Identification of the gene cluster for the dithiolopyrrolone antibiotic holomycin in Streptomyces clavuligerus

Abstract

Streptomyces clavuligerus, an industrially important producer of clavulanate as well as cephem antibiotics, also produces the N-acylated dithiolopyrrolone antibiotic holomycin, a reported inhibitor of RNA synthesis. The genome sequence of S. clavuligerus ATCC 27064 was examined for a potential biosynthetic gene cluster, assuming that holomycin arises from some derivative of an l-Cys-l-Cys dipeptide that has undergone eight-electron oxidation, fused five-five ring formation, and decarboxylation. ORFs 3483–3492 comprise a candidate cluster, with a predicted acyltransferase, a stand-alone nonribosomal peptide synthetase (NRPS) module, and four flavin-dependent oxidoreductases. Deletions of ORF3488, the NRPS module, and ORF3489, a phosphopantothenoylcysteine decarboxylase homolog, abolished holomycin production both in wild type and in a holomycin-overproducing mutant. Heterologous expression and purification of ORF3488 allowed demonstration of l-Cys-AMP formation and subsequent covalent tethering of Cys to the phosphopantetheinyl arm of the thiolation domain of this NRPS protein. Purified ORF3483 shows acyltransferase activity, converting holothin to holomycin and longer acylated homologs as the last step in antibiotic assembly.

The N-acylated dithiolopyrrolone antibiotic thiolutin was first isolated from a streptomyces strain in the late 1940s, and its structure was solved in 1955 (1). The N-desmethyl congener, holomycin (Fig. 1), is found in Streptomyces clavuligerus (2), named for its importance in the industrial production of clavulanate, the β-lactamase inhibitor that is part of the two component antibiotic augmentin (3). Other members of the dithiolopyrrolone class include aureothricin (1), the n-propionyl homolog of thiolutin, and a variety of xenorhabdins (4) with a longer acyl chain replacing the acetyl moiety of thiolutin and holomycin. Thiolutin and holomycin have been reported to display bacteriostatic activity against both Gram-positive and Gram-negative bacteria (5, 6). Although the exact mechanism of action has not been determined, they appear to inhibit RNA synthesis and are active against rifamycin-resistant strains (7, 8). It has been suggested that holomycin and congeners could be prodrugs, activated by reduction of the disulfide bridge, but no reports on this hypothesis have appeared. The holomycin family of compounds also exhibits cytotoxicity to mammalian cells (9) although the target(s) have not been identified. Additionally, the dithiolopyrrolone framework is found in thiomarinols (10) as the amine component in amide linkage to an 8-hydroxy-octanoyl chain provided by a pseudomonic acid variant, itself an antibiotic.

Fig. 1.

Structures of dithiolopyrrolone-containing compounds holothin, holomycin, n-propionyl holothin, and thiomarinol A.

The biosynthetic origin of these dithiolopyrrolones has not been uncovered to date. The holomycin producer S. clavuligerus ATCC 27064 was chosen for genome sequencing at the Broad Institute, because it is a known producer of both clavulanate and cephem β-lactam antibiotics from adjacent gene clusters (11). Based on the genome sequence, we utilized bioinformatic genome mining, targeted gene disruption, heterologous protein expression, and enzyme activity assays to identify a holomycin biosynthetic gene cluster. It has been reported that disruption mutant in ORF10 in the clavulanate biosynthetic cluster had differential effects on clavulanate or cephem biosynthesis, suggesting some rerouting of flux of intermediates (12). This mutant and two other mutants in ORF12 and ORF15 of the clavulanate pathway dramatically overproduced a yellow compound, identified as holomycin (13), though the underlying molecular mechanism for the overproduction is currently unclear. These results not only validated the presence of biosynthetic genes for holomycin in S. clavuligerus ATCC 27064, but also indicated that these holomycin-overproducing mutants would be useful for targeting putative holomycin biosynthetic genes for disruption.

Results

Bioinformatic Analysis.

