Introduction
For more than half of the twentieth century, natural products formed a central pillar of the modern pharmaceutical industry. For example, 49% of the new chemical entities introduced into clinical use between 1981 and 2002 were of natural-product origin or inspiration, with the number rising to around 75% when applied to those drugs used for the treatment of severe and life-threatening indications1. The diversity of natural-product structure and activity has also made them indispensable tools for deciphering cellular processes. This is particularly the case with the advent of chemical genetics approaches, in which, increasingly, libraries of small molecules, including natural products, are screened to identify ligands that can modulate a specific function of a gene product in vivo, which need not necessarily be a target for therapeutic intervention2. Despite this, natural-product research was de-emphasized by the industry during the 1980s and 1990s. Increasingly, natural products were perceived as an obsolete technology, unable to compete with the speed and efficiency of combinatorial chemical library synthesis and screening. In response to this challenge, natural-product scientists were compelled to obtain more value from their efforts; this 'evolving role' for natural-products drug discovery has been reviewed by Koehn and Carter3.
The rich functionality of natural products is without doubt one of their great strengths, providing potency and selectivity. However, difficulties associated with their synthesis and semi-synthesis can present a barrier to their acceptance as leads for optimization. Nevertheless, the ability to rationally alter natural-product structures through the genetic modification of their biosynthetic machinery represents a potential solution to this problem. The application of biosynthetic engineering to this end has become an intensive area of research, driven by our increasing understanding of natural-product biosynthesis coupled with advances in molecular genetics, especially by work with the prolific Streptomyces species. These efforts have led to the production of many 'non-natural' natural products, and such work is often reported under the term 'combinatorial biosynthesis', indicating that the biosynthetic machinery of more than one pathway is combined to generate hybrid products4. It can also imply a multiplicative approach akin to combinatorial chemistry, especially if precursor-directed biosynthesis forms a part of the process. Engineering approaches have also been used to great effect for optimizing the production of a single key compound by an amenable heterologous host in what is better classed as metabolic engineering5. In addition, advances in structural biology have further facilitated the development and application of protein-engineering methods for altering the specificity of biosynthetic enzymes, which powerfully complement the combinatorial approaches. Here we demonstrate how the application of molecular genetics approaches and other technical advances have continued to invigorate the discovery, use and perception of natural products in drug discovery as well as basic biochemical research. In addition, we suggest ways by which recent advances in biosynthetic engineering may be best integrated into the discovery platform.
Mining natural-product diversity
Recently, there has been a drive to improve screening technologies that allow access to, and identification of, new chemical diversity from natural sources. Traditional screening campaigns would routinely identify known structures, giving an impression that the 'low-hanging fruit' had already been picked and that the resource was spent. The strategy of phenotypic screening and bioassay-guided fractionation of natural-product extracts, which traditionally formed the basis for the discovery of most natural-product pharmaceuticals, has become increasingly sidelined with the advent of high-throughput screening (HTS). A new twist on this approach was demonstrated recently with the successful 'target-biased' phenotypic screening of natural-product extracts in the discovery of platensimycin6. Platensimycin has an unprecedented carbon framework that prompted intense speculation as to its biosynthetic origins (Fig. 1). Platensimycin is effective against important drug-resistant pathogens and has a hitherto underexploited molecular mode of action, inhibiting lipid biosynthesis by targeting the fatty acid elongation-condensing enzyme FabF. The target-biased screen depends on the use of the antisense knockdown of fabF to increase, and the overexpression to of fabF to decrease, sensitivity to FabF inhibition. The power of this approach was exemplified when the authors subsequently found that platensimycin is produced by numerous actinomycetes, many of which had been screened numerous times previously.
