Abstract
Plants are colonized by phylogenetically diverse microorganisms that affect plant growth and health. Representative genome-sequenced culture collections of bacterial isolates from model plants, including Arabidopsis thaliana, have recently been established. These resources provide opportunities for systematic interaction screens combined with genome mining to discover uncharacterized natural products. Here, we report on the biosynthetic potential of 224 strains isolated from the A. thaliana phyllosphere. Genome mining identified more than 1,000 predicted natural product biosynthesis gene clusters (BGCs), hundreds of which are unknown compared to the MIBiG database of characterized BGCs. For functional validation, we used a high-throughput screening approach to monitor over 50,000 binary strain combinations. We observed 725 inhibitory interactions, with 26 strains contributing to the majority of these. A combination of imaging mass spectrometry and bioactivity-guided fractionation of the most potent inhibitor, the BGC-rich Brevibacillus sp. Leaf182, revealed three distinct natural product scaffolds that contribute to the observed antibiotic activity. Moreover, a genome mining-based strategy led to the isolation of a trans-AT polyketide synthase-derived antibiotic, macrobrevin, which displays an unprecedented natural product structure. Our findings demonstrate that the phyllosphere is a valuable environment for the identification of antibiotics and natural products with unusual scaffolds.
Natural products (NPs) are a remarkably diverse group of metabolites with relatively simple to highly complex structures. Most are classified as specialized metabolites that, unlike primary metabolites, are usually not directly involved in growth, reproduction, or survival.1 These compounds are typically restricted to a few producers and serve to interact with the local environment. The structural diversity of NPs is reflected in a wealth of different bioactivities, many of which have been exploited as drugs in human and veterinary medicine. The majority of approved small-molecule drugs, and especially antibiotics, are either NPs, NP-derived, or inspired by NPs.2 The ecological role of specialized microbial metabolites is in most cases only poorly understood. Some play a role in microbial warfare,3 whereas others function as developmental signals,4,5 act as signaling metabolites in host-microbe or microbe-microbe interactions,3,6,7 or serve as chelators to acquire metals from the local environment.8 Recent studies have shown that so far neglected ecological niches are treasure troves for the discovery of chemical novelty.9–11 Thus, the analysis of microorganisms in underexplored habitats represents a promising path to identify NPs and to investigate their role in situ. Metagenomics-based studies of such environments have resulted in the successful isolation of NPs.12–14 Yet, this approach does not provide direct access to the producing organisms.
Diverse culture collections of bacteria are a valuable resource to examine the potential of NP formation both via bioactivity assays and genome-mining.15 Recently, the combination of culture-independent microbiota profiling with large-scale bacterial isolation efforts resulted in the successful cultivation of exemplars of the majority of species that are reproducibly associated with leaves and roots of natural Arabidopsis thaliana populations.16 These strain collections, and their respective draft genome sequences, are now available for systematic experimental and in silico analyses.
To assess the biosynthetic potential of this resource, we conducted a large-scale interaction screen, together with bioinformatic and chemical analyses, focusing on the leaf microbiota. From a biological perspective, the limited availability of nutrients in this habitat17,18 might have selected for highly competitive strains engaged in chemical warfare. We present a comprehensive interaction screen of more than two hundred strains (the At-LSPHERE collection)16 tested against each other and selected pathogens. Our results reveal numerous inhibitory interactions and show that some strains inhibit a large phylogenetic diversity of strains. Genome analysis identified about 1,000 biosynthetic gene clusters (BGCs) covering a large variety of compound classes. Brevibacillus sp. Leaf182, a “talented producer” identified by the interaction screen as well as in silico predictions, was selected for in-depth chemical analyses in order to characterize metabolites responsible for the observed high inhibitory activities against ~50% of all tested strains.
Results
Binary interaction network of phyllosphere bacteria
To survey the network of inhibitory interactions within a collection of more than 200 bacterial leaf isolates, we screened the At-LSPHERE strain collection16 for inhibitions in binary combinations. Each strain was tested for potential growth interference against all members of the strain collection (Fig. 1a). In total, we monitored more than 50,000 pairings of the 224 strains belonging to the main bacterial phyla inhabiting the A. thaliana phyllosphere. Notably, 725 inhibitory binary interactions (1.4% of all possible pairings) were observed (Supplementary Table 1). Overall, 196 strains (88%) engaged in inhibitory binary interactions (Fig. 1b, Supplementary Fig. 1). Of these, 160 strains (71%) were sensitive against other phyllosphere isolates and 79 strains (35%) inhibited one or multiple strains, with 26 strains showing inhibition of five or more strains (Supplementary Table 2); notably, many of these inhibited isolates belonging to different phylogenetic clades (Supplementary Table 3). Most isolates, however, inhibited only one or two other strains, suggesting specific interactions. Cluster analysis of all bipartite interactions indicated that interactions occur across all phylogenetic clades (Supplementary Fig. 2) with the ten most active strains (> 20 inhibitions each) forming the central nodes of the interaction network (Fig. 1, Supplementary Fig. 2, Table 1).
Fig. 1.
Bacillales and Pseudomonadales dominate the binary interaction network. a, Exemplary phenotypes of interaction classifications. Inhibitory halos are labelled with "1" for strong and "2" for weak inhibitions, based on halo size. b, Phylogenetic tree of all isolates (n = 224) representing the observed directional interactions. Numbers in the inner ring indicate strain numbers, and each node is color coded according to phylum or class annotation, while the annotation background color represents phylum assignment. Directional inhibitory interactions (725 of 50,176 pairwise combinations tested, 1-2 biological replicates) are indicated from red to yellow. The number of sensitivities per strain is summarized in the innermost circle [S], the number of inhibitions caused in the outermost ring [I]. For a list of all interactions see Supplementary Table 1.
Table 1. The ten At-LSPHERE strains showing most inhibitory interactions in the binary interaction screen of 224 strains against 224 strains (1-2 biological replicates).
Rank | Strain | Phylum | Genus | Number of inhibitions | ||
---|---|---|---|---|---|---|
total | strong | weak | ||||
1 | Leaf182 | Firmicutes | Brevibacillus | 111 | 32 | 79 |
2 | Leaf98 | Proteobacteria | Pseudomonas | 67 | 17 | 50 |
3 | Leaf434 | Proteobacteria | Pseudomonas | 56 | 6 | 50 |
4 | Leaf58 | Proteobacteria | Pseudomonas | 54 | 4 | 50 |
5 | Leaf15 | Proteobacteria | Pseudomonas | 47 | 10 | 37 |
6 | Leaf49 | Firmicutes | Bacillus | 32 | 9 | 23 |
7 | Leaf2 | Proteobacteria | Novosphingobium | 31 | 5 | 26 |
8 | Leaf408 | Proteobacteria | Methylophilus | 31 | 11 | 20 |
9 | Leaf75 | Firmicutes | Bacillus | 22 | 9 | 13 |
10 | Leaf82 | Bacteroidetes | Flavobacterium | 21 | 9 | 12 |
|
Inhibitory interactions might be context-dependent and rely on sufficient growth, specific stimuli, or other environmental conditions. We thus selected ten strains belonging to different phylogenetic clades and tested whether inhibitions of these by all At-LSPHERE strains are robust to alternative growth media. We used six media differing in their complexity with some mimicking the nutrient sources available on leaf surfaces (Materials and Methods). Although not all interactions could be scored due to poor growth, overall, the observed inhibitions were rather congruent among all media tested (Supplementary Fig. 3, Supplementary Fig. 3 Supplementary Tables 4, 5).
