More than half of the known natural products with antimicrobial, antitumour or antiviral activity are of bacterial origin1. Most of these compounds were isolated from cultivated representatives of only five bacterial groups: filamentous actinomycetes, Myxobacteria, Cyanobacteria, and members of the genera Pseudomonas and Bacillus. Uncultivated bacteria, which are proposed to form 70% of all known prokaryotic phyla2, represent a particularly promising source for new, chemically prolific taxa. However, except for individual biosynthetic pathways reported from environmental sources3,4, the true metabolic potential of these microbes remains unexplored. Two such pathways, involved in the production of onnamide- and theopederin-type polyketides5 and ribosomal peptides of the polytheonamide group6 (Fig. 1), were previously discovered in the marine sponge Theonella swinhoei. Like many other sponges, this animal harbours a massive consortium of uncultivated bacteria belonging to hundreds of distinct phylotypes7,8,9. T. swinhoei is the source of exceptionally diverse natural products and forms distinct chemotypes; samples of the sponge collected from different locations have largely non-overlapping metabolite profiles. From the onnamide and polytheonamide chemotype occurring at Hachijo Jima, Japan, here termed T. swinhoei Y (Y referring to the yellow interior of the sponge), in total more than 40 bioactive polyketides and modified peptides belonging to seven structural classes were isolated (Fig. 1)10. As previous work on onnamides and polytheonamides has produced only metagenomic DNA fragments lacking taxonomically diagnostic features, it was unknown which members of the bacterial community are the producers of these compounds.

Figure 1: Representative bioactive natural product families isolated from the sponge Theonella swinhoei.
figure 1

Polytheonamides A and B differ in the stereochemistry of the sulphoxide moiety (polytheonamide A shows S chirality; polytheonamide B shows R chirality).

PowerPoint slide

Attribution of metabolic genes to ‘Entotheonella

Single-cell analysis has recently emerged as an efficient strategy to correlate the phylogenetic identity of environmental microorganisms with their functional gene repertoire11,12,13. To pinpoint producers in T. swinhoei Y, samples enriched in bacteria of different cell densities were prepared by differential centrifugation after sponge collection. When a fraction of higher density (Fig. 2a) was microscopically examined, we found that it contained a highly enriched population of large filamentous bacteria that fluoresce when excited with ultraviolet light (Fig. 2b). The bacteria were morphologically similar to the symbiont ‘Candidatus Entotheonella palauensis’ previously reported from a Palauan Theonella swinhoei chemotype and suspected as producer of antifungal peptides14,15. Scanning electron micrographs (Fig. 2c) revealed the presence of approximately 2- to 3-μm cells linked to each other. These bacteria, as well as those from the low-density fraction containing mostly unicellular bacteria, were sorted individually into 96-well plates by fluorescence-assisted cell sorting (FACS) (Extended Data Fig. 1a), resulting in filamentous and unicellular plates. Subsequently, multiple displacement amplification (MDA) of single bacterial genomes was performed on each well, resulting in DNA product sizes of approximately 10 kb (Extended Data Fig. 1b).

Figure 2: Single-cell analytic studies.
figure 2

a, Differential interference contrast micrograph of the filamentous fraction after differential centrifugation (n = 3). b, Fluorescence micrograph of filamentous bacteria without (top) and with (bottom) ultraviolet excitation (n = 3). c, Scanning electron micrograph of a single filamentous bacterium (n = 3). d, Nested PCR of natural product gene clusters from whole-genome amplification samples of wells sorted with single filaments (n = 48). Wells showing positive amplification for ‘Entotheonella’ sp. 16S rRNA gene, onnamide (onn) and polytheonamide (poy) gene clusters (n = 3) were used for the identification of the other enzyme clusters (n = 1). Each lane represents a single well defined by the well identifier above the top row. cth, cyclotheonamide; ker, keramamide; kon, konbamide; naz, nazumamide. PC, metagenomic DNA from filamentous fraction; pts, unknown proteusin.

PowerPoint slide

To detect wells containing DNA from the onnamide or polytheonamide producer, primers specific for onn and poy genes encoding the respective pathways were used in diagnostic PCRs. For both gene clusters, a large number of positive wells were detected among the filamentous plates (Fig. 2d and Extended Data Fig. 1c). Subsequent PCRs with eubacterial and ‘Entotheonella’-specific 16S ribosomal RNA gene primers showed that about half of the wells contained DNA originating from ‘Entotheonella’ phylotypes. Overall, from 48 wells of an analysed filamentous plate, 22 wells were positive for the onnamide, 34 for the polytheonamide, and 27 for the ‘Entotheonella’ sp. 16S rRNA gene, as confirmed by sequencing of each amplicon. Sixteen of the positive wells showed amplification for all three of the onn, poy and ‘Entotheonella’ sp. primer pairs in one or more out of three repetitive PCRs (Fig. 2d). For further analysis, wells positive for all three primer sets were subjected to PCR using eubacterial 16S rRNA gene primers. 16S rRNA genes from ‘Entotheonella’ sp. as well as Escherichia coli were identified. The E. coli amplicon was discarded, as it was also identified in MDA-treated wells that only contained water. Thus, the data suggested ‘Entotheonella’ as the source of both the onnamide-type compounds and polytheonamides.

