Lessons learned from the transformation of natural product discovery to a genome-driven endeavor

Abstract

Natural product discovery is currently undergoing a transformation from a phenotype-driven field to a genotype-driven one. The increasing availability of genome sequences, coupled with improved techniques for identifying biosynthetic gene clusters, has revealed that secondary metabolomes are strikingly vaster than previously thought. New approaches to correlate biosynthetic gene clusters with the compounds they produce have facilitated the production and isolation of a rapidly growing collection of what we refer to as “reverse-discovered” natural products, in analogy to reverse genetics. In this review, we present an extensive list of reverse-discovered natural products and discuss seven important lessons for natural product discovery by genome-guided methods: structure prediction, accurate annotation, continued study of model organisms, avoiding genome size bias, genetic manipulation, heterologous expression, and potential engineering of natural product analogs.

I. Introduction

Natural products have played a significant role in medicine, and fittingly, the discovery of novel natural products continues to hold great promise for the development of new drugs [108]. Forward natural product discovery relies on the presence of an observable phenotype or chemical property, such as biological activity, color, or a known mass, which can be tracked through successive rounds of isolation. Successful “grind and find” isolation of a natural product can be followed by identification of the genes responsible for the biosynthesis of the compound of interest. This method was remarkably productive in its early years, but has been met with increasing frustration by researchers as the rediscovery of known compounds has become commonplace, as many of the easily accessible natural products have already been isolated [7].

Meanwhile, the increasing ease of DNA sequencing has led to an explosion of fully sequenced genomes and identified biosynthetic pathways over the last 20 years. In many cases, the biosynthetic genes responsible for well-known and long-studied natural products have been identified only recently (Fig. 1). Furthermore, the elucidation of biosynthetic pathways for natural products is by no means complete, as a number of important compounds (e.g. morphine, paclitaxel) still have major gaps in what is known of their biosynthesis (Fig. 1). Biosynthetic genes for plant natural products are often particularly difficult to identify because plants do not commonly cluster the responsible genes as bacteria and fungi often do, and plant genomes are much larger than those of bacteria and fungi [157]. That is not to say the task is easy in lower organisms, as history has shown us that characterization of bacterial and fungal pathways requires immense time and effort. A noteworthy example is the bottromycin family, whose biosynthetic gene cluster was identified by four independent groups in 2012, which was 55 years after the first report of its isolation [35,53,66,68]. Likewise, the genes responsible for the production of the thiopeptide antibiotics remained unknown until being reported upon nearly simultaneously by four independent research groups in 2009 [73,89,156,106]. In the case of thiostrepton, a well-known thiopeptide, this report came 53 years after its initial isolation.

Fig. 1.

Timeline of traditional “forward” natural product discovery, highlighting the frequently decades-long gap between isolation of a novel natural product and identification of its biosynthetic genes. A date above the arrow indicates when isolation of the indicated natural product was first reported, with the date of structure elucidation (if different) in parentheses. A date below the arrow indicates when the biosynthetic genes were first identified for the natural product, with the years since isolation in parentheses. This figure is not meant to be exhaustive but instead to provide a representative sample of natural products spanning multiple biosynthetic classes, biological targets, and producing organisms. NRP, non-ribosomal peptide; RiPP, ribosomally synthesized and post-translationally modified peptide; *, majority of biosynthetic pathway remains unknown

With the advent of inexpensive, massively parallel sequencing, a new route to discover natural products has emerged. The genomic revolution has fueled a shift in methods for identifying natural products from the traditional, phenotype-driven “forward” procedure to a genotype-driven, “reverse” discovery process, during which identification of the biosynthetic genes precedes and informs isolation of the natural product. With the availability of the of Streptomyces coelicolor and Streptomyces avermitilis genome sequences [114,11], it became clear that even some highly-studied strains whose biosynthetic capabilities had appeared to be exhausted harbored a surprising number of previously unknown biosynthetic gene clusters. For example, only three natural product gene clusters had been identified on the S. coelicolor chromosome prior to the completion of its genome sequence: those for actinorhodin [96,97], prodiginine [46], and calcium-dependent antibiotic [29]. Genome sequencing of S. coelicolor revealed a number of “cryptic” gene clusters without known associated natural products (sometimes called orphan gene clusters). In total, S. coelicolor carries the potential for 29 structurally complex natural products (Fig. 2) [11,34]. Many of these biosynthetic gene clusters remain cryptic to the present day, over a decade later, indicating that the biosynthetic capabilities of this and other organisms still have yet to be fully understood.