The entirety of all bacterial interactions revealed that the majority of observed inhibitions were due to the activity of two bacterial orders, Bacillales and Pseudomonadales (Fig. 1b and Supplementary Figs. 1,2). These two taxa make up 8% of the strain collection but confer over 60% of observed inhibitions. Bacillales isolates frequently inhibited Sphingomonadales (22% of possible inhibitions), Caulobacterales (17%), and Actinomycetales (12%), while strains of the order Pseudomonadales often inhibited Methylophilales (28% of possible inhibitions), Flavobacteriales (19%), Sphingomonadales (18%), Xanthomonadales (17%) and Rhizobiales (14%) (Supplementary Table 6). Sphingomonadales were significantly more often inhibited by Bacillales and by Pseudomonadales than expected based on the overall observed inhibitions (Supplementary Table 6). A closer inspection of the top ten inhibiting strains (Table 1) revealed that the two Firmicutes Brevibacillus sp. Leaf182 and Bacillus sp. Leaf49, as well as Flavobacterium sp. Leaf82 inhibited more members of the genus Sphingomonas (Alphaproteobacteria) than expected (one-sided Fisher's exact test, Bonferroni-adjusted P = 0.003, P = 0.01, and P = 0.0007, respectively), while Pseudomonas sp. Leaf15 was active against many members of the genus Methylobacterium (Supplementary Table 7, Supplementary Fig. 4).
Overall, only a minority of interactions were observed between members of the same family or genus. One exception were members of the genus Aeromicrobium, which inhibited the Nocardioides and Marmoricola within the Nocardioidaceae (Supplementary Table 6). Furthermore, fourteen inhibitory activities between strains belonging to the same genus (0.4% of 3,264 possible combinations) were observed, significantly less than expected by chance (two-sided Fisher's exact test, P = 8.8 x 10-9). Overall, this result indicates that leaf isolates tend to inhibit distinct phylogenetic groups rather than closely related strains.
To assess the inhibitory potential of phyllosphere isolates against well-studied bacterial phytopathogens, we also tested Pseudomonas syringae pv. tomato DC3000, P. syringae pv. syringae B728a, Ralstonia solanacearum AW1, Agrobacterium tumefaciens C58, and Xanthomonas campestris pv. campestris LMG 568. Several of the top inhibitory strains also showed activity against these model pathogens (Supplementary Table 8). Pseudomonas sp. Leaf58 showed inhibitory activity against all tested pathogens except R. solanacearum AW1, while Pseudomonas sp. Leaf434 inhibited all but P. syringae DC3000 and R. solanacearum AW1. In consequence, strains of the collection isolated from healthy plants have potential to help in defense against pathogens via direct inhibition.
Identification of putative natural product biosynthesis gene clusters
The identification of a large inhibitory interaction network prompted us to investigate the full biosynthetic potential of each strain in the At-LSPHERE16 collection. We thus mined the genome-sequenced strains for biosynthetic gene clusters (BGCs) putatively involved in the production of NPs. The antiSMASH 4.0 toolkit19 was used for the prediction and annotation of NP BGCs. In total, 1,053 BGCs were identified and separated into eight distinct NP (sub)classes (Fig. 2, Supplementary Tables 9-11). In order to identify BGCs that might encode the biosynthesis of uncharacterized NP scaffolds, antiSMASH results were subsequently compared against the MIBiG database20 of characterized BGCs using BiG-SCAPE (https://git.wageningenur.nl/medema-group/BiG-SCAPE) (Supplementary Table 12), a recently established analysis platform to compare BGCs based on distance metrices. BiG-SCAPE analysis revealed a large set of 766 BGCs (73%) that putatively encode biosynthetic novelty (Fig. 3, Supplementary Fig. 5, Supplementary Table 13), as indicated by orphan BGCs that are not connected to characterized entries in the MIBiG database. In total, 97 clusters that did not contain any MIBiG BGCs and 176 singletons were found.
Fig. 2.
Biosynthetic potential of the At-LSPHERE strain collection. Phylogenetic representation of the genome-sequenced At-LSPHERE strain collection (n = 207) with circular heat-map representation of number of detected BGCs and bar representation of total number of BGCs per isolate (Supplementary Table 9). Inner ring labels indicate strain numbers. Outer rings correspond to the following BGCs: 1) Modular polyketide synthase (PKS), 2) Modular PKS non-ribosomal peptide synthetase (NRPS) hybrid, 3) Non modular PKS, 4) NRPS, 5) ribosomally synthesized and post-translationally modified peptide (RiPP), 6) Quorum sensing, 7) Terpene, 8) Other. Numbers in parentheses correspond to total number of BGCs in each category. Heat-map colors correspond to number of BGCs detected.
Fig. 3.
BiG-SCAPE analysis of BGCs detected by antiSMASH in 207 genomes of the At-LSPHERE strain collection and comparison with the MIBiG database of characterized BGCs. a, RiPP BGCs (n = 246), b, type I PKS BGCs (n = 277), c, NRPS BGCs (n = 338). Nodes and singletons are color coded according to their origin (MIBiG database or phylum/class) and represent individual BGCs. The widths of interconnecting edges indicate the degree of relatedness between two BGCs, with connections up to a raw distance of 0.75 retained (Supplementary Table 12). Numbers of singletons are indicated. Black labels denote compounds associated with selected MIBiG BGCs. Blue labels highlight clusters of related BGCs or individual BGCs of interest by phylogenetic distribution.
Among the most important NP classes that contain specialized metabolites with inhibitory effects are ribosomally encoded and post-translationally modified peptides (RiPPs), nonribosomally synthesized peptides (NRPs), type I polyketide synthase (PKS) products, type II PKS-derived polyketides, aminoglycosides, and peptide-polyketide hybrids.
Our analysis of RiPP BGCs revealed a broad variety (103 BGCs) in the genomes of the At-LSPHERE strain collection (Fig. 3a). According to antiSMASH and BiG-SCAPE analyses, these belong to the RiPP families bacteriocins, lanthipeptides, lassopeptides, microviridins, linaridins, thiopeptides, thiopeptide-linaridin hybrids, and lantipeptide-proteusin hybrids indicating a broad structural diversity of ribosomally encoded peptides. Only a few lantipeptides and one bacteriocin show distant similarity to known RiPP BGCs. All other identified RiPP BGCs were almost exclusively family- or genus-specific and did not cluster with any characterized BGCs, thus indicating a large potential for the identification of different RiPP types. Many RiPPs are narrow-spectrum antibiotics21 and hence might be responsible for selectively inhibiting only a few members of the strain collection.
AntiSMASH and BiG-SCAPE analyses suggested the presence of mono-modular type I PKSs particularly in members of the Rhizobiales and Nocardiaceae (Fig. 3b). Notably, in the genus Rhodococcus, a mono-modular type I PKS BGC is conserved across all strains of the collection, and a related BGC is present in Willamsia, another member of the Nocardiaceae (Supplementary Fig. 6a). All other type I PKSs and PKS-NRPS hybrids were only identified in individual genera, with strains of the phyla Firmicutes and Bacteroidetes being particularly rich in architecturally unusual multi-modular PKSs that suggest polyketides with unprecedented scaffolds (Fig. 3a, Supplementary Figs. 5a,6b). Trans-AT PKS or hybrids of trans-AT PKS and NRPS BGCs were detected in the genomes of Duganella sp. Leaf61 and Brevibacillus sp. Leaf182 (Supplementary Figs. 5b,6c). The latter seems to be a prolific producer of trans-AT PKS-derived polyketides, harboring at least four trans-AT PKS gene-containing loci, one of the highest numbers for any known bacterium.22,23
In addition, we identified one type II PKS BGCs in Bacillus sp. Leaf406, which is rare since type II PKSs are almost exclusively reported from filamentous actinomycetes.24
NRPS BGCs were abundant in genomes of the Beta- and Gammaproteobacteria (Fig. 2), as well as in the top inhibitors of the phyla Firmicutes and Bacteroidetes, which contain one or more NRPS or NRPS hybrid BGCs. Of all 135 NRPS and 18 NRPS-hybrid BGCs, the majority (72 and 15 respectively) do not cluster with any MIBiG entry and thus suggest biosynthetic novelty (Supplementary Table 13, Fig. 3c). The family Nocardiaceae is noteworthy, being the only rich actinobacterial source of NRPS BGCs in the At-LSPHERE. A portion of the Rhodococcus NRPS BGCs is distantly related to characterized BGCs such as the erythrochelin biosynthetic locus,25 with which they share predicted adenylation (A) domain substrate specificities. However, the majority of Rhodococcus NRPSs are i) shared among members of the At-LSPHERE and ii) predicted to encode unusual peptides that are rich in Ser or Cys residues and iii) only show weak similarity to any MIBiG entry. Some of these shared BGCs encode up to 17 NRPS modules, of which ten (60%) contain A domains with predicted specificities for Ser or Cys (Supplementary Fig. 6d). One of these conserved BGCs shows distant similarity to the significantly smaller BGC for the Cys-rich cytotoxin thiocoralin,26 which in contrast encodes only five modules.
