This study used a total of 23 A. echinatior colonies collected during fieldwork in the Gamboa area of Soberania National Park, Panama, between 2001 and 2014. Colonies were subsequently maintained at the University of Copenhagen, in climate-controlled rooms (ca. 70% humidity, 25°C), with the exception of the Ae707 sub-colony, which was set up in the laboratory for ∼12 months at the University of East Anglia, and fed bramble leaves, apple, and dry rice, similar to the colonies maintained in Copenhagen. Pseudonocardia strains were isolated from 1–2 A. echinatior worker ants from each of the 23 captive colonies by direct contact of a needle with the laterocervical plates. Bacteria were cultured from 1–2 single worker ants taken from each colony (Figure 1). They were plated onto Lennox agar (20 g agar, 10 g Tryptone, 5 g Yeast Extract, 5 g NaCl made up to 1 L with distilled water and then autoclaved) and incubated at 30°C until small white Pseudonocardia colonies were visible. Isolates were re-streaked until pure and then spread for confluent lawns on MS agar (20 g agar, 20 g mannitol, 20 g soya flour made up to 1 L with tap water and autoclaved twice) to induce sporulation. Spores were collected by gentle washing in 3 mL sterile 20% (v/v) glycerol using sterile cotton buds and then transferred to 2 mL screw cap tubes and stored at -20°C. Phylotype identification of Pseudonocardia was performed by PCR amplification of the 16S rRNA gene using primers PRK341F (5′-CCT ACG GGR BGC ASC AG-3′) and MPRK806R (5′-GGA CTA CNN GGG TAT CTA AT-3′).

To simulate challenges to Pseudonocardia dominated leafcutter ant cuticles and thereby assess invasion of Pseudonocardia microbiomes by other bacteria, we devised the following bioassays. Pseudonocardia spores were spread onto sterile cellophane disks placed on top of MS agar plates and grown at 30°C for 7 days until they covered the plates in a confluent lawn. The cellophane disks were then peeled off to remove the Pseudonocardia and the agar plates were replenished with nutrients by adding 200 μl Lennox broth and then air dried to get them ready for the bioassays. Unicellular bacterial test strains were grown from frozen glycerol stocks overnight in 10 mL Lennox broth at 37°C. They were then sub-cultured 1:100 in 10 mL fresh Lennox broth. Subcultures were grown for ∼4.5 h, after which an OD₆₀₀ was measured, where OD₆₀₀ = 1 was assumed to represent 8 × 10⁹ cells. Working stocks were then made of the test strains to 10⁸ cells. Equivalently, titred spore stocks of Streptomyces (Kieser et al., 2000) were diluted to a working concentration of 10⁸ spores. Subsequently, 10 μl of the working stocks of all 10 bacterial test strains (five unicellular and five Streptomyces) were added to the plates by pipetting and the plates were incubated for 2 days at 30°C.

High quality draft genomes of strains Ae150A_Ps1, Ae168_Ps1, Ae263_Ps1, Ae356_Ps1, Ae331_Ps2, Ae406_Ps2, Ae505_Ps2, and Ae706_Ps2 were obtained using Illumina technology. The two strains Ae707_Ps1 and Ae717_Ps2 were sequenced using PacBio to obtain high quality draft genomes to use as a reference for comparison. Pseudonocardia spores were inoculated onto Lennox agar plates covered with sterilized cellophanes. After 2–3 weeks of incubation at 30°C the cellophanes were removed and the mycelium was scraped off into sterile tubes. The salting out method was used to prepare genomic DNA from these mycelium samples (Kieser et al., 2000). Minor modifications included the addition of achromopeptidase and DNase-free RNase (Qiagen) to the lysozyme step, as well as longer incubation steps for lysozyme and proteinase K treatment. DNA was quantified and quality checked using a Nanodrop 2000c spectrophotometer. Illumina sequencing of DNA was carried out at the DNA Sequencing Facility, Department of Biochemistry, University of Cambridge, UK, using TruSeq PCR-free and Nextera Mate Pair libraries and a MiSeq 600 sequencer. Genome assembly was performed using Roche Newbler v3.0, scaffolds were polished using PILON version 1.13, and reads were mapped using Burrows–Wheeler transformation version 0.7.12-r1039 (Walker et al., 2014). PacBio sequencing of DNA was carried out at the Earlham Institute, Norwich Research Park, Norwich, UK, using SMRT bell adaptor libraries and PacBio standard large insert conditions, with 2–3 SMRT cells per sample. PacBio data were assembled using SMRT analysis software (Pacific Biosciences of California, Inc.) incorporating HGAP3 (Chin et al., 2013). Assembled genomes were submitted to antiSMASH for BGC analysis and individual BGCs were compared across all strains using MultiGeneBlast (Weber et al., 2015). Each BGC was downloaded as a GenBank file from the antiSMASH output and amino acid sequences from those clusters were used to generate a database. The database was then searched using individual BGCs as query sequences using the MultiGeneBlast algorithm. 16S rRNA, rpsL and rpoB sequences were aligned using ClustalW 2.1 (Larkin et al., 2007). Blast analysis was carried out using ncbi-blast-2.2.31+ (Camacho et al., 2009).

