Introduction

Microbial genome mining is an alternative approach to more traditional methods for the discovery of novel secondary metabolites, which continue to serve as scaffolds for further embellishment by medicinal chemistry and combinatorial biosynthesis for the development of products for human medicine, animal health, crop protection, and numerous biotechnological applications. Historically, natural products have made a very large impact in these areas [25, 33, 59], and the large fraction of current investigational new drug applications based on natural products strongly suggests that they will continue to impact human therapeutic discovery and development in the future. The concept of genome mining has been maturing as a discipline since the discovery by David Hopwood and coworkers by whole-genome sequencing that Streptomyces coelicolor encodes many more secondary metabolites than had been anticipated from decades of study [15, 20]. In recent years, this observation has been generalized in many reports [1, 8, 40, 42, 58, 74]. It is fitting that we dedicate this Special Issue of the Journal of Industrial Microbiology and Biotechnology on Microbial Genome Mining to Sir David Hopwood on the occasion of his 80th birthday, August 19, 2013.

What is genome mining and why is it important?

For much of its history, secondary metabolite discovery has been a process driven in large part by chance. In most cases, discovery of new natural products has been driven either by bioactivity-guided fractionation of crude fermentation broth extracts, or via chemical screening (isolation of chromatographically resolvable metabolites with ‘interesting’ spectroscopic properties). As the natural pharmacopeia has grown and the preponderance of readily identified compounds has been catalogued, the long-term success of secondary metabolite discovery campaigns has generally been determined by the degree to which this dependence on blind chance can be minimized. Historically, this has been accomplished via a number of strategies including the exploration of new ecologies, judicious selection of genera [14, 43, 69], and by the development of new analytical methodologies with improved analytical separation and sensitivity [55].

Genome mining is a radical re-envisioning of the process of secondary metabolite discovery, which has the theoretical potential to eliminate all chance from secondary metabolite discovery. In the context of this special issue, genome mining may be defined as the process of technically translating secondary metabolite-encoding gene sequence data into purified molecules in tubes. In comparison to the historical ‘grind and find’ mode of natural product discovery, the success of genome mining methods will be defined by the degree to which they unleash secondary metabolic gene clusters within a given system (Fig. 1a, b) and identify encoded metabolites. In recent years, the easy and inexpensive access to genomic sequence data, resulting from the advent of next-generation sequencing technologies [27], has created a potential embarrassment of riches regarding the starting point of genome mining. Indeed, most sequenced microorganisms with relatively large genomes, and plants contain dozens or more blueprints for the biosynthesis of secondary metabolites. Moreover, automated bioinformatics platforms now facilitate the semi-automated prediction of natural products encoded by secondary metabolic blueprints [16, 17]. However, the identification of genome-encoded secondary metabolism is only the first step in the process of genome mining. Indeed, genome mining now spans the full spectrum of the updated central dogma of molecular biology (Fig. 1c) including bioinformatic prediction of gene and pathway function, the control of gene expression and translation, and the identification and structural elucidation of new metabolites from within the metabolome of the producing organisms. As a consequence, genome-mining studies often become more than solely natural product discovery as they entail comprehensively understanding and manipulating cellular molecular systems. This issue contains articles that seek to address this navigation of the central dogma of genome to metabolites.

Fig. 1
figure 1

Strategies for natural product discovery. a The historical ‘grind and find’ mechanism of secondary metabolite discovery. b Post-genomic discovery now seeks to leverage prescience of gene sequence data to improve the yield of discovery. c The extended central dogma as it relates to genome mining for secondary metabolism