Fig. 2.

Schematic representation of the linear 8.7 Mb Streptomyces coelicolor A3(2) chromosome, highlighting the locations of all known natural product biosynthetic gene clusters. Names in brackets indicate putative compounds for which no predicted structure appears in the literature [34]

The number of complete bacterial genome sequences available in the National Center for Biotechnology Information (NCBI) database increased approximately 25-fold, to over 2500, between 2003 and 2013. With the ready availability of sequenced genomes, particularly those from bacteria, it is increasingly possible to identify putative biosynthetic genes, use their sequence to predict the structure and properties of the potential product, and use those predictions to guide efficient isolation and characterization. One interesting example of this is the case of bacillaene, whose structure eluded scientists due to its chemical instability [115]. The genetic sequence of the bacillaene producer led to identification of the protein sequences responsible for its biosynthesis, which in turn enabled a structural prediction for bacillaene that guided purification based on the perceived physical properties. Ultimately, this procedure was successful, and the structure was solved [24,18]. In addition, the prototypical reverse-discovered natural product, coelichelin had its structure predicted with a reasonably high degree of accuracy several years prior to its isolation [21,83].

In Table 1, we present an extensive list of natural products that have been reverse discovered. For more detailed discussion of many of these natural products, we direct the reader to a number of reviews on this topic [20,55,150], in addition to those contained in the current issue of this journal. Many of the natural products listed in this table were discovered in 2008 or later, reflecting the exponential rise in available genome sequences. A number of these natural products are from well-studied model organisms (S. coelicolor, Aspergillus nidulans). Most are from bacteria, consistent with the larger number of sequenced bacterial genomes relative to fungi and plants.

Table 1.

Reverse-discovered natural products, organized by year of reported isolation. In cases where more than one species is known to produce a particular natural product, only the strain from which the compound was first isolated is listed

