Non-enzymatic pyridine ring formation in the biosynthesis of the rubrolone tropolone alkaloids

The pyridine ring is a potent pharmacophore in alkaloid natural products. Nonetheless, its biosynthetic pathways are poorly understood. Rubrolones A and B are tropolone alkaloid natural products possessing a unique tetra-substituted pyridine moiety. Here, we report the gene cluster and propose a biosynthetic pathway for rubrolones, identifying a key intermediate that accumulates upon inactivation of sugar biosynthetic genes. Critically, this intermediate was converted to the aglycones of rubrolones by non-enzymatic condensation and cyclization with either ammonia or anthranilic acid to generate the respective pyridine rings. We propose that this non-enzymatic reaction occurs via hydrolysis of the key intermediate, which possesses a 1,5-dione moiety as an amine acceptor capable of cyclization. This study suggests that 1,5-dione moieties may represent a general strategy for pyridine ring biosynthesis, and more broadly highlights the utility of non-enzymatic diversification for exploring and expanding natural product chemical space.

R ubrolone A (1) was first reported in 1978 as a red pigment produced by Streptomyces enchinoruber 1 , and subsequent structure elucidation revealed a nearly coplanar and highly conjugated aglycone consisting of a tropolone ring, a five-membered cyclopentanone, and a pyridine ring 2 (Fig. 1). A unique carbon-carbon bond links C-2 of the deoxysugar to the tropolone ring of the aglycone. This highly oxygenated hybrid structure with distinct ring assembly has inspired chemists and microbiologists, but only two total synthetic routes to the rubrolone aglycone have been devised 3,4 and the biosynthetic pathway has not been elucidated.
Although rubrolone A has shown no significant biological activity in cytotoxicity and antimicrobial assays 1 , we recently isolated rubrolone B (2) (Fig. 1) as a new potentially cardioprotective rubrolone from the endophytic strain Streptomyces sp. KIB-H033 (S. sp. KIB-H033) 5 . Importantly, feeding experiments with [ 13 C]-acetate demonstrated aglycone assembly via type II PKS chemistry followed by complex oxidative rearrangements to afford the tropolone, which differs from the pinacol rearrangement-based ring expansion during fungal stipitatic acid tropolone biosynthesis 6 . Moreover, the additional benzoic acid moiety of rubrolone B (2) compared with rubrolone A (1) provides an example of a unique cationic inner salt moiety, the presence of which raises questions about biosynthetic pathway logic and timing; for example, is rubrolone A the biosynthetic precursor to rubrolone B, or vice versa?
The rubrolones are members of the pyridine alkaloid family of natural products ( Fig. 1) found widely in nature and possessing antimalarial, anticancer and antibacterial activities [7][8][9][10][11] . The biologically significant pyridyl moiety is assembled via diverse biosynthetic pathways that primarily originate from amino acid precursors. For example, the pyridine rings of virginiamycin S (ref. 12) and nikkomycin 13 derive from L-lysine, while aspartate generates the pyridyl moiety of nicotine 14 and pyridomycin 15 and L-tryptophan is the precursor for the pyridyl moiety of thiocoraline 16 (Fig. 1). Only in rare cases are pyridine alkaloids derived from other sources; for example, one of the collismycin pyridine rings is derived from L-cysteine and acetate ( Fig. 1) 17 . However, because most of these insights have been achieved from feeding isotope-labelled precursors, the detailed mechanisms of alkaloid pyridine ring formation are poorly understood.
To understand the mechanism of rubrolone biosynthesis in S. sp. KIB-H033, and pyridyl construction in particular, we herein report the biosynthetic gene cluster together with heterologous expression experiments that enable us to propose a biosynthetic pathway. Critically, inactivation of rubrolone sugar biosynthetic genes led to the accumulation of a key intermediate that was shown to react non-enzymatically with either ammonia or anthranilic acid to afford the respective aglycones of rubrolones A (1) and B (2). This study provides important new insight into the non-enzymatic and non-amino acid origins of the pyridyl moiety in alkaloid natural products.

