Using natural products for drug discovery: the impact of the genomics era

Abstract

Introduction

Evolutionarily selected over billions of years for their interactions with biomolecules, natural products have been and continue to be a major source of pharmaceuticals. In the 1990s, pharmaceutical companies scaled down their natural product discovery programs in favor of synthetic chemical libraries due to major challenges such as high rediscovery rates, challenging isolation, and low production titers. Propelled by advances in DNA sequencing and synthetic biology technologies, insights into microbial secondary metabolism provided have inspired a number of strategies to address these challenges.

Areas covered

This review highlights the importance of genomics and metagenomics in natural product discovery, and provides an overview of the technical and conceptual advances that offer unprecedented access to molecules encoded by biosynthetic gene clusters.

Expert opinion

Genomics and metagenomics revealed nature’s remarkable biosynthetic potential and her vast chemical inventory that we can now prioritize and systematically mine for novel chemical scaffolds with desirable bioactivities. Coupled with synthetic biology and genome engineering technologies, significant progress has been made in identifying and predicting the chemical output of biosynthetic gene clusters, as well as in optimizing cluster expression in native and heterologous host systems for the production of pharmaceutically relevant metabolites and their derivatives.

1. Introduction

Humans have long recognized the rich repertoire of molecules produced by microorganisms as a fertile source of therapeutics. Evolutionarily selected for interactions with biomolecules, the remarkably diverse and complex core chemical scaffolds of natural products have a higher chance to be biologically active than compounds from combinatorial synthesis approaches. In fact, more than half of clinical drugs from 1981 to 2014 are derived from or inspired by natural products [1]. In the pre-genomic era, most natural product discovery efforts employed a ‘top-down’ approach driven by the screening biological samples for desirable bioactivities, followed by compound isolation and characterization [2]. Yet by the 1990s, such strategies had largely failed to uncover new natural products as pharmaceutical companies struggled with high rediscovery rates and de-emphasized their natural product discovery efforts. Driven by the rapidly decreasing cost and increasing throughput of DNA sequencing technologies [3], significant progress in genomics has renewed interest in natural product discovery [4]. Rapidly expanding microbial genomic and metagenomic datasets reveal a vast number of biosynthetic gene clusters in nature, which are predicted to encode far more natural products than what we have characterized to date [5]. Uncovering novel natural products is not only fascinating but also highly pertinent in face of today’s burgeoning drug resistance and health problems.

The conventional ‘top-down’ discovery approach can only provide us access to a small fraction of microbial biosynthetic potential given that majority of microorganisms cannot be isolated or cultured [6, 7]. Furthermore, even in culturable organisms, the encoded secondary metabolites of many biosynthetic gene clusters (BGCs) are unknown [8, 9]. This may be due to strong down-regulation of product biosynthesis at the transcriptional, translational and/or post-translational levels in the absence of the right activating cues in the laboratory. Alternatively, secondary metabolites that are produced at very low yields may escape detection and characterization. The ability to efficiently and strategically access these vast unexplored chemical resources will be invaluable to drug discovery. Besides relying on serendipitous discoveries of bioactive compounds, natural product discovery is now increasingly driven by genomics and focused on BGCs that are predicted to encode novel biomedically relevant molecules [4]. A better understanding of genotype-chemotype relationships informs researchers about the chemical logic of natural product biosynthesis and guides ‘bottom-up’ approaches that begin with genetic manipulation for the “detectable” production of target metabolites. Natural product discovery, which used to be predominantly an analytical chemistry problem, has become a challenge in genomics and molecular biology on how to manipulate relevant genes and sequences to produce the desired encoded metabolites.

In this review, we highlight key technical and conceptual advances in genomics-driven natural product discovery. These include bioinformatics-guided identification of BGCs in genomes and metagenomes, pathway and host engineering strategies for the activation of silent gene clusters in native and heterologous systems, as well as combinatorial biosynthesis for generating natural product analogs. This is not meant to be an exhaustive review and we apologize if key studies were inadvertently left out.

2. Genome-mining for natural product BGCs

The development of next-generation sequencing has significantly accelerated the sequencing of microbial genomes at much reduced costs, leading to an exponential growth of genomic sequencing data with over 30,000 sequenced bacterial genomes currently deposited in public archives [10]. However, an immediate bottleneck is our current ability to readily analyze and process such large volumes of data. For natural product drug discovery, this means to identify potential secondary metabolites BGCs that encode for novel bioactive metabolites from microbial genomes. Toward this end, a variety of genome-mining tools have been developed.

