Optical rotation was determined on a 341 polarimeter (Perkin Elmer, Inc.). UV spectrum was recorded with a U-2900 spectrophotometer (Hitachi). IR spectrum was obtained using a Nicolet^*6700 FT-IR spectrometer (Thermo Scientific). ¹H, ¹³C, and 2D NMR spectra were recorded on Bruker 700 spectrometer with tetramethylsilane (TMS) as the internal standard. Low-resolution mass spectrometric data were determined using an amaZon SL ion trap mass spectrometer. High-resolution electrospray ionization mass spectrometric (HRESIMS) data were measured on a MaXis 4G UHR-TOFMS spectrometer (Bruker Daltonics Inc.). Column chromatography (CC) was performed with silica gel (100–200 mesh, Jiangyou Silica Gel Development, Inc., Yantai, P. R. China). Thin layer chromatography (TLC, 0.1–0.2 or 0.3–0.4 mm) was conducted with precoated silica gel GF254 (10–40 nm, Yantai) glass plates. Preparative TLC (pTLC) was conducted with precoated glass plates (silica gel GF254, 10–40 nm). Sephadex LH-20 (40–70 μm; Amersham Pharmacia Biotech AB, Uppsala, Sweden), and YMC^*gel ODS-A (12 nm S-50 μm; Japan). MCI gel CHP-20P (75–150 μm, Mitsubishi Chemical Corp., Tokyo, Japan). Medium pressure liquid chromatography (MPLC) was performed on automatic flash chromatography (Cheetahtmmp 200, Bonna-Agela Technologies Co., Ltd.) with the monitoring wavelength at 220 nm and the collecting wavelength at 254 nm. Semi-preparative HPLC was carried out on a Hitachi-L2130 HPLC (equipped with a Hitachi L-2455 diode array detector) using a Phenomenex Luna C18 column (250 × 10 mm, 5 μm). Analytical HPLC was performed on an Agilent 1260 Infinity series instrument (equipped with a quaternary pump, a vacuum degasser, an autosampler, a thermostatic column compartment, and a DAD detector) using an Agilent ZORBAX SB-C18 column (150 × 4.6 mm, 5 μm) under the following program: solvent system (solvent A, 10% acetonitrile in water supplementing with 0.08% formic acid; solvent B, 90% acetonitrile in water); 5% B to 80% B (linear gradient, 0–20 min), 80% B to 100% B (20–21 min),100% B (isocratic elution, 21–24 min), 100% B to 5% B (24–25 min), 5% B (isocratic elution, 25–30 min) with the monitoring wavelength at 400 nm. Small scale production process was performed in a compact sterilizable-in-place fermentation system BioFlo 510 fermentator (Eppendorf AG, German).

The recombinant strain S. albus J1074/pCSG5033 has been previously described (Huang et al., 2018).

The recombinant strain S. albus J1074/pCSG5033 was cultured in the seed medium (3% tryptic soya broth, pH 7.0) and carried on a rotary shaker (200 rpm) at 28°C for 1–3 days. A 20 L scale fermentation was performed by inoculating 1.5 L of the seed culture into a 40 L fermentator (stirring rate:120–230 rpm; dissolved oxygen: 30%; ventilation volume: 15–20 L min⁻¹; pH: 7.2–7.4; pressure: 3 psi; fermentation temperature: 28°C) containing 20 L of the production medium (0.1% peptone fish, 1% starch soluble, 0.6% corn powder, 0.2% bacterial peptone, 0.5% glycerol, 0.2% CaCO₃, 3% sea salt), and cultured at 28°C for 3–5 days. A total of 20 L fermentation cultures were harvested and centrifuged to supernatants and mycelium. The supernatants and mycelium were extracted 3 times with equal volume of butanone and acetone, respectively. And then both were evaporated to dryness under reduced pressure and combined to obtain the crude extract (10.0 g).