Given the unusual heterobicyclic dithiolopyrrolone scaffold of holomycin, it might be difficult to recognize constituent ORFs even if the biosynthetic genes were clustered. We intuited a potential biosynthetic pathway, by working backward: The last step is likely to be conversion of the free amino group of desacetyl holomycin, also known as holothin (Fig. 1), to the N-acetylated end product holomycin by an acyltransferase, because holothin can be converted to holomycin in vitro by crude extracts of S. clavuligerus (13). The bicyclic scaffold of holothin contains five carbons, two nitrogens, and two adjacent sulfur atoms in a disulfide linkage and could arise from two cysteine building blocks after losing one carbon as CO₂. Subsequently and/or concomitantly, a net loss of eight electrons, perhaps in four 2-electron steps, could generate the dithiolopyrrolone scaffold. Thus, we anticipated the involvement of multiple oxidoreductases, an amino acid decarboxylase, a holothin acetyltransferase, and some enzymatic machinery that catalyzes the formation of a Cys-Cys peptide bond. Among the routes to amide bond formation could be one or more nonribosomal peptide synthetase (NRPS) modules.

With this reasoning as context for bioinformatic mining of the S. clavuligerus ATCC 27064 genome, we identified a candidate gene cluster (Fig. 2A) on a supercontig (accession no. NZ_DS570652), encompassing ORFs 3478–3496, with particular focus on six ORFs: 3483, 3485, 3487, 3488, 3489, and 3492. ORF3483 is a putative acetyltransferase and could be involved in the last step, acetylation of holothin to holomycin. The four ORFs 3485, 3487, 3489, and 3492 are all predicted to be flavin-dependent oxidoreductases, one of them (3489) to be in the subfamily of phosphopantothenoylcysteine decarboxylase, which would release CO₂ from an amino acid-derived intermediate (14). ORF3488 presents the characteristics of a single module NRPS with three domains: cyclization (Cy), adenylation (A), and thiolation (T) in the canonical order Cy-A-T (15).

Fig. 2.

(A) Proposed holomycin biosynthetic gene cluster with gene annotations. (B) Proposed biosynthetic pathway for holomycin.

Gene Disruption Strategy.

Based on the bioinformatic analysis, the NRPS module ORF3488 and the putative decarboxylase ORF3489 are likely to catalyze key steps in the biosynthesis of holomycin. Therefore these two genes were selected for replacement with an apramycin-resistance cassette in the wild-type strain of S. clavuligerus to evaluate the effects on holomycin production. Because of the variability of holomycin production in the wild-type strain, in-frame deletions in the holomycin-overproducing strain ORF15::apr were carried out in parallel. Under current growth conditions, the ORF15::apr mutant strain consistently produces a high level of holomycin, such that nearly 30 mg of holomycin was purified from 2 L of ISP4 agar culture, enabling structure validation by mass spectrometry and proton and carbon NMR (SI Appendix, Figs. S1–S3). Purified holomycin also served as a source of the presumed penultimate intermediate holothin via chemical deacetylation (16). A hygromycin-resistance cassette was employed in the gene deletions in the ORF15 mutant strain as the apramycin resistance was already in use. Replacement of ORF3488 and ORF3489 with antibiotic-resistance cassettes in both wild-type S. clavuligerus strain and the ORF15::apr overproducer was achieved via homologous recombination and validated by PCR analysis (SI Appendix, Fig. S4). Such deletions completely abolished holomycin production in both wild-type and holomycin-overproducing strains (Fig. 3), indicating that both ORF3488 and ORF3489 are required for holomycin production.

Fig. 3.

High-resolution ESI-Q-Tof mass spectrometric analysis of holomycin production by ORF3488 and ORF3489 deletional mutants in S. clavuligerus wild type, and ORF15::apr holomycin-overproducing mutant strain. Shown are extracted ion chromatograms for holomycin (214.9949, [M + H]⁺) with 20-ppm mass error tolerance. It is noteworthy to point out that the production level of holomycin in wild type is highly variable. The top trace shown here represents a particular case when wild type produced a comparable amount of holomycin as the overproducing strain.

Purification and Assay of ORF3488.