Name	Class	Producer	Year isolated (Year predicted, if different)	Genome sequenced	Reference(s)
hopene	Polyketide	Streptomyces coelicolor A3(2)	2000	2002	[121,11]
collinone	Polyketide	Streptomyces collinus DSM 2012	2001	n/a	[98]
geosmin	Terpenoid	Streptomyces coelicolor A3(2)	2003	2002	[19,11]
desferrioxamine	Other	Streptomyces coelicolor A3(2)	2004	2002	[9,11]
halstoctacosanolides	Polyketide	Streptomyces halstedii HC34	2004	n/a	[146,147]
SapB	RiPP	Streptomyces coelicolor A3(2)	2004	2002	[77,11]
tetrahydroxy-naphthalene	Polyketide	Streptomyces coelicolor A3(2)	2004	2002	[4,11]
thalianol	Terpenoid	Arabidopsis thaliana	2004	2000	[45,2]
aurafurons	Polyketide	Stigmatella aurantiaca DW4/3-1	2005	2011	[81,67]
coelichelin	NRP	Streptomyces coelicolor A3(2)	2005 (2000)	2002	[21,83,11]
ECO-02301	Polyketide	Streptomyces aizunensis NRRL B-11277	2005	n/a	[100]
myxochromides S	NRP	Stigmatella aurantiaca DW4/3-1	2005	2011	[155,67]
aspoquinolones	Alkaloid	Aspergillus nidulans	2006	2005	[132,49]
bacillaene	Polyketide/NRP	Bacillus amyloliquefaciens FZB42	2006	2007	[24,18,22]
DKxanthenes	Polyketide/NRP	Myxococcus xanthus DK1050	2006	2006	[103,51]
ECO-501	Polyketide	Amycolatopsis orientalis ATCC 43491	2006	n/a	[8]
germicidins	Polyketide	Streptomyces coelicolor A3(2)	2006	2002	[138,11]
haloduracin	RiPP	Bacillus halodurans C-125	2006	2000	[102,84,142]
penocin A	RiPP	Pediococcus pentosaceus ATCC 25745	2006	2006	[39,94]
terrequinone A	NRP	Aspergillus nidulans	2006	2005	[16,49]
trichamide	RiPP	Trichodesmium erythraeum IMS101	2006	2006	[141]
aeruginosides	NRP	Planktothrix agardhii CYA126/8	2007	Draft	[70]
aspyridones	Polyketide/NRP	Aspergillus nidulans	2007	2005	[12,49]
CBS-40	Polyketide	Streptomyces sp. CB2544	2007	n/a	[64]
orfamide A	NRP	Pseudomonas fluorescens Pf-5	2007	2005	[56,116]
aerucyclamide C	RiPP	Microcystis aeruginosa PCC7806	2008	Draft	[122]
albaflavenone	Terpenoid	Streptomyces coelicolor A3(2)	2008	2002	[163,11]
capistruin	RiPP	Burkholderia thailandensis E264	2008	2005	[76,75]
diazepinomicin (ECO-4601)	Other	Micromonospora	2008	n/a	[101]
emericellamide	Polyketide/NRP	Aspergillus nidulans	2008	2005	[28,49]
2-methylisoborneol	Terpenoid	Streptomyces coelicolor A3(2)	2008	2002	[78,11]
microcyclamide 7806A, B	RiPP	Microcystis aeruginosa PCC7806	2008	Draft	[164]
thailandamides	Polyketide	Burkholderia thailandensis E264	2008	2005	[109,75]
vibi A-K	RiPP	Viola biflora	2008	n/a	[62]
asperfuranone	Polyketide	Aspergillus nidulans	2009	2005	[27,49]
atrochrysone	Polyketide	Aspergillus terreus	2009	Draft	[5]
emodin	Polyketide	Aspergillus nidulans	2009	2005	[15,49]
F9775 A/B (blecanoric acid)	Polyketide	Aspergillus nidulans	2009	2005	[15,129,49]
lecanoric acid	Polyketide	Aspergillus nidulans	2009	2005	[135,49]
lichenicidin	RiPP	Bacillus licheniformis	2009	2004	[10,40,137,126]
monodictyphenone	Polyketide	Aspergillus nidulans	2009	2005	[15,49]
Mra4, 5	RiPP	Melicytus ramiflorus	2009	n/a	[148]
nygerone A	Polyketide/NRP	Aspergillus niger CBS 513.88	2009	2007	[61,119]
orsellinic acid	Polyketide	Aspergillus nidulans	2009	2005	[135,129,49]