Indeed, with ever-increasing amounts of genome sequence data available, it has become clear that those compounds that have been identified and isolated represent only the tip of the iceberg and that there is a wealth of natural-product chemistry still to be mined. The published genome sequences of several important actinomycetes indicate that as much as 90% of the chemical potential of these organisms remains undiscovered7, 8. For whatever reason, the compounds encoded by many of these genetic loci are not produced under laboratory conditions, even though the biosynthetic machinery appears functional. As we will describe, it appears that some chemical or environmental signal that is required for their expression cannot be replicated. Genome mining thus offers a powerful new method for tapping into this cryptic natural-product diversity, providing new biosynthetic mechanisms and 'genetic hardware'.
Advances in DNA sequencing technology will prove increasingly important, making genome-mining enterprises faster and more cost effective9. This was made startlingly clear by the recent discovery of the moenomycin biosynthetic gene cluster, whose identification was not initially possible using assumptions based on sequence similarity to other related glycosylated secondary metabolites, but yielded rapidly to shotgun sequencing of the S. ghanaensis genome10. Many other important applications of genome mining have been reported during the last few years, clearing up several intriguing biosynthetic puzzles in the process11. Notably, the ability to directly correlate primary genetic information with protein function and thus chemical structure is now relatively straightforward for certain biosynthetic paradigms such as modular thiotemplate-derived polyketide synthase (PKS)12 and nonribosomal peptide synthase (NRPS) products13, thus allowing a decision to be made concerning their value before expression and isolation efforts. The genomic-guided approach is also useful for identifying genetic loci that may encode products with new biosynthetic potential14. This was the case for ECO-04601, a farnesylated dibenzodiazopine discovered after scanning the genome of a Micromonospora species. ECO-04601 (Fig. 1) is an apoptosis inducer now being tested in phase 1 clinical trials against solid tumors15. Very recently, Hertweck and co-workers have uncovered new PKS-NRPS hybrids from Aspergillus nidulans through genomics-driven discovery16. The expression of a silent gene cluster was induced after the ectopic expression of a specific regulatory gene. Interestingly, the resulting natural product contained a pyridone moiety, which could not have been anticipated from primary sequence analysis (Fig. 1). This pragmatic method obviates the need for heterologous gene expression and uses readily available genetic tools to induce the production of new chemical species. Particularly exciting is the prospect of applying chemical genetic studies to this and other systems using promoter-reporter methods to identify the chemical signals required to induce native expression of such gene clusters. Indeed, it is tempting to suggest that there are likely to be a far larger number of signaling molecules capable of inducing gene expression in microorganisms than the limited number of molecules that have so far been implicated in quorum sensing17. Similarly exciting is the potential chemical diversity encoded within the microbial metagenome. As various estimates suggest that less than 1% of microbial diversity has so far been sampled through laboratory-cultured organisms18, this approach offers great potential, and several biotechnology companies have formed to exploit these possibilities. Technical issues still limit the full exploitation of this resource. For example, a single or even a few heterologous expression hosts are unlikely to be sufficient for expression of all the DNA, and problems remain regarding regulation, genetic compatibility and the screening power required to leverage the numbers of clones available. Despite this, several successful examples of environmental DNA (eDNA) expression in heterologous hosts have been reported, leading to the identification of several new natural products19, 20.
The engineer's tool kit
Many of the organisms that produce natural products are slow-growing or genetically intractable, and it is thus desirable to establish expression of their entire biosynthetic gene clusters in a more suitable host. Not only can this allow increased production levels, but it can also afford the opportunity to manipulate the biosynthetic genes more rapidly to produce new products. After earlier successes with heterologous expression of PKS genes in Escherichia coli21, the production of several further polyketide and peptide products has been established. Notable are the heterologous biosynthesis of echinomycin, a nonribosomal peptide with antitumor activity22, and the patellamides, ribosomally encoded peptides from an ascidian obligate (unculturable) symbiont (Fig. 2)23, 24. The use of E. coli is advantageous, as its cell-doubling times are much shorter and more efficient fermentation methods can be used. However, problems associated with metabolic background and limiting titers in E. coli mean that there is a need for alternative hosts that are more amenable to heterologous expression of a wide range of secondary metabolic gene clusters. One alternative may be Pseudomonas species, for which a comprehensive set of genetic tools exists. Pseudomonas species grow rapidly, have been validated as exceptional protein-production hosts and are prolific producers of natural products.
