Variable genetic architectures produce virtually identical molecules in bacterial symbionts of fungus-growing ants

Significance

Bacterially produced natural products comprise a group of molecules with highly diverse and generally complex structures that possess a remarkable array of biological activities. These molecules are separated into families sharing a common structural core and, accordingly, conserved sets of genes encoding the biosynthetic enzymes required to generate these shared structural features. Genomic characterization of related bacteria that produce remarkably similar molecules led to the surprising discovery that gene context was not conserved for the respective biosynthetic pathways. A comparison of these variable arrangements documents one way in which closely related symbiotic bacteria acquire the capacity to produce new molecules with new functions.

Abstract

Small molecules produced by Actinobacteria have played a prominent role in both drug discovery and organic chemistry. As part of a larger study of the actinobacterial symbionts of fungus-growing ants, we discovered a small family of three previously unreported piperazic acid-containing cyclic depsipeptides, gerumycins A–C. The gerumycins are slightly smaller versions of dentigerumycin, a cyclic depsipeptide that selectively inhibits a common fungal pathogen, Escovopsis. We had previously identified this molecule from a Pseudonocardia associated with Apterostigma dentigerum, and now we report the molecule from an associate of the more highly derived ant Trachymyrmex cornetzi. The three previously unidentified compounds, gerumycins A–C, have essentially identical structures and were produced by two different symbiotic Pseudonocardia spp. from ants in the genus Apterostigma found in both Panama and Costa Rica. To understand the similarities and differences in the biosynthetic pathways that produced these closely related molecules, the genomes of the three producing Pseudonocardia were sequenced and the biosynthetic gene clusters identified. This analysis revealed that dramatically different biosynthetic architectures, including genomic islands, a plasmid, and the use of spatially separated genetic loci, can lead to molecules with virtually identical core structures. A plausible evolutionary model that unifies these disparate architectures is presented.

Fungus-growing ants of the tribe Attini originated ∼50 Mya in the Amazon and have since evolved into more than 200 species that collectively are the major herbivores of the New World Tropics (1, 2). A typical agricultural system contains at least four interacting organisms: the ants, their fungal crop (phylum Basidiomycota), a symbiotic bacterium (Actinobacteria), and a specialized fungal pathogen (phylum Ascomycota). The ants maintain their fungal gardens by providing plant material to the fungal crop, whereas the bacterial symbionts, which often localize with specialized anatomical structures on the ant, provide small-molecule chemical defenses that suppress microfungal pathogens in the genus Escovopsis and other potential competitors (3, 4). In an earlier study on Apterostigma dentigerum, we characterized dentigerumycin (1) as the major antifungal agent produced by the symbiotic bacterium that selectively inhibited the Escovopsis pathogen (4). Our subsequent studies on the molecular diversity to be found in this well-defined ecological niche featured both a metabolomics approach to structurally characterize the molecules produced and a genomic approach to characterize the biosynthetic genes that produced them.

To initiate a systematic study on small molecules and their biosynthetic gene clusters (BGCs) found in the symbiotic bacteria, we returned to isolates from both Apterostigma spp. and the more highly derived Trachymyrmex cornetzi with a focus on dentigerumycin-like molecules. This study reports three previously unidentified cyclic depsipeptides named gerumycins A–C, which are structurally related to the originally discovered selective antifungal molecule dentigerumycin, the first three finished ant-associated Pseudonocardia genomes, and the three BGCs encoding the cyclic depsipeptides. The BGCs are encoded within variable genetic architectures in the three producing bacteria and are most plausibly accommodated by a model in which each has been recently acquired by the respective symbiont. The results, which describe both plasmid and chromosomal genomic islands, illustrate a useful strategy for symbionts of the fungus-growing ants to acquire and evolve new molecules.

Results

Dentigerumycin and the Gerumycins.