The large callow workers of A. echinatior are covered in filamentous Pseudonocardia that are visible as a whitish covering on the cuticle (Figure 1) and can be cultured on nutrient agar. For a few weeks after hatching the filamentous bacterial growth can also be seen all over the worker ant cuticles and the rest of the body using scanning electron microscopy (Figure 1B; Poulsen et al., 2003). The A. echinatior workers typically have a concentration of Pseudonocardia on their laterocervical plates, so we scraped the plates of 1–2 workers taken from 23 separate A. echinatior colonies to isolate their mutualist strains. We consistently isolated either a single Ps1 or Ps2 Pseudonocardia strain from 22 out of the 23 ant colonies but for colony Ae_707, the only colony reared in Norwich rather than Copenhagen, we isolated both Ps1 and Ps2 strains. This result matched an earlier survey (Andersen et al., 2015), which had a single double infection in a sample of 19 colonies of Panamanian A. echinatior and A. octospinosus. In total, we isolated 24 strains of Pseudonocardia. We then selected five Ps1 strains and five Ps2 strains for genome sequencing (Table 1); these included six strains originating from colonies also used in the Andersen et al. (2015) study. Under laboratory culture conditions, Ps2 strains generally grew faster and generated more biomass than Ps1 strains, but all the Pseudonocardia isolates grew weakly compared to Streptomyces species and to Pseudonocardia strains isolated previously from A. octospinosus worker ants (Barke et al., 2010). Only one of the Pseudonocardia strains sequenced grew in liquid culture, and we could not conjugate any vectors into the strains.

Culture-dependent and independent studies have shown that Pseudonocardia dominate the cuticle of A. echinatior ants, but closely related Streptomyces albus strains have also been isolated from Acromyrmex ants collected from different locations in South America (Haeder et al., 2009; Sen et al., 2009; Barke et al., 2010). It is still not clear if these are mutualist strains, but they have been shown to produce antibiotics on the surface of worker ants using MALDI-TOF, which means they are growing and metabolically active on the ant cuticle (Schoenian et al., 2011). Domination of the ant cuticle by Pseudonocardia is facilitated by vertical transmission and inoculation of worker ants within 24 h of eclosing (Marsh et al., 2014). Subsequently, the Pseudonocardia blooms over the entire surface of the ant before eventually shrinking back to the laterocervical plate when large workers are ca. 6 weeks old and start leaving the nest to become foragers (Poulsen et al., 2003). One way in which the Pseudonocardia mutualist might inhibit other bacteria from gaining a foothold on the cuticle is by making antibacterial compounds in addition to the antifungals used to suppress Escovopsis (Barke et al., 2011; Scheuring and Yu, 2012). To test this hypothesis, we devised resistance bioassays in which Pseudonocardia strains were grown on sterile, porous cellophane disks placed on top of agar plates, allowing secondary metabolites produced by Pseudonocardia to permeate the agar.