In addition, BiG-SCAPE analyses also revealed a putative mangotoxin BGC encoded in different Pseudomonas species (Supplementary Figs. 5c,6e). This is noteworthy because mangotoxin is an antimetabolite that inhibits ornithine/arginine biosynthesis in plants and causes apical leaf necrosis.27 We compared the total number of BGCs and the number of observed inhibitions for each strain (Supplementary Fig. 7). Most strains with potent inhibitory activities contain four or more of these potentially inhibitory BGC classes. An exception is Exiguobacterium sp. Leaf196 that is devoid of any PKS, NRPS, or RiPP BGCs, yet caused three strong and 13 weak inhibitions, potentially based on either undetected BGCs, antibacterial proteins, or competition for nutrients rather than antibiosis. Other examples include the Nocardiacae that based on our antiSMASH and Big-SCAPE analysis are particularly prolific NRPS producers, yet do not show any inhibitory activities, or a Paenibacillus sp. Leaf 72 that also harbors multiple BGCs classified into groups with putative antibiotic potential, yet no inhibitory activity was observed in our screen. This discrepancy might either be due to BGCs that are silent under the growth conditions used, or the products of the BGCs have a different ecological function beyond bacterial inhibition.
Besides BGCs that might be involved in inhibitory microbe-microbe interactions, clusters potentially involved in bacterial communication were detected (e.g., homoserine lactone- and butyrolactone BGCs,28). Furthermore, we detected BGCs for compounds involved in environmental metal chelation (e.g. NRPS-derived siderophores in addition to xanthoferrins, desferrioxamin and petrobactin-like siderophores) (Supplementary Fig. 5c,6f).29–32
Terpenes are the most abundant BGC class, with at least one cluster detected in 187 (91%) strains (Fig. 2, Supplementary Fig. 5d) with the exception of Gammaproteobacteria that were almost completely devoid of detectable terpene BGCs. The high prevalence of BGCs for terpene biosynthesis partly relates to carotenoid and pigment biosynthesis. It has been postulated that pigmentation is favorable for leaf-colonizing bacteria, as they are exposed to solar radiation in their natural habitat.17,33,34 For this reason, we examined whether genomes of strains isolated from leaves are enriched in pigment BGCs (carotenoid BGCs and arylpolyene PKSs35) compared to isolates obtained from the below-ground root and soil compartments of the At-SPHERE collection.16 Candidates for pigment biosynthesis were detected in 181 leaf isolates (88% of collection; Supplementary Fig. 8, Supplementary Table 14) compared to 111 soil and root isolates (50% of collection), confirming that they are indeed significantly overrepresented in genomes of leaf isolates (two-sided Fisher's exact test, P = < 2.2 x 10-16). In addition to carotenoid BGCs, BiG-SCAPE analysis of leaf isolates revealed other terpene BGCs and a wealth of genus-specific terpene BGCs mainly in Actinobacteria and Proteobacteria (Supplementary Fig. 5d). Generally, a number of BGCs (mainly terpene-, type II PKS- and mono-modular type I PKS-, and NRPS BGCs) are conserved amongst all members of a given genus, or even crossing genus boundaries. However, the top inhibitor strains contain BGCs that are unique to the producers. This distribution indicates that the corresponding metabolites could be responsible for the strong inhibitory potential observed for a particular strain.
Chemical analysis of the top inhibitor strains by MALDI imaging
The top inhibitors Brevibacillus sp. Leaf182, Pseudomonas sp. Leaf98, Bacillus sp. Leaf49 and Flavobacterium sp. Leaf82 were responsible for more than 30% of all inhibitory interactions. Each of these strains inhibited bacteria belonging to at least six different classes (Supplementary Table 3) and harbored a large number of BGCs putatively involved in the production of antibiotics (Supplementary Fig. 7). We thus selected these strains for chemical analysis using MALDI imaging mass spectrometry (IMS). Plate assays against Gram-positive and Gram-negative bacteria affirmed a multitude of colony-associated and secreted metabolites (Fig. 4a,b), supporting the notion that these strains are prolific producers of specialized metabolites. We analyzed the IMS dataset for diffusable metabolites confined to the zone of inhibition (Fig. 4b, Supplementary Fig. 9). These analyses revealed the top inhibitor Brevibacillus sp. Leaf182 as the most conspicuous producer of candidate metabolites with various masses confined to the zone of inhibition. We therefore selected this strain for in-depth chemical analysis using a bioactivity-guided and IMS-assisted purification strategy.
Fig. 4.
MALDI imaging results of selected top inhibitors against different sensitive strains and antibiotics isolated from Brevibacillus sp. Leaf182. a, False color representation of selected metabolites (indicated by m/z values) produced by four of the top inhibitors against Sphingomonas sp. Leaf205 detected by MALDI imaging (1 biological replicate). b, False-color heatmap (low intensity: blue to high intensity: red) representation of selected metabolites produced by the top inhibitor Brevibacillus sp. Leaf182 on Nocardioides sp. Leaf285 or Sphingomonas sp. Leaf67 confined to the zone of inhibition as detected by MALDI imaging (1 biological replicate). c, Structures of bioactive metabolites isolated in this study. 1, streptocidin D; 2, streptocidin A; 10, phosphobrevin; 11, marthiapeptide A; 12, macrobrevin; n.i., not identified.
Bioactivity-guided identification of antibiotics from Brevibacillus sp. Leaf182
To characterize the putative antibiotics produced by Brevibacillus sp. Leaf182, we fractionated liquid culture extracts and subsequently tested them for antibiotic activity against a panel of sensitive leaf strains. Four bioactive fractions were identified that contained metabolites belonging to the same compound family based on molecular network analysis7,36 (Supplementary Fig. 10). A combination of MS2 and NMR experiments followed by database search revealed streptocidin D (1) of the tyrocidin family as the bioactive principle in one of the fractions (Supplementary Figs. 11-16, Supplementary Table 15). Searching the genome for an NRPS BGC with matching adenylation domain substrate specificities resulted in the identification of an NRPS BGC with strong homology to the tyrocidin NRPSs (Supplementary Fig. 17, Supplementary Tables 12 and 16).37 In agreement with a previous study,38 streptocidin D (1) showed inhibition against most Gram-positive but not Gram-negative strains tested (Supplementary Table 17). Analysis of the MS network data revealed at least nine congeners (Supplementary Figs. 10-11, 18-25). In addition to the known congeners streptocidin A (2), B/C (3), D (1), G (4), and M (5), we identified four previously undescribed variants that we termed streptocidins Q (6), R (7), S (8), and T (9) (Supplementary Figs. 18-25, Supplementary Table 18).