The technical requirements for engineering the large and often repetitive pieces of DNA involved in biosynthetic engineering experiments are challenging. To address this, the Red/ET recombineering approach in E. coli represents a technological advance that significantly facilitates DNA manipulation efforts as it minimizes restriction enzyme cutting-ligation requirements25. Notably, the myxochromide S (Fig. 2) biosynthetic gene cluster from Stigmatella aurantiaca was reassembled in E. coli, which required the 'stitching together' of >50 kbp of DNA from multiple cosmids and involved specific promoter exchanges before expression in Pseudomonas putida25. This led to levels of production of the hybrid polyketide–nonribosomal peptide myxochromide S that were significantly greater than those for the native organism, and in a shorter time. This methodology has rapidly become a method of choice for engineering experiments. For example, it was used to introduce an operon required for methylmalonyl-CoA biosynthesis into P. putida26, which will enable future efforts to establish heterologous expression and engineered biosynthesis of polyketides (and hybrid products) requiring methylmalonyl-CoA.
The speed of gene synthesis, coupled with its decreasing cost, has made synthetic genes an invaluable resource in biosynthetic engineering27. The ability to introduce tailored restriction sites and optimize codon usage for the host organism makes the manipulation of synthetic genes far more efficient than for wild-type genes. A powerful demonstration of this technique came with the synthesis of the entire set of erythromycin PKS genes (DEBS1-3), which represent a contiguous 32 kb of sequence, and their expression in E. coli28. This approach was extended to generate genes encoding 14 type I PKS modules with differing N- and C-terminal linker variations, allowing them to be assembled into 154 possible bimodular combinations encompassing more than 1.5 mbp of sequence29 (Fig. 3). These were expressed in E. coli, and 72 of the 154 combinations produced the anticipated polyketide product, the majority in abundances in the range of 0.02–10 mgl- 1. Although the ability to produce combinatorial PKS libraries in this way represents a major step in biosynthetic engineering, its extension to building larger assembly lines for the de novo production of tailored polyketides remains a formidable challenge. Multimodular PKSs rely not just on the juxtaposition of modules bearing the correct domain architecture but also on complex kinetic and molecular-recognition mechanisms. This has been brought into focus with the realization that portions of many modular PKSs can themselves operate iteratively and with complete fidelity, and attempts to insert additional extension modules into existing PKS frameworks can be accompanied with their functional 'skipping' while the original PKS machinery functions normally and leads to synthesis of the original product30. Moreover, it is evident that both PKS and NRPS assembly lines employ multiple proofreading mechanisms, which can reduce or block production of engineered multienzymes. In addition to type II thioesterases (TEs), which can hydrolyze misprimed carrier proteins31, 32, the specificity of domains within modular assembly lines can prevent the extension of noncognate substrates, which leads to premature hydrolytic release of intermediates33.
Figure 3: Combinatorial assembly of PKS modules through synthetic biology.
Building blocks comprising loading module (LMery), intrapeptide linker (LI), N-terminal interpeptide linker (LN), synthetic donor and acceptor (extender) modules, C-terminal interpeptide linker (LC) and thioesterase (TEery). The LM and TE are derived from the erythromycin PKS (ery). The restriction sites allowing their ready assembly are annotated. All of the genes are introduced into the E. coli expression host on separate and compatible plasmids, and typical examples of a final protein arrangement and chemical (triketide lactone, TKL) product are shown below. AT, acyltransferase; ACP, acyl carrier protein; KS, 
-ketoacylsynthase; KR, -ketoacylreductase. Modified with permission from ref. 29.
