Streptomyces coelicolor M1146 was used as a host for the heterologous expression of the type II gene cluster. Escherichia coli EPI100 and the pTG19-T vector were used for general cloning. E. coli JTU007/pUZ8002 and the pOJ436 vector were used for the conjugation of E. coli/Streptomyces. E. coli BW25113/pKD47 was used for gene knockout. The pET-30a vector and E. coli BL21 were used for protein expression. Biochemicals and media were purchased from Sangon Biotech Co., Ltd. (Shanghai, China) and Wanqing Co., Ltd. (Nangjing, China) unless otherwise stated. Restriction enzymes were purchased from TaKaRa Biotechnology Co., Ltd. (Beijing, China).

Cosmid DNA was isolated from the Yunnan soil library as the template to amplify the KS_α sequence using the 540F&1100R primer, which was designed based on the KS_α domain in type II polyketide synthase (Wawrik et al., 2005; Wang et al., 2017). Amplicons of the correct predicted size [560 base pair (bp)] were gel-purified, sequenced, and compared with deposited KS_α genes in the National Center for Biotechnology Information (NCBI) database. Unique KS_α genes were used as probes to recover type II PKS-containing clones by a serial dilution method using the following touchdown protocol: denaturation (95°C, 4 min), 10 touchdown cycles (95°C, 40 s; 65°C [−1°C per cycle up to 55°C], 40 s; 72°C, 1 min); 30 standard cycles (95°C, 40 s; 55°C, 40 s; and 72°C, 40 s); and a final extension step (72°C, 10 min).

Open reading frames (ORFs) were deduced from the sequence with the assistance of the RAST server program (https://rast.nmpdr.org/). The corresponding deduced proteins were compared with other known proteins in the databases using available BLAST methods (http://www.ncbi.nlm.nih.gov/blast/).

Cosmid YN1903 was retrofitted with the oriT- and Amp^R-containing Dra I fragment from pOJ436 (Bierman et al., 1992). Retrofitted cosmid was then conjugated from E. coli JTU007/PUZ8002 into S. coelicolor M1146 for the heterologous expression via a standard intergeneric conjugation protocol (Feng et al., 2011; Musiol et al., 2011).

The seed culture of the S. coelicolor YN1903 strain was prepared by inoculating 0.5 ml of the spore with 50 ml of R5 liquid medium, and the resultant solution was incubated for 3 days at 28°C with shaking at 225 revolutions per minute (rpm). Then, the seed culture was inoculated in 50 ml of ISP4 liquid medium (1:100), to which 5 g of HP-20 resin was added. The cultures were incubated for 7 days at 28°C with shaking at 225 rpm. After fermentation, HP-20 resin was rinsed with water and then dried by air. Next, the resin was extracted three times with 100% methanol. Methanol extracts were combined, concentrated, and subjected to high-performance liquid chromatography (HPLC) analysis. Methanol eluents were analyzed by HPLC (1 ml/min) using a linear gradient from 80:20 H₂O:MeOH to 100% MeOH over 40 min.

Metabolites obtained in the methanol extraction were subjected to silica gel column chromatography for the first-round isolation. The crude extract was subjected to a silica gel column by elution with a mixture of CH₂Cl₂ and MeOH, with a gradient from 100:1 → 50:1 → 20:1 → 10:1 → 5:1, and the elution was detected by HPLC. The extract was further purified by semi-preparative HPLC separation (Fisher Wharton C18, 5 μm, 10 mm × 250 mm) on Shimadzu LC-20A. The column was equilibrated with 40% solvent A (H₂O containing 0.1% formic acid)/60% solvent B (MeOH) and developed with the following program: 0 −30 min, a linear gradient increase from 40% A/60%B to 100%B. The flow rate was 4 ml/min, and the detection wavelength was 435 nm.

For the detection of high-resolution mass spectrometry (HR-MS), the extracts were dissolved in chromatographic-grade methanol and centrifuged for 10 min. The resultant clear supernatant (10 μl) was used for MS analysis. HR-MS was performed on Agilent Q-TOF 6520A. A nuclear magnetic resonance (NMR) study was performed on a 600-MHz NMR (BRUKER AVANCE III 600). Data were generated for triplicate experiments.