AntiSMASH (antibiotics and secondary metabolite analysis shell) allows in silico identification of BGCs in bacterial and fungal genomes [11, 12]. Studies using antiSMASH have revealed numerous BGCs in bacteria and fungi [13, 14, 15, 16], many of which encode novel bioactive metabolites. Notable features of antiSMASH version 3.0 include more comprehensive analysis of gene clusters, annotation of key residues in biosynthetic enzymes, and the inclusion of a lantipeptide-focused module [12, 17]. In addition, there are genome-mining tools specific for particular biosynthetic pathways or species. SMURF (secondary metabolite unknown region finder) focuses on genome-mining in fungi [18], while BAGEL3 is developed for the analysis of bacteriocin encoding-genes and relatively unexplored classes of posttranslationally modified peptides [19]. Multiple tools including NP.searcher and ASMPKS are available for analyzing PKS (polyketide synthase)/NRPS (non-ribosomal peptide synthetase) pathways and predicting their substrates [20, 21, 22]. Recognizing knowledge gap between isolated PKs/NRPs and their gene clusters (i.e. only 10% of PKs/NRPs are associated with gene clusters), Dejong et al. introduced a retro-biosynthetic in silico analysis platform to link known PKS and NRPS metabolites to emergent gene clusters [23]. The platform includes a retro-biosynthetic analysis component (GRAPE) and an alignment algorithm for cheminformatics (GARLIC), and is useful for identifying clusters encoding for known and novel molecules. Furthermore, a consortium effort to standardize BGC annotation yielded the Minimal Information about a Biosynthetic Gene cluster (MIBiG) format, which encourages more systematic and consistent characterization of BGCs by the community [12, 24]. Readers are referred to recent reviews for detailed discussion and comparison of genome-mining strategies focused on natural product discovery [25, 26, 27, 28].

Plant-derived natural products are also important drug sources [1], but compared to bacterial and fungal systems, genomics-driven natural product discovery in plants is in the early stages. Genome and transcriptome analyses led to the recent recognition of BGCs as common genomic features in plants and supported the regulated production of many bioactive natural products, including alkaloids and terpenoids [29, 30]. With an expanding list of plant genomes, natural product biosynthesis pathway discovery is increasingly genomics-driven compared to the pregenomic approach of starting with the biochemical characterization of a biosynthetic gene. There are already efforts to standardize the analyses and curation of plant BGCs with their microbial counterparts to support future genome mining endeavors [24, 31].

As predictive software programs become increasingly robust and streamlined for genome mining (Figure 1), researchers will not need expertise in bioinformatics to perform a variety of analyses, yet one should be mindful of the limitations. Since most tools employ predictive rules based on characterized pathways, they are geared towards identification of major families of biosynthetic gene clusters such as PKSs/NRPSs, they are less reliable with regard other pathways. Moreover, most software programs have been unable to accurately predict metabolite structures [25]. Nevertheless, BGCs identified in silico are excellent starting points for downstream experimental investigations.

Figure 1. Workflow of genome-mining for natural product drug discovery.

Genome-mining software such as antiSMASH is able to analyze the sequenced genome in silico, identify potential biosynthetic gene clusters and predict core structures of encoded metabolites. The predicted BGCs are starting points for downstream experimental activation and validation.

3. Accessing silent natural product gene clusters

Secondary metabolism is tightly regulated [32] and the fact that many BGCs remain “silent” under laboratory conditions presents a major challenge to “activate” these BGCs and assess the therapeutic potential of their encoded natural products. As metabolic profiling using advanced analytical methods continue to uncover new compounds that have eluded detection due to low production yields (Section 4.2), the activation of BGCs that are not or inadequately expressed under laboratory conditions have also resulted in the discovery of many diverse natural products exhibiting a range of bioactivities [5]. Here we highlight the major strategies to activate BGC expression with a focus on natural product discovery. Readers are referred to recent reviews for details on the metabolic, pathway and genome engineering approaches in native and heterologous host systems for secondary metabolite production [33, 34, 35].

3.1. Activation of silent BGCs in native hosts

Native hosts are endogenous producers of the natural products of interest. Logically, they possess sufficient and necessary cellular factors for metabolite biosynthesis, including relevant precursors, pathway regulators, and transporters. In principle, under the ‘right’ conditions with the appropriate biological and/or environmental cues, metabolite production can be elicited in native hosts. Some native hosts can successfully express silent gene clusters when grown under alternative conditions, while others require targeted genetic manipulations such as the introduction of heterologous promoters [36, 37] or perturbation of transcriptional and post-transcriptional regulatory mechanisms (vide infra). The development of genome editing and genome engineering technologies in natural product-relevant organisms will greatly enhance the scalability of these activation strategies in native hosts [34]. Notably, genetic manipulation can be challenging for non-model native producing hosts. While developing genetic tools for uncharacterized wild type host organisms may be tedious and time consuming, the chances of obtaining BGC-encoded secondary metabolites from native producers, which likely possess all the metabolic and biosynthetic requirements, are high.

