INTRODUCTION

Top Previous Next References

The explosive growth in the number of cloned and sequenced genes has led to an enhanced need for robust heterologous gene expression methods. This need exists, in part, because the proteins encoded by these numerous genes of interest promise to have an important impact in areas ranging from basic biochemical and biophysical research to the practical use of proteins as pharmaceuticals, animal health products, and industrial enzymes. Notwithstanding the enormous efforts that have gone into the development of the heterologous-expression toolbox, the production of a desired protein in a heterologous host remains an empirical and often unpredictable process.

Although there is no universal solution for heterologous protein production, certain generalizations can be made. A few prokaryotic and eukaryotic systems such as Escherichia coli, Saccharomyces cerevisiae, Sf9 insect cells, and Chinese hamster ovary (CHO) cells have emerged as choice hosts due to their simplicity of use, excellent growth characteristics, and a plethora of readily accessible genetic tools. In turn, this has fueled studies aimed at understanding the cellular machinery responsible for protein synthesis, posttranslational modification, protein folding, trafficking, and degradation in these model host cells. Another feature common to good heterologous hosts is the availability of good fermentation protocols that maximize protein productivity, especially those in which cell growth can be decoupled from recombinant gene expression. Finally, it is generally understood that a balance exists between product quality and quantity, regardless of the host used. High levels of gene expression are often accompanied by phenomena such as inclusion body formation, increased amino acid misincorporation, incomplete or inaccurate posttranslational modification, reduced recombinant cell line stability, and a range of other features associated with a greater metabolic burden as a result of heterologous protein production.

In most cases, the ultimate goal of heterologous gene expression is to produce the desired protein in reagent quantities. Over the past 20 years, however, there has been a growing interest in the potential for harnessing the intrinsic metabolic activity of proteins in heterologous hosts. Multiple genes are often coexpressed in metabolic engineering, where the goal is not simply to synthesize large quantities of the target proteins themselves but to optimally interweave their activities among each other as well as among functional protein networks in the host. Examples of metabolic engineering include the bacterial biosynthesis of indigo (19) and the conversion of 3-dehydroshikimic acid, a key intermediate in aromatic amino acid metabolism, into a variety of value-added products such as vanillin (47).

Over the past decade, the study of bioactive natural-product biosynthesis has benefited significantly from the use of heterologous hosts. Notably, the rapid growth in understanding and manipulating polyketide biosynthesis closely parallels developments in the ability to reconstitute these multistep catalytic processes in genetically (and now genomically) friendly heterologous hosts. This review focuses on the development of heterologous expression systems for polyketide synthases (PKSs), and discusses their impact on the field of natural-product biosynthesis and drug development. Reconstitution of polyketide biosynthesis in heterologous hosts demands that large multienzyme assemblies be functionally expressed, their posttranslational modification needs be adequately met, their substrates be available in vivo in reasonable quantities, and the producer cell be protected against the toxicity of the biosynthetic products. Two particular heterologous hosts will be the focus of most of our discussionStreptomyces coelicolor and Escherichia coli. However, the relative merits of a variety of other heterologous hosts will also be discussed.

Polyketides, as the name implies, are synthesized from repetitive condensation reactions that link small carbon precursors (typically, 2- and 3-carbon acyl groups derived from coenzyme A (CoA) thioesters) (Fig. 1A) (63). The process is similar in many respects to bacterial and mammalian fatty acid synthesis. In fact, PKSs are classified into type I or II categories based on how closely they mimic the architecture of type I (vertebrate) or type II (bacterial and plant) fatty acid synthases (35). However, unlike fatty acids, the structures of polyketides are far more diverse due to variations in the fatty acid synthesis theme (Fig. 1B) and post-PKS modifications. This structural diversity is also reflected in diversity in their biological modes of action. A number of polyketides have been clinically approved as drugs for treating disorders such as infections, cancer, cardiovascular diseases, and inflammation. Approximately two-thirds of the known bioactive polyketide natural products originate from the actinomycetes. Other major microbial sources of polyketides include the myxobacteria and filamentous fungi. The exact role of polyketides in the life cycles of producing organisms remains unknown, but these secondary metabolites are presumably synthesized to ward off competing microbes during periods of nutrient limitation. Some polyketides are also synthesized as spore pigments (14, 55).