anacylamides	RiPP	Anabaena sp. 90	2010	2012	[87,153]
aspernidine	Alkaloid	Aspergillus nidulans	2010	2005	[133,49]
aureusimine	NRP	Staphylococcus aureus	2010	2008	[160,166,6]
Bsa	RiPP	Staphylococcus aureus	2010	2004	[36,63]
coelimycin P1	Polyketide	Streptomyces coelicolor A3(2)	2010 (2007)	2002	[118,117,11]
csypyrone B1	Polyketide	Aspergillus oryzae	2010	2005	[136,92]
diorcinol	Polyketide	Aspergillus nidulans	2010	2005	[129,49]
erythrochelin	NRP	Saccharopolyspor a erythraea NRRL 2338	2010	2007	[85,127,113]
gerfelin	Polyketide	Aspergillus nidulans	2010	2005	[129,49]
Globa A, B	RiPP	Gloeospermum blakeanum	2010	n/a	[17]
microcin H47	RiPP	Escherichia coli (various strains)	2010	n/a	[149]
microcin M	RiPP	Escherichia coli (various strains)	2010	n/a	[149]
microviridin L	RiPP	Microcystis aeruginosa NIES843	2010	2007	[165,72]
pneumococcin	RiPP	Streptococcus 20neumonia R6	2010	2001	[93,65]
prochlorosins	RiPP	Prochlorococcus marinus MIT9313	2010	2003	[88,128]
TP-1161	RiPP	Nocardiopsis sp. TFS65-07	2010	n/a	[43]
venezuelin	RiPP	Streptomyces venezuelae ATCC10712	2010	2011	[54,124]
austinol	Polyketide/terpenoid	Aspergillus nidulans	2011	2005	[110,49]
desmethylbassianin A	Polyketide	Beauveria bassiana	2011	Draft	[60]
grisemycin	RiPP	Streptomyces griseus IFO 13350	2011 (2010)	2008	[32,31,112]
(iso)flavipucine	Polyketide/NRP	Aspergillus terreus	2011	Draft	[125]
koranimine	NRP	Bacillus spp.	2011	n/a	[44]
plantazolicin	RiPP	Bacillus amyloliquefaciens FZB42	2011 (2008)	2007	[86,134,22]
stambomycins	Polyketide	Streptomyces ambofaciens ATCC23877	2011	Draft	[82]
thailandepsin	Polyketide/NRP	Burkholderia thailandensis E264	2011	2005	[152,75]
alternariol	Polyketide	Aspergillus nidulans	2012	2005	[1,49]
astexin-1	RiPP	Asticcacaulis excentricus	2012	2010	[95]
azanigerones	Polyketide	Aspergillus niger	2012	Draft	[162]
burkholderic acid	Polyketide/NRP	Burkholderia thailandensis E264	2012	2005	[48,75]
catenulipeptin	RiPP	Catenulispora acidiphila DSM 44928	2012	2009	[154,33]
cichorine	Polyketide	Aspergillus nidulans	2012	2005	[1,49]
curvopeptin	RiPP	Thermomonospor a curvata	2012	2011	[80,26]
elgicin	RiPP	Paenibacillus elgii B69	2012	Draft	[145]
fusarielins F-H	Polyketide	Gibberella zeae	2012	Draft	[139]
geobacillin	RiPP	Geobacillus thermodenitrifica ns NG80-2	2012	2007	[50,47]
luminmycin	Polyketide/NRP	Photorhabdus luminescens subsp. laumondii TT01	2012	2003	[13,41]
malleilactone	Polyketide	Burkholderia thailandensis E264	2012	2005	[14,75]
O-methyldiaporthin	Polyketide	Aspergillus oryzae	2012	2005	[107,92]
pre-shamixanthone	Polyketide	Aspergillus nidulans	2012	2005	[131,49]
rhizopodin	Polyketide/NRP	Stigmatella aurantiaca Sg a15	2012	Draft	[120]
caulosegnins	RiPP	Caulobacter segnis	2013	2010	[59]
flavipeptin	RiPP	Kribbella flavida	2013	2010	[151,123]
flavopeptins	NRP	Streptomyces flavogriseus ATCC 33331	2013	2011	[25]
fumicycline A/neosartoricin A	Polyketide	Aspergillus fumigatus	2013	2005	[79,30,111]
neosartoricins B-D	Polyketide	Trichophyton tonsurans	2013	2012	[161,99]
thailanstatins	Polyketide/NRP	Burkholderia thailandensis MSMB43	2013	Draft	[91]

