Combinatorial Biosynthesis of Polyketides – A Perspective

Abstract

Since their discovery, polyketide synthases have been attractive targets of biosynthetic engineering to make “unnatural” natural products. Although combinatorial biosynthesis has made encouraging advances over the past two decades, the field remains in its infancy. In this enzyme-centric perspective, we discuss the scientific and technological challenges that could accelerate the adoption of combinatorial biosynthesis as a method of choice for the preparation of encoded libraries of bioactive small molecules. Borrowing a page from the protein structure prediction community, we propose a periodic challenge program to vet the most promising methods in the field, and to foster the collective development of useful tools and algorithms.

Introduction

Polyketides are a structurally diverse but biosynthetically related family of natural products that includes a number of medicinally important substances such as lovastatin (a cholesterol-lowering agent), erythromycin (an antibiotic), and FK506 (an immunosuppressant) [1]. Their structural and stereochemical complexity makes systematic chemical manipulation a formidable undertaking. Consequently, there has been considerable interest in the potential of harnessing combinatorial biosynthesis to introduce novel functionality into these bioactive compounds and to produce altogether new chemotypes.

In this review, combinatorial biosynthesis is defined as the genetic manipulation of two or more enzymes involved in polyketide biosynthesis. According to this enzyme-centric definition, combinatorial biosynthesis could even yield the natural product itself, as long as the corresponding polyketide synthase (PKS) harbors two or more genetically modified enzymes. These enzymatic modifications can be accomplished by either genetic manipulation of the original enzyme or by replacing it with a homolog (although the latter approach is more common at the present time). By contrast, a product-centric definition of combinatorial biosynthesis would encompass natural product analogs with two or more functional group transformations, regardless of how these modifications are achieved. For example, products of combinatorial biosynthesis could be derived via precursor directed biosynthesis or through other metabolic engineering strategies [2]. We have chosen an enzyme-centric definition because, in our opinion, it highlights the fundamental and technological challenges to exploiting the functional modularity of PKSs [3]. Specifically, combinatorial biosynthesis can be achieved by manipulating enzymes responsible for primer unit incorporation, chain elongation, and chain termination.

The State of the Art

The feasibility of combinatorial biosynthesis has been demonstrated in the context of different types of multifunctional PKSs [4-8]. Whereas the architectures of these PKS sub-families are variable, all multifunctional PKSs harbor one or more ketosynthases (KS), acyltransferases (AT) and acyl carrier proteins (ACP). In addition, most PKSs also include auxiliary enzymes such as reductases, dehydratases, transferases, cyclases, and thioesterases. To highlight the scope of combinatorial biosynthesis, here we primarily focus on multimodular PKSs with assembly line architectures. Like an automobile assembly line, these PKSs have multiple way stations (called “modules”), each of which harbors distinct protein domains. Except in a few rare cases, each module is deployed only once in the PKS catalytic cycle; this one-to-one correspondence facilitates convenient mapping of each enzyme domain in the PKS to a unique reaction in the polyketide biosynthetic pathway. A prototypical example of this PKS sub-family is the 6-deoxyerythronolide B synthase (DEBS) (Figure 1) [9]. A particularly impressive showcase for the enzymatic complexity of assembly line PKSs is the FR901464 biosynthetic synthase. (Figure 2) [10]. FR901464 is synthesized by an assembly line encompassing a PKS with several atypical architectural and enzymatic features, including a nonribosomal peptide synthetase and an HMG-CoA reductase.

Figure 1.

The 6-deoxyerythronolide B synthase (3 genes, 32 kbp) is a canonical multimodular PKS [9]. The growing chain is shown as it moves down the assembly line. N and C terminal linkers are also shown. KS: ketosynthase, AT: acyltransferase, DH: dehydratase, ER: enoyl reductase, KR: ketoredutase, ACP: acyl carrier protein. The ketoreductase domain in module 3 is inactive (shown in lower caps).

Figure 2.

Modules in biosynthetic pathway of FR901464. FR901464 synthase is encoded by 20 ORFs spanning 93 kb. 3 ORFs encoding accessory tailoring proteins are not shown here [10]. The growing polyketide chain is shown attached to individual ACP domains. Acronyms in lower case refer to non-functional domains. The trans-acting AT acts on eight modules, modules 1 and 3-9. KS: ketosynthase, AT: acyltransferase, DH: dehydratase, ER: enoyl reductase, KR: ketoredutase, ACP: acyl carrier protein. GAT: glyceryl transferase/phosphatase, ECH: enoyl-CoA reductase OX: FAD-dependent monooxygenase, MT: methyltransferase, PCP: peptidyl carrier protein, C: condensation, A: adenylation.

Phylogenetic and structural analysis of assembly line PKSs suggests that nature has harnessed gene duplication, mutation, and recombination to pursue combinatorial biosynthesis over evolutionary time [11]. In a presumably analogous laboratory investigation, ca. 50 analogs of 6-deoxyerythronolide B were produced by engineering two or more domains of DEBS [5]. The latter study was enabled by the establishment of tools for the reconstitution of complete PKS pathways into genetically amenable hosts such as Streptomyces coelicolor [12, 13].