View larger version (20K): [in this window] [in a new window]   FIG. 1.   Key reactions and catalysts in polyketide biosynthesis. (A) The decarboxylative condensation reaction that defines a polyketide. The electrophile is attached to the ketosynthase (KS), whereas the nucleophile is attached to an ACP. (The only known exceptions to this rule are the chalcone synthase-like PKSs [see Fig. 3], where the nucleophile remains attached to CoA) (B) In addition to the above-mentioned C---C bond-forming reaction, PKSs catalyze all, some, or none of the following reactions. The -carbonyl generated on C---C bond formation can be reduced by an NADPH-dependent enzyme called a ketoreductase (KR). The resulting alcohol can be dehydrated by a dehydratase (DH). The resulting olefin can be hydrogenated by another NADPH-dependent enzyme called an enoylreductase (ER). Other, less commonly occurring reactions such as C-methyl transfers (28) (see also Fig. 4) are not shown.

Figure 2 shows examples of familiar polyketides and their representative biological sources. It also highlights differences in the architectures of PKSs that synthesize the carbon skeletons of these natural products. Four broad architectural varieties of PKSs have been isolated thus far from microbes. The first category of PKSs is perhaps the simplest and actually resembles plant PKSs such as the chalcone synthase. An example is shown in Fig. 3 (24). This is the only variety of PKS in which the nucleophilic group involved in each C---C bond-forming reaction is attached to a CoA instead of an acyl carrier protein (ACP). Although the primary amino acid sequences of these PKSs are only weakly related to those of other PKSs, X-ray crystallographic analysis (22) and site-directed mutagenesis (38) of a prototype of this family, a plant chalcone synthase, have clearly established the close evolutionary connection between these PKSs and others. Fungal PKSs represent a second subclass of PKSs that closely resemble vertebrate (type I) fatty acid synthases. The proteins are multidomain and act in an iterative fashion. An example is the lovastatin LNKS, which catalyzes the formation of dihydromonacolin, the polyketide precursor for lovastatin biosynthesis (Fig. 4) (41). Bacterial PKS systems with an architectural relationship to type II fatty acid synthases represent a third PKS category. Also acting in an iterative fashion, these multienzyme systems produce aromatic polyketide products. However, individual active sites occur as distinct polypeptides rather than as domains within a single multifunctional polypeptide. An example is the doxorubicin PKS, which synthesizes the tetracyclic skeleton of this anthracycline antibiotic (Fig. 5) (2). Finally, a fourth category of PKSs, known as modular PKSs, is exemplified by the deoxyerythronolide B synthase (Fig. 6) (13, 17). Here, polyketide biosynthesis proceeds in a processive fashion on a very large megasynthase in which each active site is used once during the overall catalytic cycle.

View larger version (36K): [in this window] [in a new window]   FIG. 2.   Representative polyketide natural products and PKS systems. This figure illustrates the diversity in polyketide natural products and the PKSs that catalyze their formation. The block arrows represent ORFs encoding different PKS components. The given scales indicate approximate ORF sizes, although individual catalytic domains are not shown. In general, the larger proteins encode several catalytic domains whereas the smaller polypeptides possess a single enzymatic activity. Protein abbreviations: KS, ketosynthase; CLF, chain length factor; KR, ketoreductase; ARO, aromatase; CYC, cyclase; Dps, doxorubicin polyketide synthase; DEBS, deoxyerythronolide B synthase; MSAS, methyl salicylic acid synthase.

View larger version (16K): [in this window] [in a new window]   FIG. 3.   Chalcone synthase-like PKS. The RppA PKS iteratively condenses five malonyl-CoA units, which then cyclize to form 1,3,6,8-tetrahydroxynaphthalene (THN). THN most probably represents an intermediate in the production of antibiotics such as napyradiomycin A. RppA resembles a ketosynthase (KS) domain in other PKS paradigms. However, in this case, RppA condenses malonyl-CoA units directly as opposed to accepting a malonyl unit from an ACP.