Given genetic evidence for involvement of ORF3488, the proposed tridomain NRPS module, in holomycin production, it was overproduced and purified from Escherichia coli as an N-terminally His-tagged soluble protein (at ∼20 mg/L yield, SI Appendix, Fig. S5). We anticipated that the predicted A domain of this single NRPS module would activate cysteine as the mixed Cys-AMP anhydride. The reversible formation of such aminoacyl–adenylates is classically detected by amino acid-dependent exchange of radioactivity from ³²PPi into ATP (ATP-[³²P]PPi exchange assay). Based on this assay, the A domain shows robust activity for l-Cys but not the other proteinogenic amino acids, including l-Ser, l-Ala, or l-Met (SI Appendix, Fig. S6). Kinetic characterization with l-Cys indicated a K_m of 1 mM and a k_cat of 98 min^-1 (SI Appendix, Fig. S7). Such robust exchange is consistent with l-Cys as the physiological substrate.

To assess the function of the thiolation domain in the Cy-A-T tridomain, ORF3488 was overproduced in the holo form in E.coli BAP-1 strain (17), where sfp is integrated into the chromosome and coexpressed on induction. This in vivo posttranslationally modified Cy-A-T protein could then load [¹⁴C-l-Cys], as covalent Cys-S-pantetheinyl-ORF3488, in the presence of ATP, Mg²⁺, and tris(2-carboxyethyl)phosphine (Fig. 4A). A compound with a mass corresponding to l-Cys-l-Cys was detected in solution as shown by mass spectrometric analysis (Fig. 4B). When ATP was omitted from the loading assay mixture, no such peak was observed. The presence of l-Cys-l-Cys in solution could be a result of the hydrolysis of that dipeptidyl-ORF3488 intermediate. Alternatively, the excess free l-Cys in solution could lead to thiolytic release of the l-Cys-S-ORF3488 intermediate, and a subsequent nonenzymatic S to N shift would generate the l-Cys-l-Cys dipeptide.

Fig. 4.

l-Cysteine loading by ORF3488. (A) Covalent loading of 1-¹⁴C-l-Cys onto the holo-T-domain of ORF3488 at eight different time points between 1 min and 120 min as shown by autoradiograph of SDS-PAGE gel. (B) Extracted ion chromatograms of Cys–Cys dipeptide (225.0368, [M + H]⁺) with 20-ppm mass error tolerance in the loading assay of l-Cys by ORF3488 in the presence of ATP (Top) and without ATP as negative control (Bottom).

Purification and Assay of ORF3483.

Recombinant ORF3483 (SI Appendix, Fig. S5) was obtained from E. coli, and the predicted acyltransferase activity was assayed toward the proposed penultimate intermediate holothin. Holothin was generated from purified holomycin by microwave-assisted deacetylation. As shown in Fig. 5, addition of purified ORF3483 and acetyl-CoA quantitatively transformed holothin into holomycin. Kinetic analysis (SI Appendix, Fig. S8) confirmed the anticipated Michaelis–Menten kinetics with an apparent K_m of 6–15 μM and k_cat of 80–100 min^-1 for acetyl- and propionyl-CoAs at a 20-μM holothin concentration, agreeing with the fact that both holomycin and n-propionyl holothin are the naturally occurring products from S. clavuligerus (18). Kinetic parameters could not be obtained for holothin, as the increase in holothin concentration led to a decrease in enzyme activity (potentially resulting from substrate or product inhibition). This in vitro enzymatic assay supports the involvement of ORF3483 in the last step of holomycin production by acetylation/acylation of the free amine of the dithiolopyrrolone holothin. An initial survey of longer chain acyl-CoAs revealed that ORF3483 was able to utilize hexanoyl-, octanoyl-, and palmitoyl-CoA as substrates (Fig. 5 and SI Appendix, Fig. S9), albeit much less efficiently. For example, an apparent K_m of 30 μM and apparent k_cat of 0.07 min^-1 was obtained for octanoyl-CoA in the presence of 20 μM holothin (SI Appendix, Fig. S10). These data are consistent with fermentation reports on alternative acyl chain incorporations on the amino group of the holothin or N-methyl holothin scaffold (19, 20) and the detection of longer acyl chain variants as natural products, notably the octanoyl–holomycin framework in thiomarinols (21).

Fig. 5.