Two chemically distinct ‘Entotheonella’ symbionts

As not all wells were positive for all three primer pairs and bacteria might have been overlooked owing to incomplete genome coverage during MDA16, we wished to validate further our results by metagenomic sequencing. On the filamentous bacterial cell sample, several rounds of Illumina, 454, PacBio, and Sanger sequencing were performed (Supplementary Table 1). Of the sequencing reads, 78.3% assembled to longer contigs, resulting in 18,093 contigs of at least 500 bp. The remaining reads did not show significant overlap, suggesting that the corresponding phylotypes were present only at low concentration. This hypothesis was backed by the observation of a high variance in coverage, ranging between 3.3- and 1,564.7-fold for contigs of at least 2 kb length. Basic Local Alignment Search Tool X (BLASTX) analysis of the contig and scaffold sequences followed by binning based on sequence depth and G + C content revealed two large populations of bacterial DNA with a G + C content around 55% (Supplementary Table 2). A third set of low coverage and low G + C contigs delivered hits against various eukaryotic genomes and was therefore excluded from further analyses (Extended Data Fig. 2a). A more detailed analysis of the filtered data set revealed for most bacterial genes the existence of two highly similar versions (approximately 85–91% nucleotide identity) that resided in virtually syntenous genomic environments encompassing over 4.5 Mb (Extended Data Fig. 3). The overlapping genomic regions included exactly two orthologues of 35 single-copy genes often used as bacterial phylogenic markers (Supplementary Table 3)17. These features suggested that the large majority of assembled bacterial sequences belonged to two closely related ‘Entotheonella’ variants, termed TSY1 and TSY2, with 97.6% identical 16S rRNA gene sequences and an average G + C content of 55.79% (Extended Data Fig. 2b). The identity of the 16S rRNA genes to that of ‘E. palauensis’ was about 97%. Depth analysis also suggested the presence of about 236 kb of DNA belonging to at least one large plasmid (G + C content: 55.11%). Coverage was 60.3-, 24.5- and 278.5-fold for the TSY1 and TSY2 chromosomes and the plasmid(s), respectively (corresponding to a ratio of 1:0.4:5), indicating that TSY1 is the dominant strain (Extended Data Fig. 2b). Both strains possess genomes of similar size that exceed 9 Mb, thus belonging to the largest known prokaryotic genomes (Supplementary Table 2). A remarkably large number of repetitive elements, some present in about 25 to 100 copies, as well as the high degree of similarity of the two genomes prohibited further assembly. To determine completeness of genomes, a core gene group analysis18 was performed, identifying 62 of 66 core groups for both TSY1 and TSY2. Thus we assume that the protein inventory of both strains was almost completely established.

The search for metabolic genes in this data set revealed complete sets of onn and poy genes on the plasmid-derived contigs. In addition, an unexpectedly high number of further gene clusters for polyketide and ribosomal or non-ribosomal peptide biosynthesis were identified on the chromosomal sequences. To allow for prediction of the corresponding metabolites, sequence gaps within most of these loci were filled by paired-end sequencing of 3- and 8-kb libraries and by combinatorial or targeted PCR, resulting in at least 28 biosynthetic gene clusters on 31 scaffolds (Extended Data Fig. 4 and Supplementary Table 4). For many non-ribosomal peptide synthetase (NRPS) clusters, bioinformatic predictions based on enzyme colinearity rules19, substrate recognition motifs20,21,22, and the presence of genes for non-proteinogenic amino acid biosynthesis (Supplementary Table 5), revealed known bioactive peptides from Japanese T. swinhoei as the best structural hits. Specifically, we identified virtually perfect matches for the cyclotheonamides, keramamides and nazumamide A. In addition, we identified a konbamide A-type23 cluster in which five of the six NRPS modules are present and colinear with the compound structure, but two ORF insertions disrupt the NRPS architecture, suggesting that the cluster is an inactive evolutionary relic. Consistent with this, members of the onnamide, polytheonamide, keramamide, and cyclotheonamide compound families were detected using high-resolution mass spectroscopy (HRMS) in extracts of our sponge specimens and enriched filamentous cell fractions, but we were unable to detect the konbamides (Supplementary Tables 6 and 7, and Extended Data Fig. 5). Taking together the combined bioinformatic and chemical analyses, candidate gene clusters existed for all known peptide and polyketide families including onnamides and polytheonamides, except for the aurantosides. In addition to these attributable genes, loci for at least 14 peptides of unknown identity were found (Extended Data Fig. 4). Notably, this also includes four further gene clusters for proteusins, a recently discovered new natural product family with polytheonamides as the only members known to date6,24. Tandom mass spectrometry (MS–MS)-based molecular networking25 suggested a high diversity of previously unknown metabolite families, indicating that at least some of these orphan pathways are likely to be active (Extended Data Fig. 6). The gene candidates for konbamides (kon), keramamides (ker), nazumamide A (naz), and an unknown non-ribosomal peptide formed a supercluster of 129 kb. The binning data suggested that this region, the putative cyclotheonamide (cth), and the two unassigned proteusin loci all belong to the chromosome of the dominant ‘Entotheonella’ sp. TSY1 (Extended Data Fig. 2b). The chromosome of TSY2 contained fewer (at least seven) metabolic gene clusters (two polyketide, at least two NRPS, and a further proteusin cluster) that could not be assigned to known compounds. Except for a small NRPS and a type III polyketide synthase (PKS) system present in both genomes, there was no overlap in the natural product gene repertoires of TSY1 and TSY2, indicating that significant chemical variation exists among members of ‘Entotheonella’, even within the same sponge individual. To validate further the source of the plasmid-based polytheonamide and onnamide genes, we conducted additional single-cell experiments (Fig. 2d). All MDA samples previously analysed positive for onn and poy genes were tested again with PCR primers for various genes of the kon, naz, cth, ker and one unknown proteusin pathway. For all cases, positive wells were identified, suggesting that TSY1 carries the plasmid and produces the entire set of metabolites.

