Engineering Broader Specificity into an Antibiotic-Producing Polyketide Synthase

See allHide authors and affiliations

Science  09 Jan 1998:
Vol. 279, Issue 5348, pp. 199-202
DOI: 10.1126/science.279.5348.199


The wide-specificity loading module for the avermectin-producing polyketide synthase was grafted onto the first multienzyme component (DEBS1) of the erythromycin-producing polyketide synthase in place of the normal loading module. Expression of this hybrid enzyme in the erythromycin producer Saccharopolyspora erythraea produced several novel antibiotic erythromycins derived from endogenous branched-chain acid starter units typical of natural avermectins. Because the avermectin polyketide synthase is known to accept more than 40 alternative carboxylic acids as starter units, this approach opens the way to facile production of novel analogs of antibiotic macrolides.

The emergence of pathogenic bacteria multiply resistant to antibiotics represents a growing threat to human health (1) and has given added impetus to the search for new drugs and for more effective versions of those in current use. The antibacterial erythromycin A, like other macrocyclic polyketides, is derived from simple carboxylic acid precursors by stepwise chain assembly on a modular polyketide synthase (PKS). Such PKSs are giant multienzyme complexes (2-5) containing a different set or “module” of enzyme domains to accomplish each successive cycle of polyketide chain extension. The erythromycin-producing PKS [6-deoxyerythronolide B synthase (DEBS)] of Saccharopolyspora erythraea contains six such extension modules, together with a “loading module” comprising an acyltransferase (AT) and an acyl carrier protein (ACP), and an “off-loading” domain, a thioesterase/cyclase (TE) (2). The modularity of such systems suggests that splicing together of the structural genes for different PKSs should produce hybrid multienzymes that might synthesize novel molecules incorporating elements of different polyketide natural products (6).

We report the production of novel macrolide antibiotics by this method. Our strategy was to replace the loading module of DEBS, which accepts propionyl coenzyme A (CoA) and acetyl-CoA in vivo (7), with the loading module from the avermectin-producing polyketide synthase of Streptomyces avermitilis, which has an unusually broad specificity for branched carboxylic acids as starter units in vivo (8). The viability of the approach was demonstrated initially in a truncated PKS consisting of the first two extension modules of DEBS and the chain-terminating thioesterase (9). The equivalent module swap was then carried out on the intact modular PKS in S. erythraea, whereupon novel, biologically active erythromycin derivatives containing branched-chain starter units were produced. These results suggest that the promiscuity of the avermectin PKS loading module can be transferred to other modular PKSs, opening the way to the wholesale and convenient biosynthesis of novel analogs of important polyketide antibiotics, antifungals, and immunosuppressants.

The donor module in these experiments was the NH2-terminal AT-ACP di-domain of AVRS1, the first multienzyme component of the avermectin-producing polyketide synthase from S. avermitilis. The avermectins are a family of 16-membered macrocyclic lactones which are potent antiparasitic agents through their action as agonists of a γ-amino butyric acid–insensitive chloride channel (10). The pentacyclic ring structure of avermectin B1 is assembled from five propionate units and seven acetate units, with either an isobutyrate (B1b) or a 2-methylbutyrate (B1a) starter unit (11). The clustered genes encoding the biosynthetic enzymes have been cloned, and the locations of most of the constituent domains of the PKS have been reported (5). Although the natural avermectins normally incorporate just two alternative starter acids, the avermectin loading module has been shown previously to accommodate a wide range of nonnatural starter units, and this has allowed the facile production of avermectin analogs with novel starter units (8).

The loading module of the erythromycin PKS, DEBS, is highly selective for propionate and acetate units both in vivo and in vitro (12). Recently, domain swaps have been used successfully to alter the extender specificity of a modular PKS from propionate to acetate (13), and to alter the starter unit specificity from acetate to propionate (14). Independently, a precursor-directed approach has recently succeeded in the production of erythromycin D analogs altered in the region of the starter unit, by aberrant incorporation of synthetic partial chains into the polyketide synthase (15). It is not known why these experiments did not give rise to fully active erythromycin A analogs. In contrast, the genetic alteration we describe was aimed at the conversion of DEBS into a hybrid enzyme of broader specificity, potentially capable of accepting the wide range of alternative branched-chain starter units characteristic of the avr PKS within a host organism potentially capable of fully processing the new macrolides into erythromycin A analogs.

