expressed as N-terminal 6× His-tagged fusion proteins. Seed cultures were prepared by inoculating individual colonies into 4 mL of LB solution, supplemented with the corresponding antibiotics (SI Appendix, Table S4), and incubated for 16 h. Then, 0.4 L LB was inoculated with 4 mL of seeds and incubated at 37 °C until OD₆₀₀ reached 0.5 to 0.8. The cells were then cooled to 16 °C, and 0.1 mM isopropyl-β-D-thiogalactopyranoside (IPTG) was added. The cells were incubated for 24 h and harvested by centrifugation. The harvested cells were resuspended in 30 mL lysis buffer [25 mM HEPES, pH 7.5, 300 mM NaCl, 5 mM imidazole, 10% (v/v) glycerol]. The cells were lysed by sonication, followed by centrifugation (12,000 rpm, 30 min, 4 °C). The supernatant was mixed with Ni-NTA agarose resin and loaded onto a gravity flow column. Proteins were eluted using increasing concentrations of imidazole (25 mM, 50 mM, 100 mM, and 300 mM) in Buffer A [25 mM HEPES, pH 7.5, 300 mM NaCl, 10% (v/v) glycerol]. Protein was further concentrated by centrifugation using Amicon Ultra-4 (3 kDa and 10 kDa, GE Healthcare). Protein purity was determined by 12.5% (w/v) Sodium Dodecyl Sulfate-Polyacrylamide Gel Electrophoresis (SDS-PAGE), and protein concentration was determined by Bradford's method using a BSA calibration curve. The sizes of the obtained proteins were consistent with those observed in 12.5% SDS-PAGE (SI Appendix, Fig. S1).

For FasAB, the gene cassette containing fasAB was amplified from the genomic DNA of Streptomyces kanamyceticus CGMCC 4.1441 and inserted into the NdeI and EcoRI sites of pWY45 to yield the plasmid pJTU102 (47). FasAB was expressed as a heterodimer with a 6×His tag at the N-terminus of FasA. Following a standard protocol described previously, the pJTU102 was transferred to S. lividans TK24 via conjugation, and exconjugants with an apramycin (Am)-resistant (50 µg mL⁻¹) phenotype were selected to generate the recombinant strain mJTU26 (47, 48). PCR-validated positive exconjugants were cultured in 100 mL trypsin soybean broth (TSB) medium and shaken at 220 rpm at 30 °C for 3 d. Seed cultures (1 mL) were inoculated into 4 L (100 mL × 40) YEME medium and cultured for 3 d. Thiostrepton (10 µg mL⁻¹) was added to the medium to induce FasAB expression, and the culture was maintained for 24 h. The cells were harvested by centrifugation and lysed by UH-24 high-pressure homogenizers (600 bar, 5 min, 400 mL min⁻¹) and the subsequent procedures followed the protein purification established in E. coli. The calculated sizes of the obtained proteins were consistent with the sizes observed in 12.5% SDS-PAGE (SI Appendix, Fig. S1).

In Vitro Biochemical Assay.

Malonyl-CoA was synthesized in situ by combining malonic acid (2 mM), HSCoA (2 mM), MgCl₂ (2 mM), ATP (4 mM), MatB (10 µM), and 0.23 mL HEPES Buffer (100 mM HEPES, 100 mM NaCl, pH 7.5) at 30 °C for 30 min. Then, SAM (5 mM) was added to the malonyl-CoA solution, followed by the addition of FabD (20 µM) and various combinations of (+)/(−)-ABXs or FAS enzymes (such as, 30 µM FasAB, 50 µM Abx₍₊₎S, 25 µM Abx₍₊₎C, 20 µM Abx₍₊₎D, 20 µM Abx₍₊₎R, 30 µM Abx₍₊₎M) to the mixture. A control experiment was performed by replacing FasAB with HEPES buffer. The enzyme mixture was incubated at 37 °C for 16 h. After the incubation, an equal volume of EtOAc was added, and the mixture was centrifuged at 12,000 rpm for 10 min, repeated twice. EtOAc was removed by rotary evaporation, and the sample was dissolved in 50 µL MeOH, then further analyzed by a UPLC on a Shimadzu system with an SPD-M30A detector and shim pack XR-ODS III (2.0 mm × 75 mm, 1.6 μm). The solvents used were (A) H₂O containing 0.1% formic acid and (B) MeCN, with a gradient of 20 to 100% B over 15 min, 100 to 20% over 5 min, and a hold at 20% B for 2 min, with the detector set at 365 nm. The flow rate was 0.2 mL min⁻¹, and the injection volume was 50 μL.