3.1.1. Variation of growth conditions and small molecule inducers

Systematic variation of growth conditions, also known as the OSMAC (one-strain, many compounds) strategy, has traditionally been employed to explore the biosynthetic capabilities of isolated strains. For Aspergillus nidulans, simple variation of fermentation conditions such as media and culturing period can dramatically affect its metabolite profile, leading to the identification of several potential anti-cancer compounds [38, 39]. But more often than not, specific conditions are required to elicit the expression of target gene clusters in microbes. In the discovery of lugdunin, a novel peptide antibiotic produced by the human commensal Staphylococcus lugdunensis, Zipperer et al. noted that metabolite production was only elicited when S. lugdunensis was cultivated under iron-limiting conditions on solid agar [40]. Other physical triggers of metabolite production such as rare earth elements have been reported in Streptomyces spp. [41, 42]. Nonetheless, screening of different conditions in multiple strainsoften suffers from limited reproducibility and can quickly become impractical when considering a range of variables. In addition, obscure conditions cannot be predicted and require empirical studies. Notably, innovative culture strategies have been developed for natural product discovery. Designed to isolate single microbial cells in diffusion chambers that mimic the natural environment, iChip allows the recovery of up to 50% of the microorganisms in environmental samples [43]. Use of iChip led to the identification of lassomycin, a ribosomally synthesized cyclic peptide antibiotic with activity against Mycobacterium tuberculosis [44], and teixobactin, a new class of antibiotics made by a newly identified species Eleftheria terrae that shows broad-spectrum antimicrobial activity without detectable resistance [45]. In addition, a high-throughput assay to identify small molecule elicitors for targeted BGCs in Burkholderia thailandensis was recently reported [46]. Since microorganisms naturally reside in a complex environment and constantly interact with other species, co-cultivation of different microorganisms can also induce expression of silent BGCs [47, 48, 49, 50].

3.1.2. Manipulation of regulators

Genome mining studies have uncovered global and pathway-specific transcriptional regulators that can be manipulated for BGC activation. Overexpression of transcriptional activators leads to production of “hidden” metabolites [51, 52]., including the production of an unprecedented 51-membered glycosylated macrolide [53]. On the other hand, deletion of pathway-specific repressors has been demonstrated to activate a number of silent BGCs and trigger metabolite production [54, 55, 56]. Likewise, manipulation of transcription factors in plants has yielded the overproduction of desired metabolites [57]. For biosynthetic pathways without obvious regulators, global alteration of gene expression may be employed. In streptomycetes, specific mutations in RNA polymerase or ribosomal proteins can affect gene expression and production of new antibacterial compounds [58]. In addition, a reporter strain-based mutant selection strategy has been developed, enabling unbiased screening of activation conditions. Using this selection, two novel aminoglycosides were discovered in Streptomyces sp. PGA64 [59].

3.1.3. Perturbation of epigenetic control

The importance of epigenetic modifications such as DNA/histone methylation and acetylation in regulation of eukaryotic gene expression has been recognized in recent years [60]. Manipulation of fungal epigenome potentially allows BGC activation. In A. nidulans, deletion of epigenetic modification enzymes such as histone deacetylase (HDAC) or histone methyltransferase has been shown to activate several silent BGCs and dramatically change its metabolite profile [61]. In addition, chemical perturbation of the fungal epigenome and led to the discovery of new metabolites [62, 63]. However, current strategies for epigenetic modulation result in global changes, which cannot be predicted. With advances in DNA sequencing and editing technologies, the precise location and function of an epigenetic marker in a fungal genome are likely to be recognized and manipulated to enable targeted BGC activation.

3.2. Activation of silent BGCs in heterologous hosts

To access silent BGCs in genetically recalcitrant or uncultured microorganisms, heterologous expression of BGCs in a genetically tractable organism represents an attractive approach. For functional expression in heterologous hosts, target BGCs often require additional refactoring, such as replacement of the native promoters with well-characterized promoters, insertion of appropriate ribosomal binding sites or terminators. Since de novo synthesis and genetic manipulation of large gene clusters (>40 kb) can be challenging, a number of synthetic biology tools have been developed to facilitate the reconstruction of biosynthetic pathways in heterologous hosts.

3.2.1. Direct capturing and refactoring of gene clusters

Advances in recombinant DNA and synthetic biology technologies allow BGCs to be directly captured into compatible vectors and refactored for expression in heterologous hosts [33]. For example, RecET-mediated linear-linear recombination in E. coli is employed to capture large (>50 kb) megasynthases into expression vectors [64]. Similarly, yeast-based transformation-associated recombination (TAR) allowed the capture and activation of a 67 kb NRPS cluster from Saccharomonospora in Streptomyces coelicolor to produce a new antimicrobial lipopeptide, taromycin A [65]. Use of the RNA-guided Cas9 nuclease circumvents the requirement for unique restriction sites flanking the target BGCs and facilitates direct cloning of large clusters (up to 100 kb) by Gibson assembly or TAR [66, 67]. Recently, the combined use of TAR with CRISPR/Cas9 in a yeast-based promoter-engineering platform mCRISTAR enabled the efficient multiplex replacement of eight promoters to activate the tetarimycin A cluster in S. albus [68].