In the broader field of enzymology, the advent of directed evolution has allowed the selection of redesigned enzymes with improved properties, altered substrate specificity and, in some cases, new catalytic function34. Despite this, the difficulties of designing assays capable of screening the large numbers of mutants required for the directed evolution of enzymes involved in biosynthesis is often all too limiting. An alternative and illuminating approach is that of introducing specificity into the highly promiscuous enzymes. Using the terpene cyclase 
-humulene synthase as an example35, it was suggested that promiscuous enzymes evolve divergently to acquire higher specificity and activity through the mutation of a few key 'plasticity residues' located in and around the active site. By homology modeling, based on the solved crystal structure of a close relative, 19 potential plasticity residues were identified in -humulene synthase and subjected to saturation mutagenesis. A subset of these (four) were identified as catalytically significant, affecting product ratios but not (significantly) overall productivity. The effects of recombining variations of these mutations was systematically analyzed using a mathematical model, allowing the authors to preselect potentially valuable combinations of amino acid substitutions for the production of specific products. From this work, seven highly active synthases were identified, targeted for investigation and subsequently shown to operate through distinct reaction pathways to give specific products. This required the screening of only 2,500 mutants, a number that can be quite adequately addressed through the detailed assays used. Clearly, the application of a mathematical approach to preselect combinations of mutations likely to give desired outcomes is extremely powerful and could be harnessed, along with related methods, as a means of improving the activity of specific catalytic domains when used 'out of context' in an engineered system.
Structural insights for biosynthetic machinery
It has been more than a decade since pioneering studies demonstrated how modular PKS and NRPS enzymes could be reprogrammed36, 37, 38. Typical efforts to engineer these biosynthetic assembly lines are based on homology comparisons and predictions of secondary and tertiary structure based on sequence alignments, and on comparison to structures for related enzymes, notably fatty acid synthases (FAS). With this in mind, the successes obtained are remarkable, although the productivity of resulting hybrid enzymes has been variable.
Our lack of insight into the organization, interactions and structures of domains within intact multienzyme ensembles has no doubt been one of the main reasons for the often low product yields associated with reprogramming PKS and NRPS in vivo. In this respect, the recent X-ray crystallographic structures of the modular type I mammalian and fungal FAS offer a tantalizing, albeit low-resolution, glimpse of how we might expect PKS and NRPS assembly lines to be organized39, 40. Notable for its absence from these structures, and also from the recently described high-resolution structure of a modular PKS didomain comprising the ketosynthase and acyltransferase domains41, is the key acyl carrier protein (ACP). This is most likely due to the highly dynamic nature of the entire multienzyme and, in particular, the intrinsic mobility of the ACP, which must 'shuttle' significant distances to engage, through protein-protein interactions, with the other catalytic domains. In fact, the more recent 3.1-Å structure of the Thermomyces lanuginosus FAS indicates that the ACP is connected through flexible polypeptide linkers to two anchor points; this connection allows the ACP to rotate around the catalytic domains in a circular motion42 (Fig. 4). For both modular and nonmodular multienzyme ensembles, including type II PKS and other in-trans activities, an understanding of the molecular recognition processes that guide the 'shuttling' carrier proteins will prove essential in optimizing the activity of engineered biosynthesis.
Figure 4: The fungal FAS is another biomolecular motor.
(a) The central wheel of the 
66 heterododecameric FAS from T. lanuginosus. (b) The fungal FAS contains two reaction chambers, one above and the other below the central wheel, each containing three sets of active sites. The ACP domain (shown in blue) encircles the catalytic domains, rotating about two fixed anchor points through flexible peptide linkers. Domains from the -chains are shown in pink and blue, and domains from the -chains are shown in green, yellow and gray. (c) The circular motion of the ACP perpendicular to the vector connecting the anchor points resembles the motion of the 'flyballs' in the centrifugal governor designed by James Watt in 1788, which was used to control the speed of a steam engine. In this sense at least the fungal FAS could be considered to be a biomolecular motor, in which the negative free energy of the reactions catalyzed by the FAS drives the motion of the ACP in a fashion similar to the way that ATP hydrolysis provides the mechanical driving force for the ATPase and DNA helicase motors. Panels a and b from ref. 42. Reprinted with permission from AAAS. Panel c adapted with permission from Encyclopædia Britannica, copyright 1996 by Encyclopædia Brittanica, Inc.