After 7 days, the cellophane and Pseudonocardia lawns were removed, the agar was replenished with nutrients by applying Lennox broth, and the surfaces of the agar plates were dried before inoculation with test strains of unicellular bacteria or filamentous Streptomyces species. Some of these were isolated previously from fungus-growing ant nests whereas some were soil-isolated Streptomyces strains (Table 1). The cellophane disks allowed exchange of nutrients and secondary metabolites, including antibiotics that might inhibit the growth of other bacteria. Results showed that both Ps1 and Ps2 strains could partially inhibit the growth of Gram-negative and Gram-positive bacteria, but the Ps1 strains appeared to be more potent, at least under the conditions used here. This suggests that both strains are producing antibacterial compounds, and since we know that typically ≤25% of BGCs are expressed in vitro it seemed likely that in nature they would exhibit more potent activity against other strains, particularly since the genomes of all 10 sequenced Pseudonocardia strains encode multiple potential bacteriocins. Consistent with our predictions (Barke et al., 2011), only the unicellular bacterial strains were inhibited, whereas growth of the Streptomyces strains was unaffected or even enhanced compared with the control plate (Figure 2). This is consistent with the fact that Streptomyces strains carry multiple antibiotic resistance genes and can invade the ant cuticular microbiome (Wright, 2007; Barke et al., 2011).

For all Ps1 and Ps2 strains, their draft genomes consisted of one large contig representing the majority of the chromosome (Supplementary Table S1). The additional, smaller contigs per strain could represent either unassembled parts of the chromosome or extrachromosomal plasmids. Alignment of full length 16S rRNA and rpoB and rpsL genes (Supplementary Material) showed that the Ps1 and Ps2 strains are more similar within (99.93–100%) than between strains (97.61–97.74%) for 16S rRNA (Supplementary Table S2). For rpoB (Supplementary Table S4), Ps1 conservation was 99.74–100.00%, Ps2 conservation was 99.91–100.00%, and between Ps1 and Ps2 there was 94.96–95.10% identity. For the relatively short rpsL gene (Supplementary Table S3), both Ps1 and Ps2 conservation were 100% similar within each lineage, whereas between Ps1 and Ps2, mean similarity was 92.8%. We therefore conclude that the Ps1 and Ps2 phylotypes represent two different species of Pseudonocardia. Since the nystatin P1 producer (Ps1 phylotype) was first characterized in detail from the A. octospinosus system, we suggest the name Pseudonocardia octospinosus for Ps1 and Pseudonocardia echinatior for Ps2.

To determine which secondary metabolites are encoded by the Pseudonocardia mutualists of A. echinatior, we used antiSMASH 3.0 to identify the BGCs in all 10 sequenced genomes (Figure 3; Supplementary Figures S1–S10; Weber et al., 2015). Although, in some cases we know that antiSMASH has called the BGCs bigger than they really are, we have adopted a consistent approach of presenting all the data from the antiSMASH analysis because for most BGCs we have no idea where they start and end. AntiSMASH outputs indicated that the Ps1 strains encoded for the production of 14–15 secondary metabolites, while the Ps2 strains encoded 11–15 (Figure 3). In order to verify the similarity of the clusters between strains, we used MultiGeneBlast (Figures 4–6; Weber et al., 2015). Through shared homology and gene architecture, we deduced that six of the BGCs were shared between all 10 strains, seven are unique to Ps1, and five are unique to Ps2 (Table 2). The BGCs shared by all 10 Pseudonocardia strains (clusters A–F) include the ectABCD operon (Figure 4F, cluster F) which encodes biosynthesis of the osmoprotectants ectoine and 5′-hydroxyectoine which are common in both Gram-positive and Gram-negative bacteria and often found in actinomycetes (Galinski et al., 1985). There was a distinct bacteriocin BGC present in all Ps1 and Ps2 strains (Figure 4E, cluster E), and these proteinaceous molecules usually have antimicrobial activity, which can range from narrow to broad spectrum (Gross and Morell, 1971; Cotter et al., 2013). Encoded within the cluster is a peptide with a TIGR04222 domain predicted to be processed as a ribosomally encoded peptide, several nucleases and a sulfur transferase plus accessory protein that could modify the peptide backbone. Ps1 and Ps2 strains share two terpene encoding BGCs (Figures 4B,D, clusters B and D), hydrocarbons comprised of isoprene-derived units that have diverse bioactivities including antimicrobial activity (Gershenzon and Dudareva, 2007; Gallucci et al., 2009). One of these terpene BGCs (cluster B) has similarity to carotenoid BGCs, and encodes a polyprenyl synthetase and a phytoene synthase (Richter et al., 2015), while the other shared terpene BGC (cluster D) contains a terpene cyclase closely related to a lycopene cyclase found in carotenoid biosynthesis. This cluster also has similarity to a conserved cluster in Rhodococcus bacteria. The Ps1 and Ps2 strains also share an oligosaccharide cluster which encodes for enzymes associated with the biosynthesis of deoxysugars and glycosyltransferases, but for which we were unable to predict a product (Figure 4A, cluster A).