View larger version (17K): [in this window] [in a new window]   FIG. 4.   Fungal type I PKS. (A) The lovastatin nonaketide synthase (LNKS; LovB) protein, with its individual enzymatic domains denoted within the PKS. KS, ketosynthase; AT/MT, acetyl/malonyl transferase; DH, dehydratase; MeT, methyltransferase; KR, ketoreductase; ACP, acyl carrier protein. (B) LNKS works in an iterative fashion to produce dihydromonacolin L. The LNKS catalyzes a decarboxylative condensation, and the reaction product (the polyketide chain) then transfers to a new LNKS unit. The reaction arrows indicate condensation reactions between an extender malonyl unit and either a priming acetate unit or the growing polyketide chain (eight condensations in total). The KS domain accepts the starter acetate unit or a growing polyketide chain and catalyzes the condensation with a malonyl unit loaded onto the ACP domain by the AT/MT. The level of reduction applied for a particular iteration is denoted near the arrows. In this case, a separate lovastatin PKS protein (not shown) provides the enoylreductase (ER) activity.

View larger version (14K): [in this window] [in a new window]   FIG. 5.   Bacterial type II PKS. The doxorubicin polyketide synthase (Dps) proteins iteratively condense one propionyl-CoA unit and nine malonyl-CoA units. Initially, DpsC, DpsD, and DpsG (analogous to the KS, CLF, and ACP) catalyze a diketide formation between propionyl-CoA and malonyl-CoA. The DpsA, DpsB, and DpsG (again, analogous to the KS, CLF, and ACP) domains catalyze eight more condensation reactions, starting with the previous diketide and adding successive malonyl units to eventually produce a decaketide. The DpsE domain (encoding a KR activity) then reduces the polyketide chain before DpsF cyclizes the chain. DpsH appears to provide another cyclase activity.

View larger version (26K): [in this window] [in a new window]   FIG. 6.   Bacterial type I PKS. The deoxyerythronolide B synthase (DEBS) system catalyzes the formation of the erythromycin derivative 6-deoxyerythronolide B (6-dEB) by processively combining one propionyl-CoA primer unit and six methylmalonyl-CoA extender units. Each DEBS protein contains two modules; modules represent a physical location for a Claisen-like condensation reaction. For example, once loaded onto the KS of module 1 by the loading acyl transferase (AT) and ACP domains, propionyl-CoA condenses with a methylmalonyl unit loaded onto module one ACP by the module 1 AT domain. After the reaction takes place, reductive domains, specific for each module, reduce the resulting ketone group. The chain then passes in a processive manner from module 1 ACP (via the phosphopantetheine arm) to the KS domain of the next module. In this fashion, the polyketide chain grows and diversifies before being released and cyclized by the C-terminal thioesterase (TE) domain.

There are several features associated with microbial polyketide biosynthesis that make these pathways particularly well suited for heterologous expression. First, PKSs are remarkably similar in primary sequences and (by inference) tertiary structures, giving hope that, as improved heuristics for heterologous expression in a given host emerge, they can be applied with greater confidence to newer systems. Second, in all known bacterial and fungal examples, PKS genes have been found to exist as gene clusters (together with transcriptional regulators and self-resistance genes), a feature that greatly simplifies their genetic isolation as well as heterologous expression (at least in bacteria, where multigene operons can be constructed). Third, polyketides are naturally produced as secondary metabolites, suggesting that they can be readily produced in two-stage fermentations in which cell growth can be decoupled from product formation. Fourth, an extraordinary diversity of chemotypes can be synthesized by the PKS paradigm from a relatively small subset of intracellular precursors such as acetyl-CoA, propionyl-CoA, malonyl-CoA, and methylmalonyl-CoA. Moreover, each of these precursors can be derived via multiple metabolic routes from exogenously available carbon sources, thereby presenting a range of options for any potential heterologous host. Finally, as the time lines for converting pharmacologically interesting leads into clinically useful molecules continue to shrink, the earlier paradigm of developing the biology of new microbes for every new polyketide natural product of interest is no longer practical. Lateral transfer of a pathway of interest into a well-developed surrogate host therefore becomes an attractive alternative, both with respect to overproducing the parent natural product itself and for generating novel analogs via biosynthetic engineering. Notwithstanding these advantages, there are several unique characteristics of PKSs that make heterologous expression a definite challenge. Notable PKS characteristics include their relatively large size (100 to 10,000 kDa), the high G+C content of many PKS genes (especially those coming from the actinomycetes, where the G+C content typically exceeds 70%), their requirement for posttranslational modifications, and the relatively rudimentary understanding of the special regulatory and metabolic features associated with the transition from primary to secondary metabolism in any organism.

