Introduction

Type II polyketide synthases (PKSs) catalyze the biosynthesis of the carbon skeletons of numerous polyfunctional aromatic natural products from the actinomycetes, including the clinically important antibiotic oxytetracycline and the anticancer agent doxorubicin.1, 2 Recent analysis has demonstrated considerable functional modularity within these multienzyme assemblies, which in turn can be exploited to alter priming units and change polyketide backbone length.3, 4 At the same time, the substrate flexibility of late-stage tailoring enzymes in these natural product pathways can also be exploited to create designer molecules.4, 5 Here, we have harnessed each of these modular elements to reconstruct a chimeric pathway to the antiparasitic antibiotic frenolicin B (1b), as well as targeted analogs in a heterologous host. Biological evaluation of our analogs against Toxoplasma gondii has revealed the strong sensitivity of this family of antibiotics to the identity of the C-15 substituent, thereby highlighting a new direction for future antiparasitic drug design.

The benzoisochromanequinone (BIQ) antibiotic frenolicin B, produced by the actinomycete Streptomyces roseofulvus, has been extensively investigated as an anticoccidial agent owing to its potent activity against Eimeria tenella.6, 7, 8, 9 Nonetheless, low fermentation titers10 and challenging synthetic procedures11, 12 appear to have limited further development of this promising anti-infective lead substance. We sought to address both of these problems by reconstructing a biosynthetic pathway to frenolicin B in the heterologous host Streptomyces coelicolor CH999, from which the entire actinorhodin biosynthetic gene cluster has been deleted.13

Materials and methods

General

S. coelicolor CH999/pBOOST*,14 which lacks the complete actinorhodin (act) gene cluster, was used as the host for production of polyketides. The plasmid pBOOST* has been shown to cointegrate with vectors containing the SCP2* origin of replication, thereby resulting in higher copy numbers and correspondingly improved antibiotic titers.14 The transformation of shuttle vectors bearing BIQ pathway genes into S. coelicolor CH999/pBOOST* were performed following standard procedure.15 Deuterated solvents for NMR experiments were purchased from Cambridge Isotope Laboratories (Andover, MA, USA) and all other solvents were purchased from Fisher Scientific (Pittsburgh, PA, USA) at the highest available grade. 1H and 13C NMR spectra of purified polyketide products were recorded on Varian Inova 500 or 600 MHz instrument (Varian, Walnut Creek, CA, USA) in CD3OD or DMSO-d6. The 1H NMR spectra were referenced to the solvent peak at 3.31 p.p.m. for CD3OD or 2.50 p.p.m. for DMSO-d6. The 13C NMR spectra were referenced to solvent at 49.0 p.p.m. for CD3OD or 39.52 for DMSO-d6. Single-bond 1H–13C, multiple-bond 1H–1H and 1H–13C connectivity was determined by HSQC, COSY and HMBC, respectively on Varian Inova-600 NMR instrument. Mass spectra were obtained by ESI at the Vincent Coates Foundation Mass Spectrometry Laboratory at Stanford University.

Construction of shuttle vectors

All cloning steps were performed in either E. coli XL1-Blue (Stratagene, La Jolla, CA, USA) or E. coli DH5α (MCLab, South San Francisco, CA, USA). Plasmids used for cloning included pUC18 (New England Biolabs, Ipswich, MA, USA) and pET28b (Novagen, now EMD Bioscience, Madison, WI, USA). Expression vectors were constructed using standard molecular biology techniques. See Supplementary Information for a detailed description of the primers and construction.

Production, isolation and characterization of polyketide products

Each transformed strain was grown on R5 agar plates containing 50 mg l−1 thiostrepton and 100 mg l−1 of apramycin at 30 °C for 7 days,15 after which metabolites were extracted with either 100% EtOAc or EtOAc/MeOH/acetic acid (89:10:1). This crude material was purified by a variety of methods including preparative HPLC and NaHCO3 extraction. Structural determination was performed using a variety of 2D heteronuclear NMR experiments and LC/MS. See Supplementary Information for full details.