Notwithstanding encouraging progress over the past two decades [7, 14-17], the promise of rationally guided combinatorial biosynthesis remains unrealized. In the sections that follow, we discuss key ecological, enzymological, and technological challenges that must be addressed in order to efficiently synthesize libraries of “unnatural” natural products.

Ecological challenges

Until recently, a major obstacle to combinatorial biosynthesis was the availability of DNA sequences of an adequately large number of cloned PKS genes. Less than 20 multifunctional PKS gene clusters had been fully sequenced by the turn of the millennium. As high-throughput sequencing techniques gained momentum, this number increased exponentially. Whereas the growth in PKSs corresponding to structurally characterized natural products has remained modest, the emergence of whole genome sequencing methods has resulted in the discovery of cryptic gene clusters at an explosive pace (Figure 3). Not only has there been an immense growth in the repertoire of enzyme domains and modules, but new assembly line architectures have also been discovered (e.g., “AT-less” PKSs [18, 19]). Today, an aspiring biosynthetic engineer has access to a virtually infinite palette of genetic raw material, although much of it remains to be functionally decoded.

Figure 3.

Growth in the number of cloned and sequenced multimodular PKS gene clusters. The cumulative increases over the past decade in the number of orphan PKS gene clusters and structurally characterized polyketides are shown. Data was obtained by calculating the number of “polyketide synthase” entries published each year in the nucleotides database (pubmed, URL: http://www.ncbi.nlm.nih.gov/nuccore). The entries were manually screened and sorted into orphan versus characterized PKSs.

Notwithstanding breathtaking advances in mining nature's PKS gene clusters [20], the ability to identify complete PKSs from unculturable microorganisms remains seriously constrained. The development of resource-efficient strategies for cloning and sequencing large (20-100 kb) contigs from metagenomic sources will enable at least two related types of opportunities in combinatorial biosynthesis. First, the DNA encoding unprecedented chemotypes could become accessible. For example, close structural analogs of marine natural products such as discodermolide [21] or spongistatin [22] have not yet been isolated from cultured microorganisms. Combinatorial biosynthesis of discodermolide or spongistatin analogs is therefore predicated upon cloning their complete gene clusters. Second, thus far, the vast majority of cloned PKS genes have been isolated from terrestrial bacteria, primarily the actinomycetes, bacilli and myxobacteria. As the genetic content of the earth's oceans is mined for PKSs, new biocatalytic strategies will surely emerge, which in turn could be exploited through combinatorial biosynthesis. For example, an enzyme that catalyzes a Favorskii rearrangement was found in the enterocin biosynthetic pathway from a marine actinomycete [23]. This has led to the engineering of new types of polyketide analogs.

Enzymological challenges

At its core, combinatorial biosynthesis of assembly line PKSs is an exercise in enzyme engineering that rests upon two crucial assumptions. First, individual enzymes along the assembly line must have relaxed substrate specificity. Second, the mechanisms that promote channeling of biosynthetic intermediates from one enzyme to the next must be sufficiently conserved in order to permit the engineering of chimeric assembly lines. Available evidence suggests that both hypotheses are plausible, but lack thorough validation. In the remainder of this section, we review the experimental evidence supporting these hypotheses.

At least three different lines of evidence can be cited in support of the hypothesis that PKS enzymes have broad tolerance for the growing polyketide chain supplied to them. First the substrate specificity of a few PKS modules has been quantified, and is known to be relatively modest (i.e., many substrate analogs have k_cat/K_M values within 10-100 fold that of the natural substrate) [24, 25]. Therefore, unless the relevant reaction is a major rate-limiting step in the biosynthetic pathway, a structurally altered biosynthetic intermediate should be well tolerated. Second, precursor directed biosynthesis has been used to convert unnatural primer units or diketides of a number of natural PKSs into the corresponding polyketide analogs [26-28]. Third and perhaps most intriguingly, successive modules of certain PKSs, such as the mycolactone synthase [29], show exceptionally high conservation (>90% identity) in KS and ACP domain sequences, suggesting that module duplication may have been sufficient for the evolution of long, variably functionalized polyketide backbones.

Available structural models for DEBS suggest that the growing polyketide chain is channeled across extraordinary lengths (50-100 nm) before the product is released. In contrast to model systems such as tryptophan synthase and carbamoyl phosphate synthetase, where intermediates are channeled through relatively short (1-10 nm) mostly buried tunnels [30, 31], PKSs rely on selective and dynamic protein-protein interactions. For example, at least three types of protein-protein interactions have been identified in DEBS that strongly influence unidirectional movement of the growing polyketide chain. As shown in Figure 1, the N/C-terminal linkers flanking the three DEBS proteins dock together to constrain the growing chain to a path where module 3 follows module 2 and module 5 follows module 4 [32, 33]. At the same time, the ACP domain of each module preferentially docks onto the downstream module of the assembly line, thus ensuring that the growing chain does not iteratively pass through any module more than once every catalytic cycle [34]. Last but not least, chain elongation by each module also relies on selective intramodular domain-domain interactions [35]. The design of catalytically active PKS chimeras requires each type of non-covalent interaction to be preserved.

