Natural products - small molecules derived from living organisms - have long been objects of fascination and utility, and they have provided most of the motivation for developing organic chemistry 1. An example is given by morphine, the most active of the sleep-inducing compounds in opium, which was isolated in pure form in 1806 but was known thousands of years earlier 2. Collaboration between chemists and biologists led to the identification of the opioid receptor and the isolation of its endogenous ligands (enkephalins). The story of morphine and related compounds has been repeated many times, and natural-products research still contributes important small molecules to medicine. Between 2000 and 2003, 15 new drugs derived from natural products were introduced for the treatment of disorders such as malaria, fungal infections, bacterial infections, cancer, blood clots, premature labor, infertility, and stimulation of the central nervous system, such as Alzheimer's disease 34. Two recent papers 56 describe the identification and cloning of genes encoding the biosynthetic pathway of patellamide, a potential anticancer agent, highlighting the profound changes that genomic approaches are bringing about in what is arguably the oldest scientific discipline.

Natural-products research was transformed in the 1940s by the establishment of the actinomycete group of Gram-positive filamentous soil bacteria as the premier source of medically useful natural products. The actinomycete group produces the antibiotics streptomycin, actinomycin, erythromycin, and vancomycin; the antifungal agents nystatin and amphotericin; the anticancer agents doxorubicin and calicheamicin; the immunosuppressive agents FK506 and rapamycin; and many other useful molecules. In addition to their ability to produce this staggering array of important natural products, the biosynthetic genes of bacteria have an organization that has greatly simplified genetic studies: all of the instructions for making a product from simple metabolites - and to avoid being killed by it - are usually found on a continuous stretch of DNA, and heterologous expression of this region in an alternative host confers biosynthetic competence (for example, see 7). This revelation undoubtedly reflects the evolutionary history of natural-product biosynthesis pathways: inheriting only a fraction of a pathway, or the complete pathway without the gene encoding resistance to the molecule produced (so that the organism risks poisoning itself), confers no survival advantage. The clustering of biosynthetic, resistance and regulatory genes in prokaryotic pathways has proved to be a general rule.

As the biosynthesis pathways were probed in greater depth, it became clear that many bacterial natural products are made by 'assembly lines' of enzymes and that the order of assembly could be read from the order of the biosynthetic genes 1. Two large and related chemical families produced by these assembly lines - the polyketides and the nonribosomal peptides - include most of the important actinomycete drugs. These assembly lines have been identified in many sequenced genomes, and we now realize that there are large numbers of 'cryptic' metabolites: natural products whose existence can be inferred from genomic analysis but which have never been isolated 8. In one recent report 9, a group at Ecopia Biosciences was able to predict the properties of a natural product from the genome alone with enough precision that it could be isolated.

The structural similarity of natural products from widely different organisms led to the suspicion that they might in fact be produced by similar bacteria associated with the various organisms. One example is provided by pederin (Figure 1a), a toxic compound from the blister beetle, Paederus fuscipes. In addition to raising the blisters that give the beetle its name, pederin is also a powerful inhibitor of protein synthesis and mitosis, and in some model systems it has been shown to extend the lives of animals with tumors, even at subnanomolar concentrations. Compounds with very similar structures and biological activities, such as theopederin A and mycalamide A, are found in sponges, especially Theonella swinhoei (Figure 1a). If any of these molecules were to be developed into a therapeutic agent, it would have to be supplied either by collection from the animal or by chemical synthesis. Isolating them from either beetles or sponges could prove difficult, as they are minor constituents of these animals found in inconsistent amounts; and practical large-scale synthesis would be challenging given their complex structures. Recent reports from the Piel laboratory 101112 make a convincing case that, in both beetles and sponges, an associated bacterium - not an actinomycete but an uncultured species of Pseudomonas - is responsible for the biosynthesis of pederin-like compounds.

Because the molecular structure of pederin-like compounds suggests a polyketide-type assembly line, Piel and coworkers 10 guessed the biosynthetic genes likely to be part of the pathway and used PCR to clone them from the collective DNA (beetle and associated microbes) of P. fuscipes. They found the 54 kilobase (kb) ped cluster, which includes genes encoding an assembly line for a mixture of polyketides and nonribosomal peptides flanked by transposase pseudogenes. A more detailed analysis of the cluster provided strong evidence that it was from an uncultured Pseudomonas species and that it was responsible for pederin biosynthesis. Additional evidence was provided by the tight correlations between the ped cluster's occurrence in an organism and the isolation of pederin from that organism. A similar approach starting with the collective DNA from T. swinhoei revealed an almost complete biosynthetic pathway for the shared part of the pederin-like molecules 11. Comparison of the genes for the putative biosynthetic pathways from the two organisms 12 added confirmatory evidence that the true biosynthetic pathways had been identified. Although the combined evidence - gene analysis, correlation of pederin production and the ped cluster, and sequence comparison of the two pathways - made a strong case that the pathway had been identified, the failure to identify or culture the bacterial symbiont and the inability to express the pathway heterologously in an alternative host left the story incomplete. The problem of providing a reliable supply of a potentially useful therapeutic compound thus remained unsolved by this work.