Because most of the metabolites identified in IMS experiments were not produced in liquid culture, we tested organic extracts from agar plates inoculated with Brevibacillus sp. Leaf182. Two different bioactive metabolites were purified in consecutive rounds of bioactivity-guided chromatographic separation, followed by MS- and NMR-based structure elucidation. We identified an unusual lysophospholipid that we termed phosphobrevin (10) (Supplementary Figs. 26-31, Supplementary Table 19). The compound acts against Gram-negative bacteria (Supplementary Table 17), for which activities are particularly difficult to find.39 The second antibiotic was identified as marthiapeptide A (11) (Supplementary Figs. 32-37, Supplementary Tables 17,20), which was previously isolated from the deep-sea bacterium Marinactinospora thermotolerans. 40 Despite genome sequencing efforts, the BGC for marthiapeptide A had remained enigmatic. We thus screened the genome of the reported producer and Brevibacillus sp. Leaf182 for a candidate BGC responsible for marthiapeptide biosynthesis. After genome re-sequencing of Brevibacillus sp. Leaf 182 and de novo assembly, we identified an NRPS BGC as a candidate for marthiapeptide biosynthesis (Supplementary Fig. 38, Supplementary Table 21). To test whether the candidate BGC is indeed involved in marthiapeptide biosynthesis, the core biosynthetic gene marB was disrupted and the wildtype and knockout subjected to comparative metabolomic analysis (Supplementary Fig. 39). These analyses confirmed that the mar BGC is responsible for marthiapeptide A biosynthesis.
Genome mining-guided identification of unusual bioactive metabolites
In addition to the activity-guided metabolite identification approach described above, we used a genome-mining strategy to exemplarily correlate genome to phenome. Again, we focused on Brevibacillus sp. Leaf182 due to its high number of loci encoding trans-AT PKSs, a family of natural product enzymes known for its potential to generate chemically diverse bioactive molecules.22 Closer inspection revealed three of these are PKS-NRPS hybrid clusters and the fourth displays a trans-AT PKS and a NRPS in close vicinity. We concentrated on the latter trans-AT PKS BGC (named bre PKS, Supplementary Fig. 40, Supplementary Table 22) and subjected its protein sequences to the automated genome mining tool TransATor (transATor.ethz.ch/Trans-AT-PKS) (Supplementary Results 1). TransATor annotates trans-AT PKSs and predicts the structure of corresponding metabolites. The predicted structure (Supplementary Fig. 40b) was used to search for similar metabolites in NP databases. Since no matching candidates were identified, the BGC was a strong candidate for the biosynthesis of a so far undescribed polyketide scaffold. Brevibacillus sp. Leaf182 cell extracts were used for MS-guided purification, which was highly challenging, as the compound was prone to degradation. Structure determination using MS and 2D NMR experiments identified a previously unknown polyketide that we named macrobrevin (12) (macro for macrolactone and brevin to indicate the producer Brevibacillus sp. Leaf182), displaying a unique chemical scaffold (Supplementary Figs. 41-47, Supplementary Table 23). Disruption of breB abolished macrobrevin production, as expected (Supplementary Fig. 48). We tested 12 against a panel of different At-LSPHERE isolates and identified the compound to be active against Bacillus sp. Leaf49 (Supplementary Table 17). Our genome mining-based polyketide discovery highlights the hidden metabolic diversity of the At-LSPHERE strain collection and suggests, together with the BGC data, that the phyllosphere harbors an untapped potential for previously unknown specialized metabolites, as exemplarily shown for Brevibacillus sp. Leaf 182.
Discussion
Nature is a prolific resource for specialized metabolites of pharmaceutical importance, such as antibiotics and anticancer drug leads.2 In times of increased need for antimicrobial drugs, it is essential to move beyond traditionally screened habitats and microorganisms for the discovery of so far unknown NP scaffolds.11 For example, in an in-silico and taxon-based approach, a recent study on members of the genus Bacillus spp., which also included various plant-derived strains, revealed diverse BGCs that suggested biosynthetic specialization.41 Another comprehensive study revealed the presence of BGCs for thiopeptides, a common family of antibiotics, in human-associated bacteria.42 Here, we applied a habitat- and interaction-centered strategy to evaluate the potential for unprecedented antimicrobials. Microbial inhabitants of the phyllosphere have to compete for the scarcely available nutrients and trace elements, are exposed to UV radiation, oxidative stress, drought, and fast changing conditions.17 The adaptation to this environment might be a driving force for the selection of microorganisms that produce potent specialized metabolites to compete with other bacteria and cope with harsh environmental conditions. To date, the leaf environment has not been systematically analyzed in this regard. We combined a high-throughput binary interaction screen of a representative strain collection of more than 200 leaf isolates (At-LSPHERE16) and genome mining. We identified more than 1000 BGCs for compounds putatively involved in bacterial interactions and niche adaptation, with a majority lacking similarity to characterized BGCs. These belong to diverse biosynthetic classes including RiPP and terpene systems, NPRSs, and trans-AT PKSs, suggesting high potential for uncharacterized NPs. This hypothesis is supported by isolating two unknown antibiotics, the lysolipid phosphobrevin and the trans-AT PKS-derived macrobrevin, from one selected strain of the At-LSPHERE collection.
We observed that a few strains of the At-LSPHERE collection showed a wide inhibition spectrum, which is consistent with two binary interaction screens using 67 bacterial isolates from eight environments43 and 140 bacterial isolates from different compartments of Echinaceae purpurae 44, respectively. The total number of antagonistic interactions observed ranged from 1.4% of all combinations (this study) to 3.4%43 and to 11.3%.44 However, a direct comparison should be taken with caution because a different method to examine inhibitions was used in the latter study that consisted of cross-streaking (might also detect inhibition principles other than NPs, such as modification of pH or nutrient depletion) versus direct plating with detection of inhibition halos.
We explored unknown compounds in an exemplary phyllosphere habitat. In addition to implications for NP discovery, the data provide a valuable basis for testing the relevance of NPs for community assembly. It is assumed that antimicrobial compounds contribute to the plant microbiota composition.45,46 Our collection of cultivated, genome-sequenced bacteria will permit in planta studies to test the impact of individual BGCs in wild-type and mutant strains within complex synthetic communities.15 The Brevibacillus mutants generated here represent a good starting point to study how specialized metabolism shapes plant microbiomes.47 NPs produced by the plant microbiota might act on the producer or other microbes and/or the plant.48 In addition, NP effects might be dose-dependent insitu.49 In this context, in planta studies will also be informative with regard to BGC-rich strains that did not show an antagonistic interaction.
In conclusion, we have shown that the phyllosphere is a promising resource for bacteria with a large and distinct biosynthetic repertoire that provides the basis for the isolation of bioactive metabolites with unusual structural scaffolds and in planta testing. Overall, a combined strategy of ecosystem-, activity-, and genome-guided analysis as conducted here is a promising path for NP research in the context of ecological studies and compound discovery.