As is now clear from several studies involving reconstituted PKS systems (reviewed in reference 42), PKSs are soluble, cytosolic multienzyme systems that do not require any intracellular substructure or organelle to maintain activity. Individually purified protein components (or in some instances proteins purified as heterodimers) can be mixed in vitro to yield PKS activity. While not completely ruling out the need for additional cellular machinery to enhance PKS activity, these experiments indicate that, if introduced into a foreign cellular host, the DNA encoding PKS genes, when actively expressed, should support polyketide production provided that the required substrates are available. Below we review the important considerations for reconstituting polyketide pathways in a heterologous host.

It has long been known that the active site of an ACP involves the thiol moiety of a 4'-phosphopantetheine group that is covalently attached to a conserved serine residue in the polypeptide (79, 83). This pantetheinyl group is posttranslationally derived from intracellular CoASH (Fig. 7) (18). Until recently, very little was known about the enzymatic basis for this modification. Over the past few years, the work of Walsh and coworkers has been instrumental in identifying an evolutionarily related superfamily of enzymes, the phosphopantetheinyl transferases (PPTases), that catalyze this reaction (44, 45). Although these enzymes are known to exist in all organisms except perhaps for the archaea (which do not make fatty acids or polyketides), individual members of this superfamily have significantly distinct configurations and substrate preferences. For example, the E. coli genome contains at least three different PPTase genes (44), each existing as an individual open reading frame (ORF), whereas in Saccharomyces cerevisiae the PPTase that recognizes the cognate fatty acid ACP is a distinct domain within one of the fatty acid synthase subunits (44). Likewise, the E. coli PPTase, which ordinarily pantetheinylates its fatty acid ACP, has relatively tight selectivity for its cognate substrate (44). In contrast, the sfp gene product, which is part of the surfactin biosynthetic gene cluster in Bacillus subtilis, is among the most tolerant PPTase discovered to date and can effectively modify ACPs from all PKS subclasses as well as related peptidyl carrier protein and aryl carrier protein domains from nonribosomal peptide synthetases (NRPSs) (11, 32, 40, 44, 69). (Although a thorough discussion of NRPSs is outside the scope of this review, NRPSs will be occasionally referred to below, given the close relationships between these systems. For further details, the reader is directed to recent reviews on the subject [10, 61].) Therefore, in deciding on a strategy for heterologous PKS gene expression, the choice of a partner PPTase is an important consideration, from both the substrate specificity and the gene regulation perspectives. Depending on the host and expression system chosen, this posttranslational modification may or may not happen.

View larger version (10K): [in this window] [in a new window]   FIG. 7.   PPTase-catalyzed reaction. The PPTases catalyze the transfer of the 4'-phosphopantetheine arm from free CoASH to a conserved ACP serine. This "swinging arm" facilitates polyketide transfer during biosynthesis.

Once functionally expressed and posttranslationally modified within a cellular host, the PKS will require a substrate pool to draw upon for polyketide production. PKSs are known to utilize a broad range of substrates including acetyl-CoA, propionyl-CoA, isobutyryl-CoA, isovaleryl-CoA, malonyl-CoA, methylmalonyl-CoA, ethylmalonyl-CoA, propylmalonyl-CoA, and hydroxymalonyl-CoA (or its methylated counterpart, methoxymalonyl-CoA) (37). When the substrates are chiral, such as methylmalonyl-CoA, the corresponding acyltransferases exhibit strict stereospecificity (52). The -carboxylated substrates (e.g., malonyl-CoA and methylmalonyl-CoA) are sources of extender units, whereas neutral substrates such as acetyl-CoA are sources of primer units for polyketide chain synthesis. Finally, many "hybrid" natural products are derived from the tandem action of PKSs and NRPSs (9). These multifunctional enzymes utilize an even broader range of substrates, including carboxylic acids such as p-aminobenzoic acid, 3-amino-5-hydroxybenzoic acid, cyclohexenoyl carboxylic acid, and dozens of - and -amino acids. Preactivated forms (e.g., CoA thioesters) of these free acids are not required; rather, the acids are activated in situ by ATP-dependent adenylation domains that are intrinsic components of NRPS modules (61).