Results
Identification of the rubrolone biosynthetic gene cluster. Our strategy for identifying the rubrolone (rub) biosynthetic gene cluster was based on our hypotheses that (i) the aglycone of rubrolones A and B is of polyketide synthase origin 5 , and (ii) a sugar 4,6-dehydratase installs the C-6 methyl group of the deoxysugar. The dehydratase-encoding gene has been used as a probe to successfully identify the gene cluster of the enediyne antibiotic C-1027 (ref. 18). To search for these two gene loci, the genomic deoxyribonucleic acid (DNA) of S. sp. KIB-H033 was subjected to Illumina sequencing. Using KSa and sugar 4,6-dehydratase genes as sequence queries to scan the assembled scaffolds, we identified a 20-kb region of the genome possessing both genes. To validate the involvement of this DNA region in rubrolone biosynthesis, a genomic library of S. sp. KIB-H033 was constructed based on cosmid pJTU2554 (ref. 19), from which the cosmid p9B10 containing this DNA region was identified through library screening. Integration of p9B10 into the chromosome of Streptomyces albus J1074 (ref. 20) by conjugation generated strain S. albus 9B10, from which heterologous expression of the putative rubrolone gene cluster could be monitored by a colour change on an agar plate; the recombinant strain produced a ruby red pigment, whereas control strain containing the relevant empty vector pJTU2554 did not ( Supplementary Fig. 1). To confirm the identity of this red metabolite, liquid culture extracts of recombinant strain S. albus 9B10 and appropriate control were analysed by high-performance liquid chromatography (HPLC). The putative rub-containing clone produced a peak with  retention time identical to that of rubrolone A (Fig. 2, trace I), and this peak was absent from the S. albus control strain containing the empty vector (Fig. 2, trace II). The presence of rubrolone A (1) was further confirmed by high-resolution electrospray ionization mass spectrometry (HRESIMS) and 1 H NMR spectroscopy. Overall, these results demonstrated that cosmid 9B10 possesses the requisite gene cluster encoding the biosynthesis of rubrolone A.
Determining the rub gene cluster boundaries. The sequence of the 37 open reading frames identified in cosmid 9B10 (Fig. 3a) has been deposited in Genbank with accession number KX218108. To determine the rub gene cluster boundaries, we first constructed cosmid p9B10-1 (from genes rubS1 to orf37, Fig. 3a) using l red-mediated polymerase chain reaction (PCR) targeting mutagenesis 21 , and this strain still produced rubrolone A ( Supplementary Fig. 2, trace I). In contrast, inactivation of the oxidoreductase-encoding rubS1 gene abolished production of 1 in the new recombinant strain S. albus 9B10-DS1, but resulted in the accumulation of two new compounds 3 and 4 (Fig. 2, trace III). Therefore, the left boundary of this gene cluster was determined to be rubS1. To locate the right boundary of the gene cluster, cosmid p9B10-2 (genes from orf1 to rubE9, Fig. 3a) was constructed and was then conjugated into S. albus generating the engineered strain S. albus 9B10-2, which did not affect the production of 1 ( Supplementary Fig. 2, trace II). The gene at the right boundary of cosmid 9B10-2 is the putative ketosynthase (KS) chain length factor (CLF) encoding gene rubE9, in-frame deletion of which in strain S. albus 9B10-DE9 abolished production of 1 (Fig. 2, trace IV), confirming type II PKS involvement in rubrolone biosynthesis. Having established rubS1 and rubE9 as the boundaries of the B22.5 kb rub gene cluster (Fig. 3a), we then analysed all 22 ORFs using a basic local alignment search tool to assign putative functions (Supplementary Table 1). In summary, the rub biosynthetic genes encode: (i) PKS and associated enzymes (RubE1-E9), (ii) sugar biosynthesis enzymes (RubS1-S7), (iii) oxygenases (RubA-C) and the regulatory protein RubR and (iv) two unknown proteins (RubN1 and RubN2  Enzymes in the biosynthesis of the rubrolone scaffold. Our previous 13 C-acetate feeding experiments demonstrated that the aglycone originates from type II PKS chemistry followed by complex oxidative rearrangements 5 . Bioinformatic analysis yields the expected minimal type II PKS encoded by rubE8 (KSa), rubE9 (CLF/KSb) and rubE4 (acyl carrier protein, ACP), which share high sequence similarity with the type II PKS involved in R1128 biosynthesis (ZhuB, 79% identity; ZhuA, 71% identity; and ZhuN, 68% identity) 22 . Additionally, rubE1, rubE2 and rubE3 encode b-ketoacyl/ACP synthase III (KSIII), ACP and an acyltransferase, respectively, which share high sequence similarity with a set of starter unit butyryl-ACP generating enzymes (ZhuH, 61% identity; ZhuG, 53% identity; and ZhuC, 56% identity) in R1128 biosynthesis 23,24 . Two putative polyketide cyclases, RubE5 and RubE6, are also found within the rub gene cluster and share high sequence similarity with the cyclase/dehydratase protein ZhuJ (69% identity) and cyclase ZhuI (79% identity),     Fig. 4, comparison of the two gene clusters), which is consistent with common structural features of a propyl side chain and the same PKS chain length in both R1128 and rubrolones (Fig. 3b). Interestingly, the putative reductase RubE7 is not found in the R1128 biosynthetic gene cluster, and may function as the missing ketoreductase (KR) responsible for generating the butyryl starter unit (Fig. 3b). Inactivation of rubE7 slightly reduced the production of 1 to B60% of production in S. albus 9B10 ( Supplementary Fig. 2, trace III), consistent with previous assumption upon partial complementation by the KR from fatty acid biosynthesis 22 .