Functional evidence for the identity of ‘Entotheonella’ gene clusters was obtained by biochemically characterizing gene products from several pathways. Two selected NRPS adenylation domains encoded within the putative cth and ker pathways were overproduced in E. coli and analysed using a γ-18O4-ATP pyrophosphate exchange assay26 to investigate their amino acid substrate specificity (Extended Data Fig. 7). For the cth NRPS, the adenylation domain of module 2 (CthA2) exhibited high selectivity for the rare amino acid 2,3-diaminopropionate (DAP), consistent with the cyclotheonamide structure (Extended Data Fig. 7). The incorporation of this building block is also supported by the presence of two genes in the cluster that encode homologues of SbnAB-type DAP synthases27. KerA5 showed greatest substrate specificity for Leu, in agreement with known keramamides and the bioinformatic prediction (Extended Data Fig. 7). Thus, taking the colinearity rule of NRPSs into account, the data support the proposed function of these gene clusters.

We also obtained functional support for a biosynthetic role of the unknown proteusin pathway TSY1_14 by co-expressing the putative nitrile-hydratase-like precursor peptide with a predicted lanthionine synthetase encoded directly adjacent to the precursor gene. Up to three dehydrations of the core peptide were observed by HRMS for the co-expression product compared to the unmodified peptide from expression of the precursor peptide alone. Subsequent alkylation of reduced cysteine residues and tandem MS–MS indicated for each dehydration, one lanthionine bridge was formed within the predicted core peptide (Extended Data Fig. 8 and Supplementary Table 8). These experiments demonstrated that the putative proteusin gene cluster TSY1_14 encodes a functional precursor peptide and modifying lanthionine synthetase. Considering the high complexity of the sponge microbiome, which contains hundreds of ribotypes, the accumulation of metabolic genes in two variants of ‘Entotheonella’ is remarkable. Owing to the extraordinary biosynthetic repertoire, we propose the name ‘Candidatus Entotheonella factor’ (latin, factor; the producer) for these bacteria.

Entotheonella’ species are ubiquitous

These findings raised the question whether ‘Entotheonella’ spp. also inhabit other sponges and could have a general role in natural product biosynthesis. It was previously shown that an enriched fraction of ‘E. palauensis’ from a Palauan chemotype of T. swinhoei contained high concentrations of the hybrid polyketide-peptide theopalauamide14,15. ‘Entotheonella’ members were also detected in another lithistid sponge, Discodermia dissoluta, that contains the anticancer polyketide discodermolide28. To analyse the distribution of ‘Entotheonella’ spp. in depth, 37 taxonomically diverse sponge species collected at 20 locations (Supplementary Table 9) were tested by PCR based on conserved, unique regions of ‘Entotheonella’ 16S rRNA genes. Of the 37 sponges, 28 yielded amplicons with sequences exhibiting 95.5–99.9% nucleotide identity to the homologues of ‘E. factor’ (Extended Data Fig. 9a, b). Thus, ‘Entotheonella’ spp. seem to be widely distributed in marine sponges from distant geographical regions. ‘Entotheonella’ amplicons were also obtained from various seawater samples; however, contamination from sponges growing nearby cannot be excluded. For further insights into the discovery potential and chemical variability of these bacteria, we initiated studies on another chemotype of T. swinhoei (type W1, referring to the white sponge interior) that contains the actin inhibitor misakinolide A (Extended Data Fig. 10b), a complex polyketide not present in the Y chemotype. PCR detection of PKS genes using the total sponge DNA generated exclusively amplicons that were phylogenetically attributed to the sup type (Extended Data Fig. 10c), a putative fatty acid synthase that is widespread and dominant in most sponge microbial consortia and not involved in the production of complex, bioactive polyketides29. In contrast, a highly enriched ‘Entotheonella’ fraction (Extended Data Fig. 10a) prepared from this sponge yielded a completely different set of amplicons consisting of six gene fragments all belonging to PKSs associated with complex polyketide production (Extended Data Fig. 10c). None of these had a close homologue in TSY1 or TSY2, thus further supporting a diverse chemistry of ‘Entotheonella’ phylotypes.

A new candidate phylum, ‘Tectomicrobia’

To obtain insights into the taxonomic position of ‘Entotheonella’, an initial 16S rRNA-based phylogenetic analysis was conducted (Extended Data Fig. 9c). Altogether, 243 16S rRNA gene sequences were analysed from marine sponges in this study and from public databases. As the 16S rRNA sequences were only 82% identical to representatives from known bacterial phyla and form a well-separated clade, we suggest the status of a new candidate phylum30. The name ‘Tectomicrobia’ (latin, tegere; to hide, to protect) was chosen to reflect their uncultured status as well as the capability to produce bioactive compounds that are likely used as chemical defence. The closest relatives to ‘Tectomicrobia’ are Nitrospina spp., which were recently proposed to belong to a new phylum, Nitrospinae31. The known sequences belonging to ‘Tectomicrobia’ comprise at least three discrete phylogenetic clades. The largest encompasses all ‘Entotheonella’ sequences sensu stricto, which were largely recovered from marine sponges but also seawater (138 sequences total, of which 107 sequences were produced in this study), a second clade includes related, non-‘Entotheonella’ 16S rRNA gene sequences from various marine sponges (36 sequences), and a third group contains 16S rRNA gene sequences from terrestrial soils (18 sequences). For further validation of the phylogenetic data, we calculated trees using up to 38 concatenated, universally conserved single-copy marker proteins17 of TSY1, TSY2, and 2,509 bacterial and archaeal taxa to determine the position of ‘Entotheonella’ in the tree of life. Recalculations with data sets from closely affiliated phyla (Fig. 3) supported ‘Entotheonella’ as belonging to a new sister phylum to Nitrospinae, in agreement with the 16S rRNA data.