Besides direct capture, various methods have been developed for scarless assembly of BGCs from de novo synthesized DNA fragments and PCR products [69]. These assembly techniques should also facilitate the engineering of biosynthetic gene clusters for combinatorial biosynthesis (Section 6) as well as cluster refactoring for heterologous expression. For example, Luo et al. utilized yeast DNA assembler to simultaneously assemble and refactor a PKS-NRPS cluster from Streptomyces griseus by inserting a constitutive promoter in front of each of the six biosynthetic genes. Heterologous expression of the refactored pathway in Streptomyces lividans yielded new polycyclic tetramate macrolactams [70].

3.2.2. Optimization of heterologous hosts

When choosing a heterologous host for expression, it is important to consider whether the production host possesses the necessary metabolic precursors, enzymatic machinery and appropriate regulatory systems for target BGCs. Uncoordinated expression of biosynthetic genes may result in imbalanced metabolic flux and production of toxic intermediates. For instance, Escherichia coli, which is one of the widely used Gram-negative bacteria for heterologous expression, has been engineered to express phosphopantetheine transferases (PPTases) for PKS/NRPS activation and production of PK/NRP natural products [71]. In order to improve secondary metabolite production, it is often necessary to optimize the heterologous production hosts. Multiplexed automated genome engineering (MAGE), which enables facile introduction of genetic diversity at targeted loci during DNA replication, has been used to effectively increase lycopene production titer in E. coli [72]. The CRISPR/Cas9 system was recently reconstituted and used to delete genes and entire BGCs in multiple Streptomyces [73, 74, 75, 76], paving the way towards genome-minimized hosts for heterologous expression of biosynthetic pathways from actinomycetes. Conceivably, extension of these valuable technologies to optimize heterologous host systems will facilitate BGC characterization and natural product drug discovery.

4. Integrated strategies to access genome-encoded small molecules

4.1. Bioinformatics-guided synthetic approach

Given the fact that genetic manipulation of native and heterologous hosts can be time-consuming, a new paradigm in natural product drug discovery which bypasses the need for strain cultivation, gene cluster expression and product isolation from fermentation broths was reported recently. Based on the primary sequence of NRPS clusters in the human microbiome, Chu et al. predicted and chemically synthesized a small library of peptides for the encoded NRP and identified humimycin, a potent anti-MRSA (methicillin-resistant Staphylococcus aureus) peptide with a new mechanism of action [77]. This work represents the first example of the synthetic-bioinformatic natural product (syn-BNP) approach for drug discovery and may be especially helpful in metagenomics studies (Section 5). However, as mentioned earlier, in contrast to well-annotated information on PKS/NRPS clusters, prediction algorithms for other types of BGCs are relatively limited; in those cases, structures of the synthetic molecules can significantly deviate from the encoded natural products and hence may fail to elicit the desired bioactivity. Whether the syn-BNP approach can be generalized to most BGCs depends on future development for more accurate bioinformatics predictions.

4.2. Mass spectrometry-guided genome mining

Driven by advanced analytical methods, mass spectrometry-guided genome mining involves the iterative matching de novo MSⁿ sequence tags to genomic features, using retro-biosynthetic logic in order to connect secondary metabolites to their BGCs [78, 79]. Matched BGC sequence information may be harnessed to further elucidate compound structures and/or to identify additional molecular features for searching. Mass spectrometry-guided genome mining has been extended to connect groups of structurally related molecules with entire BGC families [80]. Notably, the genetic tools and pathway engineering strategies described in Section 3 will be invaluable to metabolomics-guided genome mining efforts by the functional validation and characterization of BGCs and their metabolites. Furthermore, when coupled with principal-component analysis of metabolite profiles, genetic mutants can be used to identify key molecular features that correspond to a BGC of interest [37]. While metabolomics-guided genome mining has been largely employed for the discovery of peptide (ribosomal and non-ribosomal) natural products due to relatively well-characterized biosynthetic logic, the approach can be extended to other groups of secondary metabolites such as glycosylated natural products [81] and possibly polyketides.

5. Metagenomics-driven drug discovery

More than 99% of microorganisms in the environment have resisted laboratory cultivation and this uncultured microbial majority represents a vast chemical treasure trove [6, 7]. Metagenomics, which involves the direct capture and analysis of environmental DNA (eDNA), allows culture-independent and unbiased access to microbial biosynthetic potential otherwise missed by traditional methods requiring the isolation and cultivation of pure microbial cultures [82, 83]. In this section, we highlight the technical and conceptual advances in metagenomics that led to the discovery of new natural products from diverse environmental niches including soil, marine environments and the human body. Readers are referred to recent reviews for more details [28, 84].

Given that BGCs constitute only a small fraction of microbial genomes [85, 86], the ability to efficiently search metagenomic libraries for rare clones containing relevant biosynthetic genes involved in secondary metabolism and encoding for novel natural products is imperative. Screening is especially challenging for large complex metagenomes such as soil, which can contain up to 10⁵ unique species and require multimillion-membered mega-libraries for sufficient coverage [87, 88]. Conventional function- and sequence-based methods involve screening libraries for easily observable phenotypes or the presence of target DNA sequences, respectively (Figure 2a).