Full size image

In the Sanger sequencing era (pre ~2005), genome mining efforts were primarily enabled by the genome of only two model Streptomyces and by gene clusters discovered using oligonucleotide gene sequence probes gene sequence tags (GSTs) based on known secondary metabolism. In the former case, S. coelicolor and Streptomyces avermitilis, revealed the apparently untapped potential of secondary metabolism and resulted, over the course of a decade, in the discovery of many new metabolites from these organisms that had previously been considered to be mined to exhaustion [20, 40]. The efforts of Ecopia Biosciences [30] generated thousands of high-quality gene clusters encoding the biosynthesis of secondary metabolites identified via a twofold process of (1) low-resolution shotgun genome scanning of potential microbial secondary metabolite producers (1 read/5–20 kpb) to identify short sequences with homology to sequences in annotated secondary metabolism databases and (2) follow-up sequencing of cosmids hybridizing to secondary metabolic GSTs. Regardless of the source of sequence data, the early days of genome mining efforts generally capitalized on the prescience of secondary metabolic potential to effectively ‘look harder’ in the producing organisms for the predicted metabolites. For instance, the prediction of a siderophore in S. coelicolor [21] prompted growth in low-iron media and application of siderophore assay for isolation and structural elucidation [48]. The prediction of an antifungal polyene in Streptomyces aizunensus prompted the use of antifungal screens using a range of growth conditions for the producing organism [4, 53]. Similarly, the observation of enediyne-encoding gene clusters prompted producing-organism growth-condition screening in combination with a DNA damage assay screen for detection of putative enediyne natural products [73].

The importance of genome mining extends well beyond its potential to completely circumvent the chance component of the process of secondary metabolite discovery. For instance, understanding the connection between metabolites, which represent one of the end points of the central dogma, and the gene sequences that encode them, can provide insight into the basic biology of producing organisms as discrete individuals, and as members of the microbiota of their environment. It is becoming increasingly clear that many if not most secondary metabolites play roles in interspecies, intergeneric, and/or interkingdom chemical ecological associations. In this special issue, Crawford summarizes exciting developments in microbial ecology as they relate to genome mining of Photorhabdis and Xenorhabdis species [67]. It is becoming increasingly apparent that understanding the roles of secondary metabolites in their endogenous contexts has the potential to reveal new strategies for controlling undesirable interkingdom relationships [57]. For instance, bacterial infections in humans may be addressed through the discovery of new antibiotic substances or bioactive metabolite antibiotic combinations discovered via gene mining methods. Beyond antibiosis, applications for interrogating interkingdom cell signaling provide inroads to new therapeutics for cancer and other human diseases. The discovery of the antifungal compound rapamycin from a strain of Streptomyces hygroscopicus resulted in the revelation of a whole area of cell signaling in the mammalian target of rapamycin (mTOR, for which over 11,000 PubMed entries are available) [18] and identification of new therapeutic targets cascading from this central signaling kinase [64, 71].

What microbes should be mined?

There has been an ongoing debate as to which microorganisms are the best sources for current and future discovery of natural products. Some scientists have suggested that unculturable microorganisms might serve as “untapped” sources for novel secondary metabolites [41]. With the advent of inexpensive microbial DNA sequencing, it became possible to explore the genetic capacity of different groups of microorganisms, and to ask: (1) which microbial taxa have the highest potential to produce large numbers of complex secondary metabolites with drug-like properties; (2) which taxa have moderate potential; and (3) which have the lowest potential. It stands to reason that effort should be focused on the microorganisms with the highest potential, and those with the lowest potential should not be heavily emphasized. If we use the numbers of type I polyketide synthase (PKS) and non-ribosomal peptide synthetase (NRPS) genes per microbe as yardsticks to measure the potential to produce secondary metabolites (i.e., pathways that contain type I PKS, NRPS, or mixed PKS-NRPS account for well over 60 % of important secondary metabolites discovered over the past 50 years [14], then it is clear that microbes with large genomes are generally more productive sources of secondary metabolites, and that among these the actinomycetes are the most productive [12, 28, 75]. It has been shown that the number of functional NRPS pathways can be estimated by counting the number of mbtH homologs in a microbial genome. MbtH homologs are generally small chaperone proteins (65–75 amino acids) that enhance adenylation reactions of some adenylation (A) domains during peptide assembly by NRPS proteins [12, 38, 54]. Because of the diversity of A domains encountered in NRPS genes, MbtH homologs can be orthologous for related pathways and paralogous for unrelated pathways. This dichotomy renders MbtH homologs ideal surrogates or “beacons” to count the numbers of NRPS pathways, and to triage known and unknown pathways by using low pass sequencing [12, 14]. Twenty-four internal segments of diverse MbtH homologs were concatenated to generate a probe for BLASTp analyses of MbtH-like proteins. The relative homologies to the individual 24 MbtH homolog probes were converted into numerical MbtH codes, which can facilitate the triage process. The MbtH code analysis confirmed that among the actinomycetes with large genomes, there are gifted, average, and not-so-gifted species. The MbtH codes for several gifted actinomycetes suggested that there are many new and novel NRPS pathways to be unraveled and possibly exploited for natural product discovery and development. Some other microbial groups are essentially devoid of mbtH (and NRPS) genes [12]. However, within the Proteobacteria, mbtH homologs are observed in Burkholderia, Photorhabdus, and Xenorhabdus species, and the full extent of secondary metabolite biosynthetic capabilities of these genera are being revealed by genome mining [49, 67].