From the above shortlist, it can be appreciated that a good heterologous host must be endowed with the ability to synthesize an impressive range of substrates (in addition, of course, to NADPH and ATP, which are routinely available). Moreover, the supply of these precursors must be coordinately regulated with polyketide biosynthesis so as to ensure availability when required while at the same time avoiding imbalances in tightly controlled metabolite pools (such as those corresponding to various CoA derivatives). Although this might appear to be a formidable problem for the metabolic engineer, several simplifying factors deserve to be noted. (Some of these points are also elaborated below.) First, many polyketide natural products are derived from a subset of acetyl-CoA, propionyl-CoA, malonyl-CoA, and (2S)-methylmalonyl-CoA alone. Therefore, the ability of a candidate host to supply just these four metabolites can make it an attractive environment for heterologous PKS expression. Second, since multiple metabolic routes are known to exist for many of these precursors (Fig. 8), a given host needs to be endowed with only one such pathway in order to facilitate polyketide synthesis. For example, at least four pathways to (2S)-methylmalonyl-CoA are known to exist in bacteria (7, 8, 16, 33, 36, 71, 87). (It should be noted that, from a quantitative viewpoint, not all biosynthetic pathways are energetically or kinetically equivalent.) Third, the broad specificity of certain enzymes that synthesize some of the above metabolites facilitates intracellular production of more than one PKS substrate from a given pathway, depending on the exogenous supply of carboxylic acids. For example, a single enzyme, malonyl-CoA synthetase, can be used to synthesize numerous -carboxylated CoA thioesters from their corresponding exogenously supplied 1,3-dicarboxylic acids (N. Pohl and C. Khosla, unpublished data). Fourth, the modularity of many PKSs makes it feasible to tailor their substrate preferences to the available intracellular pool of CoA thioesters in a given host (48, 65, 73). A similar approach can also be used to constrain the substrate range of NRPSs, if desired. Fifth, the biosynthetic pathways for relatively uncommon substrates, such as hydroxymalonyl-CoA or 3-amino-5-hydroxybenzoic acid, are often encoded as part of PKS gene clusters that utilize these substrates (1, 75). Therefore, these auxiliary genes can also be harnessed for heterologous expression in very much the same way as the target PKS genes. Finally, since the actinomycetes are the most prolific producers of polyketides, the complete genome sequence of Streptomyces coelicolor (http://www.sanger.ac.uk /Projects/S_coelicolor/), together with its concomitant development as a heterologous host of choice for PKS gene expression, makes the task of the metabolic engineer considerably easier. Orthologs of many known precursor biosynthetic enzymes have already been identified in the genome (unpublished results), and many more can be expected to emerge as functional genomic approaches are applied to investigate the metabolome of this genetically friendly bacterium.

View larger version (26K): [in this window] [in a new window]   FIG. 8.   Polyketide synthase substrate routes. Potential substrates are boxed. (A) Enzymes performing one enzymatic conversion: 1, acetyl-CoA synthetase (alternatively, 1' represents a two-enzyme pathway, acetate kinase followed by acetylphosphotransferase); 2, acetyl-CoA carboxylase; 3, malonyl-CoA decarboxylase; 4, malonyl-CoA synthetase. (B) Enzymes performing one enzymatic conversion: 1, propionyl-CoA synthetase (1', propionate kinase followed by propionylphosphotransferase); 2, propionyl-CoA carboxylase; 3, methylmalonyl-CoA decarboxylase; 4, methylmalonyl-CoA epimerase; 5, methylmalonyl-CoA mutase; 6, isobutyryl-CoA mutase.

Given the potential cytotoxicity of most bioactive polyketides, transmembrane proteins are required for their export. Although putative export proteins (often ATP binding cassette transporter homologs) are often found associated with PKS gene clusters, very little is known about their mechanism or selectivity. For example, inactivation of the gene encoding the actinorhodin exporter in S. coelicolor leads to intracellular accumulation of this isochromanequinone antibiotic (20). However, even in a mutant lacking the entire actinorhodin gene cluster, a number of novel polyfunctional aromatic compounds have been produced and are efficiently secreted into the extracellular medium (56). While the possibility of passive diffusion of these compounds cannot be excluded, the lack of observed toxicity due to polyketide production (even in cases where antibacterial activity of the products has been demonstrated) suggests that other relatively tolerant transporter proteins might be encoded by the S. coelicolor genome. Here, too, functional genomic approaches may be useful in providing clues that will lead to the identification of these transporters.