In addition to crystallography, mutagenesis and NMR studies with PKS43 and NRPS44, 45 enzymes have helped to define the protein-protein interfaces and conformational changes that occur when carrier proteins interact with catalytic domains within the assembly line or with external enzymes that are essential for function. Nevertheless, for the unrestrained redesign of these enzymes through the creation of new modular and domain fusions to be perfected, it would be desirable to possess the elusive high-resolution structures of carrier-protein domains engaged with other catalytic domains. One way to do this may be to tether mechanism-based inhibitors to carrier proteins through the natural phosphopantetheine chain using synthetic chemistry46 or the promiscuous phosphopantetheinyl transferase Sfp and related enzymes47. In this way, the carrier protein–tethered inhibitor can bind with higher affinity or irreversibly cross-link to the active site of a neighboring catalytic domain or another enzyme. Provided that the geometry and distance of the inhibitor from the carrier protein are correct, this should stabilize complexes between carrier proteins and other domains or enzymes, increasing the probability of obtaining meaningful structures. To this end, new classes of inhibitors for each of the main catalytic domains within PKS and NRPS will undoubtedly also prove useful.
Reprogramming of natural-product biosynthesis
The reprogramming of modular PKS and NRPS enzymes has been a primary focus in efforts to engineer the combinatorial biosynthesis of natural products. For example, early successes in exchanging or replacing modules within NRPS systems36 offered great promise for the biosynthesis of new peptide natural products. Despite this, it was not until recently that scientists at Cubist were able to show how this approach can be used for the practical in vivo combinatorial biosynthesis of acidic lipopeptides of therapeutic value48, 49. Allied to this, the use of recombinant NRPS TE domains to catalyze regiospecific cyclization of synthetic peptides is proving to be a powerful complementary method with which to generate libraries of lipopeptides and related structures with improved activity50. Having selected interesting peptide leads, one could then use the in vivo genetic approach to engineer strains to produce sufficient quantities of lead molecules for clinical studies48, 49. One provocative alternative to re-engineering existing NRPSs is the de novo design of a new catalyst that mimics them. In this respect, a supramolecular approach based upon coiled-coil peptide mimics of NRPS active sites was recently introduced51, 52. Although it remains to be seen whether such catalysts can effect programmed peptide assembly, unquestionably such studies will lead to a greater appreciation of nature's prowess in effecting such catalysis.
Nonribosomal peptide extension involves condensation reactions with hundreds of proteinogenic and unusual amino acids. On the other hand, polyketide extension reactions are mainly limited to condensations involving methyl-, ethyl-, and methoxymalonyl-CoA as well as malonate units. As a result, the recent clarification of the role of hydroxymalonyl-ACP and aminomalonyl-ACP extender units involved in the biosynthesis of zwittermicin A53 offers significant further scope for the biosynthesis engineering of type I polyketides and hybrid PKS-NRPS products (Fig. 5). The identification of amino acid motifs responsible for the selectivity within AT domains has also been reported54, 55 and provides the ability to use site-directed mutagenesis rather than domain replacements to generate engineered PKSs and new products, although published efforts to date have provided only product mixtures (derived from both the original and engineered domain substrates), indicating the production of more promiscuous enzymes in terms of selectivity. More recent studies have also revealed the molecular basis by which ketoreductase (KR) domains control the configuration of the alcohol group formed from reduction of -ketoacyl-ACP intermediates during type I PKS assembly56. This valuable insight should prove important in engineering new polyketides with altered stereochemistry.