While prioritization using NRPS and PKS potential is likely to enrich efforts in new molecule discovery in the near future, caution is warranted in focusing exclusively on large modular biosynthetic systems deriving from highly characterized systems. If we look beyond NRPS and type I PKS pathways, which require substantial coding capacity, it is becoming apparent from recent genome mining successes that many microbes with smaller genomes encode ribosomally synthesized and post-translationally modified peptides (RiPPs) or phosphonates [24, 44, 52]. Perhaps these discoveries will contribute to a more robust drug discovery process in the coming years. Moreover, more traditional microbial natural product discovery efforts continuously reveal new molecular diversity that is only associable with gene clusters after the fact, due to a lack of biosynthetic precedence. For instance, considering the structures of platensimycin, an oxidatively modified cyclic terpenoid [68], and merochlorin, a mixed polyketide-terpenoid natural product [45], both would unlikely be prioritized by comparison to well-characterized biosynthetic systems. These discoveries underline the continued importance of biosynthetic studies to understand newly discovered natural products.

How should microbial genomes be mined?

Genome-mining campaigns can range in complexity from simple trial-and-error approaches to breathtakingly ambitious programs in synthetic biology. Currently, these campaigns can be organized into one of two major categories: those involved in eliciting expression in the encoding producing organism (homologous expression) and those endeavoring to recapitulate pathways in non-producing hosts (heterologous expression) (Fig. 2). In the case of homologous expression, the power of secondary metabolic prescience alone should not be underestimated and foreknowledge of metabolic potential has forever changed the process of natural product discovery. Simply ‘looking harder’ via growth condition parameterization with structural guidance by (bio) analytical chemistry has unlocked a significant fraction of unknown metabolites [23, 32]. Indeed, this strategy saturated the discovery pipeline at Ecopia Biosciences and lead to farnesylated benzodiazepinone ECO4601, the first genome mining-derived natural product entered into human clinical trials in 2003 [5, 36]. However, there are limits to eliciting gene cluster expression and product detection by media formulation and analytical foreknowledge. Genetic regulation of secondary metabolism remains poorly understood across the diversity of secondary metabolite-producing organisms and identifying a low-abundance discrete predicted metabolite from within a crude metabolome is a non-trivial challenge. Consequently, more recently, a host of methods for activating regulated secondary metabolism have been developed with the ability to unlock tightly regulated clusters [11, 26, 51, 61, 76]. In combination with ‘looking harder’, and given the rate-limiting steps of isolation and structural elucidation, these approaches can likely occupy discovery efforts for some time. Heterologous expression can often aid in the production of adequate levels of compound for evaluation, particularly when expression hosts are derived from industrial production strains or highly engineered laboratory strains [9, 11, 34, 40]. Synthetic biology approaches are also being developed which focus on refactoring secondary metabolic gene clusters, either in producing hosts via genetic recombination, or in ‘clean’ heterologous hosts via gene synthesis [22, 62]. Heterologous expression approaches have many advantages in that they are not limited to gene clusters derived from cultivatable microbes, that the products of heterologous expression can be identified by comparatively straightforward differential metabolomic analysis of the clean and transformed host, and that heterologous systems, once established, can be readily genetically manipulated to diversify the encoded natural products. This special issue addresses both major categories of approaches. The work of Zhu et al. [76], Ochi et al. [61], and Yoon and Nodwell [72] focus specifically on creative new methods to activate secondary metabolism in actinomycetes and the work of Gomez-Escribo and Bibb [34], Ikeda et al. [40] and Cobb et al. [22] discuss cutting-edge approaches to heterologous expression.