The formation of biologically active polyketides sometimes requires the activity of various "tailoring" enzymes that act on the PKS-derived intermediate to yield the final natural product. Tailoring enzymes are evolutionarily diverse entities that commonly include cyclases, group transferases (e.g., C-, O-, and N-methyltransferases, glycosyltransferases, and acyltransferases), NADP(H)- or FAD(H)-dependent oxidoreductases, and cytochrome P450-type oxygenases. These enzymes are invariably encoded by genes adjacent to PKS genes and can therefore be readily cloned. In most cases, heterologous expression of these monofunctional enzymes is relatively straightforward. Sometimes, however, cosubstrate availability can be an issue. For example, glycosyltransferases associated with polyketide pathways often utilize specialized TDP-deoxysugars, themselves the products of multistep biosynthetic pathways (26, 66, 80). As with PKS genes, the clustering of the genes responsible for TDP-sugar biosynthesis facilitates their expression in heterologous hosts (64).

If biologically active polyketides are to be heterologously produced, a final consideration would be the need for a resistance mechanism to inhibit the effect of the natural product on the heterologous host. Fortunately again, self-resistance genes invariably exist within the natural PKS gene cluster (15, 20, 25). However, coexpression of these genes adds another level of complexity to the heterologous production of polyketides and can be particularly challenging in cases where novel polyketides are engineered for which no known resistance mechanisms have been identified.

For many reasons, S. coelicolor is an ideal host for the heterologous production of polyketides. As the first few genes encoding polyketide biosynthesis were identified (3, 5, 13, 17, 21, 78), a suitable host was desired for expressing these genes (in either wild-type or altered forms). The extensive knowledge about the biology of S. coelicolor, coupled with its membership of the Streptomyces genus, made this host a desirable choice. S. coelicolor naturally produces at least two known polyketides of its own (Fig. 2), actinorhodin (50) and the whiE spore pigment (14), although the latter is exclusively spore associated and is not produced in liquid culture. To eliminate background polyketide "noise" due to the actinorhodin pathway, a genetically engineered "clean" host strain, CH999, was constructed, in which the entire actinorhodin (act) gene cluster was surgically deleted via homologous recombination and replaced with an ermE marker gene (56). Concomitantly, a low-copy shuttle vector, pRM5, was engineered, which carries the actI-actIII bidirectional promoter together with the actII-ORF4 gene, which encodes an activator for the actI-actIII promoter (56). Introduction of heterologous PKS genes under the control of this expression system allows the production of polyketides in a secondary metabolite-like manner in a wide range of actinomycete strains including S. coelicolor (56), S. lividans (88), S. parvulus (43), and S. erythraea (72). The list of bacterial and fungal polyketides produced using this host-vector system includes products derived from the frenolicin (57), tetracenomycin (56), oxytetracycline (23), R1128 (53), erythromycin (39), picromycin/methymycin (81), oleandomycin (77), megalomicin (84), 6-methylsalicylic acid (4), and epothilone (82) gene clusters. (In some cases, for technical convenience, a "clean" derivative of S. lividans, K4-114, which also lacks the native act gene cluster, has been used as a heterologous host [88].) PKS proteins are produced at ~1% of the total cellular protein levels (67), and the titers of the resulting polyketide products are typically in the range of 1 to 100 mg/liter of culture.

The introduction of polyketide pathways into S. coelicolor has also influenced the ease with which PKSs can be genetically engineered for the production of "unnatural" natural products. Type II and modular PKSs have been particularly fertile targets of manipulation in this regard. Libraries of novel aromatic polyketides (58, 86) as well as macrolides (59) have been generated by the combinatorial manipulation of PKS domains and subunits. In turn, this has led to the emergence of heuristics for regioselective modification of polyketide structures. While considerable work lies ahead to increase the predictability of these heuristics to the level of the "codon table" for polypeptide biosynthesis, it is reasonable to anticipate that this enhanced biosynthetic capability will draw heavily from advances in heterologous expression technologies for PKS gene clusters.