First, a hybrid PKS was constructed in which the broad specificityavr PKS loading module was fused, in place of the natural propionate-specific loading module, to the NH2-terminus of the truncated PKS, DEBS1-TE (9), which contains only the first two extension modules of the ery PKS and the chain-terminating thioesterase. The predicted δ-lactone products for this hybrid PKS would have the normal erythromycin triketide lactone structure (9), but the alkyl substituent at carbon 5 (C-5) would be specified by the avr PKS loading module and therefore would be predicted to be derived from either isobutyrate or a 2-methylbutyrate (Fig. 1A) (11), because isobutyryl- and isovaleryl-CoA are the only C-2 branched acyl-CoA esters normally found in actinomycete cells (16). The region of the avr gene cluster ofS. avermilitis encoding the loading module was identified from published data (5), cloned, and sequenced. The loading module was confirmed to consist of an AT-ACP di-domain showing significant sequence similarity to the loading module of DEBS1. Theavr PKS loading module was inferred to start within the first multienzyme AVRS1 at the sequence 108VVFVFPGQ (17) and to end some 150 amino acid residues later at the sequence HGGTAAAD556 (numbering refers to corresponding amino acid residues in DEBS1) (18). The DNA encoding the loading module was amplified by polymerase chain reaction (PCR) and cloned in the place of the equivalent fragment in the DEBS1-TE expression plasmid pRMTE (13) to create plasmid pIG1 (19), which was used to transform Streptomyces coelicolor CH999 (20).

Figure 1

The products of hybrid modular PKSs containing an altered loading domain. (A) Expression plasmid pIG1 in S. coelicolor CH999 (20) encodes the loading domain from the avermectin PKS (avr-lm) in place of the natural loading module, fused to a truncated PKS (DEBS1-TE) consisting of modules 1 and 2 and the chain-terminating thioesterase (end) of DEBS (9). (B) Mutant strain NRRL2338/pAVLD (ERMD1) of S. erythraea contains the hybrid avr-lm-DEBS1 multienzyme, DEBS2, DEBS3, and the auxiliary genes required to convert 6-deoxyerythronolide into erythromycin A (6). The active sites in each domain of DEBS are ketosynthase (KS), acyl carrier protein (ACP), acyltransferase (AT), dehydratase (DH), ketoreductase (KR), enoylreductase (ER), and thioesterase (TE). The erythromycins produced include analogs of erythromycins A, B, and D.

In this system, the expression of the hybrid DEBS1-TE is under the control of the actI promoter and the specific activator protein ActII-ORF4 (21). When S. coelicolor CH999/pIG1 was grown for 5 days in liquid medium and the fermentation broth was extracted with ethyl acetate, several polyketides not produced by a control fermentation of S. coelicolor CH999 were identified and were subsequently purified from the extract. The products included the triketide lactones3 and 4 arising from the use of the isobutyrate and 2-methylbutyrate starter units, respectively. The loading module swap does indeed generate a functional hybrid PKS and one that accepts starter units characteristic of the transplanted broad-specificityavr PKS loading module. The triketide lactones 1through 4 (Fig. 1A) were obtained in the approximate ratio of 2:8:2:5 (total yield 10 mg/liter). The fermentation conditions were not optimized, and yields might also have been improved by addition of suitable carboxylic acids to the fermentation (8). All of the compounds were fully characterized by gas chromatography, high-pressure liquid chromatography (HPLC), two-dimensional (2D) nuclear magnetic resonance (NMR), electrospray mass spectrometry (ESMS), and high-resolution ESMS (22), and by comparison with synthetic standards (23). The production of 1 and2 by the hybrid PKS was not unexpected: The avrPKS loading module accepts both acetate and propionate starter units (24).