In this review, we focus on seven lessons for the community to keep in mind during the reverse-discovery of natural products: 1) structure prediction, 2) accurate annotation, 3) continued study of model organisms, 4) avoiding genome size bias, 5) genetic manipulation, 6) heterologous expression, and 7) potential engineering of analogs. To illustrate the utility of these lessons, we highlight five natural products from the recent literature (aureusimine, plantazolicin, astexin-1, the prochlorosins, and asperfuranone; Fig. 3) representing three prominent natural product classes: polyketides, non-ribosomal peptides (NRPs), and ribosomally synthesized and post-translationally modified peptides (RiPPs), as well as a variety of producing organisms (bacteria and fungi).

Fig. 3.

Biosynthetic gene clusters (A) and structures (B) of select reverse-discovered natural products. Gene clusters in panel A are shown to scale and are shown with the name of the organism in which they were first identified

II. Lesson 1: Structure prediction

Access to a predicted structure facilitated the isolation of the aureusimines, founding members of the pyrazinone class of natural products, by enabling mass spectrometry-guided isolation. These compounds are biosynthesized by a non-ribosomal peptide synthase (NRPS) pathway in the human pathogen Staphylococcus aureus (Fig. 3A) [160,166]. The aureusimine biosynthetic gene cluster was discovered by genome mining to identify possible NRPSs that are highly conserved among sequenced S. aureus strains. Homologous gene clusters have been found in over 50 S. aureus strains as well as other human pathogenic Staphylococcus species [160]. Prior to its isolation, the structure of aureusimine A was predicted based on the sequence of the NRPS gene and the previously established amino acid specificities for NRPS adenylation domains. The presence of a putative reductase domain at the C-terminus of the S. aureus NRPS was also factored into the structural prediction, as it indicated that the dipeptide was likely released from the synthetase as an aldehyde with the potential to spontaneously cyclize. This prediction was confirmed with the solved structures of aureusimines A and B (Fig. 3B) [160].

Structural prediction and mass spectrometry-guided isolation also proved useful in the isolation of plantazolicin, a member of the thiazole/oxazole-modified microcin (TOMM) subclass of RiPP natural products [3]. TOMM biosynthesis is characterized by the post-translational modification of ribosomally synthesized precursor peptides to generate thiazol(in)e and (methyl)oxazol(in)e heterocycles from the side chains of select cysteine, serine, and threonine residues [86]. The gene cluster for plantazolicin (Fig. 3A) was identified in 2008 during a search for genes with homology to the biosynthetic gene cluster responsible for the production of streptolysin S and microcin B17, which are TOMMs from Streptococcus pyogenes and Escherichia coli, respectively [86]. Like other RiPPs, the sequence of a predicted precursor peptide gene enabled mass spectrometric-guided separation to isolate plantazolicin from organic surface extracts of modified strains of B. amyloliquefaciens FZB42 [134] and aided significantly in the elucidation of its structure (Fig. 3B) [71,105].

A concluding example of the utility of structure prediction during natural product discovery is provided by coelichelin, a prominent early example of a reverse-discovered natural product, which had its structure predicted several years prior to its isolation from S. coelicolor [21]. Although the initial structure was not entirely correct [83], the ability to predict structure based on genetic sequence is playing a growing role in reverse natural product discovery, whether by providing insight into possible extraction conditions or by simplifying identification by mass spectrometry.

III. Lesson 2: Accurate sequence annotation

Regardless of how many genome sequences are available, they are useless without interpretation. Complete and accurate sequence annotation also play crucial roles in the efficient reverse discovery of natural products. RiPP precursor peptide genes, in particular, are typically short and are often unannotated in published genomes [3,104], complicating their identification during genome mining. For example, the open reading frame (ORF) encoding the plantazolicin precursor peptide was unannotated at the time of the initial report [86]. The same has also proven to be the case for the precursors of several lasso peptides, an emerging subclass of RiPP natural products [76,59].

Despite the potential difficulty of identifying unannotated precursor genes, the discovery of the lasso peptide astexin-1 through a precursor-guided screen provides a promising avenue for future genome mining in the pursuit of novel RiPP natural products. Lasso peptides are characterized by the post-translational formation of a polypeptide loop structure with a covalent bond between the N-terminus of the precursor peptide and a specific aspartate or glutamate side chain elsewhere on the peptide. The C-terminal end of the precursor peptide is then threaded through the loop and often held in place by the side chains of bulky residues in the tail region [3]. The resulting threaded structure typically confers unique heat stability on the lasso peptides [167]. Genome mining for RiPP biosynthetic gene clusters commonly focuses on identifying homologs of a known modifying enzyme, as was the case for the lasso peptide capistruin [76], the prochlorosins, and plantazolicin. However, in the case of astexin-1, a genetic screen for precursor peptides was employed. This screen identified short ORFs, some unannotated, containing amino acid patterns consistent with the sequences of known lasso peptides. When a potential lasso peptide precursor was found, nearby regions of the genome were then searched for homology to the maturation enzymes [95]. This endeavor resulted in the identification of 79 putative lasso peptide gene clusters distributed across nine bacterial phyla and one archaeal phylum [95].