Polymerase chain reaction- (PCR-) targeting system was used for in-frame deletion of the genes in the YN1903 gene cluster (Gust et al., 2002). The primers used in this study are listed in Supplementary Table S1; all primers used for gene deletion contained a Bcu I-recognized site at their 5′ end. Using apramycin-resistance gene as the template, Apr fragments with Bcu I cleavage sites on both sides were amplified. Then, 39-bp bases were selected from the upstream and downstream regions of the target gene as the homologous arm, and the amplified fragment in the above step was used as the template to amplify the apramycin-resistance gene fragment with homologous arms. The PCR amplification protocol was as follows: denaturation at 95°C for 3 min, 5 touchdown cycles at 98°C for 1 min; 68°C [−2°C per cycle until 58°C was reached] for 15 s; and 72°C for1 min 30 s); 30 standard cycles at 98°C for 1 min; 58°C for 15 s; and 72°C for 1 min 30 s; and a final extension step at 72°C for 10 min. The obtained gene fragment was then electrically transferred into the competent cell E. coli BW25113/pKD47/YN1903. The recombinant plasmid was digested by the Bcu I enzyme and then ligated with the T4 DNA ligase to obtain the plasmid knockout of the target gene. The mutant plasmid was confirmed by sequence analysis and was subsequently conjugated into S. coelicolor M1146 for fermentation.

Deoxyribonucleic acid (DNA) was extracted from cosmid YN1903 as the template, and a DNA fragment (amplified with the primers coeI protein-F and coeI protein-R) containing the coeI gene was cloned into the pTG19-T vector; after verification by sequencing, the 1.48-kb Nde I/Xhol I fragment was recovered from the pTG19-T vector and then ligated into the same site of pET-30a to yield the pETcoeI plasmid, which was then introduced into the bacterial expression strain E. coli BL21 to obtain E. coli BL21/pETcoeI for expressing CoeI to give a C-terminal 6 × His-tagged protein.

Escherichia coli BL21/pETcoeI was inoculated into 4 ml of LB medium containing 50 μg/ml of kanamycin. The culture was grown at 37°C overnight and then transferred to 400 ml of Lysogeny Broth (LB) medium containing 50 μg/ml of kanamycin and 0.25 mM of ammonium ferrous sulfate. When OD₆₀₀ of the culture reached ~0.3, 50 μl of 0.3 M cysteine and 100 μl of 0.1 M ammonium ferrous sulfate were added, and the culture was grown at 37°C until an OD₆₀₀ of ~0.6 was reached. Then, 50 μl of 1 M isopropyl β-d-1-thiogalactopyranoside (IPTG) was added, and the culture was incubated at 18°C for another 20 h.

The cultures were harvested by centrifugation at 4,000 × g for 10 min. The collected cells were resuspended in 10 ml of lysis buffer (NaH₂PO₄ 300 mM, NaCl 50 mM, imidazole 10 mM, and pH 8.0) containing 1 mg/ml of lysozyme, and the resuspended cells were disrupted by ultrasonication. The bacterial suspension was centrifuged at 4,000 × g for 30 min. The supernatant was then subjected to affinity purification on a column by elution with different concentrations of imidazole prepared by mixing lysis buffer with elution buffer. The protein was eluted in 3 ml of 250 mM imidazole elution buffer, subjected to a desalination column to eliminate salt ions, and finally dissolved in a 3-ml Tris•HCl buffer (Tris 50 mM, NaCl 100 mM, glycerol 10%, and pH 8.0). The purified protein was confirmed on a 12% sodium dodecyl sulfate-polyacrylamide gel electrophoresis (SDS-PAGE) gel. The obtained protein was concentrated again to 200 μl using a 50-kDa ultrafiltration tube, snap-frozen in liquid nitrogen, and stored at −80°C until further use.

The radical SAM enzyme needs to be reconstituted under anaerobic conditions before enzymatic assays. Reconstitution was performed at 4°C (Jin et al., 2018). Approximately 30 μl of 1 M dithiothreitol (DTT) was added to 3 ml of protein dissolved in Tris•HCl buffer, and the mixture was inoculated for 15 min. Then, 30 μl of 50 mM ammonium ferrous sulfate was added to make a final concentration of 0.5 mM; then, after 45 min, 10 μl of 50 mM sodium sulfide was added every 30 min for three times to make a final concentration of 0.5 mM and reconstituted for at least 3 h. Finally, the reconstituted mixture was treated on a desalination column to obtain the dark-brown protein in 3-ml Tris•HCl buffer.