All five Ps1 strains share three non-ribosomal peptide synthetase (NRPS) BGCs that are not present in the Ps2 strains (clusters I, K, and M). We propose that these BGCs synthesize metal binding siderophore molecules, likely to sequester ferrous iron or other metal ions, particularly since these BGCs include transporters and other siderophore associated genes. The cluster for Ps1 siderophore 1 (Figure 5C, cluster I) is predicted to have the initial peptide arrangement of 2,3-DHB-ser-ala-(ala)-orn (second ala missing from Ae150A, Ae168, Ae263, and Ae356). Due to the predicted incorporation of a catechol unit (2,3-dihydroxybenzoic acid; 2,3-DHBA), a feature commonly found in siderophores and amino acids likely to interact with metal ions such as serine and ornithine, we propose a role in iron-acquisition for the secondary metabolite related to this cluster. The cluster for Ps1 siderophore 2 (Figure 5E, cluster K) has similarity with the erythrochelin BGC which was first identified in Saccharopolyspora erythraea (Lazos et al., 2010; Robbel et al., 2010). The cluster for Ps1 siderophore 2 has an NRPS that we predict generates a peptide arrangement of orn-ser-orn-orn, similar to the sequence for erythrochelin. However, there are a further 2–3 additional adenylation domains encoded by additional genes. Finally, the cluster for Ps1 siderophore 3 (Figure 5G, cluster M) we predict makes a tripeptide, with arrangement 2,3-DHBA then a polar residue either ser or thr, and lastly orn or arg. The last two amino acids we predict are epimerized to the D-configuration. We presume that iron is limited on ant cuticles, so siderophores would potentially play an important role for resource acquisition in these mutualist strains; efficient scavenging of iron and other metals by these species may also function to inhibit the growth of other microbes and therefore contribute to an anti-microbial phenotype. All Ps1 strains shared a novel cluster that was not present in any of the sequenced Ps2 strains (Figure 5A, cluster G), which contains several glycosyl transferases and enzymes capable of activating carboxylic acids (adenylation-like domains), and makes an unknown product. Ps1 strains also had an additional BGC (Figure 5B, cluster H) that we are uncertain makes a secondary metabolite and may be an artifact of the antiSMASH search algorithm. All the Ps1 strains encoded a further two BGC clusters that encode possible NRPSs. One cluster (Figure 5D, cluster J) encodes two adenylating enzymes, one of these shows similarity to an NRPS. This cluster is classified by antiSMASH as a bacteriocin cluster on account of containing a peptide that has similarity to the bacteriocin lincocin M18 produced by Brevibacterium linens (Valdés-Stauber and Scherer, 1994). The other BGC (Figure 5F, cluster L) that was maintained across all Ps1 strains has NRPS-like genes and possibly encodes an adenylation domain, a carrier protein and a cyclase. We predict it catalyzes synthesis of a dipeptide.