Overall, bioinformatic analysis of the rub gene cluster has identified PKS genes with high sequence similarity to genes responsible for the biosynthesis of R1128 (Supplementary Table 1) 22 . This supports our previous proposal of KSIIIcatalysed formation of the butyryl-ACP starter unit for type II PKS-catalysed extension with seven malonyl-CoA additions to generate the highly reactive poly-b-ketone backbone (Fig. 3b).
Enzymes for oxidative formation of the tropolone scaffold. Although clearly of polyketide origin, the rubrolone carbon skeleton possesses a labelling pattern consistent with a series of complex oxidative rearrangements 5 . There are three putative oxygenases in the robulone gene cluster: RubA belongs to the bacterial luciferase-like monooxygenase superfamily 26 ; RubB possesses 55% identity to ZhuM, a cytochrome P450 oxygenase catalysing anthraquinone formation in R1128 biosynthesis 22 ; RubC is similar to naphthocyclinone hydroxylases like ActVA-orf5 (59% identity) that catalyse ring hydroxylation in the actinorhodin biosynthetic pathway 27 .
To determine if any of these three oxygenases may be involved in tropolone ring formation, we constructed the respective in-frame gene deletion mutants S. albus 9B10-DA, S. albus 9B10-DB and S. albus 9B10-DC for metabolite analysis by HPLC. Although inactivation of rubA did not affect rubrolone production ( Supplementary Fig. 2, trace IV), both S. albus 9B10-DB and S. albus 9B10-DC shared a similar metabolite profile that included abolished production of 1 together with three new peaks 5-7 (Fig. 2, traces V and VI). Large-scale fermentation of S. albus 9B10-DB enabled the purification of sufficient quantities of 5-7 for structure elucidation by mass spectrometry (MS) and NMR. The known structure of R1128A (5) was determined by comparison to the 1 H and 13 C NMR data in the literature 28 . The structures of the new compounds 6 and 7 were elucidated by a combination of HRESIMS and onedimensional (1D) and 2D NMR spectroscopic analyses (Supplementary Table 3; Supplementary Fig. 3).
Because deletion of either rubB or rubC completely abolished production but resulted in the accumulation of R1128A (5) and the related compounds 6 and 7 (Fig. 3b), both RubB and RubC are required for tropolone ring formation, and without both of these enzymes present RubE5 and/or RubE6 can catalyse cyclization and dehydration reactions analogous to those catalysed by their homologues ZhuJ and ZhuI to furnish R1128A. In addition, RubB, RubC and/or an oxidase in the host strain may catalyse oxidation and possibly also the dehydration reactions that transform 7 to R1128A (5) (Fig. 3b); a similar bifunctional oxygenase/dehydratase has been found in jadomycin biosynthesis 29 . The isolation of 5-7 from mutant strains is unexpected but consistent with the presence of homologues of all R1128 biosynthetic genes in the rub gene cluster ( Supplementary  Fig. 4). Taken together, the data suggest that RubB and RubC together may catalyse the complex oxidative rearrangements leading to tropolone ring formation, and that these oxygenases may work on the ACP-bound intermediate before the second cyclization to direct biosynthetic divergence between rubrolones and R1128 (Fig. 3b).
To determine if these compounds are biosynthetic intermediates or shunt products, 3 and 4 were fed to a mutant with rubrolone biosynthesis blocked at an early stage. RubE8 (KSa) and RubE9 (CLF) form a heterodimer catalysing Claisen-like condensation of malonyl-ACP extender units. The rubE9inactivated mutant S. albus 9B10-DE9 generated a clean background with no rubrolone-related metabolites detected by HPLC (Fig. 2, trace IV), and therefore was used as the host to carry out feeding experiments with 3 and 4. HPLC chromatograms revealed that both compounds were completely converted to 1 (Fig. 2, traces VII and VIII), suggesting that both 3 and 4 are biosynthetically competent intermediates en route to rubrolone A (1).