The identification of precursor peptide genes located outside of the primary gene cluster further underscores the advantage of searching widely for all possible biosynthetic genes during reverse natural product discovery [57]. Such was the case for the prochlorosins, a group of lanthipeptide RiPPs produced by some marine cyanobacteria species, most notably Prochlorococcus marinus [88]. The hallmark of lanthipeptide biosynthesis is the formation of (methyl)lanthionine rings between select cysteines and dehydroalanine or dehydrobutyrine residues, which arise from the dehydration of serines and threonines, respectively. Dehydration and lanthionine ring installation on the ribosomal precursor peptide may be catalyzed by one or more biosynthetic enzymes [3]. The gene clusters responsible for prochlorosin biosynthesis (Fig. 3A) were identified in P. marinus and related species through a bioinformatic search for bifunctional lanthipeptide synthetase homologs [88] and, independently, through the homology of prochlorosin precursor peptides to the Nif11 nitrogen-fixing proteins [58]. However, only seven of these 29 putative precursor peptide genes are clustered with the lanthipeptide synthetase gene [88], contrary to the norm for biosynthetic genes in bacteria and fungi.

IV. Lesson 3: Continued study of model organisms

The sequencing of the S. coelicolor genome, which brought to light the silent majority of biosynthetic gene clusters, also provides a reminder not to discount continued investigation into well-studied model organisms during the search for new natural products. These strains may contain the genetic potential to synthesize many more natural products than are superficially detectable. The fungus Aspergillus nidulans [27] was a highly-studied model organism even prior to the sequencing of its genome, and since that time, it has been determined that A. nidulans FGSC A4 contains 24 PKS genes and 3 PKS-like genes [49,130]. Several pairs of PKS genes are adjacent to each other (Fig. 3A), indicating a strong possibility that the pairs might work in concert, though A. nidulans was not previously known to produce any polyketides arising from the combined action of two PKS enzymes. The producer of plantazolicin, B. amyloliquefaciens FZB42, also produces a number of natural products [22], including the polyketide/NRP hybrid bacillaene, which likewise is a reverse-discovered natural product [24,18].

V. Lesson 4: Avoiding genome size bias

One common assumption in the natural products field has been that organisms with small genomes are unlikely to produce structurally complex natural products. Indeed, bacteria with larger genomes, such as S. coelicolor (8.7 Mb), are recognized as prolific producers of secondary metabolites [11]. However, species with small genomes should not be ignored in the search for natural products, as exemplified by the isolation of structurally complex compounds from organisms whose genomes are a fraction the size of the traditionally recognized “natural product powerhouses”. These species with smaller genomes, such as S. aureus (2.8 Mb) [6], are presumed to be less able to devote genome space to large biosynthetic gene clusters such as NRPSs, making the identification of a single NRPS gene that occupies 0.25% of the S. aureus genome even more surprising [160].

Maintenance of multiple precursor genes and the ability to modify them with a single enzyme enables organisms with compact genomes, such as prochlorosin producer P. marinus (2.4 Mb), to produce a variety of structurally complex molecules without the need for large biosynthetic gene clusters. Indeed, the precursor hypervariability of the prochlorosin system endows P. marinus MIT9313 with the potential to produce as many natural products as S. coelicolor, which bears a genome nearly four times as large [88,11].

VI. Lesson 5: Genetic manipulation

In many instances, natural products are difficult to detect under laboratory cultivation, due to low production levels or masking by the presence of other, more abundant, natural products. In such cases, genetic manipulation may be used to activate these silent gene clusters. The reverse discovery of the polyketide asperfuranone (Fig. 3B) illustrates how silent gene clusters can be activated via promoter engineering within the native host. In this case, the native promoter for a predicted transcriptional activator gene within the PKS cluster (Fig. 3A) was replaced with an inducible promoter [27]. As asperfuranone was not produced to any significant extent during laboratory cultivation, it would have been unlikely to identify this compound via any phenotype-driven effort. Other silent biosynthetic gene clusters that have been successfully activated by promoter engineering include those responsible for producing aspirydone [12], malleilactone [14], and stambomycin [82]. A limitation of this strategy is that it requires a genetically tractable host.