All five Ps2 strains share five BGCs that are not present in Ps1 strains (Table 2; Figure 6), including one bacteriocin BGC (Figure 6A, cluster N) which also has similarity to the bacteriocin lincocin M18 but with only 74–78% similarity to the gene in the Ps1 strains (cluster L). This cluster does not have the same surrounding genes, including the adenylating enzymes. All Ps2 strains have a BGC (Figure 6B, cluster O) that has two adenylating proteins that could act as an NRPS as well as several oxidative proteins. All Ps2 strains also contained a genomic island containing BGCs with Type 1 PKS (T1PKS) and NRPS genes (Figure 6C, cluster P). In Ae717_Ps2 and Ae505_Ps2, this genomic island was called as two separate and adjacent BGCs by antiSMASH, but they were called as one TIPKS-NRPS hybrid cluster in Ae331_Ps2, Ae406_Ps2, and Ae706_Ps2. It is likely that these are in fact three separate BGCs. In Ae717_Ps2 cluster 5 (cluster P, part 1) makes a pentapeptide (predicted sequence; ala-thr-orn-ser/thr-orn). Being rich in threonine and ornithine residues suggests the resulting peptide is likely to bind metal. Consistent with this, proteins immediately adjacent include a siderophore interacting protein member involved in metal acquisition/utilization. Ae717_Ps2 cluster 5 (cluster P, left) is separated from cluster 6 (cluster P, center/right) by apparent primary metabolism genes. Ae717_Ps2 cluster 6 can be split into two subclusters (cluster P, center/right); subcluster 6A (cluster P, center) is an NRPS-PKS hybrid consisting of modules for assembly of a glycine starter unit followed by five PKS extension steps. The PKS appears to have an unprecedented hybrid in cis and in trans acyltransferase architecture. Ae717_Ps2 subcluster 6A also has a type 1 glycosyl transferase and other deoxysugar genes associated with it. Subcluster 6B (cluster P, right) encodes three co-linear NRPS genes that likely make a hepta/hexa-peptide including a 2,3-DHBA unit followed by cys-ala-ser/thr-orn-(D-orn) residues (the second orn is absent from Ae331_Ps2, A406_Ps2, and Ae706_Ps2). This arrangement strongly suggests a metal binding peptide likely to act as a siderophore. Ps2 strains also had a unique terpene BGC (Figure 6D, cluster Q) with some similarity to the brasilicardin BGC. Brasilicardin is a diterpenoid molecule possessing immunosuppressive activity produced by Nocardia brasiliensis (Shigemori et al., 1998; Hayashi et al., 2008). The Ps2 cluster could make a molecule with a cyclic core and additional elaboration. The last Ps2-specific BGC is predicted to encode biosynthesis of a lassopeptide (Figure 6E, cluster R), a ribosomally encoded peptide that is post-translationally modified. Lassopeptides form a distinct topology where the N-terminus is bound covalently to an aspartate or glutamate side chain further back in the peptide. The C-terminal end of the molecule is then threaded through the ring forming the “lasso” structure. The Ps2 BGC encodes an asparagine synthase that could perform the peptide cyclization, and there is a GCN5 N-acetyltransferase that could perform the proteolytic action of peptide processing. We also found a small open reading frame upstream of the asparagine synthase that could provide the starting peptide molecule. Lassopeptides belong to a class of natural product known as ribosomally-synthesized and post-translationally-modified peptides (RiPPs) and have been recognized as an important class of molecules because their intrinsic protease resistance gives them great potential as scaffolds for drug design (Hegemann et al., 2015; Piscotta et al., 2015). This BGC is encoded on a fragment of the genome sequence that we predict is a plasmid; this lassopeptide BGC could therefore be passed around a population of Pseudonocardia strains in a functionally dependent manner in response to environmental circumstances.