Figure 3: Phylogenetic inference of the ‘Tectomicrobia’ and affiliated phyla.
figure 3

RAxML inference of 991 taxa with 100 bootstrap iterations based on up to 38 marker genes. Sequences are collapsed on the phylum level and the number of collapsed sequences is shown for each clade. The two ‘Tectomicrobia’ variants TSY1 and TSY2 are highlighted in bold. Bootstrap support values of equal or greater than 70% are shown for each node. The scale bar represents 10% estimated sequence divergence. PV-1 and 3-11 are strain names; OP8 is the former name of the (then candidate) phylum Aminicenantes.

PowerPoint slide


Owing to the high frequency of structurally distinct, bioactive metabolites in sponges, these animals have an important role in drug discovery. Compound localization studies suggested Bacteria as producers of individual metabolites14,15,32,33, but remained ambiguous owing to the possibility of sequestration or transport. The true source of sponge natural products has therefore been a long-standing and, with the exception of metagenomic data providing kingdom-level information5,6,34, unanswered question. Here we provide evidence that a single member of the highly diverse microbiome of T. swinhoei Y, ‘E. factor TSY1’, is the source of almost all polyketides and peptides that have been isolated from its sponge host. The bioinformatic assignment to known compounds is further supported by functional studies for polytheonamides6, onnamide-type compounds35,36, keramamides, cyclotheonamides and an orphan proteusin. Our data on TSY1, TSY2, and a highly enriched ‘Entotheonella’ preparation from a second T. swinhoei chemotype, indicate that members of this candidate genus contain producers with a rich and, so far, unique secondary metabolism. Reports on ‘Entotheonella’ spp. from two other chemically rich sponges15,28,37 and our detection of these bacteria in many additional species hint at their more widespread role in the chemistry of their hosts. This study adds the first uncultivated prokaryotes to the taxonomically limited canon of metabolically talented bacteria. ‘Entotheonella’ spp. exhibit interesting parallels to streptomycetes and some other well-known producer groups38,39,40,41,42; for example, expanded genome size, biosynthetic superclusters43 and multiple modular assembly lines, high metabolic variability among closely related organisms, and complex morphology. For ‘Entotheonella’ spp., complex morphology is particularly noteworthy, as it affords attractive opportunities to systematically study chemical interactions in marine symbioses and to exploit uncultivated bacteria in a targeted way for drug discovery.

Methods Summary

An adapted differential centrifugation protocol14 was used to sediment filamentous and unicellular bacteria from the sponge tissue. Single bacteria cells and filaments were sorted into micro-titre plates by flow cytometry with a BD FACSAria II cell sorter (BD Biosciences). Genomic DNA was amplified using an Illustra Genomiphi V2 DNA Amplification Kit (GE Healthcare) and subjected to PCR analysis. Sequence information was obtained using the GS-FLX (454) and MiSeq (Illumina) platforms, using whole-genome sequencing and long mate-pair libraries. Additional sequence reads were obtained by PacBio sequencing (GATC) and Sanger sequencing (IIT). Reads were assembled using the Newbler (v2.6) de novo assembler. Automated annotation was performed with Rapid Annotation and Subsystem Technology (RAST)44 and validated manually. PKS and NRPS domain architecture and substrate specificities were based on sequence alignments and prediction-based software22,45,46. Adenylation domains overexpressed in E. coli were characterized using a γ-18O4-ATP pyrophosphate exchange assay as previously described26. The TSY1_14 proteusin precursor peptide was overexpressed in E. coli with and without the putative modifying LanM-like lanthionine synthetase from the same gene cluster. The resulting peptide products were analysed by liquid chromatography (LC)–electrospray ionization (ESI)–HRMS after TCEP (tris-(2-carboxyethyl)-phosphine) treatment, tryptic digest and derivatization. Extracts of T. swinhoei and enriched ‘Entotheonella’ were analysed by ultra-performance liquid chromatography (UPLC) and nano-LC heated ESI (HESI)–HRMS followed by eMZed47 data analysis and molecular networking25.

Online Methods

Sponge collection

A list of analysed sponges and their collection sites is provided in Supplementary Table 6. Specimens were placed separately into plastic bags and brought to the surface. Immediately after collection, sponge tissues were cut into pieces and stored at −80 °C in a transportable liquid nitrogen freezer (Bahamas collection) or in 70% aqueous ethanol.

Isolation of bacteria

To prepare enriched bacterial fractions, a protocol adapted from a previous paper was used14. From freshly collected T. swinhoei, the thin red ectosome layer of a 500-g sponge piece was removed. The remaining portion was cut into smaller pieces, cleaned of other animals and processed in a National MJ-C28 juicer. Liquids and the solid residues were transferred into a 2-l graduated cylinder, and the volume was adjusted to 1.5 l with Ca- and Mg-free artificial sea water (CMF ASW)48. The mixture was incubated at room temperature for 15 min, while stirring every 2 min to dissociate sponge cells, then left undisturbed for additional 10 min to allow settling of particles. Supernatants were decanted into another graduated cylinder, and CMF ASW was added to the residues to a final volume of 1.5 l, followed by a second dissociation and settling period. The collected supernatants were passed through a 32-μm Nitex mesh and centrifuged at 1000g for 10 min to pellet filamentous cells. The supernatants were subsequently centrifuged at 4500g to pellet unicellular bacteria. Each bacterial fraction was washed once with 200 ml CMF ASW, pelleted again, resuspended in 200 ml CMF ASW and stored at 4 °C.