During our investigation of the bacterially produced small molecules that contribute chemical defenses to the fungus-farming ants, the metabolomic profiles of three ant-associated bacteria revealed that piperazic acid-containing cyclic depsipeptides are commonly observed within this ecological niche. Specifically, we observed that dentigerumycin (1) (Fig. 1 and SI Appendix, Table S1), the selective inhibitor of Escovopsis that was initially discovered from a bacterial symbiont of A. dentigerum (4), was also produced by Pseudonocardia sp. AL041005-10, a symbiont of T. cornetzi from Peru. In addition to 1, the chemical structures and BGCs of three newly discovered metabolites named gerumycins A–C (2–4) (Fig. 1 and SI Appendix, Table S1) from Apterostigma symbionts in Panama and Costa Rica, Pseudonocardia sp. EC080625-04 and HH130629-09, respectively, form the basis of this report.

Fig. 1.

Cyclic depsipeptides from ant-associated Pseudonocardia spp. The atoms highlighted in red in molecules 2 and 3 are derived from a spatially separated BGC.

Gerumycin A (2), the major small molecule produced by EC080625-04 was purified by reverse-phase chromatography to give an amorphous white powder having a molecular formula of C₂₅H₃₉ClN₈O₉, as determined by high-resolution mass spectrometry (MS) and nuclear magnetic resonance (NMR) analyses. One-dimensional proton and carbon NMR spectra, along with a suite of 2D NMR experiments (COSY, HSQC, HMBC), enabled complete assignment of the proton and carbon chemical shifts (SI Appendix, Table S2). Analysis of the COSY and HMBC correlations defined the spin systems of each amino acid, which led to a proposed cyclic depsipeptide structure for gerumycin that featured unusual residues, including piperazic acid, γ-hydroxypiperazic acid, γ-chloropiperazic acid, and α-methyl-l-serine. The chloro- and hydroxy-substitutions distinguished the aliphatic proton signals of each piperazic acid, giving rise to distinct patterns in the COSY for each of the three similar residues. A weak correlation between the methyl group protons and the β-protons of the α-methyl-l-serine, along with the HSQC and HMBC data, helped corroborate the identity of this unusual amino acid. Facile crystallization of gerumycin occurred in 50% acetonitrile/water to form colorless block crystals. An X-ray single-crystal diffraction experiment was performed using CuKα radiation, and the crystal structure was successfully solved and refined in space group P2₁2₁2₁. The absolute configuration of 2 was unequivocally determined through anomalous dispersion effects with a Flack x parameter of 0.021(3) (SI Appendix, Fig. S1 and Table S3).

Next, we isolated and characterized two additional gerumycin analogs (SI Appendix, Tables S1, S4, and S5), including a minor product from EC080625-04 and the major cyclic depsipeptide produced by HH130629-09 (for experimental details, see the SI Appendix). In addition to gerumycin A, EC080625-04 produces a number of congeners in relatively limited quantities, and gerumycin B (3) was among the most abundant of these minor constituents and differed from 2 in that the lactic acid of the ester-forming depsipeptide linkage is replaced by 2-hydroxybutanoic acid (Fig. 1). The HH130629-09 product gerumycin C (4) differed from 2 by a single hydroxylation of a second piperazic acid moiety. In addition to the MS and NMR characterization of 4, crystals of 4 formed by vapor diffusion with dichloromethane/benzene were analyzed by an X-ray single-crystal diffraction experiment using MoKα radiation. The structure was successfully solved and refined in space group C2. The absolute configuration of 4 was unequivocally determined through anomalous dispersion effects with a Flack x parameter of 0.02(2) (SI Appendix, Fig. S1 and Table S3) revealing that 2 and 4 have the same absolute configuration at every shared stereocenter.

Compounds 1–4 differ in several respects. Most significantly, in all three gerumycins, the inclusion of an α-hydroxy acid instead of a β-hydroxy α-amino acid has two consequences: the 19-membered macrocyclic core observed in 1 becomes an 18-membered core in 2–4, and the amino group of the β-hydroxyleucine of 1, which forms the attachment of the polyketide-derived fragment, is absent. Additional tailoring chemistry in 2–4 also differentiates these molecules from each other and 1. The trivial name gerumycin reflects both the similarity to dentigerumycin and the missing polyketide-derived fragment.