Although some progress has been made towards understanding the structure and mechanism of assembly line PKSs, the list of major unanswered questions is considerably longer. Foremost amongst them is the elucidation of PKS structural dynamics. The occurrence of large conformational changes in a multimodular PKS, as it progresses through its catalytic cycle, is all but certain. However, atomic-level insights into this remarkable phenomenon are limited. Other fundamental challenges include the development of broadly applicable experimental strategies to identify rate-limiting steps in natural or engineered PKS assembly lines, and the contribution of intramolecularity to the reaction rates on the assembly line. Last but not least, decoding the function of orphan gene clusters is becoming feasible, as relationships between PKS structure and function become clearer [36]. Deeper insights into PKS enzymology could eventually enable complete automation of this capability.

Technological challenges

In addition to the above challenges in basic biology and chemistry, the toolbox for combinatorial manipulation of PKSs must also be improved. The rapid growth in PKS sequences (Figure 3) will likely accelerate further, as methods for automated assembly of genome-sized contigs from GC-rich organisms are improved. Together with the rapidly decreasing cost of oligonucleotides, this could enable assembly of expression constructs for full-length (i.e. 30-100 kb) natural or modified PKS genes [37]. Of course, heterologous hosts capable of functionally expressing these PKS pathways must be available. Whereas no single host is capable of expressing all types of PKS pathways, Escherichia coli and hosts such as Streptomyces coelicolor and its close relative Streptomyces lividans appear to have a broad scope for this purpose. The key remaining challenge is to improve the polyketide productivity of these hosts. Our own experience suggests that productivity in heterologous hosts is most often not limited by the PKS itself, because specific polyketide productivity is higher in native hosts, even though PKS protein levels are higher in the heterologous host. This is an important challenge for the metabolic engineer [38].

Along with improved methods for PKS cloning and expression, superior methods for detecting and characterizing polyketide products are also needed. Advances in microscale NMR spectroscopy already allow structure elucidation of new natural products with as little as a few nanomoles, thus reducing the sample size by 2-3 orders of magnitude [39]. Similarly, new techniques for ionization and detection of natural products via mass spectrometry open the door to the characterization of very small samples [39]. In both approaches, the problem of contaminants can be particularly vexing. Therefore, robust workflows need to be developed for analyzing trace quantities of a new compound made by an engineered bacterium. Here too, the use of heterologous hosts is an advantage, as isotope-tagging methods could be implemented to differentiate polyketide products from background contaminants.

For combinatorial biosynthesis to gain widespread acceptance as a method for producing small molecule libraries, automation at all levels of experimental design is essential. In recent years, programs such as antiSMASH [40] and Clustscan [41] have been developed to accurately annotate PKSs by identifying domains and even predicting the structures of their natural products. Large databases of sequenced PKSs could also spawn new evolutionary strategies for designing assembly lines with unnatural specificity [11, 42], Perhaps most intriguingly, newer computational tools (such as SBSPKS [43], [44]) are attempting to perform structure-based analysis of PKSs [45]. With further development and installation of appropriate user interfaces, these programs could eventually serve as portals for combinatorial library design.

Combinatorial biosynthesis: A call to action

After nearly two decades, combinatorial biosynthesis remains in its infancy. In the future, large-scale combinatorial biosynthesis will require a catalog of validated domains, didomains, modules, and linkers capable of performing the spectrum of catalytic chemistry that is observed in nature's assembly lines. Their salient characteristics will be well documented in order to allow the engineer to rationally choose the right components for rationally designing a chimeric assembly line. The catalog will be accompanied by an instruction manual, and reference data from a set of control experiments.

As is perhaps obvious to any natural products biosynthetic chemist, this is an ambitious undertaking. How might one get from here to there? We suggest borrowing a page from the protein structure prediction community and its longstanding CASP challenge program intended to advance automated methods for protein structure prediction. Started in 1994, CASP is approaching its tenth biennial competition, and has been a major driving force for the emergence of the most widely used methods for structure prediction and homology modeling [46]. Importantly, efforts such as CASP have not only fostered useful tools for protein engineers, but have also facilitated scientific progress in the field. An analogous format for combinatorial biosynthesis could be contemplated. In each competition cycle, two (or more) well-defined problems would be presented to the community, and all resources (DNA, vectors, hosts, sequences, reference compounds, etc.) would be made available to labs wishing to participate in the competition. At least two types of problems are envisioned in each competition round;

• The first set would require prediction of the product structure of an orphan PKS, where the actual structure is known but not yet published.

• The second set would require engineering a PKS that produces the highest titer of a target synthon in a defined heterologous host.

The best solutions will not only be widely publicized, but are also likely to gain rapid acceptance as benchmarks for next-generation challenges. We welcome suggestions regarding how such an approach might be further tailored to catalyze rapid advances in combinatorial biosynthesis.