T. gondii growth inhibition

Tachyzoites of T. gondii strain RHΔhxgprt YFP16 were harvested from human foreskin fibroblast (HFF) monolayers grown in 25 cm2 T-flasks. These parasites were then inoculated into a 96-well culture plate containing a confluent layer of HFF cells with ∼8 × 103 tachyzoites/0.32-cm2 well. Parasites were pre-incubated with HFF cells for 6 h at 37 °C, 5% CO2. After pre-incubation, the media was exchanged for fresh DMEM complete ((Dulbecco's modified eagle medium) supplemented with 1% PS (penicillin–streptomycin) and 10% fetal bovine serum) containing a given concentration of antiparasitic agent in DMSO. Cultures were then incubated for 72 h at 37 °C, in 5% CO2. Parasite growth inhibition was monitored in a SpectraMax Gemini EM fluorescent plate reader (Sunnyvale, CA, USA) at 510 nm excitation and 540 nm emission. ED50s were determined using GraphPad Prism (GraphPad Software, La Jolla, CA, USA).

Cytotoxicity assay

HFF cell cytotoxicity was determined using a MTS (3-(4,5-dimethylthiazol-2-yl)-5-(3-carboxylmethoxylphenyl)-2H-tetrazolium) assay (CellTiter 96 Aqueous One Solution Assay; Promega, Madison, WI, USA). HFF cells (2 × 103 cells per well) were grown in a 96-well plate until confluence (∼48 h) in phenol red-free DMEM complete at 37 °C, 5% CO2. Varying concentrations of antiparasitic agent were then added to each well and incubated for 48 h at which point the medium was replaced with 100 μl phenol red-free DMEM and 20 μl MTS. After an additional 1.5 h of incubation at 37 °C, and the absorbance at 490 nm was calculated and the ED50 was determined using GraphPad Prism.

Results and Discussion

Determining the tailoring enzymes necessary for BIQ core formation

Before embarking on the reconstruction of the frenolicin pathway, we first attempted to complete the biosynthesis of the simplest BIQ, deoxykalafungin 2a (Scheme 1),17 which does not require an initiation module and is an intermediate in the actinorhodin (4) pathway.18 Previous studies have shown that introduction of the plasmid pRM5 (which encodes act KS/CLF, ACP, C-9 KR, ARO and CYC genes; for details, see Scheme 1) into CH999 leads to the production of 3,8-dihydroxy-1-methylanthraquinone-2-carboxylic acid (6a) and aloesaponarin II (7a), as summarized in Table 1.13, 14 The next step in BIQ production is stereoselective reduction of the C-3 ketone into an alcohol capable of cyclizing the third ring of the scaffold (Scheme 1 in red). To accomplish this, we cloned the C-3 ketoreductase gene, encoded by actVI-ORF1, onto pRM5, yielding the plasmid pJF6 (Table 1). As predicted, S. coelicolor CH999/pJF6/pBOOST* yielded 5a after cyclization and loss of water (Supplementary Figure S16).5, 19

Table 1 Plasmid constructs and resulting polyketide products
Full size table

The next step in the pathway is enoyl reduction of 5a followed by oxygenation. Mutants of the actinorhodin-producing strain S. coelicolor A3(2), in which the enoyl reductase gene actVI-ORF2 has been inactivated, produce no actinorhodin, and mutants of the secondary enoyl reductase gene actVI-ORF4 produce only 18% of the actinorhodin of the native strain.20 We therefore constructed two derivatives of pJF6, pCR66 and pCR67, harboring actVI-ORF2 and actVI-ORF2/actVI-ORF4, respectively (Table 1). The product profiles of S. coelicolor harboring either plasmid was identical, and included 2a as the next isolable intermediate (Supplementary Figure S16). This finding has two interesting implications for BIQ biosynthesis. First, overexpression of actVI-ORF2 in a pRM5-derived vector obviates the need for a supplementary enoyl reductase in a heterologous system. Second, and more interestingly, the inferred product of the enoyl reductase reaction (Scheme 1 bottom left) was not observed, nor was there a large buildup of 5a. It therefore appears that, although the act gene cluster harbors dedicated oxygenase genes,21, 22 the S. coelicolor genome either has a nonspecific oxygenase capable of catalyzing the same reaction, or that quinone formation proceeds spontaneously in air.