Methods
Pairwise interaction screen of phyllosphere bacteria
The 224 leaf isolates of the At-LSPHERE16 were grown on R2A agar (R-2A, Sigma Aldrich) supplemented with 0.5% (v/v) methanol (R2A+M) or minimal medium (MM)50 containing 0.5% (v/v) methanol and 25 mM glucose (MM+M+G) as carbon sources at a pH of 6.5 at room temperature (RT, for plates) or 28 °C (for liquid cultures) (Supplementary Table 24). Strains tested for sensitivity were re-streaked from cryostocks, incubated on solid media used for the inoculation of 5 mL liquid cultures in round-bottom cultivation tubes (Falcon). High-throughput cultivations of inhibitory isolates were performed in 96-well plates containing solid media. To assess binary interactions between the different strains, stationary liquid cultures of strains tested for sensitivity were diluted to a total of 50 mL with melted R2A+M agar at 45 °C (final agar concentration ≈ 1.35% (w/v)) and 25 mL each were poured into two square plates to form a uniform layer containing the test strain. For strains that did not grow sufficiently well in either liquid medium or when poured into R2A+M agar, the sensitives were grown on R2A+M agar plates and resuspended in 5 mL liquid R2A+M prior to pouring the lawn. The inhibitory interaction partners were resuspended from 96-well plates and incubated for 2 h while shaking. Subsequently, roughly 1 µL of each strain on the 96-well plate was printed onto the solidified agar layer containing the putative sensitive strain using a replicator. Results of the binary interactions were evaluated after 2-3 days of incubation at RT. Interactions were classified as “strong inhibitions” if the inhibition zone, the distance of the outside border of the halo to the edge of the colony, exceeded 3 mm, while “weak inhibitions” either showed a smaller or partially turbid halo (Fig. 1a). Data analysis, statistics and visualization was done using R, Python, including the NetworkX package, and GraPhlAn.51
Pairwise interaction screen of selected phyllosphere bacteria on different media
A subset of 10 indicator strains (see Supplementary Table 25) was tested for sensitivities against the At-LSPHERE collection on six different media. Briefly, these were a barley extract medium (BEM), a plant-mimic medium (PMM) at two different pH, half-strength lysogeny broth (½ LB) and R2A with and without methanol supplementation (for details on media composition see Supplementary Methods). These media were chosen to mimic different growth conditions. BEM and PMM were designed to mimic the plant surface, the first being a full plant extract and thus containing all its nutrients only supplemented with methanol and the latter a synthetic medium containing sugars, small organic acids and amino acids likely encountered on plant surfaces18,52 as well as glycerol. Half strength LB is richer in nutrients and contains more salt compared to the standard medium R2A+M. Strains grown on R2A+M agar were suspended in 10 mM MgCl2 solution and inoculated onto each medium to be tested in 96-well plates ("inhibitior strains", complete At-LSPHERE collection) or on round agar plates ("sensitive strains", see Supplementary Table 25). Strains were grown at 22 °C for 4 d (BEM, R2A and R2A+M) or 5 d (PMM 6.1, PMM 7.1 and ½ LB). "Sensitive strains" were resuspended in each medium and an amount corresponding to a final OD600 of 0.01 was added to 15 mL top agar (42 °C, 1% agar) and immediately poured on top of square plate containing 25 mL of the same medium (1.5% agar). "Inhibitor strains" were resuspended by adding liquid medium to each well of the 96-well plate, incubating for 15 min followed by vortexing for 5 min. Bacterial suspensions were transferred to empty 96-well plates and printed onto the solidified agar layer containing the putative sensitive strain using a replicator as described above. Appearance of inhibitory halos was scored after incubation at 22 °C for 2-3 days as described above. Interactions were not scored when either one of the interaction partners had not grown or when the strain spotted on top showed a contamination.
Genome sequencing of Pseudomonas sp. Leaf98 and Brevibacillus sp. Leaf182
Genomic DNA of Pseudomonas sp. Leaf98 was extracted using the MasterPure™ DNA Purification kit (Epicentre). For Brevibacillus sp. Leaf182, genomic DNA was extracted as described by Yamanaka et. al.53 Libraries were prepared form purified genomic DNA and sequenced using the Illumina HiSeq platform. Draft genomes were de novo assembled using SPAdes 3.11 for Pseudomonas sp. Leaf98 and CLC Genomics Workbench 10 for Brevibacillus sp. Leaf182 (Supplementary Methods).
Secondary metabolite cluster prediction
We employed the antiSMASH (antibiotics & Secondary Metabolite Analysis Shell) standalone toolkit v.4.0.219 to mine the genomes of all genome-sequenced leaf isolates (BioProject PRJNA297956, QFZI01000000, LMPN02000000) for the presence of putative NP BGCs. The identified BGCs were grouped (Supplementary Table 10), summarized for each strain (Supplementary Table 9) and visualized using GraPhlAn.51
BiG-SCAPE analysis
BGCs from antiSMASH analyses were compared to the MIBiG database20 v 1.3 using the Biosynthetic Genes Similarity Clustering and Prospecting Engine (BiG-SCAPE; BiG-SCAPE-master-596cfbe25056305379e0a05e8442492891197c32; downloaded Februrary 19th, 2018) with PFAM database 31.0.54 Analysis was conducted using default settings with mode "auto" and retaining singletons. Networks were computed for raw distance cut-offs of 0.10 to 0.80 in increments of 0.05. The lower the cut-off, the fewer connections are kept between clusters.55 Results (Supplementary Table 12) were visualized as a network using Cytoscape 3.6 for a cut-off of 0.75 (Fig. 3, Supplementary Fig. 5).
TransATor-based analysis of trans-AT PKS BGCs
Protein sequences of trans-AT PKSs identified by antiSMASH were extracted and subjected to the automated trans-AT PKS genome mining and structure prediction pipeline TransATor (transATor.ethz.ch/Trans-AT-PKS). The structures predicted by TransATor were used for similarity searches in MarineLit (http://pubs.rsc.org/marinlit/), ChemSpider (http://www.chemspider.com/), and the Dictionary of Natural Products (http://dnp.chemnetbase.com/faces/chemical/ChemicalSearch.xhtml).
Analysis of carotenoid and arylpolyene BGCs
Arylpolyene BGCs were identifiefd by antiSMASH analysis (v.4.0.2)19 for the genomes available from the At-SPHERE collection16 (BioProjects PRJNA297956, PRJNA297942, PRJNA298127; QFZI01000000). Strains putatively containing carotenoid biosynthesis genes were predicted based on the presence of homologous of at least two of the three proteins: phytoene synthase, phytoene desaturase, and lycopene cyclase. To identify potential phytoene synthase and lycopene cyclase proteins, we queried the At-SPHERE strain collection database using the HMMER toolkit (http://hmmer.org/, v3.1b2) with the Hidden-Markov-Models PF00494.18 and PF05834.1, respectively. Proteins with an e-value threshold of 1E-20 and 1E-25 were considered as similar for the phytoene synthase and lycopene cyclase, respectively. To identify potential phytoene desaturase proteins, we used blastp of the BLAST+ standalone software (v.2.2.31) to query the sequence ACS41376.1 of Methylobacterium extorquens AM1 against the At-SPHERE database. Proteins with an e-value threshold of 1E-25 were considered a hit. Fisher's exact test was used to compare the presence of either an arylpolyene BGC or carotenoid biosynthesis genes in the At-LSPHERE and the remaining At-SPHERE.
Imaging Mass Spectrometry (IMS) of bacterial colonies
Sensitive indicator strains were grown on R2A+M, resuspended in 10 mM MgCl2, diluted into 25 mL melted R-2A+M top agar (1% agar, 42 °C) to a final OD600 of 0.01 and immediately poured into a petridish (diameter 13.5 cm). Inhibitory strains were resuspended in 10 mM MgCl2 at an OD600 of 4 and 2 µL each spotted on top. After incubation at RT for 2 days, samples were prepared and imaging experiments conducted as described (for details see Supplementary Methods).56
Isolation, purification, and structure elucidation of streptocidin D and macrobrevin from liquid culture extracts
Brevibacillus sp. Leaf182 was cultured in liquid R2A+M medium in 1 L ultra-high yield™ flasks (Thomson Instrument Company) for 2-4 days at 28 °C. Brevibacillus sp. Leaf182 culture (3 L) was harvested, and the pellet was extracted with acetone and the resulting extract subjected to Reversed Phase High Performance Liquid Chromatography (RP-HPLC) (Phenomonex Kinetex C18, 10 x 250 mm, UV detection at l = 280 nm, RT, with 5% MeCN for 5 min, then a gradient from 5% MeCN to 100% MeCN for 30 min, and 100% MeCN for 25 min).