Acyltransferase activity of ORF3483. High-resolution ESI-Q-Tof mass spectrometric analysis of acylation of holothin by ORF3483 in the presence of acetyl-CoA (B), hexanoyl-CoA (C), octanoyl-CoA (D), and palmitoyl-CoA (E). A represents the negative control without enzyme. Shown are the extracted ion chromatograms for holothin (A, 172.9843), holomycin (B, 214.9949), hexanoyl–holothin (C, 271.0569), octanoyl–holothin (D, 299.0882), and palmitoyl–holothin (E, 411.2134) with 20-ppm mass error tolerance.

Additional ORFs in the Gene Cluster.

ORF3485, ORF3487, ORF3489, and ORF3492 were similarly expressed as N-His-tagged proteins in E. coli. All but 3485, the putative acyl CoA dehydrogenase, were purified as yellow proteins with absorption maxima at 450 nM, consistent with the predictions as flavin-dependent oxidoreductases. ORF3489 was of particular interest because of its essential role in holomycin production as indicated by deletional mutagenesis. BLAST search revealed homology of ORF3489 to the subfamily of flavoenzymes that carry out β-oxidation and subsequent decarboxylation, occurring both in CoASH biosynthesis and at the C-terminal Cys of several lantibiotics on the way to an aminovinyl–cysteine residue (22). ORF3489 was proposed to encompass comparable oxidoreductase-coupled decarboxylase activity in holomycin biosynthesis, for example, in the oxidation of a Cys–Cys–dipeptide species to an enethiol to set up decarboxylation. We assayed both free l-Cys-l-Cys and N-acetyl-l-Cys-l-Cys for decarboxylation by ORF3489 in the presence of NAD or NADP, but no loss of CO₂ was detected.

Discussion

The dithiolopyrrolone class of antibiotics has been known for over 50 years, and much interest has been shown in this class as antibiotics that target RNA polymerase that became resistant to rifamycins. However, the unusual framework of the dithiolopyrrolones has rendered the bioinformatic search of putative biosynthetic gene clusters in producer microorganisms challenging. After surveying the genome sequence of holomycin-producer S. clavuligerus, we proposed a candidate gene cluster including a single module NRPS module (ORF3488), an acyltransferase, a phosphopantothenoylcysteine decarboxylase (ORF3489), and three other oxidoreductases. Genetic deletion of the ORF3488 and ORF3489 genes abolished detectable holomycin/holothin production, strongly supporting the involvement of this gene cluster in holomycin biosynthesis. The discovery of the gene cluster allowed further biochemical evaluation of the underlying chemical logic for the compact yet chemically complex dithiolopyrrolone scaffold.

Function of the A and T domains of stand-alone single NRPS module 3488 has now been established. In line with the view that two molecules of cysteine may be condensed and morphed to the holothin scaffold, the A domain of 3488 reversibly activates l-Cys as the Cys-AMP mixed anhydride, and ¹⁴C-l-Cys could be covalently tethered to the phosphopantetheinylated holo form of the T domain. As yet unresolved is the function of the 50 kDa Cy domain, predicted to be a variant of condensation domains to make peptide bonds during chain elongation on NRPS assembly lines (15). Given the genetic deletion result establishing NRPS essentiality and the l-Cys activation and loading, we have surmised that two molecules of Cys-S-ORF3488 might dock and undergo a Cy domain-catalyzed formation of the l-Cys-l-Cys-S-ORF3488. Alternatively, the stand-alone condensation domain, ORF3495, may be responsible for the peptide bond formation. Interestingly, the Cys–Cys dipeptide was detected in solution of the loading mixture by MS analysis, which may be a result of spontaneous hydrolysis. Inter alia, thioester release from ORF3488 is presumed to be a prerequisite to yield the free carboxylate for oxidative decarboxylation by ORF3489. Subsequent studies will involve biochemical analysis of heterologously expressed ORF3486 and ORF3494, predicted thioesterases, to assess their functions for example in thiolytic release of a Cys-S-3488 by a second l-Cys.