Another common genetic manipulation used for natural product discovery is biosynthetic gene disruption. Arguably, the most powerful method to establish gene function is to evaluate the phenotypes of the parent (wild-type) and the genetic deletion (mutant) strains. This reverse genetics method is often employed during the genome-guided discovery of natural products. The phenotype compared between parent and mutant is typically either the metabolite profile (usually obtained by mass spectrometry) or the bioactivity of chromatographic fractions. The detection and isolation of plantazolicin demonstrated a secondary use of such a method to identify a novel compound whose production had been masked by other, more abundant compounds. When production of all known bioactive compounds was eliminated, the remaining bioactivity indicated that at least one additional biosynthetic gene cluster had yet to be identified [134,23].

VII. Lesson 6: Heterologous expression

In cases where genetic manipulation in the native host is not feasible, the transfer of biosynthetic genes clusters to a genetically tractable host has the potential to facilitate compound production. The biosynthetic gene cluster for the lasso peptide astexin-1 (Fig. 3A) was identified in the freshwater α-proteobacterium Asticcacaulis exentricus, although compounds with the expected mass could not be detected after laboratory cultivation of this organism, necessitating an alternate expression method [95]. Astexin-1 was successfully expressed, purified, and structurally characterized using E. coli as a heterologous host (Fig. 3B) [95,167]. Following extensive variation of culture conditions, a recent study succeeded in producing astexin-1 from A. exentricus, albeit at low levels compared to those from E. coli [167]. Likewise, successful high-level production of a number of prochlorosins required heterologous expression in E. coli and reconstitution in vitro, ultimately enabling structural confirmation of these natural products (Fig. 3B) [88,144].

For further discussion and examples of heterologous expression, the interested reader is directed to other reviews specifically focused on this topic, also in this Special Issue [69,52].

VIII. Lesson 7: Potential engineering of natural product analogs

Beyond simply providing a novel natural product, identification of new biosynthetic gene clusters also has the potential to facilitate rational engineering of unnatural natural product variants. The production of prochlorosins from P. marinus MIT9313 is notable among lanthipeptides because of the exceptionally high number of precursor peptides (29) [88]. While it is not unusual for a lanthipeptide gene cluster to harbor more than one precursor peptide, these typically require multiple lanthionine synthetases to modify all the precursor peptides [102,10]. Indeed, 17 prochlorosins from P. marinus MIT9313 have been successfully biosynthesized both in vitro and in vivo using this single lanthipeptide synthetase, [88]. The potential for a single lanthipeptide synthetase enzyme to effect modification of up to 29 precursor peptides suggests that the prochlorosin synthetase is remarkably promiscuous, indicating that it may be a prime candidate for engineering novel lanthipeptides.

In addition to the natural combinatorial biosynthesis of prochlorosin and its potential use in engineering, several other examples presented in this review have shown promise for the future engineering of natural product variants. The NRPS responsible for aureusimine production has been heterologously expressed and used to explore the pyrazinone assembly line using alternate substrates [159,158]. The enzymes responsible for asperfuranone biosynthesis have also provided a compliant system for the engineering of unnatural polyketides by domain swapping, an endeavor which has historically been challenging due to the complex nature of the requisite megasynthase enzymes [90]. The heterologous expression of plantazolicin in E. coli has also been achieved and exploited using precursor peptide replacement to generate unnatural variants [38]. In summary, the thorough understanding of biosynthetic gene clusters afforded by reverse natural product discovery has the potential to expedite the production of unnatural variants for applications in medicinal chemistry and structure-activity relationship studies.