In addition to having different structures, compounds 1–4 have different biological activity. Dentigerumycin is capable of selectively inhibiting the nest parasite Escovopsis but not the ant’s fungal cultivar at low micromolar concentrations (4). In contrast, purified 2 did not exhibit significant antifungal activity in vitro up to 1 mM against a dentigerumycin-sensitive strain, and phenotypic screening of the gerumycin-producing bacteria against Escovopsis (SI Appendix, Fig. S2) did not display marked activity, indicating that 1 is at least three orders of magnitude more potent than the gerumycins at suppressing Escovopsis.

Biosynthetic Gene Clusters.

The observation of four cyclic depsipeptides and, in particular, the three virtually identical gerumycins from two distinct bacterial isolates prompted an investigation aimed at a better understanding of the genetic basis of cyclic depsipeptide production by ant-associated Pseudonocardia species. Accordingly, we sequenced the genomes of the producing organisms using PacBio single-molecule, real-time (SMRT) cell technology and hierarchical genome assembly process (HGAP) (5, 6). In general, the bacterial symbionts possess a circular chromosome of ∼6.1 Mb with an unusually high GC content of ∼74%. The three bacteria can be distinguished as distinct species using average nucleotide identity (ANI) across the entire chromosome as a metric (7). Specifically, the ANI values ranged between ∼84% (EC080625-04 compared with the other two bacteria) and 92% (comparing HH130629-09 and AL041005-10). Each individual bacterium also differed in total replicon number (SI Appendix, Fig. S3 and Table S6) with EC080625-04 and HH130629-09, but not AL041005-10, carrying plasmids.

We next used antiSMASH (8) to predict the natural product BGCs encoded by these three ant-associated Pseudonocardia, and we identified chromosomally encoded loci for the production of 1 and 4 by AL041005-10 and HH130629-09, respectively (Fig. 2). Closer inspection of these BGCs revealed that despite their position on the chromosome, they reside in genomic islands (GIs). Specifically, each BGC is part of a sequence of DNA that differs in average GC content from the rest of the chromosome by ∼4.5% or, more locally, relative to the nearest 20 kb up- or downstream, by 3.5–5.6%. Codon use in these regions, defined as a GI, also differs from the rest of the chromosome, reflecting this difference in GC content (SI Appendix, Fig. S4). Most notably, each of these GIs is integrated at a distinct tRNA gene and could also be detected in silico using SIGI-HMM (9), predictive software integrated as part of IslandViewer3 (10).

Fig. 2.

BGCs encoding the production of 1–4. The hybrid NRPS/PKS gene cluster responsible for the production of 1 is drawn, highlighting the juxtaposition of different subclusters (NRPS in green; PKS in blue). The two distinct plasmid-borne BGCs (named primary and accessory) are separated by ∼92 kb in EC080625-04 and encode for the full suite of enzymes required to produce 2 and 3. The single chromosomal BGC in HH130629-09 that encodes for 4 is drawn, highlighting the primary and accessory plasmid-borne clusters.

The GI harboring the hybrid NRPS/PKS BGC responsible for the production of 1 in AL041005-10 was inserted at a tRNA-Pro gene and lies adjacent to a second type 1 PKS BGC. We defined the boundaries of this GI by the tRNA gene and a 48-bp direct-repeat sequence containing two substitutions that is found ∼69.6 kb away (SI Appendix, Fig. S5 and Tables S7 and S8). Notably, the dentigerumycin BGC itself was seemingly formed by the joining of NRPS and PKS gene clusters, which is apparent from the discrete spatial organization of the distinctly classed genes in the cluster (Fig. 2).