In addition to this, mutasynthesis or precursor-directed biosynthesis, which involves the combination of synthetic chemistry and targeted genetic mutation, remains an important route for producing focused libraries across several product classes57, 58. In many cases, this yields compounds with rather conservative structural changes, although more diverse libraries such as that for the coumarin family have also been reported59.
Many natural products are decorated with carbohydrates that are essential for targeting them to their biological receptors. In light of this, there has been increasing interest in understanding the pathways for their biosynthesis and in manipulating the substrate range and tolerances of the glycosyltransferases, which attach the sugars to their aglycone substrates60. This has led to the development of powerful methods for manipulating sugar biosynthesis both in vitro and, particularly, in vivo, where the utility of sugar biosynthetic gene cassettes for NDP-deoxysugar production in heterologous Streptomyces hosts has been clearly demonstrated61. Theoretically, once in hand, any glycosyltransferase can be expressed in these 'cell factories' and the appropriate aglycone added and diversified, with the main limitation being the substrate flexibility of the glycosyltransferase. So far, most glycosyltransferases examined are reasonably tolerant of substrate structural variation and are thus useful tools for compound glycodiversification. Further, the evolution of glycosyltransferase function has been reported for two such enzymes involved in biosynthesis of the aromatic polyketide urdamycin62. These efforts produced several efficient glycosyltransferases, some with a switch to the other parental activity and others with promiscuous, but efficient, activity functioning with both parent activities. Most interesting was the production of mutants showing a new activity capable of forming a branched saccharide chain. The surprising ability of glycosyltransferases to act in a reversible manner has also been demonstrated, indicating their potential for effecting in vitro transglycosylation as a method for structural diversification. In addition, this method offers a straightforward and direct route for generating unusual NDP-deoxysugars from readily available natural-product sources63. Indeed, by combining these ideas, a library of over 70 calicheamicin analogs has been produced63, a task extremely difficult to achieve by pre-existing biochemical or synthetic methods (Fig. 6).
Figure 6: General two-step strategy for the transglycosylation of calicheamicin using the native glycosyltransferase CalG1.
CalG1 transfers the native deoxysugar to TDP and then replaces it with an exogenous analog. The TDP-deoxysugars for transfer can be generated by synthesis or from other natural products through alternative transglycosylation reactions. This approach has been used to produce more than 70 calicheamicin analogs that are extremely difficult to prepare by chemical synthesis alone.
Full size image (32 KB)Much of our discussion has focused on microbial products and neglected the plant kingdom. Although the potential of plants as a source of natural products is well established, efforts to control and redirect plant biosynthetic pathways have been limited. To this end, recent notable studies include the terpene synthase engineering discussed above and the redirected biosynthesis of a modified terpene indole alkaloid in the Madagascar periwinkle (Catharanthus roseus)64, using both plant seedlings and hairy root cultures. The same group of researchers has also altered the substrate specificity of strictosidine synthase, the central enzyme in terpene indole alkaloid biosynthesis, to accept substrate analogs leading to new alkaloids65. Unlike bacteria, the genes encoding enzymes for secondary metabolism in plants are not clustered and as such are more difficult to manipulate by conventional genetic tools. For this reason, the heterologous expression of plant biosynthetic genes in more tractable fungal or bacterial hosts is essential. The recombineering and gene-synthesis methods we have described should expedite these efforts. A notable example of success here is the production of the antimalarial drug precursor artemisinic acid in an engineered yeast strain (Fig. 7)5. In addition, unnatural protein fusions have been used to facilitate the engineering of resveratrol biosynthesis in both yeast and mammalian cells66. Such technology may also have applications for medical research; for example, it is suggested that resveratrol can prolong the life span of human donor cells, improving cell-replacement therapies.
Figure 7: Production of the antimalarial drug precursor artemisinic acid in engineered yeast and its chemical conversion to the active species artemisinin.