The crude extract was subjected to normal phase silica gel (100–200 mesh) column chromatography eluted with a gradient solvent system of chloroform/methanol (from 100:0 to 0:100, v/v) to give four fractions (Fr.1–Fr.4) on the basis of initial assessment by thin-layer chromatography (TLC). Fr.1 was subjected to Sephadex LH-20 column chromatography, eluting with CHCl₃/MeOH (1:1) to afford four subfractions (Fr.1.L1–Fr.1.L4). Subfraction Fr.1.L3 was further purified using C18 reversed phase MPLC [40 × 2.5 cm ID, eluting with a linear gradient of H₂O/CH₃CN (100:0 → 0:100, 15 mL min⁻¹, 200 min)], Sephadex LH-20 column chromatography, and reversed-phase semi-preparative HPLC (H₂O/CH₃CN) successively, to yield 3 (7.0 mg), 4 (1.4 mg), 5 (9.6 mg), and 6 (7.0 mg). Subfraction Fr.2 was further separated with MCI gel column (CH₃CN/H₂O, from 0:100 to 100:0) to give five fractions Fr.2.M1-Fr.2.M5. SEK43F (2, 4.0 mg) was obtained from Fr.2.M3 by Sephadex LH-20 column chromatography and pTLC (petroleum ether/acetone 50:50), successively. FST C (1, 56.0 mg) was obtained from subfraction Fr.4.

Antimicrobial activities were measured against seven indicator strains, Staphylococcus aureus ATCC 29213, Enterococcus faecalis ATCC 29212, Escherichia coli ATCC 25922, Acinetobacter baumannii ATCC 19606, Bacillus subtilis SCSIO BS01, Micrococcus luteus SCSIO ML01, and methicillin resistant S. aureus ATCC 43300, by previously described broth microdilution method (Yang et al., 2015). MCF7 (human breast adenocarcinoma cell line), NCI-H460 (human non-small cell lung cancer cell line), HepG2 (Human hepatocellular liver carcinoma cell line), and SF268 (human glioma cell line) were kindly used for cytotoxic activities assay in vitro according to the previously established protocols (Mosmann, 1983).

Upon a large scale fermentation of S. albus J1074/pCSG5033 (containing the intact fls gene cluster (Yang et al., 2015; Supplementary Figure 1) in a 40-L fermentator and the subsequent isolation with various chromatographic methods, FST C (1) was obtained as the major product, along with several new dimeric FSTs derived from non-enzymatic reactions (Huang et al., 2018). In addition, two new decaketide derivatives SEK43F (2) and fluoquinone (3), two known non-aketide-derived anthraquinones 2-acetylchrysophanol (4) (Abdelfattah, 2009) and 4-acetylchrysophanol (5) (Shaaban et al., 2007), and a previously synthesized compound 3,3′,4,4′,5,5′-hexamethyl-2,2′-dipyrrolylmethene (6) (Figure 1), were also isolated (Guseva et al., 2008; Lund and Thompson, 2014).

SEK43F (2) was isolated as a yellow-green solid. The molecular formula of 2 was established to be C₂₈H₂₅NO₇ (m/z 486.1557 [M - H]⁻, calcd for 486.1558) by high-resolution electrospray ionization mass spectroscopy (HRESIMS), suggesting 17 degrees of unsaturation. The NMR data of 2 (Table 1, Supplementary Figure 2) revealed the presence of four methyls, one methylene, seven sp² methines, and 16 sp² quaternary carbons. HMBC correlations from H₃-7′ to C2′/C3′/C4′, H₃-8′ to C3′/C4′/C5′, H₃-9′ to C4′/C5′, and NH-6′ to C3′/C4′/C5′ revealed the presence of a 1,2,3,4-tetrasubstituted pyrrole moiety (unit A, Figure 2). The remaining NMR data of 2 (unit B, Figure 2) were highly similar to those for SEK43 (Meurer et al., 1997), including a characteristic 1,2,3-trisubstituted phenyl fragment (δ_H 6.82, 1H, d, J = 8.8 Hz; 7.24, 1H, dd, J = 8.2, 8.8 Hz; δ_H 6.83, 1H, d, J = 8.2 Hz; Table 1) and meta couplings spin system of H16 (δ_H 6.11, 1H, d, J = 2.3 Hz) and H18 (δ_H 6.07, 1H, d, J = 2.3 Hz) (Figure 2, Table 1). In addition, presence of the β-oxo-δ-lactone moiety was supported by HMBC correlations from H4 (δ_H 5.59) to C2/C3/C5, and from H1′ (δ_H 7.84) to C1/C2/C3 (Figure 2, Supplementary Figure 2). Finally, the two units A and B were linked by C2–C1′, which was supported by HMBC correlations (Figure 2, Supplementary Figure 2) from H1′ (δ_H 7.84) to C3′ and C1/C2/C3. However, the Z/E configuration of Δ^2,1′ in 2 couldn't be determined solely by NMR data. Similar phenomena were observed for hybrubins (Zhao et al., 2016).