For the isolation of streptocidin D, inhibitory fractions 9-13 (eluting at 24-36 min) were subjected to Ultra High Performance Liquid Chromatography data dependent Mass Spectrometry (UHPLC-ddMS2) analysis coupled to molecular network analysis as previously described (for details see Supplementary Methods).36,57 For the isolation of streptocidin D, fraction 11 (eluting at 30-33 min) was fractionated by RP-HPLC using a Luna 5u C18 (100A 250 mm x 21.2 mm) column on an 'Infinity' 1260 instrument (Agilent, Santa Clara CA, USA) with a water to acetonitrile gradient of 0-25% in 10 mins, 25-65% in 36 mins and 65-100% in 10 mins.
For the isolation of macrobrevin, the fraction containing the metabolite matching the structure predicted by TransATor (see above, fraction eluting at 46-48 min) was further purified by RP-HPLC (Phenomenex Luna 5u Phenyl-Hexyl, 10 x 250 mm, UV detection at λ = 280 nm, RT, 70% MeCN) to yield 0.5 mg of macrobrevin. For details on MS- and NMR-assisted structure elucidation see Supplementary Methods.
Isolation, purification, and structure elucidation of marthiapeptide A and phosphobrevin from agar plate extractions
Marthiapeptide A: Brevibacillus sp. Leaf182 was cultured on R2A+M agar (50 square petri dishes) and incubated for 2 days at RT. Agar plates were extracted with ethyl acetate. The extracts were filtered and dried under reduced pressure. Crude extracts were re-dissolved in a 2 mL methanol-water solution (1:1), centrifuged and the pellet discarded. Inhibitory activities of supernatants from Brevibacillus sp. Leaf182 agar plate extracts were determined by filter disc assays against a panel of leaf isolates. Briefly, sterilized filter discs (Macherey-Nagel, MN615) were loaded with 7 μL of extracts, dried, and placed onto R2A+M agar containing a sensitive strain as described above. Active extracts were fractioned by RP-HPLC using a Luna 5u C18 column 0-60% in 10 mins, 60-100% in 20 mins, 100% for 5 mins and a flow rate of 21.2 mL min-1 monitored by UV absorption. The collected fractions were tested for inhibitory activity as above. The second round of purification was conducted using a Luna 5u Phenyl-Hexyl column (Phenomenex Luna 5u Phenyl-Hexyl, 10 x 250 mm). The gradient was set to and 0-60% in 10 mins and 60-100% in 36 mins.
Phosphobrevin: Individual colonies of Brevibacillus sp. Leaf182 (5µleach)were spotted on R2A+M agar (170 petri dishes with approximately 40 colonies each) and incubated for 4 days at RT. Agar plates were lyophilized and extracted with 6 L of a methanol:water mixture (80:20). The extracts were filtered and dried under reduced pressure. Extracts were re-dissolved in water and consecutively extracted with hexan, chloroform, diethyl ether, and ethyl acetate. Inhibitory activities were determined by filter disc assays against a panel of leaf isolates. A quarter of the active ethyl acetate fraction was fractionated by preparative RP-HPLC using a Kinetex 5µm XB-C18 (100A 150 mm x 21.2 mm) column with a water to acetonitrile gradient (10-100% ACN in 35 mins, 100% ACN for 7 mins, and a flow rate of 21.2 mL min-1) monitored by UV absorption. Fractions were collected every minute and tested for inhibitory activities as above. The bioactive fraction 24 was subjected to a second round of reverse-phase chromatography-based separation using a Luna 5u Phenyl-Hexyl (100A 250 x 21.2 mm) column with a water to acetonitrile gradient. The gradient was set to 25% ACN for 5 mins, 25-100% ACN in 45 mins and 100% ACN for 5 mins and a flow rate of 21.2 mL min-1 monitored by UV absorption. The bioactive fraction 24 was subjected to MS- and NMR-assisted structure elucidation (Supplementary Methods).
Confirmation of marthiapeptide A and macrobrevin BGCs in Brevibacillus Leaf182 by marker exchange mutagenesis
Homologous regions (HR) up- and downstream of the genes of interest (breB nucleotides 1-1839, marB nucleotides 1-2699) were amplified from Brevibacillus sp. Leaf182 genomic DNA and introduced into a pSEVA28158 vector using the oligonucleotides and restriction sites indicated in Supplementary Table 26. The erythromycin resistance cassette (erm) was amplified from pBs2E59 and inserted into pSEVA281_hr1/2 yielding the final knock-out plasmids. Non-replicative KO plasmids were transferred into Brevibacillus sp. Leaf182 by conjugation with E.coli S17 (λ-pir) and transformants were selected on ½ LB supplemented with colistin (10 μg ml-1) and erythromycin (5 μg ml-1). Double crossover mutants were identified by PCR.
For confirmation of absence of macrobrevin and marthiapeptide, Brevibacillus sp. Leaf182 wild-type and mutants (Brevibacillus sp. Leaf182 breB::erm and Brevibacillus sp Leaf182 marB::erm) were treated as described above for the isolation of macrobrevin and marthiapeptide, respectively, and the extracts subjected to UHPLC-HR-HESI-MS experiments using the liquid chromatography solvent gradients described in the respective section for the initial detection of the compounds on a Phenomenex Kinetex 2.6 mm C18 100 A° (150 x 4.6 mm) column at at flow rate of 0.5 mL/min connected to a QExactive (Thermo scientific) mass spectrometer.
NMR
NMR spectra were recorded on a Bruker Avance III spectrometer equipped with a cold probe at 600 MHz for 1H NMR and 150 MHz for 13C NMR. Chemical shifts were referenced to the solvent peak at δH 7.27 and δC 77.23 for chloroform-d, δH 2.5, δC 39.51 for DMSO-d 6, δH 1.94 δC 1.39 for acetonitrile- d 3, and δH 3.31, δC 49.15 for methanol-d 4.
Bioactivity assay with purified compounds
Bacteria grown on R2A+M were resuspended in 10 mM MgCl2, diluted into melted R2A+M top agar (1% agar, 42 °C) to a final OD600 of 0.01 and poured immediately on top of previously prepared R2A+M agar plates. The purified compounds were dissolved in MeOH (streptocidin D, phosphobrevin), 90% MeCN (marthiapeptide A), or MeCN (macrobrevin). Solvent control and purified compound (10 µl each) were applied on sterile filter paper discs (Macherey-Nagel, MN615), dried for 5-10 min under laminar flow and subsequently placed on the top agar plates containing bacteria. Plates were incubated at RT and formation of zones of inhibition was documented for 3-5 days (Supplementary Table 17).
Supplementary Material
Acknowledgements
This work was financially supported by an SNF grant NRP72 to J.P. and J.A.V. and by the European Research Council Advanced Grants (PhyMo to J.A.V. and SynPlex to J.P.).
Footnotes
Code availability. Source code of the TransATor web application is available from the corresponding authors upon request.
Data availability. Source data for Figures 1-4 and Table 1 is available in the supplementary information files. Data supporting the findings of this study is generally available within the paper and its supplementary information files. The whole genome shotgun projects of Pseudomonas sp. Leaf98 and Brevibacillus sp. Leaf182 have been deposted at DDBJ/ENA/Genbank under the accessions QFZI00000000 and LMPN00000000. The versions described in this paper are QFZI01000000 and LMPN02000000. Raw reads have been deposited at SRA (SRP148446, SRP149408). Information on the marthiapeptide (BGC0001469) and macrobrevin (BGC0001470) biosynthetic gene clusters were uploaded to the MIBiG database. All other raw data is available from the corresponding authors upon request.