Production of frenolicin B

Having identified the genes required to produce 2a in S. coelicolor CH999/pBOOST*, we turned our attention to frenolicin B and analogs thereof. Whereas the minimal PKS responsible for frenolicin production has been previously defined,23 the initiation module and downstream tailoring enzymes have not yet been characterized. Intriguingly, homologs of the actVI-ORF1 and actVI-ORF2 genes are not known to exist in the frenolicin (frn) gene cluster.24 In fact, constructs harboring all putative biosynthetic genes from the identified frn gene cluster failed to produce frenolicin or any related BIQ antibiotic (data not shown). In the face of insufficient genetic data, we hypothesized that the tailoring enzymes identified in production of 2a could be functionally repurposed for the biosynthesis of longer chain analogs. To accommodate the longer alkyl priming chain, we replaced the octaketide act KS-CLF genes with the mixed octaketide/nonaketide frn KS-CLF genes, and also included the genes thought to encode the initiation module of the frn PKS.

The frn gene cluster harbors homologs of the genes encoding the well-characterized initiation module of the R1128 cluster.25 Specifically, frnI is a homolog of zhuH, frnK is a homolog of zhuC and frnJ is a homolog of zhuN. These genes encode a homodimeric KS responsible for chain initiation, an acyl-ACP thioesterase and a dedicated ACP for the initiation module, respectively. Co-expression of these three frn genes with the genes encoding the frn KS-CLF, the frnN ACP the act C-9 KR, the act ARO and the act CYC, as well as the actVI-ORF1 and actVI-ORF2 genes (pJF35; Table 1) yielded a mixture of butyryl-primed nonaketide derivatives that arise from off-path cyclization reactions (Supplementary Figure S17). In addition, very small quantities of 2b were also observed. Aside from 6b, the two other major shunt products were propyl analogs of the previously identified compounds dehydromutactin and SEK34.26, 27, 28 They form either due to the inability of the act ARO to aromatize the first ring or the inability of the act CYC to form the second six-membered ring (Scheme 1). Notwithstanding the relative abundance of these shunt products, the ability of actVI-ORF1 and actVI-ORF2 to replace the unidentified components of the frenolicin pathway underscores the potential for rational design of a biosynthetic scheme using a diverse PKS toolbox in the face of incomplete genetic information.

Analogs of frenolicin B

With a route in hand to the bioactive natural product frenolicin B, we turned our attention to the rational design of biosynthetic analogs. Earlier work with E. tenella had shown that extension of the C-15 methyl substituent of 1a to a propyl substituent (1b) led to improved antiparasitic activity.6 We therefore reasoned that analogs 1c and 1d could be useful to investigate the steric effect of the C-15 substituent on biological activity. As chemical synthesis of such analogs is not straightforward, we replaced the genes encoding the frn initiation module and ACP with homologs from the R1128 cluster, which incorporates bulkier primer units into the polyketide backbone.29 At the same time, to compensate for the anticipated increase in overall chain length of the polyketide skeleton, we also replaced the frn KS-CLF genes with homologs from the decaketide tetracenomycin (tcm) PKS.30 In combination with the act KR, act ARO and act CYC, this hybrid bimodular PKS gene cluster yielded a mixture of compounds, including 6c, 6d, 7c and 7d (pYT127; Table 1).23