Flow-cytometric analysis and cell sorting

Prior to sorting, both the pellet and supernatant fraction were analysed by flow cytometry using the BD FACSAria II cell sorter (BD Bioscience). Size distributions of the bacteria within both the fractions were determined using size calibration beads (Life Technologies) based on the following sizes: 1 μm, 2 μm, 4 μm, 6 μm, 10 μm and 15 μm (Life Technologies). Samples used for sorting were diluted accordingly and size distribution was analysed at 500–1,000 events per sec. Flow-cytometry results were analysed using the FACSDiva software (BD Bioscience). Sterile 96-well plates, containing 1 μl of nuclease-free ultrapure water in each well were prepared and single cells or multicellular filaments were sorted accordingly. Confirmation of the successful cell sorting was conducted by observation of each drop under a fluorescence microscope. The resulting 96-well plates were stored at 4 °C for subsequent whole-genome amplification (WGA).

Whole-genome amplification

Single isolated bacterial cells or filaments attained from the cell sorting were disrupted by heat treatment at 95 °C, 3 min, and WGA was conducted based on the phi29 polymerase-mediated multiple displacement amplification (MDA) technique using the Illustra Genomiphi V2 DNA Amplification Kit (GE Healthcare). Each WGA reaction per well was optimized and conducted at 10-μl volumes as recommended by the manufacturer. Upon MDA, each well was diluted tenfold with nuclease-free ultrapure water before storage.

Detection of biosynthetic and 16S rRNA genes

For the detection of 16S rRNA genes and the biosynthetic gene clusters (onn, poy, kon, naz, cth, ker and pts), nested PCR was performed against wells containing MDA amplified genomic DNA from single-filament bacteria. PCR was conducted using the high-fidelity PrimeSTAR Max DNA Polymerase (TaKaRa) (first amplification) and the AmpliTaq Gold 360 Master Mix (Applied biosystems) (second amplification) based on the following conditions. First amplification: 98 °C, 5 min (initial denaturation); 98 °C, 10 s; 54 °C, 15 s; 72 °C, 1.5 min (for 35 cycles); 72 °C, 7 min (final extension). Second amplification: 95 °C, 10 min (initial denaturation); 98 °C, 30 s; 59 °C, 30 s; 72 °C, 30 s (for 30 cycles); 72 °C, 7 min (final extension). For the first amplification, 1 μl of template from the tenfold-diluted MDA samples was used, and 0.5 μl was used directly from the first PCR amplification product for the second amplification. All PCRs were conducted at 25 μl final volume. The primer sets used in the nested PCRs are summarized in Supplementary Table 10. Amplicon sizes were determined by agarose gel electrophoresis and confirmed by Sanger sequencing. Nested PCR was performed in triplicate for the ‘Entotheonella’ 16S rRNA gene, onn and poy gene clusters, each on different days. For the other biosynthetic clusters, nested PCR was only conducted once. To test for the presence of contaminating bacteria, the 16S rRNA gene was amplified using the 16SU27F and 16SU1492R primers from MDA-amplified wells containing or not containing single-filament bacteria, cloned, and 20 clones were randomly selected for sequencing.

Genome sequencing and annotation

Sequencing was performed on two platforms: first, a GS-FLX (454) using a whole-genome shotgun and a 3-kb-long paired-end library, both prepared according to the manufacturer’s instructions and sequenced using the Titanium chemistry; second, a MiSeq (Illumina) using a Nextera (Epicentre Biotechnologies) whole-genome shotgun paired-end and a 8-kb mate-pair library. The latter was prepared using a hybrid protocol by replacing the 454 adapters with Illumina paired-end adapters in the final steps of an 8-kb-long paired-end library construction. The resulting library fragments were size selected by gel electrophoresis and excision to a size of 400 bp. The library was sequenced in a 2× 151-bp paired-end run, and a 2× 251-bp run was performed for the 8-kb mate-pair library. Prior to assembly, the read pairs of the 8-kb mate-pair library were joined, excluding all reads without a perfect match in the overlapping region. The joined pairs were split at the 454 long paired end linker, excluding all reads without a perfect match and the two resulting reads were reverse complemented to simulate large insert Sanger reads. In addition, sequence information was obtained by PacBio sequencing (GATC) and Sanger sequencing (IIT). All reads were then assembled using the Newbler (v2.6) de novo assembler, non-GS-FLX reads were provided in FASTQ format, assembly was performed with default parameters except using 30 bp as minimum overlap match. Contig numbers for the TSY1 and the TSY2 genome are 1,774 and 3,270, respectively. Synteny analysis was performed with r2cat49. Automated annotation was performed with Rapid Annotation and Subsystem Technology (RAST)44. Manual identification and annotation of natural product biosynthetic gene clusters were based on similarity searches (BLAST) with validated biosynthetic genes as queries. Additional automated identification of natural product gene clusters was performed with antibiotics and Secondary Metabolite Analysis SHell (antiSMASH)45. Preliminary PKS and NRPS domain architecture and adenylation domain specificities were determined using freely available software22,45,46. All manual annotation and routine bioinformatic analysis was performed using Geneious version 6.0.6 created by Biomatters (available from Scaffold gaps were closed using PCR amplification with Phusion High-Fidelity or LongAmp Taq DNA polymerase (New England Biolabs) and sequencing (GATC Biotech).