To investigate the timing and specificity of the putative glycosyltransferase RubS7, we sought to present this enzyme with large quantities of 4, which is the precursor of the pyridylcontaining aglycone. This was achieved by decreasing the amount of nitrogen source in the culture media, which dramatically decreased the production of 1 and accumulated large amounts of 4 in the heterologous expression strain S. albus 9B10 ( Supplementary Fig. 2, trace VIII). This suggests that the downstream glycosyltransferase RubS7 is specific for the pyridyl-containing aglycone, and therefore pyridyl moiety formation occurs before glycosylation.
Our collective interrogation of the sugar biosynthetic genes suggests a biosynthetic pathway from glucose-1-phosphate to dTDP-2-keto-D-fucose, which is installed on the fully maturated pyridyl-containing rubrolone aglycone by the glycosyltransferase RubS7 (Fig. 3c). An unprecedented carbon-carbon bond between C-2 of the sugar and C-10 of the aglycone is then formed by an unknown mechanism. Significantly, the compounds 3 and 4 that accumulate upon sugar biosynthetic gene inactivation are biosynthetic intermediates en route to rubrolones.
Anthranilic acid is a biosynthetic precursor to product 2.
Having established the intermediacy of 3 and 4 in the biosynthesis of rubolone A (1), we next sought to understand the biosynthetic incorporation of the benzoic acid moiety of rubrolone B (2). Although a 2-l fermentation of the wild type strain S. sp. KIB-H033 produces large quantities of 1 (60 mg l À 1 ) and 2 (50 mg l À 1 ) under even suboptimal conditions 5 , the heterologous expression strain S. albus 9B10 showed a similar titre to the WT strain for 1, but only a very small amount of 2 (B0.2 mg) was isolated from 2-l fermentation broth. The varying titre ratios of rubrolones A (1) and B (2) in different strains may arise from biosynthetic pathways in which (i) one is the precursor of the other or (ii) they arise from divergent amination of a common intermediate 4 (Fig. 5). For example, if 1 is the precursor of 2, then the benzoic acid moiety would need to be attached to the nitrogen atom of 1 by oxidative N-C coupling to afford 2 (Fig. 5). Conversely, initial production of 2 would require subsequent N-C bond cleavage to liberate the benzoic acid moiety and yield 1 (Fig. 5). Alternatively, 3 and 4 may generate 1 and 2 via divergent elaboration of 4 with either ammonia or anthranilic acid as respective nitrogen sources (Fig. 5). Interestingly, the latter hypothesis is favored by the presence of large quantities of anthranilic acid only in the wild type strain, which produced large quantities of 2. In contrast, anthranilic acid was not detected in the heterologous expression strain S. albus 9B10, which yielded only trace amount of 2.
To begin to differentiate among these three possible routes, we fed 15 N-labelled anthranilic acid to the fermentation media of wild type strain S. sp. KIB-H033 at 24 h (the time point at which 1 and 2 could first be detected) and 48 h after inoculation, and then isolated and analysed 1 and 2 by HRMS, 15 N NMR and 1 H-15 N long range correlation experiments to characterize 15 N incorporation. Consistent with the divergent pathway hypothesis, adding 200 mg l À 1 of 15 N-labelled anthranilic acid to cultures yielded 15 N enrichment in 2 but not 1 (Supplementary Fig. 5). Moreover, adding exogenous anthranilic acid to the heterologous expression strain S. albus 9B10 completely restored production of 2 (Fig. 2, trace IX). In summary, the incorporation of exogenous anthranilic acid supports its role as a biosynthetic intermediate en route to rubrolone B (2).
The pyridyl moiety arises from non-enzymatic amination of 4. The divergent biosynthetic pathways for 1 and 2 implied by anthranilic acid incorporation, together with the production of intermediate 4 in sugar biosynthetic gene knockout mutants, strongly suggested that 4 can react with free ammonia and anthranilic acid to generate the pyridine ring and pyridine inner salt moiety, respectively (Fig. 5). To further test this hypothesis, we fed 200 mg l À 1 anthranilic acid to the 4-producing mutant S. albus 9B10-DS1. HPLC analyses of the fermentation extracts revealed production of a new compound 8 (Fig. 2, trace X), of which we isolated 7.9 mg from a 2-l culture. Evaluation of the HRESIMS at m/z 416.1140 [M À H] À for C 24 H 18 NO 6 and 1D and 2D NMR spectra of 8 (Supplementary Table 3; Supplementary Fig. 3) in comparison to the NMR data of 2 and 3 (Supplementary Table 2) led to the structural assignment of 8 (Fig. 4) as the aglycone of rubrolone B (2).