A principal limitation of S. coelicolor is that its polyketide natural products (actinorhodin and the spore pigment) are exclusively synthesized from malonyl-CoA-derived building blocks. As such, its metabolic apparatus appears to be limited in the supply of other PKS substrates, such as methylmalonyl-CoA, and the productivity of heterologous polyketides derived from these substrates may suffer. Efforts are under way to better understand the pathways employed by S. coelicolor for PKS substrates. For example, genome-sequencing efforts have already led to the identification of its putative methylmalonyl-CoA mutase subunits (GenBank accession number AL138668); understanding the mechanisms by which the activity of this enzyme is regulated at both the transcriptional and posttranscriptional levels could lead to more efficient conversion of glucose into polyketides via succinyl-CoA (a key tricarboxylic acid pathway intermediate). Likewise, at least three different acyl-CoA carboxylase enzymes are encoded within the S. coelicolor genome (71). This apparent redundancy in the biosynthesis of -carboxylated CoA thioesters might suggest that different alleles have different substrate preferences (e.g., for acetyl-CoA versus propionyl-CoA) and/or regulatory features in response to the transition from primary metabolism (when only malonyl-CoA derived fatty acids are synthesized) to secondary metabolism (when polyketides are synthesized). Finally, the potential exists for the incorporation of entirely new pathways devoted to substrate production. For example, heterologous expression of the genes encoding the malonyl-CoA synthetase and dicarboxylate transporter protein from Rhizobium trifolii into S. coelicolor led to a substantial enhancement in the yield and productivity of methylmalonyl-CoA-derived erythronolide (unpublished data). Importantly, since this synthetase can activate a wide range of 1,3-dicarboxylic acids into the corresponding CoA thioesters (N. L. Pohl, Y. S. Kim, and C. Khosla, submitted for publication), it provides a potentially useful route for the intracellular generation of substrates that are ordinarily unavailable for polyketide biosynthesis.

In addition to optimizing cellular metabolism in S. coelicolor, further developments in understanding and maximizing gene expression (especially in response to the onset of secondary metabolism) are expected to have a significant impact on the utility of this host for the production of heterologous PKSs and polyketides. Of the different promoters that have been investigated thus far (which include the actI [21], tipA [62], and ermE [6] promoters), the actI promoter appears to have the most favorable characteristics with respect to both expression levels and induction at the onset of the stationary phase. However, the exact signals that lead to maximal induction of this promoter system have not yet been elucidated (54), and PKS proteins are produced only during a relatively short (ca. 24-h) window. Moreover, utilization of this expression system on high-copy-number vectors can lead to instability for reasons that are not understood, and the development of vectors that span a range of copy numbers could be valuable in this regard. Together, these features make controlled, high-level expression of PKS genes in S. coelicolor a challenging yet opportune problem.

In general, the utility of E. coli as a host for heterologous gene expression as well as metabolic engineering is unquestioned. However, in order to reconstitute polyketide biosynthesis in E. coli, several issues beyond gene expression needed to be addressed. Unlike S. coelicolor, E. coli is a significantly different heterologous host from those that naturally produce polyketide products. The robust E. coli expression systems available can induce PKS proteins to accumulate as inclusion bodies; however, the judicious control of temperature, medium composition, and other induction conditions yields substantial levels (1 to 5% of total cellular protein) of correctly folded protein for PKSs with molecular masses of >200 kDa (unpublished data). Likewise, coexpression of the sfp PPTase gene in a plasmid-borne or chromosomal format leads to stoichiometric pantetheinylation of soluble PKS proteins in E. coli (B. Pfeifer, unpublished results). Also, the high G+C content of actinomycete PKS genes leads to an inappropriate bias in codon usage that is difficult to remedy using synthetic gene approaches due to the large ORF sizes. The use of host strains that contain extra copies of the rare AGG (arginine) and CCC (proline) codons can be helpful in this regard (reference 74 and unpublished data). The availability of PKS substrates in E. coli presents another major challenge, since biosynthesis of malonyl-CoA is under tight control (40) and substrates such as propionyl-CoA and methylmalonyl-CoA are produced under poorly understood conditions (34). Reconstitution of heterologous substrate generation pathways in this host has provided a useful starting point for addressing this problem (unpublished data). Finally, the availability of well-established protocols for two-stage, high-cell-density fermentation protocols, where the growth phase can be decoupled from PKS gene expression and polyketide production, could provide rapid development of processes that yield high volumetric productivities for both PKSs and polyketides (46).