Author contributions
EJNH, CMV, RU, MS, FR, DBM, JP and JAV designed research. CMV, MS, FR, DBM and MK, performed binary interaction screens. EJNH, CMV, FR and SP performed genome mining studies. CMV and DBM conducted statistical analyses. EJNH, CMV and MS conducted MALDI imaging experiments. EJNH, CMV, MS, FR and SP conducted bioassays. EJNH, CMV, RU, FR and SP isolated and structure elucidated metabolites. MS generated Brevibacillus knock-out mutants. EJNH, CMV, DBM, JP, and JAV wrote the manuscript with contributions from all authors.
Competing Interests
The authors declare no competing interests.
References
- 1.Dewick PM. Medicinal natural products: a biosynthetic approach. 3rd edn. John Wiley & Sons Ltd; 2009. [Google Scholar]
- 2.Newman DJ, Cragg GM. Natural products as sources of new drugs from 1981 to 2014. J Nat Prod. 2016;79:629–661. doi: 10.1021/acs.jnatprod.5b01055. [DOI] [PubMed] [Google Scholar]
- 3.Mlot C. Microbiology. Antibiotics in nature: beyond biological warfare. Science. 2009;324:1637–1639. doi: 10.1126/science.324_1637. [DOI] [PubMed] [Google Scholar]
- 4.Meiser P, Bode HB, Müller R. The unique DKxanthene secondary metabolite family from the myxobacterium Myxococcus xanthus is required for developmental sporulation. Proc Natl Acad Sci USA. 2006;103:19128–19133. doi: 10.1073/pnas.0606039103. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5.Hawver LA, Jung SA, Ng WL. Specificity and complexity in bacterial quorumsensing systems. FEMS Microbiol Rev. 2016;40:738–752. doi: 10.1093/femsre/fuw014. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6.Höfer I, et al. Insights into the biosynthesis of hormaomycin, an exceptionally complex bacterial signaling metabolite. Chem Biol. 2011;18:381–391. doi: 10.1016/j.chembiol.2010.12.018. [DOI] [PubMed] [Google Scholar]
- 7.Phelan VV, Liu WT, Pogliano K, Dorrestein PC. Microbial metabolic exchange-the chemotype-to-phenotype link. Nat Chem Biol. 2012;8:26–35. doi: 10.1038/nchembio.739. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 8.Saha R, Saha N, Donofrio RS, Bestervelt LL. Microbial siderophores: a mini review. J Basic Microbiol. 2013;53:303–317. doi: 10.1002/jobm.201100552. [DOI] [PubMed] [Google Scholar]
- 9.Wilson MC, et al. An environmental bacterial taxon with a large and distinct metabolic repertoire. Nature. 2014;506:58–62. doi: 10.1038/nature12959. [DOI] [PubMed] [Google Scholar]
- 10.Lincke T, Behnken S, Ishida K, Roth M, Hertweck C. Closthioamide: an unprecedented polythioamide antibiotic from the strictly anaerobic bacterium Clostridium cellulolyticum . Angew Chem. 2010;49:2011–2013. doi: 10.1002/anie.200906114. [DOI] [PubMed] [Google Scholar]
- 11.Pidot SJ, Coyne S, Kloss F, Hertweck C. Antibiotics from neglected bacterial sources. Int J Med Microbiol. 2014;304:14–22. doi: 10.1016/j.ijmm.2013.08.011. [DOI] [PubMed] [Google Scholar]
- 12.Wilson MC, Piel J. Metagenomic approaches for exploiting uncultivated bacteria as a resource for novel biosynthetic enzymology. Chem Biol. 2013;20:636–647. doi: 10.1016/j.chembiol.2013.04.011. [DOI] [PubMed] [Google Scholar]
- 13.Rondon MR, et al. Cloning the soil metagenome: a strategy for accessing the genetic and functional diversity of uncultured microorganisms. Appl Environ Microbiol. 2000;66:2541–2547. doi: 10.1128/aem.66.6.2541-2547.2000. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 14.Banik JJ, Brady SF. Recent application of metagenomic approaches toward the discovery of antimicrobials and other bioactive small molecules. Curr Opin Microbiol. 2010;13:603–609. doi: 10.1016/j.mib.2010.08.012. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 15.Vorholt JA, Vogel C, Carlström CI, Müller DB. Establishing causality: opportunities of synthetic communities for plant microbiome research. Cell Host Microbe. 2017;22:142–155. doi: 10.1016/j.chom.2017.07.004. [DOI] [PubMed] [Google Scholar]
- 16.Bai Y, et al. Functional overlap of the Arabidopsis leaf and root microbiota. Nature. 2015;528:364–369. doi: 10.1038/nature16192. [DOI] [PubMed] [Google Scholar]
- 17.Vorholt JA. Microbial life in the phyllosphere. Nat Rev Microbiol. 2012;10:828–840. doi: 10.1038/nrmicro2910. [DOI] [PubMed] [Google Scholar]
- 18.Ryffel F, et al. Metabolic footprint of epiphytic bacteria on Arabidopsis thaliana leaves. ISME J. 2016;10:632–643. doi: 10.1038/ismej.2015.141. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 19.Blin K, et al. antiSMASH 4.0-improvements in chemistry prediction and gene cluster boundary identification. Nucl Acids Res. 2017;45:W36–W41. doi: 10.1093/nar/gkx319. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20.Medema MH, et al. Minimum information about a biosynthetic gene cluster. Nat Chem Biol. 2015;11:625–631. doi: 10.1038/nchembio.1890. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 21.Yang SC, Lin CH, Sung CT, Fang JY. Antibacterial activities of bacteriocins: application in foods and pharmaceuticals. Fron Microbiol. 2014;5:241. doi: 10.3389/fmicb.2014.00241. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 22.Helfrich EJ, Piel J. Biosynthesis of polyketides by trans-AT polyketide synthases. Nat Prod Rep. 2016;33:231–316. doi: 10.1039/c5np00125k. [DOI] [PubMed] [Google Scholar]
- 23.Blin K, Medema MH, Kottmann R, Lee SY, Weber T. The antiSMASH database, a comprehensive database of microbial secondary metabolite biosynthetic gene clusters. Nucl Acids Res. 2017;45:D555–D559. doi: 10.1093/nar/gkw960. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 24.Hillenmeyer ME, Vandova GA, Berlew EE, Charkoudian LK. Evolution of chemical diversity by coordinated gene swaps in type II polyketide gene clusters. Proc Natl Acad Sci USA. 2015;112:13952–13957. doi: 10.1073/pnas.1511688112. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 25.Lazos O, et al. Biosynthesis of the putative siderophore erythrochelin requires unprecedented crosstalk between separate nonribosomal peptide gene clusters. Chem Biol. 2010;17:160–173. doi: 10.1016/j.chembiol.2010.01.011. [DOI] [PubMed] [Google Scholar]
- 26.Lombó F, et al. Deciphering the biosynthesis pathway of the antitumor thiocoraline from a marine actinomycete and its expression in two Streptomyces species. ChemBioChem. 2006;7:366–376. doi: 10.1002/cbic.200500325. [DOI] [PubMed] [Google Scholar]
- 27.Arrebola E, et al. Mangotoxin: a novel antimetabolite toxin produced by Pseudomonas syringae inhibiting ornithine/arginine biosynthesis. Physiol Mol Plant Pathol. 2003;63:117–127. [Google Scholar]
- 28.Bassler BL, Losick R. Bacterially speaking. Cell. 2006;125:237–246. doi: 10.1016/j.cell.2006.04.001. [DOI] [PubMed] [Google Scholar]
- 29.Pandey SS, Patnana PK, Rai R, Chatterjee S. Xanthoferrin, the α-hydroxycarboxylate-type siderophore of Xanthomonas campestris pv campestris is required for optimum virulence and growth inside cabbage. Mol Plant Pathol. 2017;18:949–962. doi: 10.1111/mpp.12451. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 30.Barona-Gomez F, Wong U, Giannakopulos AE, Derrick PJ, Challis GL. Identification of a cluster of genes that directs desferrioxamine biosynthesis in Streptomyces coelicolor M145. J Am Chem Soc. 2004;126:16282–16283. doi: 10.1021/ja045774k. [DOI] [PubMed] [Google Scholar]