In contrast to the BGCs for 1 and 4, inspection of the antiSMASH-predicted BGCs in the EC080625-04 replicons precluded identification of a single locus satisfying the biosynthetic requirements for 2 and 3. Surprisingly, two loci, separated by ∼92 kb on plasmid pFRP1-1, encode for gerumycin production (Fig. 2 and SI Appendix, Table S9). The first includes ∼38 kb encoding three NRPS genes; an l-ornithine N5-monooxygenase, required for the first and only known step in piperazic acid biosynthesis (11); and a hydroxymethyltransferase that has recently been implicated in the conversion of d-alanine to α-methyl-l-serine (12). Additional tailoring, resistance, and regulatory genes are also present in this primary cluster, which is bookended by mobile genetic elements. Indeed, the plasmid pFRP1-1 is characterized by a mosaic of horizontally acquired genes punctuated with mobile genetic elements (e.g., transposases, integrases, endonucleases) or relics thereof. In addition to the primary and accessory gerumycin gene clusters, a toxin/antitoxin (TA) pair is located ∼10 kb from the primary cluster (SI Appendix, Fig. S6 and Table S9). This TA pair likely ensures plasmid maintenance or, more explicitly, continued inheritance of the gerumycin chemotype to bacteria in this particular symbiosis. Most interestingly, the accessory NRPS cluster encodes a complement of enzymes that are required for gerumycin production, including a chlorinase and a ketoreductase domain-containing NRPS required for α-ketoacid activation.

Remarkably, the distinct biosynthetic strategies used by AL041005-10 and EC080625-04 are unified in HH130629-09. First, a single chromosomal BGC encoding for the production of 4 exists on an ∼76-kb GI that encodes many mobile genetic elements and appears to have been recently inserted at a tRNA-Gly gene. The periphery of the GI includes a number of mobile elements, including a disrupted traSA-like orf similar to a gene transfer protein associated with the mobility of the actinomycete integrative chromosomal element pSAM2 in Streptomyces ambofaciens (13). Second, the primary and accessory BGCs reported for EC080625-04 are present and adjacent to one another within this GI (Fig. 2 and SI Appendix, Table S10). In comparison with EC080625-04, the major difference is the orientation of the primary cluster, which is inverted and separated from the accessory cluster by only a single predicted mobile genetic element. The nucleotide identity of the biosynthetic genes for 2–4 is ∼99%. In comparison, the nucleotide identity of five housekeeping genes (atpD, dnaA, EF-Tu, gyrB, and rpoB) that have been used for phylogenetic analysis of these bacteria (14) are between 86% (dnaA) and 96% (EF-Tu) in the gerumycin-producing bacteria.

Discussion

The discovery of the gerumycins and dentigerumycin along with their associated BGCs provides an unusual opportunity to formulate a possible scenario for the acquisition of new molecular and functional features through the evolution of biosynthetic gene collectives. Plasmids and genomic islands are commonly recognized as important factors for facilitating genome evolution in pathogens, commensals, and symbionts (15). The plasmids and genomic islands seen in this study provide molecular level support for the evolution of chemical diversification in the bacterial symbionts of the fungus-growing ant system. Dentigerumycin’s ability to selectively inhibit a microfungal pathogen rather than the ant’s fungal crop fits a proposed ecological role that is directly related to niche adaptation. The gerumycins, however, do not display the same selective antifungal activity in phenotypic screens or in vitro assays. Structurally, the gerumycins (2–4) are more closely related to the piperazimycins, which were isolated from a marine Streptomyces during anticancer drug-discovery efforts (16). At the molecular level, it seems likely that dentigerumycin, a hybrid NRPS/PKS, is derived from the gerumycins or similar cyclic nonribosomal peptides because that model would coordinate increasing chemical complexity with the development of specialized biological activity. However, at the genetic level, the BGCs for 1 and 2–4 differ too dramatically to confidently discuss a trajectory from one to the other, and the possibility of convergent evolution cannot be excluded. Moreover, the genetic architecture of the clusters, the genomic islands, and the plasmid imply that the pathways have largely evolved outside of these specific Pseudonocardia. In the future, additional chemical structures and their respective BGCs will help to clarify this situation.