Computational docking experiment was performed using Alphafold v2.4.

All solutions used in an anaerobic glove box required prior anaerobic treatment. The solution was placed in a flask and immersed in liquid nitrogen for snap-freezing. The completely frozen solution in the flask was evacuated in a vacuum and then placed in flowing water. This process was repeated three times. After replacing oxygen inside the flask three times by withdrawing nitrogen, the flask containing the solution was placed in an anaerobic glove box.

In vivo enzymatic assay was tested by whole-cell transformation experiments. Whole-cell transformation experiments were performed as follows: 400 ml of cultured cells of E. coli BL21/pETcoeI for protein expression were centrifuged and washed two times with pre-chilled phosphate-buffered saline (PBS) buffer (pH 7.0). The resuspended cell pellet was then lysed in 40 ml of PBS buffer; 10 ml of the suspension was transferred to a 50-ml centrifuge tube; and 10-mM substrate was added. The reactions were incubated at 30°C, 225 rpm, for 12 h.

All in vitro enzymatic assays were performed in an anaerobic glove box with <1 parts per million (ppm) of O₂. Enzymatic reactions were conducted in Tris•HCl buffer (Tris 50 mM, NaCl 100 mM, glycerol 10%, pH 8.0) with the following components: 1 μM reconstituted CoeI, 10 μM substrate, 1 mM SAM, 5 mM DTT, 5 mM MgCl₂, 5 mM Na₂S₂O₄, and 7% dimethyl sulfide (DMSO). The reactions were incubated at 30°C for 12 h and then quenched with methanol. Data from triplicate experiments were collected.

Liquid chromatography–mass spectrometry (LC–MS) analysis of enzymatic products was performed in a negative ion mode by a reverse-phase column (Grace Alltech Alltima, C18, 5 μm, 100 Å, 10 × 250 mm) on an Agilent 1200 series. The gradient elution was as follows: 0–29 min, a linear increase from 50% A (H₂O)/50% B (MeOH) to 5% A/95% B; 29–31.5 min, 5% A/95% B; and 31.5–34 min, a linear increase to 50% A/50% B. The flow rate was 0.8 ml/min, and the detection wavelength was 435 nm.

High-performance liquid chromatography analysis of enzymatic products was performed using a reverse-phase column (Grace Alltima, C18, 5 μm, 100 Å, 10 mm × 250 mm) on an Agilent 1200 series. The gradient elution was as follows: 0–30 min, a linear increase from 40% A (H₂O)/60% B (MeOH) to 0% A/100% B and 30–35 min, 0% A/100% B. The flow rate was 0.8 ml/min, and the detection wavelength was 435 nm. Data from triplicate experiments were collected.

The sequence data of the genes in the COE gene cluster were deposited in GenBank under accession nos. MN601984–MN601997.