Of the four predicted flavoenzymes, ORF3492 as a predicted thioredoxin–disulfide reductase homolog may be responsible for redox interconversions of the disulfide and a dithiol form of holothin/holomycin. ORF3489 may function analogously to phosphopantothenoylcysteine decarboxylase (PPC-DC) in coenzyme A biosynthesis by oxidation of a cysteine side chain, at some stage, to the thioaldehyde creating an electron sink beta to the carboxylate to enable a low-energy transition state for decarboxylation. In the PPC-DC reaction the resultant enethiol is reduced back to the CH₂-CH₂SH side chain to yield phosphopantetheine. In the related lantibiotic decarboxylases, the enethiol product is not reduced. Because no decarboxylation activity was detected for purified ORF3489 toward free l-Cys-l-Cys, it may be that formation of the aminopyrrolinone ring, set up by two CH₂-SH side chain oxidation events, occurs prior to thiolytic release and decarboxylation. One such possibility is shown in Fig. 2B, where an enethiol functionality is proposed as a carbon nucleophile equivalent to attack a neighboring thioaldehyde carbon building the aminopyrrolinone ring at a four-electron oxidized stage. To date, addition of ORFS 3483, 3485, 3487, 3489, acetyl-CoA, and exogenous FAD has not yet led to reconstitution of the dithiolopyrrolone scaffold, indicating that much is yet to be learned to understand how the biheterocycle is assembled.

The amino group of the dithiolopyrrolone ring can be found acylated with a variety of acyl groups, suggesting promiscuity in a late-stage acyltransferase. This hypothesis is supported by conversion of holothin to holomycin (13) and N-methyl holothin to thiolutin related products with longer acyl chains (23), and validated with purified ORF3483 in this work. Holomycin appeared to be fairly stable during room temperature incubation, whereas holothin displayed a half-life of ∼30 min at neutral pH, suggesting its instability/reactivity (SI Appendix, Fig. S11). Therefore, acylation of the amino group in holothin, the last step in the pathway, could be a self-protective response in producer microbes and/or stabilization of the final natural product. Although the different length acyl chains are likely to reflect some combination of ORF3483 permissivity and the intracellular concentration of acyl CoAs, no extensive analysis has been carried out on whether distinct acyl chains impart sufficiently different physical properties to confer different target selectivity, nor on the role of the acylated holothin scaffolds in the producer organisms. One hypothesis is that the disulfide within the five-membered ring fused to the N-acyl pyrrolinone can be reduced to the dithiol in vivo, thereby uncovering a potentially unstable enethiol component, which could be important for cell killing activity. Comparable reduction of the bridging disulfide in the fungal metabolite gliotoxin is imputed to be central to its toxic activity (24).

In summary, the availability of the genome sequence of the clavam and cephem antibiotic producing S. clavuligerus has enabled genome mining to identify the holomycin biosynthetic gene cluster. Given the unusual compact heterobicyclic scaffold of the dithiolopyrrolone antibiotics, we anticipated and discovered a noncanonical gene cluster. Based on the biochemical analysis and gene sequence homology, holomycin biosynthesis may involve joining two cysteine residues into an l-Cys-l-Cys dipeptide precursor along with eight-electron oxidation to generate the dithiolopyrrolone scaffold of holothin, which is then acylated to form holomycin (Fig. 2B). Four two-electron oxidations might occur at different stages of holothin assembly, some of which may take place while the substrate is still tethered to the thiolation domain of the single NRPS module. The timing of decarboxylation and of the formation of one C-C bond and one S-S bond to convert an acyclic precursor to the bicyclic holothin are not yet known. This work sets the stage for gene cluster identification in other dithiolopyrrolone producers, e.g., Xenorhabdus strains, and for examination of the complex redox enzymology likely to convert a Cys–Cys dipeptide precursor into the fused five-five dithiolopyrrolone. Also of note is the thiomarinol class of natural products, reflecting the strategy of covalent attachment of two antibiotics with distinct targets in a single amide scaffold: (i) pseudomonic acid variants, which target Ile tRNA synthetase, provide the acyl component, whereas (ii) holomycin provides the amine component. Whether the ligase ORF3483 will use not only octanoyl- but also 8-OH-substituted octanoyl CoAs en route to build up the thiomarinol scaffold from holothin will be a subject of future investigation for possible combinatorial biosynthesis of double-headed antibiotics.