PCR detection of ‘Entotheonella’ spp

DNA was isolated from sponge samples with the Fast DNA spin kit for soil (Q-Biogene) according to the manufacturer’s protocol. PCR amplification was performed with the eubacterial 16S rRNA gene specific primers 16SU27F and 16SU1492R as published previously50. The resulting PCR product was used as template in a following nested PCR using two different procedures. First, newly designed ‘Entotheonella’ 16S rRNA gene-specific primers Ento271F and Ento1290R. Conditions: 1 µl DMSO was added to 50 µl of PCR mix. An initial denaturing step for 2 min at 95 °C followed by 35 cycles of a denaturing step at 95 °C for 30 s, primer annealing at 63 °C for 30 s and elongation at 72 °C for 1 min. The program was completed with a final elongation step at 72 °C for 5 min. DreamTaq Polymerase (Fermentas) was used. Second, ‘Entotheonella’-specific primers Ento238F (ref. 15) and Ento1442R. Conditions: an initial denaturing step for 2 min at 98 °C followed by 35 cycles of a denaturing step at 95 °C for 10 s, primer annealing at 55 °C for 30 s and elongation at 72 °C for 1.5 min. TaKaRa Ex Taq polymerase was used. Different primer pairs were used to increase the diversity of detectable ‘Entotheonella’ 16S rRNA genes. The PCR products were purified, ligated into pGEM-T (Promega) and transformed into heat-competent E. coli Novablue cells. Using the vector primers SP6 and T7, colony PCRs on 20 clones of each sponge were performed. Double restriction digestion of these PCR products was performed using the enzymes HaeIII and MspI, and the insert of one representative per RFLP (restriction fragment length polymorphism) pattern was sequenced.

Detection and analysis of PKS genes in the misakinolide chemotype of T. swinhoei

Metagenomic DNA was prepared from Theonella swinhoei W1 as described previously34. Crude DNA was purified further by electrophoresis on low-melting-point agarose followed by gel extraction using the peqGOLD Gel Extraction Kit (PEQLAB). The filamentous cell fraction was prepared from T. swinhoei W1 as described above. Ketosynthase fragments were amplified from total sponge DNA and filamentous cells using the primers KSDPQQF and KSHGTGR51. Approximately 0.4 µl of the purified sponge metagenomic DNA or 0.5 µl of the rinsed cell pellet suspension was used as PCR template in a 25 µl PCR mixture that also contained 2.5 µl of 10× thermopol buffer (New England Biolabs), 0.5 µl of 10 mM dNTPs, 1 µl of 25 mM MgCl2, 2 µl each of 50 mM KSDPQQF and KSHGTGR primer, and 0.25 µl High Fidelity Hot Start Polymerase (Jena Bioscience GmbH) or 0.125 µl Taq DNA Polymerase High Fidelity (Invitrogen). The thermal cycle program was set up at 35 cycles and consisted of the following steps: lid heating at 105 °C, predenaturation at 95 °C for 2 min, denaturation at 95 °C for 1 min, annealing at different temperatures (55, 58, 61 °C), elongation at 74 °C for 1 min, and final elongation at 74 °C for 10 min. Subsequently the PCR products with the desired size (approximately 700 bp) were gel-purified and ligated into pBlueScript II (SK-) (Stratagene). Plasmids harbouring 700-bp inserts were sequenced.

Phylogenetic analysis

16S rRNA gene sequences and reference sequences identified by initial BLAST52 searches were automatically aligned to a SILVA reference alignment using the SINA Webaligner53 and merged into the SILVA version 106 database54,55. The alignment was then manually refined. A maximum parsimony tree with 1,000 bootstrap resamplings and a maximum likelihood tree with 100 bootstrap resamplings were calculated in ARB using long (≥1,200 bp) sequences only. Short sequences were added using the parsimony interactive tool in ARB without changing the tree topology.

For the whole-genome trees, both ‘Entotheonella’ variants TSY1 and TSY2 were scanned for homologues of a set of 38 universally conserved single-copy proteins present in Bacteria and Archaea17. The assemblies were translated into all six reading frames, and marker genes were detected and aligned with hmmsearch and hmmalign included in the HMMER3 package56 using HMM profiles obtained from phylosift ( Extracted marker protein sequences were used to build concatenated alignments of up 38 markers per genome. Phylogenetic inference methods used were the maximum likelihood based FastTree2 (ref. 57) and a custom RAxML bootstrap script originally provided by Christian Goll and Alexandros Stamatakis (Scientific Computing Group, Heidelberg Institute for Theoretical Studies) and modified by D. Jacobsen. The script requires two input files, the alignment file as PHYLIP format and a starting tree calculated by RAxML-Light58. The script workflow can be briefly summarized as follows. First RAxML version 7.3.5 (ref. 59) creates bootstrap replicates of the multiple sequence alignments and stepwise addition order parsimony trees as starting points for the maximum likelihood search, based on user defined rate heterogeneity and substitution models. Next RAxML-Light58 is run on every bootstrap replicate. After all RAxML-Light runs are finished the resulting replicate trees are fed into RAxML to calculate the bootstrap support values which are drawn upon the starting tree. The major advantage of this approach over simply running RAxML to sequentially perform the bootstrapping calculation is that multiple RAxML-Light instances can be used to evaluate several bootstrap replicates in parallel. In addition, the RAxML-Light implementation is both faster and more efficient than RAxML for large-scale phylogenetic inferences since it was specifically developed for use in high performance computing environments and has an efficient checkpointing and restart capability58. The rate heterogeneity and amino acid evolution models used were GAMMA and Le-Gascuel (LG) for the custom RAxML bootstrap script, and CAT approximation with 20 rate categories and Jones–Taylor–Thornton (JJT) for FastTree2. We first generated a phylogenetic tree with FastTree2 including 2,509 bacterial and archaeal taxa to verify the position of the ‘Entotheonella’ variants in the tree of life. Next we calculated phylogenies (Fasttree2, RAxML script) using a reduced data set of 991 taxa of closely affiliated phyla, including Proteobacteria, Acidobacteria, Nitrospirae, Deferribacteres, Chrysiogenetes, and the recently proposed phyla Aminincenantes and Nitrospinae17,31.