IX. Outlook

The sequencing and annotation of new genomes will continue to provide ample opportunity for the discovery of new natural products through genotype-based methods. It is possible, eventually, that natural product discovery will reach a point of diminishing returns from the investigation of new species, but the steadily increasing rate of reverse natural product discovery (Table 1) indicates that the field is still far from that point. While it is relatively straightforward to identify new putative PKS and NRPS clusters through sequence homology, other classes of natural products will require more sophisticated methods. For RiPP natural products, bioinformatics tools such as BAGEL [37] are useful in identifying genes for potential novel compounds, but less well-characterized and yet-to-be-discovered classes of natural products will require significant future investigation. The genes responsible for the biosynthesis of new or poorly understood natural products may currently be annotated as hypothetical proteins or domains of unknown function. The use of tools to group these unknown genes together by sequence homology [42], coupled with improved annotation in sequenced genomes, may allow for entirely new biosynthetic pathways to be discovered.

As the field of reverse natural product discovery moves forward, we anticipate an increasing dependence on structure prediction to facilitate isolation of interesting natural products and to provide starting points for structure elucidation. Structure prediction for new RiPP natural products is largely enabled by identification of the precursor peptide sequence, while programs such as ClustScan [140] utilize known domain specificities to predict the structure of novel PKS and NRPS gene clusters. While not strictly a method for reverse natural product discovery, peptidogenomics may also be helpful in this endeavor by identifying new peptide natural products and linking those to their respective biosynthetic genes [74]. Perhaps most widely useful, however, are comprehensive tools such as antiSMASH (antibiotics and Secondary Metabolite Analysis Shell), which incorporate information about many natural product classes beyond the comparatively well-studied PKS, NRPS, and RiPP systems. As the collective knowledge of the natural product discovery field increases, the predictive power of these tools will likewise increase and create a positive feedback loop to further enable the future isolation of novel compounds.

Natural product discovery will also be aided by the development of new genetic tools. To set a lofty goal, it would be of enormous utility to have a universal, user-friendly DNA manipulation platform for deleting genes of interest. Not only would this platform greatly enable deletion-guided approaches to natural product discovery, the whole of biology would benefit immensely from the development of a simple method for constructing genetic knockouts applicable to a broad range of organisms. If such a technique were developed and freely shared with the community, research groups focused on the chemical aspects of natural product research would now have the ability to construct useful strains without specialized knowledge in genetic manipulation.

Perhaps the most significant hurdle to isolating reverse-discovered natural products is that many gene clusters are silent (expression is undetectable) under laboratory cultivation. In such cases, no novel natural product is readily observable in extracts from cultures of the organism harboring the biosynthetic genes of interest. In order to isolate the compound in question, the genes responsible for its production must be activated. A number of methods have been developed to activate silent clusters, including variation of culture conditions, heterologous expression, promoter engineering, induction with gamma-butyrolactones, and co-culturing with other organisms to stimulate production, among others [130,55,20,143]. One persistent issue with the use of heterologous hosts is the potential for inefficient export of the natural product, which may be addressed in part by the development of greater numbers of genetically tractable heterologous hosts. These tools will help to streamline the identification and production of natural products from gene clusters which are difficult, if not impossible, to express under laboratory culture conditions.

X. Conclusion

Rather than relying on the observation of a phenotype, reverse (genotype-driven) natural product discovery bases isolation of the compound(s) of interest on knowledge of the associated biosynthetic genes. The examples of reverse-discovered natural products presented in this review represent several biosynthetic classes, producing organisms, and biological targets. It is plausible that none of the discussed examples would have been discovered by the traditional phenotype-driven method due to unknown biological activity or activity that is extremely selective so as to preclude detection during bioassay-guided isolation. Perhaps more significantly, most of these examples, as well as others listed in Table 1, required some measure of genetic manipulation (promoter engineering, comparison of genetic deletion strains, heterologous expression, etc.) in order to produce useful levels of the natural product of interest. In some cases, a low level of production was masked by other, more prominent compounds, while in others, the biosynthetic gene cluster of interest was entirely silent despite extensive modification of culture conditions. The emergence of novel natural products even from well-studied model organisms indicates that the potential for production of structurally complex compounds remains largely untapped at this point, and the constantly increasing number of publicly available genomes will certainly provide a surplus of avenues for natural product discovery.