Materials and Methods

Genetic disruption in S. clavuligerus.

In-frame deletions of ORF3488 and ORF3489 were achieved in both wild-type and ORF15::apr mutant by conjugation of E. coli WM6062 containing the disruption vectors with Streptomyces following standard protocols (25). The disruption vectors were constructed using the ReDirect PCR-targeting strategy (26). The target genes were first cloned into a PCR-Blunt vector (Kan^R, Invitrogen) with a 2-kb extension of the chromosomal sequence on each end. These vectors were then introduced into E. coli BW25113/pIJ790 strain, which allows λ red mediated recombination with the PCR fragments containing the antibiotic-resistance cassettes. The PCR fragments were generated using vectors pIJ773 (Apra^R, acc(3)IV-oriT cassette) or pIJ10700 (Hyg^R, hyg-oriT cassette) (John Innes Centre) as template with primers including a 36-nt extension homologous to the target region. Primer sequences are shown in SI Appendix, tables.

Analysis of Holomycin Production by the Deletion Mutants.

The four exoconjugants ORF3488::apr, ORF3489::apr, ORF3488::hyg/ORF15::apr, and ORF3489::hyg/ORF15::apr, confirmed by PCR analysis, were grown in tryptic soy broth media for 24–48 h for starter culture. A sample of 10–100 μL of the starter culture was used to inoculate 10 mL of liquid glycerol-sucrose-proline-glutamate medium (27), which was grown with shaking at 30 °C for 2–5 d. The resulting culture was centrifuged and the supernatant was extracted with equal volume of ethyl acetate. The ethyl acetate layer was dried by rotary evaporation and resuspended in 60 μL of methanol. A sample of 10 μL of the resuspension was injected on the Accurate-Mass Q-Tof LC/MS instrument (Agilent Technologies 6520) for detection of holomycin production via electrospray ionization (ESI).

Holomycin Purification and Acid Hydrolysis.

The holomycin-overproducing ORF15::apr mutant strain was grown on ISP4 agar plates at 30 °C for 5–7 d, and at the end of the incubation the plates turned from white to a bright yellow color. The solid media were extracted with ethyl acetate, and the yellow extracts were further purified by HPLC. The purified yellow compound was confirmed as holomycin by high-resolution mass spectrometry analysis (SI Appendix, Fig. S1a), UV spectrometric analysis (SI Appendix, Fig. S2 a and b), and ¹H and ¹³C NMR (SI Appendix, Fig. 3 a and b). Purified holomycin was dissolved in 1,3-dioxane and mixed with 12N HCl (2N final concentration of HCl). The mixture in 0.5–1 mL volume was heated in a microwave reactor (CEM Discover) at 100 °C for 10 min. The product mixture was clearly separated in a yellow dioxane layer with leftover starting material holomycin and a dark brown water layer, which contained holothin hydrochloric salt. The water layer was lyophilized, resuspended in methanol, and injected on the preparative HPLC to purify holothin. The identity of holothin was verified by high-resolution MS (SI Appendix, Fig. S1b), UV spectrometric analysis (SI Appendix, Fig. S2a), and ¹H NMR (SI Appendix, Fig. 3c).

Activity Assay for Acyltransferase ORF3483.

The concentration of purified holothin was determined by full conversion to holomycin as shown by MS analysis in the presence of 5 μM purified ORF3483, 1 mM acetyl-CoA, and 50 mM Hepes (pH 8.0) at room temperature for 10 min. The UV absorbance of resulting mixture was measured at 388 nm using UV-vis spectrometer (Varian Cary), and the concentration of the holomycin product was calculated using a extinction coefficient of 11,220 M^-1 cm^-1 as reported (18). At a fixed concentration of holothin (20 μM), the apparent k_cat and K_m for the CoA substrates were obtained by varying the concentration of acetyl-, propionyl-, or octanoyl-CoA in the presence of 10 nM (acetyl- and propionyl-) or 10 μM (octanoyl-) of ORF3483. Holomycin product formation was quantified by analytical HPLC with reference to a holomycin standard curve.