In bacteria that have been historically used for drug discovery, there are a growing number of examples in which horizontal gene transfer (HGT) drives genome evolution to yield diverse BGCs in closely related organisms. For instance, the chromosomal termini, or “arms,” of the linear Streptomyces chromosomes have long been implicated in large-scale HGT and genetic diversification, and GIs were quickly identified during the earliest comparisons of complete Streptomyces genomes, ultimately revealing that the clinically relevant small-molecule spiramycin was encoded by a GI in S. ambofaciens (17). The most comprehensive comparative analysis, which now includes nearly 100 strains in the genus Salinispora, argues that species-specific GIs carrying BGCs possess an unspecified functional or adaptive role in the marine environment (18, 19). Similar studies in Bacillus argue for the role of GIs in niche adaptation (20). Our study provides an explicit example of adaptation through HGT in which the BGCs for 1 and 4 have been integrated into the chromosome of ant-associated Pseudonocardia spp., whereas 2–3 are encoded by two plasmid-borne BGCs.

Although the biosynthetic pathways for 1 and those for 2–4 do not yet allow meaningful comparison, a more limited comparison of the gerumycins (2–4), with their virtually identical chemical and genetic structures (but not contexts), can be accommodated in a scenario for BGC acquisition and evolution. In EC080625-04, one locus encoding a seemingly complete set of NRPS genes (the primary BGC) is separated from a second locus responsible for tailoring chemistry and precursor production (the accessory BGC). In HH130629-09, these primary and accessory gene clusters are adjacent to each other and exist within a GI on the chromosome. The overall conserved identities of the genes responsible for gerumycin production between the two bacterial isolates implicate HGT. Such HGT events often confound our ability to define an evolutionary history; however, in this case, these genetic and molecular structures can be assembled into a plausible natural history for these cyclic depsipeptides (Fig. 3). This scheme describes the possible steps that would incorporate new BGCs into the chromosome after plasmid acquisition from the environment and under the appropriate selective pressure(s). Specifically, the direct observation of disparate and cooccurring subclusters—the primary and accessory BGCs of EC080625-04 that are present as a single BGC in HH130628-09—is particularly striking and highlights the potential of studying a set of Pseudonocardia isolated from a specialized ecological niche. The plausible scenario depicted in Fig. 3 describes two steps. First, a unification step employs the model of subcluster joining, which is well supported by the full spectrum of available sequence data, arguing that molecular complexity is built from smaller parts (21, 22). The second step, integration, reaffirms the importance of HGT and GIs in bacteria and echoes a growing number of examples that link this process and genomic structures to secondary metabolism.

Fig. 3.

Plausible scheme of BGC evolution in an ant-associated bacterium. We depict the stepwise unification of plasmid-borne subclusters and integration into the bacterial chromosome. The scheme is representative of the transition from a plasmid-borne split BGC in EC080625-04 to a newly integrated singular version of these BGCs in HH130629-09. A hypothetical plasmid-borne intermediate (plasmid*) is also drawn.

BGC diversification through subcluster joining and, potentially, the transition of plasmid-borne to chromosomally borne geno- and chemotypes could be the usual mode for bacterial adaptation, at least within the context of a symbiotic bacterium. These bacteria risk losing host–symbiont fidelity by welcoming random insertions into a presumably well-adapted chromosomal structure. A plasmid-based strategy of chemical diversification, at least in terms of BGC acquisition, would maintain the relationship between microbe and host without forfeiting rapid genetic adaptation. Accordingly, a plausible reconstruction illustrates a trajectory from plasmid to chromosome and, in this instance, is rationalized by three other important considerations. First, dentigerumycin (1) and its chromosomally encoded BGC represents a gain-of-function: a selective antifungal within a specialized ecological niche (4). Second, the GI for 4 shows a relic of an actinobacterial plasmid/integrative element-associated transfer protein. Finally, the plasmid encoding the BGCs for 2 and 3 encodes for a toxin–antitoxin pair that enforces plasmid maintenance. Although these observations argue for the functional relevance of a BGC-encoding plasmid and a transition to the chromosome, an alternative model in which the loss of GIs to a plasmid and the genetic separation of biosynthetic parts cannot be excluded for this small set of bacteria.