Further inclusion of the actVI-ORF1 and actVI-ORF2 genes yielded S. coelicolor CH999/pJF9/pBOOST*, which produced a spectrum of products including the desired 2c and 2d (Table 1 and Supplementary Figure S18). Although this strain produced a wide range of uncharacterized polyketide products, the ratio of the desired products 2c and 2d to the shunt products 6c and 6d was considerably higher in this recombinant strain than in S. coelicolor CH999/pJF35/pBOOST*. The lack of large quantities of butyl- and isopropyl-primed analogs of 8 and 9 indicate that act ARO and CYC interface with the tcm min PKS better than with the frn PKS. In addition, this suggests that substrate incompatibility arising from increased alkyl priming unit length is unlikely to be the cause of the large amounts of 8 and 9 observed in S. coelicolor CH999/pJF35/pBOOST*. These rationally designed analogs highlight the versatility of our approach to polyketide construction, and enabled evaluation of their antiparasitic activities.

Biological evaluation against T. gondii

Previous studies with E. tenella showed that the lactone forms of the BIQs 1a and 1b were more potent than the free-acid forms (2a and 2b).6 We therefore converted our isolated analogs 2a–d into their corresponding lactones 1a–d by previously reported methods.31 The structures of the novel compounds 1c and 1d were confirmed by NMR and mass spectrometric analysis (Supplementary Tables S3, S5 and S6). The compounds were assayed against T. gondii, the causative agent of toxoplasmosis, along with the known antiparasitic agent pyrimethamine as a reference.32 A slight modification of a previously reported yellow fluorescence protein reporter assay was used for this purpose16 (Supplementary Figures S1 and S2). As summarized in Table 2, within the C1–C4 range, the alkyl substituent has little effect on antiparasitic activity against T. gondii, but the longer branched substituent leads to a notable decrease in activity. Thus, the benefit of the longer substituent against E. tenella does not hold in our assays, and could be due to improved pharmacokinetics or differential species specificity. To assess their relative tolerability by mammalian cells, we also measured the cytotoxicity of 1a–d against HFF-1 cells using a MTS-based assay (Supplementary Figure S3). The therapeutic indices of 1a–d, reported in Table 2, appear to be comparable.

Table 2 Bioactivity of compounds 1a–d and pyrimethamine
Full size table

Conclusion

In summary, notwithstanding incomplete genetic data for the natural product biosynthetic pathway, we have successfully engineered a chimeric, heterologous system for the production of frenolicin B. Furthermore, we have exploited this heterologous system to produce novel frenolicin analogs that are difficult to access by alternative methods. Our results showcase the utility of a modular approach to the creation of medicinally relevant polyketide products of type II PKSs. Last but not least, we have shown that the new compounds have antiparasitic activity against T. gondii in vitro. Further modification of the biosynthetic system presented here could yield additional improvements in BIQ productivity as well as additional analogs of frenolicin B.

scheme 1

A biosynthetic pathway to the benzoisochromanequinones. The presumed biosynthetic route to 1a–d is shown along with isolable intermediates and shunt products. The natural product frenolicin A (3) produced by S. roseofulvus, and the natural product actinorhodin (4) produced by S. coelicolor, are also shown as forming from their corresponding carboxylic acids, 2. For details, see text. ARO, 1st ring aromatase encoded by actVII; CYC, 2nd ring cyclase encoded by actIV; C3 KR, C-3 ketoreductase encoded by actVI-ORF1; C9 KR, C-9 ketoreductase encoded by actIII; ER, enoyl reductase encoded by actVI-ORF2; IM, initiation module (ketosynthase III, acyl carrier protein, acyl-ACP thioesterase); min PKS, minimal polyketide synthase (ketosynthase/chain length factor (KS/CLF), acyl carrier protein (ACP), malonyl-CoA:ACP transacylase (MAT)); OXY, oxygenase, encoded by an unidentified enzyme in the S. coelicolor genome.