Figure 2. Metagenome-driven drug discovery with an alternative targeted sequence-based pipeline.

(a) Environmental DNA (eDNA) is extracted, cloned and ligated into a shuttle vector, and then transformed into a host cell to create a metagenomic library. Conventional function- and sequence-based methods involve screening metagenomic libraries for easily observable phenotypes or the presence of target DNA sequences, respectively. Target biosynthetic pathways are identified, assembled and functionally expressed through a variety of methods to obtain the encoded natural product for characterization. (b) Alternatively, PCR is used to profile the biosynthetic pathways present in crude eDNA samples. The PCR amplicons are phylogenetically organized to predict the chemical output of the biosynthetic pathways. Samples predicted to harbor novel or target biosynthetic gene clusters are prioritized for library construction, clone recovery, and heterologous production. Reproduced and adapted from [84] with permissions from Springer.

Function-based metagenomics does not require prior knowledge of genetic information and has been successfully applied to uncover novel chemical cores and interesting biotransformations associated with their biosynthesis. Facile visual screens for pigment production or antibiosis (zone of inhibition) led to the discovery of structurally distinct antibiotics such as turbomycin [89], N-acyl amino acids derivatives [90, 91] and isocyanide-containing compounds [92, 93], which represent some of the first natural products identified from metagenomics libraries. Depending on the target compounds and library size, other functional screens include direct metabolite analysis by LC-MS methods [94, 95], reporter/biosensor-based screens for metabolite-responsive gene expression [96, 97] and enzymatic assays [98, 99]. Enzyme activities unique to secondary metabolism can be harnessed to greatly improve the efficiency of identifying eDNA clones with functional biosynthetic gene pathways for downstream phenotypic screens. Required for the posttranslational activation of NRPSs and PKSs, PPTases were harnessed in a phage display study to recover NRPS and PKS sequences from environmental samples [100]. By coupling pigment production or cellular growth to PPTase function, complementation of PPTase activity can be used to screen or select for eDNA clones that contain PKS and/or NRPS gene clusters [87, 101]. Notably, the scope of function-based screening is limited by the insert sizes of cosmid/fosmid eDNA libraries as well as the choice of library hosts, which may not support functional expression of heterologous BGCs for various reasons including codon bias, lack of biosynthetic precursors or enzymatic activity, and incompatible regulatory systems.

In contrast, sequence-based metagenomics does not require functional reconstitution of biosynthetic pathways in library hosts and allows for the discovery of products of large BGCs (e.g. multimodular PKS and/or PKS clusters) that exceed the insert size for cosmid/fosmid or bacterial artificial chromosome libraries. Conventional sequence-based screens involve PCR with degenerate primers targeting relatively uncommon biosynthetic genes to identify eDNA clones of interest (Figure 2a). In an alternative approach, barcoded PCR amplicons of common conserved biosynthetic features (e.g. adenylation, condensation, ketosynthase domains) within environmental samples are sequenced to generate sequence tags that reflect microbial genomic diversity, which can be used to prioritize metagenome mining efforts for the discovery of novel and/or biomedically relevant BGCs and metabolites. Akin to the reconstruction of bacterial phylogeny from 16S rRNA sequences, phylogenetic analysis software such as eSNaPD (environmental Surveyor of Natural Product Diversity) and NaPDos (Natural Product Domain Seeker) classify eDNA-derived gene clusters by comparing the sequence similarity of environmental sequence tags to those from characterized BGCs [102, 103]. By recovering eDNA sequences related to known BGCs, this strategy has been successfully employed to discover new bioactive glycopeptide, epoxyketone and anthracycline congeners [87, 104, 105]. New chemical functionalities can also be discovered using this method. A new subclass of natural tryptophan dimers possessing a pyrrolinium indolocarbazole core was uncovered by focusing on eDNA-derived gene clusters with sequence tags that are not closely associated with characterized tryptophan dimer BGCs [88]. In general, while targeted metagenome mining may not be suitable for the discovery of fundamentally distinct chemical entities, it can be used to survey the biosynthetic potential of environmental samples as well as identify and recover novel variants of pharmaceutically relevant BGCs for downstream analyses (Figure 2b).