Fig. 2
figure 2

Two major categories of approaches being investigated in genome-mining efforts. In heterologous expression, whole gene clusters are mobilized into expression strains and differentially analyzed to identify new compounds. In homologous expression approaches, endogenous transcriptional, translational, or metabolic elements are manipulated, either by mutation or (bio) chemical stimulants to activate secondary metabolite production. In either case, these approaches entail the labor-intensive process of identification of new peaks within a metabolome and isolation of these compounds for structural elucidation

Full size image

Indeed, there are now no conceptual barriers towards the future unlocking of all previously cryptic and/or orphan secondary metabolic gene clusters. However, many significant technical and practical barriers must be addressed in order to realize the full potential of genome mining. Arguably, improvements in methods for isolation and structure elucidation of secondary metabolites from complex extracts have not kept pace with genomic advances, and it is likely that these will become the rate-limiting step in metabolite discovery [2, 56]. With the current state-of-the-art, isolation and elucidation of new natural products at moderate abundance (1 mg/l) require weeks to months or longer per compound for full characterization. Clearly this is an area in need of substantial innovation if the goals of genome mining are to be even partially realized. Remaining to be determined is the rate and cost per metabolite/gene cluster via genome-mining approaches. For instance, refactoring secondary metabolic gene clusters via a synthetic biology approach is highly resource intensive using standard technologies. In considering that most secondary metabolic gene clusters consist of dozens of genes (at a typical length of 20–150 kbp), and taking into account current gene synthesis costs and the time entailed in homologous recombination-based assembly of gene clusters and the likely requirement for generating multiple cluster variants, it seems probable that the cost per compound will be quite high ($50,000 USD or more per compound). Additionally, a method for universal heterologous protein expression has not yet been discovered, as indicated by the Protein Structure Initiative, in which only 18 % of human proteins could be expressed and purified in soluble form [19]. Correspondingly, in large secondary metabolite gene clusters bearing dozens of open reading frames, the potential for failure is quite high. For these reasons, homologous producer-cluster activation approaches, although they may unlock only a fraction of secondary metabolism, may prove in the short term to yield a substantially lower cost per compound and saturate isolation/elucidation pipelines in the immediate future. Ultimately however, it is expected that rapid and inexpensive gene synthesis and effective expression and more effective translation technologies will be developed to access not only the full potential of genome mining of the full natural pharmacopeia of culturable and non-culturable organisms, but also new natural chemical entities via mutasynthesis and recombineering.

How can genome mining be leveraged?

Enrichment for gifted microbes

It has been estimated that ~1026 actinomycete colony-forming units (mostly spores?) exist in the top 10 cm of soil covering the Earth, but only ~107 actinomycetes have been screened for secondary metabolite production by the pharmaceutical industry over the past 50 years [6]. Brady and coworkers [62, 63] have demonstrated that genome sequencing can be applied to environmental DNA (eDNA) extracted from soils, and the diversity of PKS and NRPS sequences assessed. This approach could be extended to include “beacon” analysis with the MbtH multiprobe and other pathway-specific probes to identify soils that contain gifted actinomycetes for cultivation and whole-genome sequencing.

Many actinomycete genera have traits that are amenable to enrichment by antibiotic selection and other nutritional methods [35]. As the most “gifted” genera are identified by sequencing many different species already identified, then specific enrichments can be used to build substantially larger collections of gifted microbes for whole-genome sequencing.

Coupling genome mining with combinatorial biosynthesis for accelerated evolution

In the past three decades, progress has been made on developing methodologies and biochemical rules for combinatorial biosynthesis of complex PKS, NRPS, and other pathways [7, 13, 22, 70]. For NRPSs, rules have been developed for coupling of domains to maintain proper upstream and downstream protein–protein interactions [13]. For instance, there are three types of condensation (C) domain for coupling fatty acid to l-amino acid, l-amino acid to l-amino acid, and d-amino acid to l-amino acid. Likewise, there are three types of thiolation (T) or peptidyl carrier protein (PCP) domain, depending on downstream interactions with C domains, epimerase (E) and C domains, or thioesterase (Te) domains. Successful genetic engineering of NRPS pathways requires the correct assembly of the right types of C and T domains in the right context (e.g., by keeping homologous C and A domains together whenever possible). Although many different NRPS modules have been already discovered, there are not many combinations of highly specialized modules (e.g., the combinations of C for fatty acid coupling have a limited number of A domain partners for initiating lipopeptide biosynthesis). Genome mining should provide a wealth of new NRPS parts and devices to facilitate expanded synthetic biology approaches to combinatorial biosynthesis of novel NRPS pathways. The same should be true for PKS and mixed NRPS/PKS pathways, and tailoring reactions (e.g., sugar biosynthesis and glycosyl transfer, hydroxylations, and transfer of methyl groups).