A variety of PKSs have been functionally expressed in E. coli. For example, coexpression of the sfp and 6-methylsalicylic acid synthase genes in E. coli was found to be necessary and sufficient for the intracellular production of 6-methylsalicylic acid, although polyketide production was substantially enhanced under conditions that ordinarily favor higher rates of malonyl-CoA formation (40). E. coli has the means to produce acetyl-, malonyl-, propionyl-, and possibly even methylmalonyl-CoA, and one might expect the ability to increase these levels based on pathway alteration. However, other options also exist. For example, a pathway leading to the formation of malonyl- and (2S)-methylmalonyl-CoA PKS substrates, which has been successfully reconstituted in E. coli, is based on the S. coelicolor acetyl-CoA carboxylase and propionyl-CoA carboxylase enzyme complexes that use biotin as a cofactor (unpublished data). The genes for acetyl-CoA carboxylase and propionyl-CoA carboxylase have recently been cloned and overexpressed in E. coli in an active form and have shown the ability to provide substrates for polyketide formation within E. coli (71). In summary, although E. coli presents a completely new environment for PKS expression and polyketide formation, it represents a promising opportunity for harnessing the biosynthetic capabilities of polyketide pathways. The ability to produce 6-deoxyerythronolide B in E. coli in a robust manner is an example of realizing this promise (unpublished data).

For reasons similar to those that prompted the utilization of S. coelicolor as a host for polyketide biosynthesis, other polyketide-producing actinomycetes could be considered. For example, as mentioned above, S. lividans has been successfully used to express several PKS systems, due to its close relationship to S. coelicolor and its greater transformation efficiency (which, for example, readily allows the introduction of multiple plasmids [85]). Similarly, S. glaucescens (76) and Saccharopolyspora erythraea (27) have also been tested as heterologous hosts for polyketide production. These hosts have naturally occurring polyketide biosynthetic pathways and may therefore be better suited for the expression of heterologous PKS genes. Finally, of particular interest in the development of heterologous hosts would be the potential utility of highly evolved polyketide-producing strains. For example, through multiple cycles of random mutagenesis and screening, derivatives of Saccharopolyspora erythraea have been selected that produce 8 g erythromycin per liter (which corresponds to a titer that is 50 to 100 times that in the wild-type strain) (60). Recent studies have demonstrated that the genetic basis for overproduction in such hosts is not encoded within the PKS genes themselves but within other loci on the genome (R. McDaniel, unpublished results). Given the multigenic nature of this trait, the development of such organisms as hosts for polyketide production may present an attractive option. Moreover, a systematic dissection of the genetic and physiological basis for overproduction could yield insights that might readily be translated to naive hosts such as S. coelicolor and E. coli.

Although the actinomycetes are perhaps the most prolific producers of polyketides, several other bacterial families are known to synthesize structurally diverse polyketides. Chief among these are the myxobacteria, pseudomonads, and mycobacteria. For example, myxobacteria such as Sorangium spp. produce polyketides including soraphen (31) and epothilone (30). Among other compounds, Pseudomonas spp. produce pseudomonic acid (51) and coronatine (70). Mycobacterium ulcerans produces the highly potent immunosuppressive agent mycolactone (29), whereas the genome of Mycobacterium tuberculosis appears to be well endowed with a plethora of as yet uncharacterized PKS pathways (12). Although genetically well-characterized strains representative of these bacterial families, such as Myxococcus xanthus, Pseudomonas putida, and Mycobacterium smegmatis, are known, the credentials of these bacteria as hosts for heterologous polyketide production have not yet been seriously evaluated.

Filamentous fungi too are prolific producers of polyketides. The utility of the fungus Aspergillus nidulans as a heterologous host has been successfully demonstrated in the context of lovastatin biosynthesis (41). A major advantage of such hosts would be the ability to splice introns that are frequently present among eukaryotic PKS genes. Moreover, given the well-established fermentation processes for lovastatin and compactin biosynthesis, it is reasonable to assume that highly evolved strains capable of overproducing these metabolites may be available. If so, these systems could present attractive options for reverse engineering into generic hosts for polyketide biosynthesis. Finally, given the premiere status of the yeast Saccharomyces cerevisiae in fungal and eukaryotic genetics and its designation as a GRAS host (Generally Regarded as Safe), its use as a host for polyketide production ought to be seriously investigated. Although yeast has no known polyketide biosynthetic pathway encoded within its genome, it does have a highly active fatty acid biosynthetic pathway. Moreover, recent studies have shown that it is capable of synthesizing polyketides at high levels (40), although several features must be introduced into the yeast genome before this can be generalized.