With advances in next-generation sequencing, bulk eDNA can be directly sequenced and assembled. This approach has been successfully employed to uncover BGCs of interest and genomes of bacterial symbionts that produce patellazoles [106] and chemotherapeutic ET-473 [107]. Yet due to challenges in assembling short reads, which may be partially addressed by pair-end sequencing or increasing sequence coverage, shotgun metagenomics is largely limited to relatively simple metagenomes or pre-enriched (e.g. filtration, differential centrifugation) samples. Rapid progress in third-generation long-read sequencing technologies circumvent the difficulties in assembling highly repetitive genomic sequences and in the near future should enable the assembly of entire BGCs, and potentially genomes, that are represented in metagenomes [108]. Until then, short-read sequencing may be applied to survey the biosynthetic genes present in the microbiome and guide the design of degenerate PCR primers to complement and expand the scope of sequence-based metagenomics. Single-cell genome sequencing allows access to genomes that are underrepresented in metagenomic samples as well as the assembly of genomes from completely uncharacterized microorganisms [109, 110, 111]. Using a combination of deep sequencing, single cell sorting and whole genome amplification, Wilson et al. uncovered a new bacterial taxon Entotheonella that accounts for the production of almost all bioactive polyketides and polypeptides isolated from its host T. swinhoei [14]. The existence of multiple additional BGCs in Entotheonella genomes suggests that the metabolic repertoire of these widely distributed symbionts may rival that of Actinobacteria.

Efficient translation of genetic sequences into actual chemical compounds is critical to the metagenomics approach for natural product discovery. For synthetically accessible natural products whose structures can be reliably predicted from nucleic acid sequences (e.g. NRPs), chemical synthesis circumvents the need for heterologous production systems and accelerates translation of genomes and metagenomes into chemical compounds that can be screened for desired bioactivities [77]. In most cases, accessing the biosynthetic potential of uncultured microorganisms will require heterologous hosts to facilitate expression of eDNA-derived BGCs and production of the encoded metabolite. Enabled by the availability of broad host range shuttle plasmids, the use of distantly related hosts including diverse Proteobacteria [112, 113] and Streptomyces [94, 114] for functional metagenomics provides access to a wider range of bioactive compounds by increasing the likelihood of matching eDNA-derived BGCs with genetically and biochemically compatible hosts for functional expression. Sequence-based metagenomics also benefits from the use of gifted heterologous hosts with the innate ability to functionally express a variety of target biosynthetic pathways. In addition, multiplex pathway and genome engineering technologies to improve heterologous expression strains and optimize cluster expression (Section 3) promise to further expand the scope of uncultured biosynthetic potential that can be accessed and harnessed.

6. Combinatorial biosynthesis

The explosion of BGC sequence information offers critical insights into how remarkably complex natural products covering vast biologically relevant chemical structure space are assembled from a limited set of simple building blocks [115]. BGCs generally encode two groups of biosynthetic enzymes – one group generates key biosynthetic precursors and assembles the core scaffold while the other group derivatizes the scaffolds [116]. Understanding nature’s logic of encoding chemical diversity will enable rational engineering of biosynthetic pathways to obtain analogs of privileged natural product scaffolds or novel natural product-like scaffolds that may be challenging to synthesize chemically for drug discovery [116, 117].

Guided by comparative genomic studies focused on biosynthetic genes, rational engineering or directed evolution of biosynthetic enzymes responsible for generating biosynthetic precursors, assembling core chemical structures and tailoring of scaffolds to expand or alter their substrate selectivity is one of the main strategies for diversification of natural product scaffolds. For example, scaffold diversification of terpenes to create non-native terpenes can be achieved by manipulating isoprenoid precursor supplies and genetic engineering of terpene synthases [118, 119]. The use of promiscuous tailoring enzymes for the derivatization (e.g. alkylation, acylation, oxidation, glycosylation) of natural product scaffolds allows exploration of a defined chemical structure-function space. Naturally occurring and engineered P450s and glycosyltransferases exhibiting broad substrate specificities have been successfully employed for late-stage derivatization of terpenes, PKs and NRPs [115]. New sulfated glycopeptide congeners were obtained in vitro and in vivo by exploiting eDNA-derived sulfotransferases frequently associated with glycopeptide BGCs [120].

The “assembly line” biosynthetic logic of modular PKS/NRPS megaenzymes makes them attractive targets for rational engineering of complex PK/NRP scaffolds [121, 122, 123]. Engineering strategies include mutagenesis, insertion/swapping of domains, modules or subunits to alter 1) type of starter and extender units, 2) extent and stereochemistry of scaffold processing, 3) chain length and release mechanism, and 4) post-PKS/NRPS modifications (Figure 3). By deleting and swapping domains, Sugimoto and colleagues reprogrammed an aureothin BGC into one that produces luteoreticulin, a related compound with distinct biological activities, and other novel derivatives [124]. Advances in synthetic biology strategies such as DNA assembler and other approaches described in section 3.2.1 greatly accelerate genetic manipulation of biosynthetic pathways [125]. Despite early promise [126, 127], the generation of large libraries of PK/NRP analogs by combinatorial biosynthesis is generally hindered by low yielding or non-functional constructs and the paucity of high throughput screens/selections for active pathways. Recently, leveraging on efficient yeast homologous recombination and short homology stretches between pikromycin and erythromycin PKS genes, Chemler et al. created a hybrid PKS library and identified active chimeric enzymes that were capable of producing novel macrolactone analogs [128]. Future application of this approach should shed light on the structure-activity relationship for these megaenzymes and inform genetic engineering efforts. With increasing structural information [121, 122] as well as the availability of algorithms to guide combinatorial library design [129] and capture beneficial variants [130], directed evolution strategies will likely be useful in generating libraries of active PKS/NRPS hybrids for PK/NRP diversification [131, 132].