- 31.Lee JY, et al. Biosynthetic analysis of the petrobactin siderophore pathway from Bacillus anthracis . J Bacteriol. 2007;189:1698–1710. doi: 10.1128/JB.01526-06. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 32.Barry SM, Challis GL. Recent advances in siderophore biosynthesis. Curr Opin Chem Biol. 2009;13:205–215. doi: 10.1016/j.cbpa.2009.03.008. [DOI] [PubMed] [Google Scholar]
- 33.Lindow SE, Brandl MT. Microbiology of the phyllosphere. Appl Environ Microbiol. 2003;69:1875–1883. doi: 10.1128/AEM.69.4.1875-1883.2003. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 34.Lindow SE, Leveau JH. Phyllosphere microbiology. Curr Opin Biotechnol. 2002;13:238–243. doi: 10.1016/s0958-1669(02)00313-0. [DOI] [PubMed] [Google Scholar]
- 35.Schöner TA, et al. Aryl polyenes, a highly abundant class of bacterial natural products, are functionally related to antioxidative carotenoids. ChemBioChem. 2016;17:247–253. doi: 10.1002/cbic.201500474. [DOI] [PubMed] [Google Scholar]
- 36.Wang M, et al. Sharing and community curation of mass spectrometry data with Global Natural Products Social Molecular Networking. Nat Biotechnol. 2016;34:828–837. doi: 10.1038/nbt.3597. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 37.Mootz HD, Marahiel MA. The tyrocidine biosynthesis operon of Bacillus brevis: complete nucleotide sequence and biochemical characterization of functional internal adenylation domains. J Bacteriol. 1997;179:6843–6850. doi: 10.1128/jb.179.21.6843-6850.1997. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 38.Gebhardt K, Pukall R, Fiedler HP, Streptocidins A-D. novel cyclic decapeptide antibiotics produced by Streptomyces sp. Tu 6071. I. Taxonomy, fermentation, isolation and biological activities. J Antibiot. 2001;54:428–433. doi: 10.7164/antibiotics.54.428. [DOI] [PubMed] [Google Scholar]
- 39.Zgurskaya HI, Löpez CA, Gnanakaran S. Permeability barrier of Gram-negative cell envelopes and approaches to bypass it. ACS Infect Dis. 2015;1:512–522. doi: 10.1021/acsinfecdis.5b00097. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 40.Zhou X, et al. Marthiapeptide A: an anti-infective and cytotoxic polythiazole cyclopeptide from a 60 L scale fermentation of the deep sea-derived Marinactinospora thermotolerans SCSIO 00652. J Nat Prod. 2012;75:2251–2255. doi: 10.1021/np300554f. [DOI] [PubMed] [Google Scholar]
- 41.Grubbs KJ, et al. Large-scale bioinformatics analysis of Bacillus genomes uncovers conserved roles of natural products in bacterial physiology. mSystems. 2017;2:e00040–17. doi: 10.1128/mSystems.00040-17. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 42.Donia MS, et al. A systematic analysis of biosynthetic gene clusters in the human microbiome reveals a common family of antibiotics. Cell. 2014;158:1402–1414. doi: 10.1016/j.cell.2014.08.032. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 43.Russel J, Røder HL, Madsen JS, Burmølle M, Sørensen SJ. Antagonism correlates with metabolic similarity in diverse bacteria. Proc Natl Acad Sci USA. 2017;114:10684–10688. doi: 10.1073/pnas.1706016114. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 44.Maida I, et al. Antagonistic interactions between endophytic cultivable bacterial communities isolated from the medicinal plant Echinacea purpurea . Environ Microbiol. 2016;18:2357–2365. doi: 10.1111/1462-2920.12911. [DOI] [PubMed] [Google Scholar]
- 45.Hassani MA, Durán P, Hacquard S. Microbial interactions within the plant holobiont. Microbiome. 2018;6:58. doi: 10.1186/s40168-018-0445-0. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 46.Venturi V, Keel C. Signaling in the rhizosphere. Trends Plant Sci. 2016;21:187–198. doi: 10.1016/j.tplants.2016.01.005. [DOI] [PubMed] [Google Scholar]
- 47.Chodkowski JL, Shade A. A synthetic community system for probing microbial interactions driven by exometabolites. mSystems. 2017;2:e00129–17. doi: 10.1128/mSystems.00129-17. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 48.Stringlis IA, Zhang H, Pieterse CMJ, Bolton MD, de Jonge R. Microbial small molecules - weapons of plant subversion. Nat Prod Rep. 2018;35:410–433. doi: 10.1039/c7np00062f. [DOI] [PubMed] [Google Scholar]
- 49.Raaijmakers JM, Mazzola M. Diversity and natural functions of antibiotics produced by beneficial and plant pathogenic bacteria. Annu Rev Phytopathol. 2012;50:403–424. doi: 10.1146/annurev-phyto-081211-172908. [DOI] [PubMed] [Google Scholar]
- 50.Peyraud R, et al. Demonstration of the ethylmalonyl-CoA pathway by using 13C metabolomics. Proc Natl Acad Sci USA. 2009;106:4846–4851. doi: 10.1073/pnas.0810932106. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 51.Asnicar F, Weingart G, Tickle TL, Huttenhower C, Segata N. Compact graphical representation of phylogenetic data and metadata with GraPhlAn. PeerJ. 2015;3:e1029. doi: 10.7717/peerj.1029. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 52.Müller DB, Schubert OT, Rost H, Aebersold R, Vorholt JA. Systems-level proteomics of two ubiquitous leaf commensals reveals complementary adaptive traits for phyllosphere colonization. Mol Cell Proteomics. 2016;15:3256–3269. doi: 10.1074/mcp.M116.058164. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 53.Yamanaka K, et al. Direct cloning and refactoring of a silent lipopeptide biosynthetic gene cluster yields the antibiotic taromycin A. Proc Natl Acad Sci USA. 2014;111:1957–1962. doi: 10.1073/pnas.1319584111. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 54.Finn RD, et al. The Pfam protein families database: towards a more sustainable future. Nucl Acids Res. 2016;44:D279–285. doi: 10.1093/nar/gkv1344. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 55.Ceniceros A, Dijkhuizen L, Petrusma M, Medema MH. Genome-based exploration of the specialized metabolic capacities of the genus Rhodococcus . BMC Genomics. 2017;18:593. doi: 10.1186/s12864-017-3966-1. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 56.Yang JY, et al. Primer on agar-based microbial imaging mass spectrometry. J Bacteriol. 2012;194:6023–6028. doi: 10.1128/JB.00823-12. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 57.Ueoka R, et al. Metabolic and evolutionary origin of actin-binding polyketides from diverse organisms. Nat Chem Biol. 2015;11:705–712. doi: 10.1038/nchembio.1870. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 58.Durante-Rodríguez G, de Lorenzo V, Martínez-García E. The Standard European Vector Architecture (SEVA) plasmid toolkit. Methods Mol Biol. 2014;1149:469–478. doi: 10.1007/978-1-4939-0473-0_36. [DOI] [PubMed] [Google Scholar]
- 59.Radeck J, et al. The Bacillus BioBrick Box: generation and evaluation of essential genetic building blocks for standardized work with Bacillus subtilis . J Biol Eng. 2013;7:29. doi: 10.1186/1754-1611-7-29. [DOI] [PMC free article] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.