A cosmid clone (YN1903) with a 42-kb insert containing PKS-II genes was identified by PCR from a soil metagenomic library (Wang et al., 2017). The cosmid DNA was retrofitted with the Dra I fragment-containing oriT and Amp^R from pOJ436 and then conjugated into S. coelicolor M1146. An HPLC analysis of the fermentation broth of S. coelicolor YN1903 (S. coelicolor M1146 harboring the COE cluster) showed that three clone special compounds were produced [Figure 1C(d)]. These compounds were purified, and their structures were elucidated by HR-MS and NMR spectroscopy. HR-MS revealed that compound 1 has the chemical formula C₁₆H₁₀O₆ ([M–H]⁻ 297.0406, observed, 297.0399, calculated). The ¹³C NMR data (Supplementary Figure S13 and Supplementary Table S2) of compound 1 showed 16 carbon signals (δ_C 142.3, 122.4, 161.2, 112.9, 136.4, 118.6, 136.7, 124.8, 161.9, 117.3, 189.7, 130.7, 182.5, 132.9, 20.43, and 168.8), which were consistent with the published data (Krupa et al., 1989). The integral value of a single peak (δ_H 2.51) in ¹H NMR of compound 1 is 3, which represents the methyl proton in the structure and belongs to the 1-CH₃ group. In the heteronuclear multiple bond correlation (HMBC) spectra, the hydroxyl groups at the C-3 and C-8 positions had a strong coupling with the quaternary carbon of δ_C 161.2 and δ_C 161.9, respectively, so δ_C 161.2 was assigned to C-3 and δ_C 161.9 was assigned to C-8. A carboxyl group (δ_C 168.8) belonging to 2-COOH was supported by HMBC correlation with H₃-1–CH₃. Therefore, compound 1 was identified as 3,8-dihydroxy-1-methylanthraquinone-2-carboxylic acid (DMAC), a polyketide intermediate that was isolated from an engineered Streptomyces strain with a known biosynthetic pathway in that strain (Javidpour et al., 2013) (Supplementary Figures S12–S16 and Supplementary Table S2). The HR-MS analysis of compound 2 gave the chemical formula C₁₅H₁₀O₆ ([M–H]⁻ 285.0406, observed, 285.0399, calculated). TheHR-MS analysis of compound 3 yielded the chemical formula C₁₆H₁₂O₆ ([M–H]⁻ 299.0563, observed, 299.0556, calculated). A careful comparison of the NMR data (Supplementary Figures S12–S23 and Supplementary Table S2) demonstrated that the structures of 2, 3 were highly similar to those of compound 1. Unlike compound 1, there was a hydroxyl group at the C-1 position of compound 2 (δ_C 165, OH-1), and a hydroxymethyl group at the C-2 position (δ_C/δ_H 51.6/4.50, CH₂OH-2). The only difference between compounds 3 and 2 was that there was a methyloxymethyl group at the C-2 position of compound 3 (δ_C/δ_H 58.1/4.43; δ_C/δ_H 61.6/2.61, CH₂OCH₃-2). The interpretation of NMR data revealed that compounds 2 and 3 were coelulatins A and B, respectively, which had previously been reported as plant metabolites (Bowie et al., 1962) (Figure 1B). Interestingly, the hydroxymethyl and methyloxymethyl groups at C-2 of compounds 2 and 3 are distinguished from the reported bacterial aromatic polyketides, implying a rare modification step at C-2 during the biosynthesis of coelulatin.

In-frame-deletion was performed to elucidate the role of various genes in biosynthesis. The gene coeB in-frame deletion mutant strain S. coelicolor YN1903ΔcoeB failed to produce polyketide compounds, supporting a corresponding relationship between polyketide compounds and COE BGC [Figure 1C(e)]. In-frame-deletions of orf10 and orf11 did not affect the production of metabolites, indicating the boundaries of the BGC [Figure 1C(b,c)]. The genes between orf10 and orf11 were analyzed, and their homologies and deduced functions are listed in Table 1. The COE gene cluster contained 13 genes, including polyketide skeleton biosynthetic genes, regulating modification genes, and transcription regulation genes (Table 1). Bioinformatic analysis revealed that the CoeI protein belongs to the family of HemN-like enzymes (Figure 2A). Phylogenetic analysis revealed that CoeI was clustered with homologs in a HemN-like escalade separated from the other radical SAM enzymes (Figure 2B). All five homologs belonged to HemN-like radical SAM enzymes and catalyzed a variety of reactions, including decarboxylation (Layer et al., 2003), hydroxylation (Jansson et al., 2003), cyclization (Layer et al., 2002; Wu et al., 2017), ring opening reaction (LaMattina et al., 2016), and cyclopropanation (Hiratsuka et al., 2014).

To determine the function of CoeI in the biosynthesis of YN1903 compounds, the coeI in-frame-deletion mutant S. coelicolor YN1903ΔcoeI was constructed. In contrast with the strain S. coelicolor YN1903, S. coelicolor YN1903ΔcoeI did not produce metabolites 2 and 3; however, it accumulated an intermediate product 4 [Figure 1C(a)]. High-resolution electrospray ionization mass spectrometry (HR-ESI-MS) analysis of the intermediate product 4 revealed the chemical formula C₁₅H₁₀O₆ ([M–H]⁻ 255.0303, observed, 255.0293, calculated). The interpretation of NMR data (Supplementary Figures S24–S28 and Supplementary Table S3) revealed that the intermediate product 4 was the C-2 de-hydroxymethyl form of 2 (Figure 1B). Thus, it was confirmed that CoeI played a critical role in C-2 modification in the biosynthesis of 2 and 3.

The ORF of coeI was amplified by PCR, and the PCR product was ligated with the pET-30a vector to obtain pETcoeI. Plasmid pETcoeI was transferred into E. coli BL21, and the transformant was cultured for CoeI expression. The recombinant CoeI protein was purified using a Ni-NTA column and detected by SDS-PAGE. A clone special band of 56.7 kDa when analyzing the supernatant of the broken cell on the SDS-PAGE corresponded to the predicted size of the recombinant CoeI well (Figure 2C). The concentration of purified recombinant CoeI was 20.16 mg/ml.