Figure 3. Diversification of polyketide scaffolds by genetic engineering of PKS clusters.

Possible modifications to polyketide structures are colored according to the responsible PKS domains or stand-alone (post-PKS) enzymes. The changes are not mutually exclusive and can be made in combination. Reproduced and adapted from [121] with permissions from The Royal Society of Chemistry.

While this section focuses on the concept of derivatization and diversification of natural product scaffolds by genetic engineering strategies guided by comparative genomics, genetic engineering of BGCs can also be used in combination with other established strategies to diversify natural product scaffolds. In mutasynthesis, strains are engineered to uptake and incorporate unusual biosynthetic precursors to generate natural product analogs with potentially improved phamacoproperties [133, 134]. Semi-synthesis melds biological and chemical synthetic routes to produce complex natural products such biomedically relevant terpenoids and FDA approved anti-cancer drug ET-743 [135, 136, 137, 138]. Notably, PKSs and NRPSs have been engineered to incorporate alternative starter/extension units harboring fluorine atoms known to improve drug pharmacoproperties [139], or orthogonally reactive handles such as terminal alkynes for combinatorial late-stage derivatization [140, 141].

7. Conclusion

Natural products are still a major source of inspiration for clinical drugs despite the scaling down of natural product discovery efforts by pharmaceutical companies in the mid-1990s due to high rediscovery rates and challenging product synthesis. Rejuvenation of natural product research in recent years can be largely attributed to advances in DNA sequencing, genomics/metagenomics, synthetic biology and genome editing technologies. Analyses of microbial genomes and metagenomes reveal that we have barely explored the chemical and functional diversity of microorganisms. Even in actinobacteria that have been the research focus for decades, majority of the BGCs has not been linked to their secondary metabolites [8], underscoring the potential for new discoveries even in the most extensively screened species. Bioinformatic tools have been developed to prioritize and identify relevant BGCs from genomes/metagenomes, predict the structures of their chemical products, integrate genomic with metabolomic and biological information to identify genotype-chemotype relationships, as well as prioritize downstream characterization efforts. New DNA capture and assembly methods as well as multiplexed genome engineering technologies facilitate the optimization of heterologous expression systems and biosynthetic pathways for the production of secondary metabolites. These developments pave a way towards more systematic and targeted discovery of novel natural products starting from genomic information.

8. Expert opinion

With the rapidly increasing amount of DNA sequence information, the promise of better quality genomes/metagenomes using third-generation and single-cell genome sequencing platforms, as well as the significant headway made in classification of BGCs and their chemical outputs, it is clear that the major bottleneck in natural product discovery lies in our ability to link BGCs with their cognate secondary metabolites [34]. Activation of silent BGCs in native producing hosts will require the development of separate sets of genetic tools for non-model organisms. Advances have been made in the development of alternative heterologous expression hosts for BGCs from unknown or inaccessible sources but success hinges upon host-BGC compatibility and often involves time-consuming capture or cloning, assembly and refactoring and/or optimization of large and unwieldy clusters. Until we better understand the determinants governing host-BGC compatibility, a diverse collection of robust, orthogonal heterologous hosts maximizes the probability of finding a host-BGC combination for successful expression. The development of new host systems will require identification and engineering of gifted strains that are genetically and metabolically predisposed to express and produce the encoded molecules from a diverse set of BGCs [112, 113, 114]. With many bacteria phyla lacking cultured representatives [14, 110], advanced culture methods and separate sets of genetic tools may be needed to obtain compatible expression hosts to more readily access the biosynthetic potential of these underexplored groups of microorganisms [45]. Significant progress has been made in the development of molecular parts and tools for the genetic manipulation of Streptomyces spp., a prolific group of bacteria known to produce bioactive metabolites, but there is still much room for improvement before “plug-and-play” natural product discovery becomes a reality, and one can expect the same for other emerging host systems. Combinatorial genome-scale engineering tools, while currently limited to model organisms like E. coli, will be invaluable towards the engineering of more robust production systems for different BGC families when transferred to natural product-relevant hosts [34]. Last but not least, entire biosynthetic gene clusters can be codon-optimized/-randomized, refactored, synthesized de novo and assembled for optimal expression in designated heterologous hosts [69, 142], although current gene synthesis costs precludes routine adoption of this approach.