For ¹³C-malonyl-CoA preparation, a reaction containing 0.1 mL HEPES Buffer, pH 8.0, 10 mM MgCl₂, 10 mM ATP, 10 mM ¹³C-malonic acid, 4 mM HSCoA, and 10 μM MatB was incubated at 30 °C for 10 min. ¹³C-SAM was prepared by a reaction containing 0.1 mL HEPES Buffer, pH 8.0, 20 mM MgCl₂, 5 mM ATP, 5 mM ¹³C-L-Met, 150 mM KCl, and 10 μM SAM synthetase at 37 °C for 10 min. ¹³C-malonyl-CoA and ¹³C-SAM reaction mixtures were then added together or individually to the PKS enzymes combination to initiate the synthesis of ¹³C-labeled product. All other treatment procedures were identical to the above.

Genomic Library Construction and Screening.

The genomic library of MA7150 was constructed using the cosmid pJTU2554 (49), which contains oriT, ϕC31-int, and attp for conjugation and integration into a Streptomyces host. The detailed procedure for genomic library construction and screening is the same as the previously reported (31).

Heterologous Expression, Isolation, and Purification of Compound 1.

The wild-type recombinant strain mJTU27 was generated by transferring the pJTU103 into S. coelicolor M1154 via conjugation using a previously described standard protocol (31). PCR-verified exconjugants were inoculated into 4 mL of TSB medium supplemented with Am and trimethoprim and shaken at 30 °C for 3 d. A 1 mL seed culture was transferred into a 500-mL conical flask containing 100 mL of ISP2 medium (0.4% yeast extract, 1% malt extract, 0.4% glucose, 2% NaCl, pH 7.2) and fermented at 30 °C for 7 d. The fermentation broth was extracted three times with an equivalent volume of ethyl acetate (EtOAc), dried by rotary evaporator, and redissolved in 2 mL MeOH. The sample was further purified by a semi-preparative HPLC with detection at 365 nm. Semipreparative HPLC was performed on the Shimadzu LC-20AT Prominence liquid chromatograph with an SPD-M20A detector using the Promosil C₁₈ column (10.0 mm × 250 mm, 5 μm) from Agela technologies. The gradient elution method was utilized with a pair of solvents containing 0.1% formic acid in H₂O (A) and MeCN (B), with the gradient set as 20 to 100% B in 15 min, 100 to 20% in 5 min, and kept at 20% B for 2 min. The flow rate was set at 2 mL min⁻¹ and the injection volume was 100 μL.

Structural Elucidation of the Isolated 1.

Compound 1 was obtained as a yellowish powder. Mass spectral analysis of compound 1 provided a molecular formula of C₂₇H₂₄O₈. The ESI-HR-MS spectrum showed a peak at m/z 477.1544 [M + H]⁺, which matched the calculated value of 477.1544 (SI Appendix, Fig. S9). This molecular formula was further supported by ¹H and ¹³C NMR data (SI Appendix, Figs. S10–S13 and Table S6). Upon comparing the ¹H NMR spectra of 1 and streptovertidione (19), it was observed that four methoxy groups (δ_H 3.45, δ_H 3.81, δ_H 3.92, and δ_H 3.94) present in streptovertidione were replaced by hydroxyl groups in compound 1 (SI Appendix, Figs. S10 and S11).

Protein Crystallization and Structure Determination.

The crystals of Abx₍₊₎D were obtained using the vapor diffusion method. Crystals were grown in 2.4 μL drops containing 1.4 μL purified Abx₍₊₎D (3.8 mg mL⁻¹), 0.8 μL reservoir solution (0.25 M HEPES, 60% Tacsimate, pH 8.1), and 0.2 μL additive (2.0 M NDSB-211) at 20 °C. The crystals were cryo-protected in the reservoir solution supplemented with 20% glycerol and flash-frozen in liquid nitrogen using nylon loops. The datasets for the native crystal were collected at the BL17U1 beamline of the National Center for Protein Sciences, Shanghai. All data were processed using the X-Ray Detector Software (XDS).

The structure of Abx₍₊₎D was determined by molecular replacement method using the program Phaser in the PHENIX package, with the atomic coordinates of 1TUW serving as the searching model. The initial model was rebuilt in COOT and iteratively refined with PHENIX. All graphic presentations of the structure were generated using the Python Molecular Viewer (PyMOL).

Supplementary Material

Appendix 01 (PDF)

Data, Materials, and Software Availability

All study data are included in the article and/or SI Appendix. The X-ray crystallographic density map and the refined model of Abx₍₊₎D have been deposited in Protein Data Bank (https://www.rcsb.org) under accession number 8IS2 (50).