The molecular formula of fluoquinone (3) was determined to be C₂₀H₁₄O₇ (m/z 365.0672 [M - H]⁻, calcd for 365.0667) by HRESIMS, requiring 14 degrees of unsaturation. The ¹H, ¹³C, and HSQC NMR data for 3 displayed a singlet methyl, two sp³ methylenes, one sp³ methine, four sp² methines, and 12 sp² quaternary carbons (Table 1, Supplementary Figure 3). The ¹H-¹H COSY spectrum of 3 (Figure 2, Supplementary Figure 3, Table 1) showed a characteristic aromatic ABC spin system (δ_H 7.70/7.79/7.40) and a CH₂-CH(O)-CH₂ fragment (δ_H 3.31/4.12/5.02 and 3.09/3.13/5.02). HMBC correlations (Figure 2, Supplementary Figure 3) from H5/H7 to C6, from H8 to C10, together with four-bond HMBC correlations (Figure 2, Supplementary Figure 3) from H5/H7/H9 to C11, and from H5 to C12 supported the presence of a 1,8-dihydroxyanthraquinone moiety, highly similar to that of the fungal metabolite dermolactone (Gill and Gimenez, 1990; Cotterill et al., 1995). Comparison of NMR data revealed that 3 was different from dermolactone by the absence of a methoxy group at C8 (COSY correlations between H7/H8/H9 in 3) and the presence of an additional acetyl group at C13 (HMBC correlations from H₃-15 to C14/C13 in 3). Thus, the planar structure of 3 was determined as shown in Figure 1. The optical rotation value of 3 ([α]25 D + 1.6, CHCl₃; c 0.22) was quite different from that of the synthesized (S)-(+)-dermolactone ([α]22 D + 169.3, c 0.21) (Cotterill et al., 1995), but comparable to that of the naturally isolated dermolactone ([α]22 D + 22.0, CHCl₃; c 0.07) (Gill and Gimenez, 1990). Natural dermolactone had been confirmed as a mixture of the (S)-(+)- and (R)-(-)-enantiomers (Cotterill et al., 1995). It was thus deduced that fluoquinone (3) should also contain a pair of racemic (S)-(+)- and (R)-(-)-enantiomers (almost in a ration of 1:1) due to its negligible optical rotation. Analysis of the ¹H NMR and ¹³C NMR data of compounds 4–6 revealed that they were identical to 2-acetylchrysophanol (4) (Supplementary Figure 4, Supplementary Table 1) (Abdelfattah, 2009), 4-acetylchrysophanol (5) (Supplementary Figure 5, Supplementary Table 1) (Shaaban et al., 2007), and 3,3′,4,4′,5,5′-hexamethyl-2,2′-dipyrrolylmethene (6) (Supplementary Figure 6, Supplementary Table 1) (Guseva et al., 2008; Lund and Thompson, 2014).

Compounds 1–6 exhibited negligible antibacterial activity against seven indicator strains: S. aureus ATCC 29213, E. faecalis ATCC 29212, E. coli ATCC 25922, A. baumannii ATCC 19606, B. subtilis SCSIO BS01, Micrococcus Luteus SCSIO ML01, and methicillin resistant S. aureus ATCC 43300. Compound 2 showed weak cytotoxic activities against four human cancer cell lines (SF-268, MCF-7, NCI-H460, and HepG-2) with IC₅₀ values at the range of 36–57 μM (Table 2).

The heterologous expression of the intact fls-gene cluster from M. rosaria SCSIO N160 in S. albus J1074 afforded a set of APKs 1–5. These aromatic compounds should be derived from the fls gene cluster-encoded enzymes because that the host S. albus J1074 is known for the absence of type II PKS gene clusters (Olano et al., 2014). Previously, a biosynthetic pathway of fluostatins has been proposed (Yang et al., 2015). Briefly, the type II PKS enzymes FlsA (the KS_α unit), FlsB (the KS_β unit) and FlsC (the dissociated ACP) assemble one unit of acetyl-CoA and 9 units of malonyl-CoA to generate a linear decaketide, which undergoes the FlsE-catalyzed C9 ketoreduction to give M1 (Figure 3A). Subsequently, the aromatic cyclase FlsD catalyzes the “norm” C7/C12 first-ring cyclization to convert M1 to M2. The downstream enzymes FlsI (a second cyclase analogous to JadI in jadomycin biosynthesis; Kulowski et al., 1999) and FlsO3 (oxygenases) converted M2 to prejadomycin (cyclization pattern I: C7/C12, C5/C14, C4/C17, C2/C19), an established precursor to undergo oxidative modifications to produce diverse FST derivatives (e.g., 1, Figure 3A) (Huang et al., 2018). Alternatively, M2 could undergo spontaneous cyclization (pattern II: C7/C12, C14/C19, C1-OH/C5) to produce SEK43 (a shunt product often encountered in bacterial type II PKS pathways) (McDaniel et al., 1995; Hertweck et al., 2007), which is a key precursor leading to SEK43F (2). The tri-methylated pyrrole unit in 2, which is also present in 6, should be biosynthesized by the host strain, because that the production of 6 was also observed in S. albus J1074 harboring the void vector pSET152 (Supplementary Figure 1). We propose that 6 is produced by the coupling of two putative subunits 6a and 6b (Figure 3B), through a condensation reaction similar to RedH catalysis in the biosynthesis of undecylprodigiosin and hybrubins (Williamson et al., 2005; Zhao et al., 2016). The formation of 2 should be produced from a “pathway crosstalk” event via a condensation reaction by connecting C2 of SEK43 (generated by the heterologous fls type II PKS) and the aldehyde group of 6b (produced by the host), either spontaneously or catalyzed by a host-derived RedH-like enzyme (Figures 3A,B) (Zhao et al., 2016).