The characterization of three genomes and a set of small molecules (1–4) from fungus-growing ant-associated Pseudonocardia spp. reveals the importance of a plasmid-based strategy of recently acquired DNA for the transmission and evolution of molecular diversity within a specialized ecological niche. The results represent critical steps in describing the evolutionary history of genetically encoded small molecules and the manner in which ant-associated bacteria acquire and evolve new molecules with new functions.

Materials and Methods

Bacterial Strains and Growth.

The Pseudonocardia spp. studied here were isolated from the exoskeleton of ants of the genus Apterostigma in Panama (EC080625-04) and Costa Rica (HH130629-09) or T. cornetzi (AL041005-10) in Peru (2, 14). For small-molecule production, each bacterium was grown on International Streptomyces Project Medium 2 (ISP2) agar for 14 d at 30 °C (see the SI Appendix for more details). For genomic DNA isolation, the bacteria were grown in liquid ISP2 medium for no longer than 10 d postinoculation from spore stocks that were preserved at −80 °C.

Isolation of Dentigerumycin and Gerumycins A–C.

In each case, 14-d-old Pseudonocardia cultures grown on ISP2 agar were cut into ∼1-cm square pieces, and the organic material was extracted with ethyl acetate. The extracts were then concentrated in vacuo and adsorbed onto Celite 503 (Sigma) for dry packing onto a 10 g C₁₈ SepPak column (Waters) that had been conditioned and preequilibrated with 15% (vol/vol) acetonitrile in water. Material was eluted from the column using increasing amounts of acetonitrile. Fractions containing dentigerumycin and gerumycin A–C were detected using liquid chromatography (LC)/MS analysis. Subsequent purification was afforded by reverse-phase HPLC using C₁₈, phenylhexyl-, and/or biphenyl-based solid phases (see the SI Appendix for molecule-specific details).

Mass Spectrometry Analysis of Gerumycin.

HPLC-MS of organic extracts was performed using an Agilent HPLC system with a diode array detector and a 6130 Series quadrupole mass spectrometer. The LC separation was performed using a Phenomenex C₁₈ (5 μm, 100 × 4.6 mm) column, with the following gradient: 0–2 min of isocratic 10% CH₃CN (0.1% formic acid), 2- to 14-min linear gradient from 10% to 100% CH₃CN (0.1% formic acid), 14–15 min of isocratic 100% CH₃CN (0.1% formic acid), and 15- to 21-min linear gradient from 100% to 10% CH₃CN (0.1% formic acid). High-resolution data for 2 was obtained at the University of Illinois Urbana–Champaign School of Chemical Sciences Mass Spectrometry Laboratory for high-resolution MS analysis on a Waters Q-TOF Ultima ESI. Samples containing 1, 3, or 4 were submitted to the Harvard University Faculty of Arts and Sciences Small Molecule Mass Spectrometry Laboratory and analyzed using a Bruker Maxis Impact LC-q-TOF Spectrophotometer.

NMR Analysis.

All NMR experiments were run on an Agilent 600-MHz NMR equipped with a 5-mm z-gradient inverse HCN probe. Dentigerumycin was dissolved in dimethyl sulfoxide for NMR analysis. Gerumycins A–C were dissolved in CD₃OD for ¹H-NMR, ¹³C-NMR, HSQCAD, gHMBCAD, and gCOSY data collection. A sample of gerumycin A was exchanged into CD₃OH to obtain proton signals for the N-H protons. Signals for O-H protons could not be definitively assigned. All spectra are included in the SI Appendix.