A Saccharomyces cerevisiae strain with attenuated sterol production was engineered to produce dimethylallyl pyrophosphate (DMAPP) and isopentenyl pyrophosphate (IPP) through the mevanolate pathway (Eng MP), which was converted to farnesyl pyrophosphate (FPP): this was cyclized by amorphadiene cyclase (ADS) with subsequent oxidation by the cytochrome P450 monooxygenase CYP71AV1 to yield the key intermediate artemisinic acid at titers of 100 mgl- 1. This can be readily converted to the active species artemisinin by two chemical steps in approximately 40% yield.
Full size image (27 KB)Perspective
In the broad sense, mining provides the raw materials, whereas new expression systems and biosynthetic engineering methodologies serve as the tool kit with which to increase structural diversity and to facilitate and enhance the production of key biologically active natural products or their intermediates. Accessing new chemical space through heterologous expression and the de novo assembly of new biosynthetic systems is provocative, and improved access to synthetic biology methods has provided a key first step along this road. However, in the face of such a bewildering possibility, one may ask what structures we should focus on and whether we are limited by our imagination. We would argue that the key practical measure of success for biosynthetic engineering and allied approaches is more prosaic: to make molecules that bind to biological targets in which we are interested and, above all, that answer fundamental scientific questions, rather than to make only those molecules that the pharmaceutical industry may desire. Thus, the use of molecular genetic and sequencing approaches for uncovering new chemistries is most effectively applied through the use of imaginative screening and engineering methods to identify and diversify new chemotypes acting against important biological targets. As the example of platensimycin illustrates, even the culturable microbial diversity, so thoroughly investigated already, has many surprises yet in store.
The output of biosynthetic engineering and related efforts to generate new compounds in quantities sufficient for screening and subsequent validation has remained predominantly limited to the rational construction of modestly sized focused libraries or specific compounds4. This is in contrast with early aspirations for the technology to produce large compound libraries. A parallel can be drawn here with combinatorial chemistry and parallel synthesis in medicinal chemistry, where the production of large libraries for screening purposes has given way to the recognition that this technology is best applied as a means of optimizing lead structures through the production of smaller, high-quality libraries3. However, unlike the starting materials of synthetic chemistry, natural products offer a privileged starting point for structural diversification67, given the structural heterogeneity between the different groups of natural products coupled with their innate ability to bind to protein-fold space. Indeed, nature has been evolving protein and small-metabolite structures for over 3 billion years. It is thus not inconceivable that, during this period, natural-product ligands may have existed that can bind to all possible protein folds67. With the likelihood that a natural-product ligand could be found for most molecular targets, the output of biosynthetic engineering programs can thus be used in lead optimization to improve binding affinities, selectivity and potency through rational structural improvements. Where the structural features of ligand-target complexes are known, this means that the production of a relatively small number of library members can correlate with success.
Without doubt, through mining, heterologous expression, biosynthetic engineering and the other methods described here, we are significantly increasing our portfolio of natural products and analogs, making these molecules an even more valuable asset not only in drug discovery but also in fundamental research, including chemical genomics2. Indeed, the argument that nature may already have evolved a natural-product ligand for most protein folds would suggest that the long-term goal of chemical genomics, which is to uncover small molecules that can specifically modulate every function of all gene products in a given cell2, could, to a large extent, be achieved using the increasingly diverse libraries of natural products and engineered variants. Of course, chemical genomics is distinct in that the value in discovering a new small-molecule ligand for a particular target is not measured solely by its therapeutic potential. Instead, the emphasis is on uncovering molecules that can be used to probe and modulate biological function across the whole organism in a systematic and fundamental sense.
In conclusion, it would be trite to suggest that natural-product research is in renaissance. Rather, we would suggest that although it may have been de-emphasized in many industrial groups, within others, as well as academia, the field has continued to flourish. Through the advent of the new technologies described here and the increasingly imaginative ways available for using natural products as probes to interrogate biological systems, there is little doubt that the impact of this field will be felt far beyond the pharmaceutical arena. Indeed, the field is not only strong enough to withstand the waning interest of some industrial sectors but more than capable of spawning new industries of the future.