One of the main underlying assumptions in genomics- and metagenomics-driven natural product discovery is that BGC sequence diversity reflects biosynthetic and chemical diversity, and by extension the pharmacological activity profiles of the metabolites. Homology to characterized clusters has been used as a proxy for “chemical dereplication” and to prioritize BGCs of interest for characterization [143, 144]. Based on phylogenetic similarity between amplicon sequences of key biosynthetic features within BGCs, the sequence-tag strategy for targeted metagenomics has been successful in identifying new congeners and subclasses of biomedically relevant natural products (Section 5). Nonetheless, there are indications that cluster sequence or architecture divergence does not always translate into chemical structure variation. For example, biosynthetic genes may not be physically clustered [145, 146]. From an evolution perspective, enzymes may already be engaged in a biosynthetic pathway before the respective genes are physically recruited into the BGC through a mechanism that is yet to be determined [116]. Acyl carrier protein (ACP)-less PKS clusters suggest an alternative biosynthetic route or the possible recruitment of ACPs from outside the cluster [117]. Additionally, highly divergent BGCs in terms of both cluster architecture and gene sequence can have similar chemical outputs. Global analysis of prokaryotic BGCs revealed that two distinct clades of BGCs with limited homology actually produce highly similar and conserved aryl polyenes [143]. On the contrary, PKS clusters containing phylogenetically related sets of AT and KS protein domains can give rise to polyketides with distinct core structures [116]. With individual BGC families evolving differently [116], the choice of biosynthetic features for phylogenetic profiling and inference of chemical output, especially for sequence-tag metagenomics, is crucial but may not be obvious without sufficient empirical information. Comparing larger regions or entire BGCs, though not always possible, will facilitate the classification of BGCs that can be linked to secondary metabolites groups and their associated pharmacological activities [143].

While this review focuses on harnessing genomic and metagenomic data to prioritize and facilitate the discovery of new chemical entities, chemical (e.g. compound isolation, structure identification) and biological characterization of natural products are just as critical to the drug discovery process. In fact, structure elucidation and determination of compound mode-of-action represent some of major bottlenecks in the natural product discovery process. Liquid chromatography-, mass spectrometry-, nuclear magnetic resonance-based technologies have progressed significantly for general chemical profiling of complex natural product samples as well as structure determination of large complex molecules [147, 148, 149, 150]. Given the inherent low-throughput nature of structure elucidation, especially of large complex molecules with unknown chemical scaffolds, it is critical that such efforts are channeled towards compounds with desired bioactivity and modes-of-action. Conceptually distinct from the conventional high throughput bioassays, multiparametric screening cell-, image- and sequencing-based strategies generate information-rich bioactivity fingerprints have been successfully used to prioritize and classify natural products based on their modes of action [150]. For effective classification, however, many of these multiparametric screening platforms require a sizable reference set of bioactivity profiles that may not be readily accessible to everyone. There are recent community-wide efforts to integrate multiple sequence databases and relevant web services, as well as standardize BGC annotation and their chemical outputs to facilitate comparative analyses of large datasets [12, 24, 151]. These repositories are publicly shared and curated to improve chemical characterization as well as foster collaboration, emphasizing the need to tackle existing challenges (e.g. high rediscovery rates, genes-to-chemical bottleneck, low production titers) in natural product discovery as a community. Similar repositories for standardized biological characterization data, integrated with genomic and chemical information data, will be invaluable towards uncovering chemical feature(s) with desired bioactivities from nature. It is possible that we will be able to reliably deduce compound pharmacological profiles from analysis of BGCs as the list and diversity of experimentally validated BGCs linked to their secondary metabolites with known bioactivity expand.

Overall, this is an exciting time for natural product discovery, which was mostly a random “hit-or-miss” endeavor in the pregenomics era. Genomics and metagenomics have given us a glimpse of the remarkable biosynthetic potential and vast chemical inventory in nature that we can prioritize and systematically mine for target natural products. As various technologies mature to clone and functionally express target BGCs in heterologous systems or activate silent BGCs in their native hosts, we will be able to more rapidly and cost effectively translate genome information into chemical compounds for drug screens. Public repositories with integrated web services establish the framework for genome-driven natural product discovery efforts and our predictive ability for the chemical output of BGCs will continue to improve as the list and diversity of experimentally characterized clusters and their encoded metabolites expand. These advances open up opportunities to understand the intricacies of natural product biosynthesis and reengineering of BGCs for the diversification and derivatization of natural products during drug discovery.

Article highlights.

• A substantial proportion of pharmaceuticals are derived from or inspired by natural products.

• Advances in DNA sequencing reveal silent biosynthetic gene clusters and uncultured microorganisms as vast untapped resources for novel bioactive chemical scaffolds.

• A collection of in silico tools supports mining of genomes and metagenomes for biosynthetic pathways that potentially encode novel pharmaceutically relevant molecules.

• Driven by advances in synthetic biology and genome engineering tools, pathway and host engineering strategies facilitate the functional expression and characterization of biosynthetic gene clusters.

• Understanding the chemical logic of biosynthetic pathways enables rational biosynthetic pathway engineering for the diversification and derivatization of privileged natural product scaffolds.

• Genomics- and metagenomics-driven natural product discovery continues to uncover new bioactive chemical entities and promises to revitalize waning drug pipelines in the near future.