Single-Crystal X-Ray Diffraction.

Colorless block crystals of 2 were obtained through overnight crystallization in a solvent mixture of 50% MeCN/H₂O in a glass tube, and colorless plate crystals of 4 were obtained through vapor diffusion where the compound was solvated in dichloromethane in a small glass tube and benzene was used as antisolvent. For 2, a crystal was picked in NVH immersion oil and was mounted in a 100 K N₂ cold stream on a Bruker APEX II DUO diffractometer with a D8 three-circle fixed chi goniometer and an APEX II CCD detector, where a sphere of data to 2θ = 133.3° was collected using φ and ω scans (CuΚα radiation used). For 4, a crystal was picked in NVH immersion oil and was mounted in a 100 K N₂ cold stream on a Bruker SMART APEX II diffractometer equipped with IμS microfocus MoKα and CuKα sealed tube sources and a Triumph monochromator, Bruker D8 three-circle fixed chi goniometer, and an APEX II CCD detector, where a sphere of data to 2θ = 50.1° was collected using φ and ω scans (MoΚα radiation used). SAINT was used to perform integration (23), and SADABS was used to apply a multiscan absorption correction to the data (24). Intrinsic phasing (SHELXTL XT) was used to solve the structure, which was subsequently refined by full-matrix least-squares on F² (SHELXL-2014) (25, 26). Hydrogens were first added through the riding model, and hydrogens on methyl groups were refined as rigid idealized groups. Hydrogens bound to heteroatoms, which initially added through the riding model or added on the basis of residual electron density, were allowed to undergo isotropic free refinement by removing the AFIX directives, which were reintroduced/added after completing free refinement. For 2, an extinction correction (EXTI directive) was applied in the refinement of the structure. The absolute configurations of 2 and 4 were confirmed on the basis of Flack x parameters of 0.021(3) and 0.02(2), respectively (27).

DNA Isolation, Sequencing, Assembly, and Analysis.

Genomic DNA was isolated from Pseudonocardia that were physically disrupted in liquid N₂ using a mortar and pestle, and then treated with lysozyme (2.5 mg/mL), protease (1 mg/mL; protease from Aspergillus melleus), and 1% SDS in TE buffer (10 mM Tris⋅HCl, 1 mM EDTA, pH 8.0). Nucleic acids were precipitated using iso-propanol, treated with RNase A, and purified using phenol:chloroform:isoamyl alcohol (25:24:1) saturated with TE buffer. Whole-genome sequencing was performed using PacBio SMRT sequencing technology (5) at the Duke University Center for Genomic and Computational Biology (GCB), Genome Sequencing Shared Resource. For each bacterium, data were collected from three SMRT cells. An 8- to 11-kb insert library (EC080625-04 and AL041005-10) or 15- to 20-kb insert library (HH130629-09) of genomic DNA was sequenced. The final assemblies reflect the result of the HGAP (6). The resulting HGAP unique contigs (unitigs) were manually curated using Geneious (28) or IGV (29) to identify overlapping/degenerate sequences between distinct unitigs or within a single unitig for circular replicons. Corrected PacBio reads were mapped onto each sequence or circular permutations thereof using GAEMR v1.0.1 (www.broadinstitute.org/software/gaemr) to verify the assemblies. The chromosome and plasmid topology for each strain, including a GC plot and the location of antiSMASH v2.0.2-detected (8) BGCs, is shown in SI Appendix, Fig. S5 [generated using DNAPlotter (30)], and a summary of the replicons is available in SI Appendix, Table S6. Sequences were annotated using the NCBI Prokaryotic Genome Annotation Pipeline; however, manual inspection and curating were performed in the particular genetic loci of interest to this work (the BGCs for 1–4), as guided by the RAST server (31, 32) and blastp (33). Genomic island prediction was performed using IslandViewer3 (10).