The chemical structure and reactivity of the biosynthetic intermediate 4, the presence of excess anthranilic acid in the fermentation culture of the wild type strain S. sp. KIB-H033, and the absence of suitable gene candidates in the rub cluster together suggest that amination may occur non-enzymatically. This would enable production of designer RUB analogues with altered benzoate moieties if the RubS7 glycosyltransferase were sufficiently promiscuous to attach different aglycones to the NDP sugar. Feeding F, Cl, Br or OH substituted anthranilic acids to the culture medium of S. albus 9B10 led to the production of several novel rubrolones based on HRMS data ( Supplementary  Figs 6-11), two of which (Fig. 4) were isolated and characterized in detail by NMR (rubrolones C (10) and D (11), Supplementary  Fig. 3; Supplementary Table 4).
Encouraged by the in vivo results, we next sought to obtain evidence for the non-enzymatic reactions by performing in vitro reactions between the isolated biosynthetic intermediate 4 and either ammonia or anthranilic acid analogues in phosphate buffer at pH 8.0 to mimic the fermentation media. As expected, incubation of 4 with ammonium acetate yielded a new HPLC peak with a retention time corresponding to that of the intermediate 3 (Fig. 6, trace II), while a peak corresponding to 8 appeared when anthranilic acid was used in place of ammonium acetate (Fig. 6, trace III). Furthermore, incubation of 4 with F, Cl, Br or OH substituted anthranilic acids yielded new rubrolone aglycones based on HRMS data ( Supplementary  Figs 12-15). These results clearly demonstrate the non-enzymatic divergent elaboration of 4 with either ammonia or anthranilic acid analogues to afford diverse rubrolone aglycones.

Discussion
Rubrolones are tropolonoid natural products with an unusual carbon skeleton of polyketide origin containing a unique 2,3,4, 6-tetra-substituted pyridine ring or pyridine inner salt moiety. In this work, we have identified a 22 ORF rubrolone biosynthetic gene cluster and performed a series of experiments that are consistent with the following features of rubrolone biosynthesis (Fig. 3): (i) a KSIII generates a butyryl-ACP starter unit, and a type II PKS catalyses seven rounds of chain extension; (ii) oxidative enzymes RubB and RubC together catalyse the complex oxidative rearrangements leading to tropolone ring formation; (iii) glucose-1-phosphate is transformed to dTDP-2keto-D-fucose, which is then attached to the fully maturated pyridyl-containing rubrolone aglycone by glycosyltransferase RubS7 and (iv) 4 is a common intermediate in rubrolone biosynthesis, from which structural divergence to rubolones A (1) and B (2) occurs by non-enzymatic condensation with respective amine donors ammonia and anthranilic acid. Several experimental results demonstrated that free ammonia and anthranilic acid can react with 4 to generate the pyridyl moieties of rubrolone A (1) and rubrolone B (2), respectively. First, only 2 was enriched with 15 N after feeding 15 N-labelled anthranilic acid to the wild type strain, demonstrating that direct conversion of 2 to 1 via N-C bond cleavage did not occur (Fig. 5). Second, exogenous anthranilic acid could completely restore production of 2 to the anthranilic acid-deficient heterologous expression strain S. albus 9B10 (Fig. 2, trace IX). Third, the rubrolone B aglycone 8 ( Fig. 4; Supplementary Fig. 3) was directly observed when anthranilic acid was supplemented in the fermentation media of the sugar biosynthesis-deficient strain S. albus 9B10-DS1. Overall, this data clearly implicates a biosynthetic pathway that diverges from 4 in the presence of ammonia or anthranilic acid to respectively afford either rubrolone A (1) or rubrolone B (2).