In addition to the BGCs that were found in either Ps1, Ps2 or both there were several BGCs specific to individual strains or subsets of strains (Table 2; Supplementary Figure S11). BGC 10 in strain Ae707_Ps1 is unique (Supplementary Figure S11A, cluster S), and antiSMASH did not find similarity with any known BGCs. Inspection of this cluster finds an AMP-dependent synthetase/ligase that may function as an NRPS. Strain Ae717_Ps2 encoded a unique BGC (Supplementary Figure S11B, cluster T) with similarity to the BGC cluster for production of the galbonolides from Streptomyces galbus (Karki et al., 2010). Galbonolides are anti-fungal macrolactones made by an iterative T1PKS. The Ae717_Ps2 BGC encodes an appropriate iterative T1PKS and other associated genes to make a related anti-fungal macrolactone. Strain Ae717_Ps2 also contained an additional BGC encoding a T1PKS (Supplementary Figure S11C, cluster U), and only Ps2 strain Ae505_Ps2 shared this cluster. We predict the product of this BGC will be a highly unsaturated pentaketide derived from a 3-amino-5-hydroxybenzoic acid (AHBA) starter unit and four extension sites using malonyl-CoA as the substrate. This is highly likely to be modified by additional post-PKS genes encoded in the cluster. Ae717_Ps2 contained a unique T2PKS BGC that is highly likely to encode for a glycosylated T2PKS (Supplementary Figure S11D, cluster V). The DNA fragment encoding this T2PKS contains similarity to plasmid sequences so could be a recent addition to the genome of this strain. A BGC predicted to make a lantipeptide (Supplementary Figure S11E, cluster W) was found in a subset of the Ps1 strains (Ae150A_Ps1, Ae168_Ps1, Ae263_Ps1, and Ae356_Ps1). Lantipeptides are RiPPs with post-translational modifications in which amino acid residues are cross-linked via thioether bridges. The Ps1 lantibiotic clusters contain lantibiotic biosynthesis enzymes including the dehydratase and cyclase activities that are required to form lanthionine bridges (Karakas Sen et al., 1999; Li et al., 2006).

Perhaps the most intriguing feature of all 10 sequenced Pseudonocardia strains is the presence of nystatin-like biosynthesis gene clusters in their genome (Figure 4C). All Ps1 strains likely encode molecules similar to Nystatin P1 and NPP, previously reported as secondary metabolites from an A. octospinosus mutualist strain (Barke et al., 2010) and from P. autotrophica (Kim et al., 2009), respectively. As we obtained high quality PacBio sequenced genomes of Ae707_Ps1 and Ae717_Ps2, we used the nystatin-like clusters from each as templates for comparison in order to compare these BGCs across all 10 strains. The nystatin cluster from Ae707_Ps1, proves to be mostly identical to that encoded by the A. octospinosus mutualist strain we identified previously (Barke et al., 2010) (data not shown). However, for comparison we performed bioinformatics analysis (Supplementary Table S5) against the NPP cluster from P. autotrophica (Kim et al., 2009) and demonstrate a high level of similarity between the amino acid sequences. In contrast, the nystatin-like cluster from Ae717_Ps2 showed less overall conservation with the P. autotropica NPP cluster (Supplementary Table S5). This prompted us to perform a detailed analysis and comparison of the polyene BGC for both phylotypes.

We have analyzed the PKS architecture and predicted the assembled polyketide product of the nystatin cluster from Ae707_Ps1 (Figure 7). The polyene cluster has a total of 19 PKS modules and contains one ketoreductase domain (module 13) and two dehydratases (modules 17 and 18) which are predicted to be inactive due to a lack of the catalytic tyrosine (ketoreductase) and histidine (dehydrogenase) motifs, respectively. This is in conjunction with the biosynthetic cluster for NPP found in P. autotrophica (Kim et al., 2009). Analysis of the KR sequence motifs (Keatinge-Clay, 2007) revealed all stereocentres being consistent with the configuration in nystatin (Brautaset et al., 2000). The KR responsible for reduction leading to the hydroxyl group at position 19 shows an LDD motif (compared to the catalytically inactive LDA motif at this position in the original nystatin producer Streptomyces noursei). Due to its location outside of the LDD loop we consider it non-functional and therefore propose an R-configuration at this position in accordance with the reported configuration of nystatin. We previously documented the presence of a second glycosyl transferase (NypY) for addition of a second hexose unit to nystatin P1 (Barke et al., 2010). The NPP cluster encodes an additional glycosyl transferase (NppY) predicted to add N-acetyl-glucosamine (Kim et al., 2015). Accordingly, Ae707_Ps1 and all the Ps1 strains encode an additional glycosyl transferase. Heterologous expression of nypY in the amphotericin producer Streptomyces nodosus led to a new amphotericin derivative with D-mannose attached to the D-mycosamine via a β-1,4-linkage (Walmsley et al., 2016). We therefore predict the same modification leading to the product shown in Figure 7, and that this is identical to nystatin P1.