Modified natural products can also be generated by mutation outside of the modular biosynthetic systems. For instance, pactamycin analogs have been generated via mutasynthetic approaches that have superior activities to the progenitor natural products [3, 50], glycovariants have been generated of a large number of natural products via enzymatic methods [31] and genetic knockouts [29].

How can progress on genome mining be accelerated?

Knowing which peak to isolate

The success of genome mining hinges on being able to correlate metabolites of interest within complex biological extracts following pathway activation or heterologous expression. This endeavor is not a trivial process for either homologous or heterologous expression categories. While in theory this process should be simplified by heterologous expression approaches, there is no guarantee that new metabolites will be easily observed (e.g., due to low abundance, lack of chromophore to simplify detection, chemical incompatibility with selected extraction conditions, etc.). In response to this limitation, more sophisticated statistical methods are being developed to identify new compounds generated by modified growth conditions or recombinant strains [26, 39, 46]. The development of genomisotopic approaches [37] is another methodology that has the potential for advancing efforts for identifying metabolites of interest.

Speeding up isolation of secondary metabolites from complex extracts

Extraction followed by chromatographic partitioning and separation remain the primary means of isolating compounds from cultures. In most cases, this process involves multiple steps with target compound losses at every step resulting in diminishing yields of compounds throughout the purification process. In nearly all discovery efforts, compound purity, which is a function of sample homogeneity and stability, is the threshold parameter for initiating structure elucidation. It is also the rate-limiting step, often requiring scale-up fermentation, extraction, and purification protocols. The isolation process alone can require weeks to months to perfect and the follow-up elucidation process can also require an equivalent amount of time, depending on the structural complexity and inherent properties of the sample. The ideal process would be small molecule ‘teleportation’ from crude extracts into tubes, a process that is surprisingly not beyond the reach of ion soft landing mass spectrometry, an analytical technique that separates ionized compounds from mixtures using a mass analyzer and lands them on a surface forming compound arrays [66]. This technology has already been successfully demonstrated in purifying and landing small molecules and even active enzymes in sufficient quantities for analysis, but has not been scaled to isolate compounds in sufficient quantities for cryogenic NMR. In the absence of the wide-scale availability of this technology, assuming isolation and elucidation workflows cannot be accelerated otherwise, genome mining efforts may grind to a snail’s pace.

Developing universal tools for gene manipulation and expression in producing organisms

Secondary metabolite-producing organisms are taxonomically diverse and, despite decades of research, generalizable tools for genetic manipulation (e.g., transformation, intergeneric conjugation, homologous recombination, gene expression) are sparse across phyla. Even within a genus or species, quirks in gene uptake, regulator and genetic marker compatibility can confound efforts at genetic manipulation required for many homologous and heterologous expression techniques. The development of reliable genus- and species-specific tools for genetic manipulation of secondary metabolic gene clusters will be essential for rapid progress in genome mining.

Along the same line, a universal strategy for up-regulating the desired genetic elements for production of new metabolites would have huge ramifications for genome mining efforts. Evidence is increasing that a significant fraction of secondary metabolism in microorganisms can be activated to isolable levels by chemical and biochemical cues [48, 51, 61, 67, 72, 76]. A molecular understanding of the connection of these cues to specific transcriptional, translational, or other metabolic elements across genera would no doubt be very valuable in this context. Ideally, a comprehensive signaling network will be mapped in response to a potential elicitor that can be extended beyond a single metabolic pathway or species, thus helping to streamline the genome-mining process.