Although our in vivo data support divergent amination of 4, the absence of suitable aminotransferase enzymes in the rub gene cluster suggested a non-enzymatic mechanism of pyridine formation. This hypothesis was proven by observing in vitro transformations of 4 to either 8 or 3 in the presence of anthranilic acid or ammonium acetate, respectively. Non-enzymatic reactions are increasingly being found in natural product biosynthetic pathways and have been exploited to expand chemical diversity of several classes of natural products. In this work, we demonstrated the facile non-enzymatic incorporation of several substituted anthranilic acids to explore rubrolone chemical space. Similar non-enzymatic condensations between carbonyl and amine functional groups are well known in natural products biosynthesis and have afforded discoipyrroles 38 , ammosamides 39 , elansolids 40 and jadomycins 41,42 . Non-enzymatic Diels-Alder reactions have been widely observed en route to natural products such as paracaseolide A (ref. 43). Additional pericyclic reactions have been observed to generate impressive diversity of fungal polyketides by exploiting the intrinsic reactivity of an isochromene intermediate 44 . In summary, our non-enzymatic incorporation of various amine donors into the rubrolone scaffold highlights the potential for further non-enzymatic elaboration of rubrolones.
Most alkaloid pyridine ring biosyntheses involve proposed cyclization between carbonyl and amine functional groups, with one recent exception being an enzyme-catalysed formal [4 þ 2] cycloaddition between two dehydroalanines and subsequent aromatization en route to the pyridyl moiety of thiocillins 45 . We propose a non-enzymatic rubrolone pyridine construction that does not involve an amino acid precursor, but instead features the capture of free amines from solution by the reactive 1,5-dione moiety of 4 ( Supplementary Fig. 16A). Specifically, we propose that the large distance between the C-15 ketone and C-3 could require hydrolytic ring opening and isomerization to the 1,5-dicarbonyl intermediate 4a, which rapidly interconverts with 4 as shown by the disappearance of the H-2 and H-4 NMR signals of 4 in D 2 O (Supplementary Fig. 16B). The reactive 1,5-dicarbonyl intermediate 4a would readily capture free amines to generate the ketimine 4b. Subsequent dehydration and aromatization would afford the rubrolone pyridyl moiety 4c (Supplementary Fig. 16A). The condensation of 1,5-diones with amines is a general approach to pyridyl formation that has been widely exploited in chemical synthesis [46][47][48] , but suffers from required extremes of temperature and pH (refs 47,48). Nature has solved this problem in rubrolone biosynthesis by generating a highly reactive 1,5-dione intermediate to non-enzymatically capture free amines en route to pyridyl assembly. We propose that such non-enzymatic capture of diverse free amine metabolites may represent a general biosynthetic strategy for furnishing pyridine rings in natural products such as cimicifugadine 49 , beatrines 50 and cassiarins 51 ( Supplementary  Fig. 17), and this study sets the stage for further discovery of an expanded set of diverse pyridine-containing natural products.

Methods
General materials and experimental procedures. All reagents, solvents and restriction enzymes were purchased from standard commercial sources and used directly. DNA isolation and manipulation in Escherichia coli (E. coli) and ARTICLE Streptomyces were performed according to standard protocols. PCR amplifications were carried out on Biometra professional thermocycler (070-851, An Analytik Jena Company, Germany) using either Taq DNA polymerase (TaKaRa) or Pfu DNA polymerase (Thermo scientific). Primer synthesis and DNA sequencing were performed at Beijing Zixi Bio Tech Co., Ltd.
Genome sequencing and genomic library construction. The genome sequencing of S. sp. KIB-H033 was performed with the Illumina Genome Analyzer (Illumina, San Diego, CA) by BGI (BGI-Shenzhen, China). High-molecular-mass genomic DNA isolated from S. sp KIB-H033 was used to construct small (200-500 bp) and large (2-3 kb) random sequencing libraries. The reads were first filtered and assembled into 85 contigs using SOAPdenovo (http://soap.genomics.org.cn/). We then used the paired-end information, step-by-step from the shortest (200 bp) to the longest (2,000 bp) insert size, to join the contigs into 17 scaffolds. Putative protein-coding sequences were predicted using the GLIMMER program, and the annotation was accomplished by BlastP analysis of sequences in the Nr, Nt, and SwissProt databases and by manual curation of the outputs of a variety of similarity searches. According to standard procedures, the genomic DNA was partially digested with MboI. Then the 30-45 kb DNA fragments were isolated and ligated to cosmid pJTU2554. MaxPlax Lambda packaging extracts were used to packaging. About 2,000 E. coli clones were picked, and stored in 20 96-well microplates at À 80°C.