The efficacy of the hybrid DEBS1-TE multienzyme encouraged us to examine whether the significant changes in starter unit would perturb either the later stages of polyketide chain assembly or the further processing steps required to produce active erythromycin antibiotics. In the first three post-PKS steps, the macrolide ring is hydroxylated at C-6, and then glycosylated successively with mycarose on the hydroxyl group at C-3 and with desosamine on the C-5 hydroxyl to give erythromycin D, the first compound on the pathway to show antibiotic activity (6). Subsequent hydroxylation at C-12 yields erythromycin C, which rapidly undergoes a final C-3" O-methylation of the mycarose moiety to give the clinically effective erythromycin A. Alternatively, the last two steps may take place in the reverse order, in which case, erythromycin B is formed as a shunt intermediate in place of erythromycin C (25).

The cloned DNA for the avr PKS loading module, flanked byery PKS sequence, was subcloned into an integrative plasmid vector and introduced into an erythromycin-producing strain of S. erythraea. DNA blot analysis confirmed that integration had occurred by homologous recombination (26), so as to create a chromosomally located hybrid DEBS multienzyme (Fig. 1B) in which the DNA encoding the avr PKS loading module replaced the loading module of DEBS1. One such integrant, strain S. erythraeaERMD1, was grown for 5 days in liquid medium, and an ethyl acetate extract of the broth supernatant was fractionated by silica-gel chromatography. A mixture of erythromycin analogs was obtained which was analyzed by reversed-phase HPLC and electrospray mass spectrometry (ESMS and ESMSMS). Erythromycins A (5), B (6), and D (7) were detected in the mixture, together with six novel erythromycins. Erythromycin analogs 8, 9,and 10 all had high-resolution masses (for both parent ion and two daughter ions) appropriate for analogs of erythromycins A, B, and D, respectively, in which the substituent at C-13 isiso-propyl rather than ethyl (Fig. 1B) (27). Analogs 11, 12, and 13 all had high-resolution masses appropriate for analogs of erythromycins A, B, and D, respectively, in which the substituent at C-13 of the macrolide ring is sec-butyl rather than ethyl (Fig. 1B) (27). The products containing the new starter units were obtained in approximately equal quantities; the ratio of D:B:A analogs was ∼50:50:5. The difference in ratio of products derived from alternative starter units for the experiments performed inS. coelicolor and S. erythraea is most probably a reflection of the variable flux of primary metabolites in these organisms. The erythromycin B analog 12, which had incorporated the alternative starter acid 2-methylbutyrate in place of propionate and therefore contained a sec-butyl group at C-13, was purified from a large-scale fermentation in amounts sufficient for 2D NMR analysis (28). The presence of the novel sec-butyl starter group at C-13 was unambiguously confirmed by the observed pattern of connected peaks in the correlated spectroscopy (COSY) spectrum. Both compounds 11 and12, an erythromycin A and erythromycin B analog, respectively, showed antibiotic activity comparable to erythromycins A and B in bioassays against an erythromycin-sensitive strain ofBacillus subtilis (29).

These results represent the engineering of novel derivatives of an important class of polyketide drugs, by substitution of key domains within the PKS by counterpart domains from an entirely different modular PKS and fermentation of the mutant strains by standard procedures. The bulky starter group typical of avermectins could still be accommodated by all of the approximately 30 active sites of the DEBS multienzyme, in accordance with recent experiments in which synthetic diketides were fed to DEBS (15). It certainly supports a model (30) in which successive modules are functionally and structurally autonomous of their neighbors. The broad specificity of the avr PKS loading module is well-documented. At least 44 different branched carboxylic acids or their precursors have been successfully incorporated into avermectins (8). The significantly improved efficacy of the avermectin analog doramectin (31), in which the starter unit is a cyclohexanecarboxylic acid, shows the potential benefits of this approach. The broad specificity avr PKS loading module can obviously be fused onto the NH2-terminus of other modular PKS for which the structural genes are available. Also, other “donor” loading modules of distinctive specificity are available (4, 32). Combinatorial libraries of aromatic polyketides have already been produced by coexpression of the genes for chain formation with genes for cyclases from other PKS clusters (33). If future work shows that many combinations of modules taken from different natural modular PKSs work productively together, then combinatorial libraries of macrolides and other complex polyketides should also become accessible (34).

  • * To whom correspondence should be addressed. E-mail: pfl10{at}


View Abstract

Navigate This Article