It is intriguing to discover the anthraquinone derivative 3 in a bacterium. It was speculated that fundamentally different cyclization strategies were employed by bacterial and fungal type II PKSs to biosynthesize structurally similar APKs with fused-rings (Thomas, 2001, 2016). The first-ring cyclization modes were classified into F-mode for fungi to contain two intact acetate units, and S-mode for bacteria to contain three intact acetate units in the first cyclized phenol ring (Supplementary Figure 7) (Thomas, 2001, 2016; Zhou et al., 2010). This fundamental difference was reinforced by the discovery of three divergent first-ring cyclization strategies to converge in generating the same metabolite chrysophanol, namely F-mode (C6/C11) in fungi and plants (Bringmann et al., 2006), and S-mode (C5/C10) or S′-mode (C7/C12) in bacteria (Bringmann et al., 2006, 2009). Several bacterial fused-ring APKs were found as exceptions to be putatively derived from fungal F-mode cyclizations, such as TW93f and Tw93g (Shen et al., 1999), piloquinone (Polonsky and Lederer, 1963), murayaquinone (Gould et al., 1997), and haloquinone (Krone et al., 1981). However, it was recently discussed that evidence should be provided for these exceptions to either validate the structures (TW93f and Tw93g), or confirm the folding pattern by detailed labeling studies (Thomas, 2016). In contrast, the presence of the six-membered lactone ring in fluoquinone (3) strongly suggested that 3 should be derived from the first-ring cyclization via C6/C11 (cyclization pattern III: C6/C11, C4/C13, C2/C15, C17-OH/C1, Figure 3C). The fungal natural product dermolactone, highly similar to 3, was confirmed to be derived from the F-mode of first ring cyclization (Figure 3D) (Gill and Gimenez, 1990). Thus, fluoquinone (3) represents an exception as a bacterial F-mode (fungal mode) to the F and S biosynthetic classifications of fused ring APKs (Thomas, 2001, 2016).

Compounds 4 and 5 were proposed to be derived from a common non-aketide, which was putatively assembled by FlsABC from one unit of acetyl-CoA and eight units of malonyl-CoA (Figures 3E,F). The discovery of 4 and 5 indicated that the CLF enzyme FlsB should exhibit a loose control in the chain length, allowing the formation of both decaketides (Figures 3A,C) and non-aketides (Figure 3E). The promiscuous chain length control has been reported for native type II PKS enzymes (McDaniel et al., 1993; Meurer et al., 1997; Shen et al., 1999), or can be altered by enzyme engineering (Burson and Khosla, 2000; Tang et al., 2003). Subsequently, the non-aketide was divergent in the third ring cyclization to provide two products 4 (cyclization pattern IV: C7/C12, C5/C14, C3/C16) and 5 (cyclization pattern V: C7/C12, C5/C14, C4/C17).

Although there are different cyclization patterns involved in the formation of 1–5, the cyclization mechanisms are quite common and well-known in type II PKS biosynthesis: (i) enzyme-catalyzed or spontaneous cyclization via aldol condensation (Shen et al., 1999; Zhou et al., 2010), e.g., C7/C12, C6/C11, C5/C14, C4/C17, C4/C13, C3/C16, C2/C15, C2/C19, and C14/C19 (Figure 3); (ii) spontaneous cyclization via lactonization (Meurer et al., 1997), e.g., C1-OH/C5, C17-OH/C1 (Figure 3).

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.