Heterologous expression and mutant construction. The positive cosmids were introduced to the selected model Streptomyces host by E. coli-Streptomyces conjugation according to the standard protocol 21 . Cosmid 9B10 which contains the whole rub gene cluster was used for the construction of gene deletion mutant. Since the cosmid pJTU2554 has the apramycin resistance gene aac(3)IV and oriT, the target gene was replaced with erythromycin resistance cassette ermE which was amplified from plasmid pJTU6722 (provided by Prof Meifeng Tao (Shanghai Jiao Tong University, China)) using l RED-mediated PCR targeting mutagenesis with corresponding primers (Supplementary Table 6). To isolate the markerless gene deletion mutant, the gene inactivation cosmids were introduced into E. coli DH5a/BT340 via electroporation; incubation overnight at 42°C to induce expression of FLP recombinase resulted in the loss of ermE cassette and the generation of an 81-bp in-frame scar, affording final markerless gene inactivation vectors. All the gene inactivation vectors used in this study were constructed by the same strategy ( Supplementary Figs 18-28).
The gene inactivation vectors were introduced by transformation into E. coli ET12567/pUZ8002 and the vectors were then transferred to S. albus J1074 by intergeneric conjugation according to the standard methods 21 . The spores of S. albus J1074 were produced in the MS medium. E. coli-Streptomyces conjugations were performed on MS solid medium freshly supplemented with 20 mM MgCl 2 . The mutants were selected with 50 mg ml À 1 apramycin after 4 days cultivation and subsequently identified through PCR with corresponding primers (Supplementary Table 6). Finally, 11 markerless gene inactivation mutants were constructed in this study.
Chemical purification of compounds 1-8 and 10-11. For HPLC analysis of the fermentation extracts of engineered heterologous expression strains, the frozen spore stocks (100 ml) of strains were inoculated into 250 ml baffled-flask containing 50 ml of seed medium and shaken at 250 r.p.m. and 28°C. After 30 h, an aliquot of seed culture (2.5 ml) was transferred into a 250 ml baffled-flask containing 50 ml of production medium and shaken at 250 r.p.m. and 28°C for 4 days. The 50 ml fermentation culture broth was separated into mycelia and supernatant by centrifugation at 3,500g for 30 min. The resulting supernatant was subsequently mixed with D101 macroporous absorbent resin (5 ml) and stirred for 20 min. After removal of supernatant, the resin was washed with 20 ml H 2 O, and then 10 ml methanol. The 10 ml CH 3 OH was evaporated to dryness, and then the residues were dissolved in 1 ml CH 3 OH. The samples were applied to reverse-phase HPLC analysis eluted with a flow rate of 1 ml min À 1 over a 28 min gradient as follows: , and the column temperature is 25°C.
A 2 -l fermentation culture broth of S. albus 9B10 was separated into mycelia and supernatant by centrifugation at 3,500g for 30 min. The resulting supernatant was subsequently applied to a D101 macroporous absorbent resin column (0.5 l), the column was washed with H 2 O and then successively eluted with 100% methanol. The CH 3 OH fraction was evaporated to dryness, and the residue was dissolved in 50% CH 3 OH/H 2 O and subjected to Sephadex LH-20 CC eluting with 50% CH 3 OH/H 2 O to yield 18 fractions. All the fractions were analysed by HPLC and combined. Selected fractions contained the rubrolone analogues were further purified by semipreparative C 18 HPLC to yield compounds 1 (26.8 mg) and 2 (B0.2 mg).
A 2-l fermentation culture broth of S. albus 9B10-DS1 was separated into mycelia and supernatant by centrifugation at 3,500g for 30 min. The resulting supernatant was subsequently applied to a D101 macroporous absorbent resin column (0.5 l), the column was washed with H 2 O and then successively eluted with 100% methanol. The CH 3 OH fraction was evaporated to dryness, and the residue was dissolved in CH 3 OH and subjected to Sephadex LH-20 CC eluting with CH 3 OH to yield 21 fractions. The fractions contained the compounds 3 and 4 were combined by HPLC analysis. Finally, the compounds 3 (15.6 mg) and 4 (33.7 mg) were purified by semipreparative C 18 HPLC. A 5-l fermentation culture broth of S. albus 9B10-DB was separated into mycelia and supernatant by centrifugation at 3,500g for 30 min. The resulting supernatant was extracted with ethyl acetate (3 l) for three times, then evaporation of EtOAc generated the residue. This residue was dissolved in CH 3 OH and mixed with an appropriate amount of polyamide, then the sample was applied to a reverse-phase C 18 column and subsequently eluted with gradient CH 3 OH-H 2 O (from 10:90 to 100:0) to afford 27 fractions, each of which was analysed by HPLC. The fractions containing the 5-7 were combined and evaporated, then subjected to Sephadex LH-20 CC eluting with CH 3 OH to yield 25 fractions, and combined by HPLC analysis. The compounds 5 (6.9 mg), 6 (11.3 mg), and 7 (4.2 mg) were finally purified by semipreparative C 18 HPLC from selected fractions.