Adenylation-domain assays

Adenylation-domain regions of the ker (KerA5) and cth (CthA2) NRPS (non-ribosomal peptide synthetase) genes were PCR amplified from genomic DNA isolated from the T. swinhoei Y filamentous fraction using Phusion High-Fidelity DNA polymerase (New England Biolabs). Primers kerA5F and kerA5R were used for domain KerA5 and primers cthA2F and cthA2R for domain CthA2. The PCR products were ligated into the vectors pGEM-T Easy or pBluescript SK(+) and transformed into E. coli DH5α for subcloning into pET28b. Based on recent findings that many A domains are only active in vitro when coexpressed with a MbtH-like protein60,61, each of the His-tagged adenylation domains were coexpressed with KerL, a MbtH-like protein from the ker pathway. KerL was cloned into the co-expression vector pCDF-DUET. For overexpression of KerA5, 1-l cultures of E. coli BL21(DE3) containing the expression plasmids were grown in Terrific Broth medium with kanamycin (50 μg ml−1) and streptomycin (50 μg ml−1) selection to a D600 nm of 1.8 at 37 °C and 250 r.p.m. The cultures were then cooled to 16 °C before being induced with IPTG (Isopropyl β-D-1-thiogalactopyranoside; 1 mM) and grown for 20 h at 16 °C and 250 r.p.m. Overexpression of soluble CthA2 was greatly improved by transforming the expression plasmids into E. coli Bl21-A1 and growing the clones as above except a final concentration of 0.2% arabinose was added at when the D600 nm reached 0.4. The protein-purification and mass-exchange-based adenylation assays were performed as reported previously26,62,63.

UPLC high-resolution HESI–MS analysis of sponge and bacterial fractions

Whole sponge samples of Theonella swinhoei and enriched ‘Entotheonella’ cell fractions were extracted with ethanol, methanol, propanol and acetone. Whole sponge extracts and ‘Entotheonella’ extracts were subjected to ultra-performance liquid chromatography (UPLC) heated electrospray ionization (HESI)–high-resolution mass spectrometry (HRMS) and nano-LC HESI–HRMS analysis, followed by eMZed47 data analysis. HESI–HRMS data were collected on a Thermo Q Exactive coupled to a Dionex Ultimate 3000 UPLC system. For the standard analysis of ‘Entotheonella’ and Theonella extracts, solvent gradients (A = H20 + 0.1% formic acid, and B = acetonitrile + 0.1% formic acid with B at 5% for 0–2 min, 5–95% for 2–14 min and 95% for 14–17 min at a flowrate of 0.5 ml min−1) were used on a Phenomenex Kinetex 2.6 µm C18 100 Å (150 × 4.6 mm) column at 27 °C (A) to 30 °C (B). The MS was operated in positive ionization mode at a scan range of (A) 200–2,500 m/z (mass-to-charge ratio) or (B) 100–1,600 m/z, respectively, and a resolution of 70,000 or 140,000 at m/z 200. The spray voltage was set to 3.7 kV and the capillary temperature to 320 °C. For the identification of polytheonamides with the UPLC setup an isocratic elution with 45% n-propanol at a flow rate of 0.5 ml min−1 was used on a Phenomenex Kinetex 2.6 µm C18 100 Å (150 × 4.6 mm) column at 45 °C. The MS was operated in positive ionization mode at a scan range from 1,000–6,000 m/z and a resolution of 140,000 at m/z 200, the spray voltage was set to 3.7 kV and capillary temperature to 320 °C. For the detection of heterologously produced and TCEP (tris-(2-carboxyethyl)-phosphine)-treated proteusins, a solvent gradient (A = H20 + 0.1% formic acid, and B = acetonitrile + 0.1% formic acid with B at 5% for 0–2 min, 5–95% for 2–14 min and 95% for 14–17 min at a flowrate of 0.5 ml min−1) was used on a Phenomenex Aeris WIDEPORE 3.6 µm C4 200 Å column (50 × 2.1 mm) at 27 °C. The MS was operated in positive ionization mode at a scan range from 600–2,000 m/z and a resolution of 70,000 at m/z 200, the spray voltage was set to 3.7 kV and capillary temperature to 320 °C. MS experiments for iodoacetamide treated samples were adjusted to a scan range from 150–2,000 m/z. The Manual Xtract function of Thermo Protein Deconvolution software version was used to obtain protein masses from the spectra, using charge states 5–50 (1–50 in tryptic digest) for mass calculation, thresholds were set to 3 for signal to noise, 3 for the minimum number of detected charge states (2 in tryptic digest), 0% for relative abundance, 25% for overlap remainder, and 1 for minimum intensity. The fit factor for isotopic pattern comparison was set to 80% and the expected intensity error to 3. MS–MS experiments on tryptic digested samples were carried out for a mass range of 400-1800 m/z at a stepwise normalized collision energy (NCE) of 24.5, 35, and 45.5. Targeted MS–MS was applied on the [M+H] ions [1051.8]3+, [1057.8]3+, [1076.54]3+, [1105.73]4+, [1110.0]4+, [1129.0]4+, [1148.0]4+, [1474.0]3+, [1480.0]3+, [1505.0]3+ and [1530.0]3+ within an isolation window of 2 m/z at a resolution of 70,000 at m/z 200. Peptide masses for MS–MS fragments were manually calculated from the observed [M+H]+ ions ([M+2H]2+ for the y37 ion). In addition, a Thermo Easy-nLC 1000 equipped with a Acclaim PepMap RSLC nano Viper column (2 µm, C18, 100 Å (15 cm × 50 µm) coupled to a Thermo Q Exactive was used increase detection sensitivity for trace compounds. The nano-LC was operated using the following solvent gradient: A = H2O + 0.1% formic acid, and B = 1-propanol + 0.1% formic acid with B at 10–80% for 0–60 min and 100% for 61–95 min at a flowrate of 0.3 ml min−1. For the identification of secondary metabolites produced by ‘Entotheonella’, as indicated by the presence of the respective gene cluster, Thermo raw files were converted into mzXML files using MSExport and further processed with the Python-based, open-source eMZed framework47. The processed data were compared to a list of known Theonella compounds and purified onnamide, cyclotheonamide and polytheonamide standards. To generate a mass-spectral molecular network25, combined data sets of data dependent nano-LC and UPLC high-resolution HESI–MS–MS experiments were used. The chromatographic separation was conducted using the aforementioned UPLC and nano-LC conditions. The top 10 most intense ions of each parent ion scan were subsequently fragmented with a resolution of 17,500 at m/z 200 and a normalized collision energy (NCE) of 35. The isolation width was set to 5 Da, the dynamic exclusion to 15 s, and the default charge state to 4 to enable fragmentation of high molecular mass secondary metabolites (for example, polytheonamides). Identical spectra were merged to form consensus spectra using MSCluster64. Cosine values were calculated for every possible pair of spectra (spectral alignment) and spectra obtained from solvent controls were removed using MATLAB scripts25,65. The cosine value threshold was set to 0.5, whereas a cosine value of one indicates identical spectra and a cosine value of 0 displays no spectral correlation. The resulting network was visualized in cytoscape66. Consensus spectra were annotated by comparison with a database of published secondary metabolites from T. swinhoei.