Two independent recently published papers from the Schmidt 5 and Jaspars 6 groups now couple the isolation of a pathway with the production of a small molecule. The patellamides and related molecules (Figure 1b) were isolated from ascidians - sac-like, marine, filter-feeding chordates - because of the pronounced anticancer activity of these compounds in biological assays. The compounds almost certainly originate from eight amino acids (for patellamide A the sequence is Ile-Ser-Val-Cys-Ile-Thr-Val-Cys or a cyclic permutation thereof; see Figure 1b). Ascidians, which produce a number of cyclic peptides and cyclic-peptide derivatives with potentially useful biological activity, harbor obligate cyanobacterial symbionts, species in the Prochloron genus, which could produce some or possibly all of the compounds isolated from ascidians.

The Schmidt laboratory 5 originally pursued an approach similar to that used by Piel and colleagues 101112 (Figure 2a). Prochloron cyanobacteria were isolated from their ascidian host (Lissoclinum patella) and used to prepare genomic DNA. The isolates consisted primarily (> 95%) of Prochloron didemni as judged by light microscopy. A search of predicted protein sequences for examples of the nonribosomal adenylation domain - a highly conserved and repetitive domain found in enzymes of the nonribosomal-peptide biosynthetic assembly line - yielded only a single candidate gene, and further analysis of its sequence indicated that the encoded protein was unlikely to function in patellamide biosynthesis. If the patellamides are not made by a nonribosomal peptide assembly line, they must be made by ribosomal synthesis of a precursor peptide followed by fusion of side chains with the main chain to form small five-member rings and joining the two ends to form a large ring (Figure 1b).

Finding a nonribosomal peptide assembly line is relatively straightforward as much is known about them, but finding a ribosomal (or possibly some other) biosynthetic pathway is much more challenging. The entire P. didemni genome was sequenced by The Institute for Genomic Research to threefold coverage, and a gene cluster that could, in principle, produce patellamide A through ribosomal translation was identified by searching for the eight possible peptides whose cyclization and subsequent alteration could generate patellamide A (Figure 2a). A single coding sequence was identified (patE, encoding a 77 amino-acid precursor peptide), and the same sequence also encoded the eight residues needed to form patellamide C, which invariably is found with patellamide A. Genes for the entire pathway (patA-G) surrounded the patE gene. In a decisive experiment, the pathway was heterologously expressed in Escherichia coli, and patellamides A and C were isolated from the culture medium; there is thus no doubt that the correct pathway has been identified. Now that the genes for the biosynthetic pathway are known, the timing and mechanism of the various steps can be analyzed.

Whereas Schmidt and colleagues 5 relied on whole-genome sequencing, the Jaspars laboratory 6 used shotgun cloning and heterologous expression, an approach that had earlier been used to identify new biologically active small molecules from cultured and uncultured bacteria 1314151617. A genomic library of cyanobacterial DNA isolated from the same ascidian as was used by Schmidt and colleagues (L. patella) but from a different location was used to construct a bacterial artificial chromosome (BAC) library in E. coli (Figure 2b). Attempts to identify clones containing nonribosomal peptide-synthase genes using Southern hybridizations revealed nothing useful, so the library was interrogated directly for the production of patellamides using liquid chromatography coupled with mass spectrometry (LC-MS). Eventually a single transformant that produced patellamide D was identified (Figure 2b). Because the article by Jaspars and colleagues 6 was rushed into publication to be roughly contemporaneous with the report by Schmidt et al. 5, no sequence information is available.

The two different approaches, complete genome sequencing 5 and shotgun cloning 6, have led to roughly equivalent results and have shown clearly that the patellamides are produced by a cyanobacterial symbiont through a pathway that can now be studied in great depth. What are the implications for natural products in general and what might we expect in the future? One obvious lesson is that DNA-based approaches have become powerful tools for finding biosynthetic pathways, both for the detailed analysis of their mechanistic details and for the production of natural compounds that would otherwise be difficult to obtain. We can confidently expect to see a great deal of similar work in the future. A subtler change could be a reorientation of natural-products research, a discipline that still retains vestiges of 19th-century exploration and natural philosophy, into a discipline focused on genes. Finally, the challenge of using the same approaches 56101112 to discover new natural products can now be faced with greater confidence.