Compound 5: Yellow powder; 1 H and 13 C NMR data was summarized in Supplementary The S. albus 9B10-DS1 was grown in seed medium for 30 h, then about 2.5 ml seed culture (5% inoculums) was inoculated into 50 ml production medium at 30°C. The anthranilic acid was added to the fermentation media at 24 h (10 mg in 100 ml DMSO) after inoculation, and then fermented for additional 3 days The extract was analysed by HPLC and found the production of compound 8. For isolation of compound 8, a 2-l fermentation culture broth of S. albus 9B10-DS1 feeding with anthranilic acid (200 mg l À 1 ) was separated into mycelia and supernatant by centrifugation at 3,500g for 30 min. The resulting supernatant was subsequently applied to a D101 macroporous absorbent resin column (0.5 l), the column was washed with H 2 O and then successively eluted with 100% CH 3 OH. The CH 3 OH fraction was evaporated to dryness, and the residue was dissolved in CH 3 OH and subjected to Sephadex LH-20 CC eluting with CH 3 OH to yield fractions, and combined by HPLC analysis. The fractions containing 8 was finally subjected to semipreparative C 18 HPLC to yield the pure compound 8 (7.9 mg).
Compound 10: Purple red powder; 1 H, 13  Feeding the S. albus 9B10-DE9 with compounds 3 and 4. The S. albus 9B10-DE9 mutant was grown in seed medium for 30 h, then about 2.5 ml seed culture (5% inoculums) was inoculated into 50 ml production medium at 30°C. After 24 h incubation, compounds 3 (6 mg) and 4 (6 mg) in 100 ml DMSO were added into culture, respectively, and fermented for additional 3 days. The extracts were analysed by HPLC.
Feeding the S. albus 9B10 with anthranilic acid. The S. albus 9B10 was grown in seed medium for 30 h, then about 2.5 ml seed culture (5% inoculums) was inoculated into 50 ml production medium at 30°C. After 24 h incubation, anthranilic acid (10 mg) in 100 ml DMSO was added into culture and fermented for additional 3 days. The extract was analysed by HPLC.
Synthesis of 15 N-anthranilic acid and feeding experiment. The 15 N-labelled anthranilic acid was synthesized from 15 N urea and phthalic anhydride by the modified Hofmann reaction 52 . Before isotope-labelled precursor feeding, the production of rubrolones was monitored and found that rubrolones was produced at around 24 h after the start of fermentation. So, the 15 N-labelled anthranilic acid at a final concentration of (200 mg l À 1 ) was fed to the fermentation media of the wild type strain S. sp. KIB-H033 at 24 and 48 h after inoculation. From 1-l fermentation culture broth, rubrolones A (10.8 mg) and B (12.6 mg) were isolated following the purification procedure previously published 5 . Both compounds were analysed by HRESIMS. Finally, the same amounts (8.0 mg) of 1 and 2 in 0.45 ml DMSO-d 6 solvent were subjected to 15 N NMR measurements to determine the 15 N incorporation ( Supplementary Figs 5 and 69).
In vitro reactions. All reactions were performed in sodium phosphate buffer (pH 8.0, 45 mmol). A mixture of compound 4 (1.5 mg, 5.03 mmol) and ammonium acetate (1.95 mg, 25.14 mmol) or anthranilic acid (3.45 mg, 25.14 mmol) was suspended in 1.0 ml of phosphate buffer and the mixture was stirring at 30°C for 4 days. After reaction, 10 ml mixture was subject to HPLC analysis eluted with a flow rate of 1 ml min À 1 over a 28 min gradient as follows: T ¼ 0, 10% B; T ¼ 20, 100% B; T ¼ 24, 100% B; T ¼ 25, 10% B; T ¼ 28, 10% B (A, H 2 O; B, CH 3 OH), and the column temperature was 25°C. The same condition was used for generation of new rubrolone aglycones 16-19 using the 5-F, -Cl, -Br or -OH substituted anthranilic acids as the amine donors. The HPLC peaks corresponding to 16-19 were collected for HRESIMS analysis. Data availability. Sequence data that support the findings of this study has been deposited in GenBank with accession codes KX218108. The authors declare that all other relevant data supporting the findings of this study are available within the article and its Supplementary Information Files and from the corresponding author upon reasonable request.