To reconstitute the tetrairon–tetrasulfur cluster necessary for the use of ferrous ammonium sulfate and sodium sulfide in the catalysis of HemN-like radical SAM enzymes, CoeI was reconstituted before the in vitro analysis in an anaerobic glove box (Jin et al., 2018). The CoeI protein exhibited a light brown color before reconstitution, but a brown–black color after reconstitution, which was identical to the color of a HemN-like SAM enzyme reported previously (Jin et al., 2018).

HemN-like enzymes contain a unique three-cysteine motif (CxxxCxxC) (Supplementary Figure S1) that binds a [4Fe−4S] cluster, which acts as a direct initiator of the enzyme reaction. These proteins also contain two SAM binding sites (Supplementary Figure S1). To understand how CoeI recognizes its substrate, computational docking was performed using a substrate (compound 4), a [4Fe−4S] cluster, and two SAM molecules as a ligand. The structure of CoeI is very similar to that of HemN (Layer et al., 2003). CoeI consists of two distinct domains. The N-terminal region bears a barrel that binds all cofactors, a 4Fe−4S cluster, and two SAM molecules. The N-terminal trip-wire and the C-terminal domain are probably involved in substrate binding (Figure 3A). The substrate (compound 4) hydroxyl is linked by the side chains of Lys16 and Thr427 in CoeI with an H-bond (Figure 3B).

Whole-cell transformation experiments were performed using proteins extracted from E. coli BL21/pETcoeI culture. When product 4 was used as a feedstock, in addition to products 2 and 3, a new product 5 was also detected (Supplementary Figure S3). The HR-MS analysis of product 5 identified the chemical formula C₁₅H₁₀O₅ ([M–H]⁻ 269.0460, observed, 269.2320, calculated). The interpretation of NMR data (Supplementary Figures S29–S33 and Supplementary Table S3) revealed that product 5 was a C-2 methylated form of product 4 (Supplementary Figure S6). Further, product 5 was also introduced into whole-cell transformation but could not be transformed into any product (Supplementary Figure S3).

The character of CoeI was studied using in vitro enzymatic assays. When product 4 was used as a substrate, it can be methylated by CoeI to form product 5 (Figure 4). Without SAM and Na₂S₂O₄, product 4 could no longer be converted to product 5 by CoeI, indicating that the reaction was both SAM- and Na₂S₂O₄-dependent (Figure 4A).

However, unlike whole-cell transformation experiments, the generation of products 2 and 3 could not be detected in the enzymatic assay system, even after increasing the amount of enzyme and prolonging the reaction time. Therefore, product 5 was used as a substrate to test the activity of CoeI, but no new peaks were detected on HPLC (Supplementary Figure S2).

In vitro biochemical activity of CoeI toward substrate 2 was tested. Many peaks appeared in the HPLC profile of the living system. The formation of substrate 3 in the reaction system could be detected ([M–H]⁻ = 299.06), but the reaction could still proceed with boiled CoeI (Figure 5A), implying that substrates 2 and 3 occurred nonenzymatically. In addition, a new peak of [M–H]⁻ = 421.24 had a higher yield than that of substrate 3 (Figure 5A). To confirm the source of the methyl group of substrate 3, we individually removed SAM and DTT from the system. The reaction could still proceed without adding SAM to the system (Supplementary Figures S4, S5), implying that the source of the methyl group of substrate 3 was not SAM. After removal of DTT from the reaction system, the peak of [M–H]⁻ = 421.24 disappeared. We speculated that the production of this compound was related to DTT (Supplementary Figure S3).

Given that the source of the methyl group of substrate 3 is not SAM, the source of the methyl group is probably methanol used to quench the reaction. To verify this conjecture, the reaction mixture was extracted with ethyl acetate instead of quenching with methanol. The production of substrate 3 could no longer be detected as a result of the reaction. In addition, perdeuterated methanol was used to quench the reaction. In accordance with expectations, substrate 3 was produced with an increased molecular weight of 3 (Figure 5B), confirming that the source of the methoxyl group in substrate 3 was methanol, which is used to quench the reaction (Figure 5C).