Functional studies on the putative proteusin gene cluster TSY1_14

The putative nitrile hydratase-like precursor gene from gene cluster TSY1_14 was PCR-amplified from genomic DNA from the T. swinhoei Y filamentous fraction using Phusion High-Fidelity DNA polymerase (New England Biolabs) and cloned into pET28b (Merck) using primers Prec-TSY1_14-F and Prec-TSY1_14-R (Supplementary Table 10). The resulting construct was digested with NcoI/HindIII and ligated into pETDuetI (Merck), producing the N-His-tagged precursor peptide expression construct pMH124. The putative LanM-like lanthionine synthetase gene from TSY1_14 was amplified using primers Lanth-TSY1_14-F and Lanth-TSY1_14-R and cloned into pCDFDuetI to obtain the untagged modifying enzyme expression construct pMH104. For functional analysis, pMH124 was transformed into E. coli BL21(DE3) alone and co-transformed with pMH104. Expression cultures (100 ml) of Terrific Broth were inoculated with 1 ml of overnight culture and incubated at 16 °C at 250 r.p.m. for 5 days after induction with IPTG. Cells were collected by centrifugation (3,220g for 10 min) and pellets frozen in liquid nitrogen. The cells were lysed by sonication in lysis buffer (50 mM potassium phosphate, pH 8.0, 300 mM NaCl, 20 mM imidazole, 10% (v/v) glycerol) and the supernatant was incubated with 300 µl Protino Ni-NTA resin (Macherey Nagel) for 1 h at 4 °C on a rocking platform. Resin was pelleted at 850g (20 min, 4°C) and applied to a Poly-Prep Chromatography column (Bio-Rad), washed with 5 ml lysis buffer, 5 ml wash buffer (50 mM potassium phosphate, pH 8.0, 300 mM NaCl, 40 mM imidazole, 10% (v/v) glycerol), and eluted with three 500-µl fractions of elution buffer (50 mM potassium phosphate, pH 8.0, 300 mM NaCl, 250 mM imidazole, 10% (v/v) glycerol). The elution fractions were adjusted to a protein concentration of 152 µM (2 µg µl−1 as measured by Roti-Nanoquant modified Bradford assay (Carl Roth)) and Tris(2-carboxyethyl)phosphine (TCEP) hydrochloride (Sigma Life Science, BioUltra) was added in 100-fold molar excess to reduce disulphide bond formation. After incubation at 25 °C for 30 min, the samples were subjected to UPLC HESI–HRMS analysis as described above. To determine the number of free thiols and putative lanthionine bridges the control and modified peptides were derivatized with iodoacetamide. In brief, the peptides were desalted (Vivaspin, 5 kDa MWCO, Sartorius) with 10 volumes of 100 mM ammonium bicarbonate (pH 7.86) and adjusted to 1 µg µl−1 (76 µM). After treatment with 9 mM TCEP and 0.09% SDS (25 min at 25 °C), the samples were treated with 16 mM Iodoacetamide (Sigma Life Science, BioUltra) in the absence of light for 30 min at 25 °C and analysed by UPLC HESI-HRMS. Sequencing grade trypsin (Promega) was added to the remainder at a trypsin:protein ratio of 1:25 and incubated for 2 h at 37 °C before HRMS analysis.