In contrast to Ps1, Ps2 strains encode a nystatin-like biosynthetic pathway that differs considerably from nystatin A1, NPP and nystatin P1 (Figure 8). First, the cluster seems to lack several modules compared to the BGC in Ae707_Ps1, including the loading module, the first two elongation modules, and a module responsible for the incorporation of a fully reduced acetate unit. This presumably results in a considerable structural re-arrangement of the polyketide scaffold and in a conjugated tetraene moiety in the southern part of the molecule. Instead of the loading KS^S, there is a CoA ligase (CAL) domain encoded in the cluster, potentially responsible for providing the starter unit. Involvement of CoA ligases in PKS assembly lines is rare and usually associated with the recruitment of aromatic starter units. The CAL domain is predicted to accept AHBA as its substrate, and consistent with this a gene belonging to the 3-amino-5-hydroxybenzoic acid synthase (AHBAS) family was found downstream of the PKS coding sequence. However, we cannot completely rule out the incorporation of a distinct but structurally related aromatic moiety into the polyene. Likewise, there appears to be a considerable re-arrangement in the later module architecture with the incorporation of an additional, fully reduced acetate unit (module 14) and the substitution of a fully reduced acetate unit with an ethylene moiety (module 12). These changes presumably lead to an altered cyclization pattern for the final product. Moreover, careful analysis of all KR sequence motifs suggests a significantly altered configuration at most of the predicted stereocenters compared to nystatin. Even though the KR in module 9 contains a VDD motif instead of LDD we suggest that catalytic activity remains after the replacement of leucine by valine, and propose the assembly of an S-configured hydroxyl group at position 17. We suggest a potential macrolactonization with the phenol group of AHBA for cleaving off the final polyketide chain but cannot rule out the formation of the alternative amide bond and macrolactamization instead. Another intriguing feature is that all polyene clusters found in the PS2 strains contain an additional methyltransferase and an O-methyltransferase, a feature rather uncommon for polyene antifungals. In contrast to the cluster in all Ps1 strains, the Ps2 polyene clusters do not encode a second glycosyltransferase and therefore potentially produce a mono-glycosylated product. Based on this analysis it is highly likely that Ps2 encodes a structurally distinct polyene macrolide with altered biological properties. Unfortunately, we have not been able to induce the expression of this silent BGC in any of the Ps2 strains so far.

To compare BGCs of the Pseudonocardia strains sequenced by Illumina we mapped the PKS amino acid sequences of either Ae707_Ps1 or Ae717_Ps2 to genome sequences of the other Ps strains using blastx (Supplementary Table S6) and closely inspected antismash outputs and multigene blast (Figure 4C). Despite all the Pseudonocardia strains containing genes for the biosynthesis of nystatin-like molecules, there is a disparity between the genomic arrangements of the nystatin-like clusters in the A. echinatior associated strains that we analyzed. An analagous evolutionary example has been observed with the variation in genomic arrangement of gene fragments of the hybrid NRPS/PKS BGCs for the gerumycins. Gerumycins are produced by Pseudonocardia mutualists of Apterostigma and Trachymyrmex ants, where the clusters are spatially separated, in some cases on plasmids (Sit et al., 2015). We suspect this represents that Pseudonocardia secondary metabolite gene clusters are under constant selective pressure to evolve new compounds. Ae707_Ps1 is the only Ps1 strain to contain all the PKS genes in a single BGC. The other Ps1 strains have clusters split into two different locations on the chromosome and are also missing two PKS modules (Supplementary Table S6) suggesting they make molecules with a smaller macrolide ring size. In contrast the Ps2 phylotype strains Ae331_Ps2, Ae406_Ps2, Ae706_Ps2 and Ae717_Ps2 all contain full intact clusters at a single locus on the chromosome. However, the Ae505_Ps2 BGC contains only half of the genes expected for the nystatin-like cluster and may encode a shorter polyketide molecule or may be inactive. The polyene PKS architectures were essentially identical except for the aforementioned Ae505_Ps2 example.