Synthetic biology tools

Heterologous expression of synthetic or cloned gene clusters is also in need of robust methods for the synthesis and assembly of small to large gene clusters in a reliable, inexpensive, and a high-throughput format. The aforementioned problems of functional gene expression will likely require the generation of multiple variants of targeted gene clusters that often consist of dozens of large biosynthetic genes such as those found in modular PKS and NRPS systems. De novo production of these genetic variants poses technological challenges in gene assembly and potential financial issues until costs per base decline. Operationally, refactoring polycistronic clusters also requires multiple orthogonal tools for selection, promoting, or otherwise marking, reassembled gene clusters, the feasibility of which has recently been described by refactoring a 20-gene, seven-operon nitrogen fixation cluster from Klebsiella oxytoca and functional expression in Escherichia coli [65].

Merge with the high-throughput model

The dominant paradigm in drug discovery, for better or worse, is via high-throughput screening (HTS) of large chemical libraries against biochemical and/or phenotypic assays. Notwithstanding the modest track record of this approach, the associated technologies are immensely powerful tools for efforts in drug discovery. Natural product discovery, which is becoming strongly associated with genome mining, would benefit greatly if natural products can be assembled in sufficient numbers, or if technology existed to assay them in sufficient numbers, to be complementary and compatible with current HTS methods and paradigms.

Investment in fundamental biosynthetic research

Bioinformatic approaches for the estimation of the secondary metabolic products of sequenced gene clusters [16, 17] and future engineering studies to generate chemical diversity are entirely dependent upon biosynthetic precedent established by basic research into the biochemistry of secondary metabolism. Indeed, decades of unraveling the molecular logic of NRPS and PKS systems has provided a sound foundation for searching genomes and predicting the chemical output (i.e., metabolite identity). As a relatively recent example, progress in understanding the biosynthesis of RiPPs has unleashed a torrent of identification of gene clusters encoding this previously poorly understood class of compounds, and created an entire new category of genome mining and synthetic biology efforts [52]. There are undoubtedly many such uninvestigated systems for currently known secondary metabolites that could create new domains for genome mining. Thus, a continued investment into unraveling the underlying biosynthetic mechanisms of structurally diverse metabolites will foreseeably refine what is meant by a “gifted” organism.

Who should fund future progress in genome mining?

In the past, natural product discovery and development has been mainly funded by large pharmaceutical companies or chemical companies with animal health or plant sciences subsidiaries. This worked well when discoveries came easily, and returns on investments were sufficient to drive the process, but most pharmaceutical companies have abandoned natural products discovery during the past two decades. More recently, biotechnology companies have been carrying much of the load, but no individual company has the resources to fully exploit the rapidly developing field of genome mining, and develop it into a robust discipline commensurate with its sizable potential. It would seem that this is an opportune time for the NIH, NSF, and DOE in the US and other funding agencies in Europe and Asia to put sizable resources into bringing this important new discipline to a technological level commensurate with its potential to generate new molecules for drug discovery not obtainable by medicinal or combinatorial chemistry. As an example, the DOE has funded a Microbial Genome Project that focuses on mission areas of alternative fuels, global carbon cycling, and biogeochemistry (http://www.jgi.doe.gov/CSP/user_guide/). This approach has generated important fundamental information on a few actinomycetes; in particular, the finished quality of the genome sequences assures a high level of confidence in the assembly of complex PKS and NRPS pathways. This in turn can serve as part of an important expanding baseline for current and future genome mining. It is intriguing that among the small number of actinomycetes sequenced so far in this program, Actinosynnema mirum [47] and Streptosporangium roseum [60] can be classified as gifted by the MbtH counting method [14], and Saccharomonospora viridis has yielded a cryptic daptomycin biosynthetic gene cluster [10], even though none of these strains were sequenced with secondary metabolite discovery in mind. To further exploit this approach of finishing subsets of microbial genomes, it would be highly valuable to develop a program to generate finished genome sequences of many actinomycetes that produce interesting secondary metabolites, and to fully annotate all known secondary metabolite clusters. Having a baseline of all known secondary metabolite pathways will accelerate the discovery of novel secondary metabolite pathways, while streamlining the de-replication of known pathways, the bane of natural products discovery in industry that has impeded progress for the past three decades.