Metabolic and evolutionary origin of actin-binding polyketides from diverse organisms

Abstract

Actin-targeting macrolides comprise a large, structurally diverse group of cytotoxins isolated from remarkably dissimilar micro- and macroorganisms. In spite of their disparate origins and structures, many of these compounds bind actin at the same site and exhibit intriguing structural relationships reminiscent of modular, combinatorial drug libraries. Here we investigate biosynthesis and evolution of three compound groups, misakinolides/swinholides, tolytoxin/scytophycins, and luminaolides. For misakinolides from the sponge Theonella swinhoei WA, our data suggest production by an uncultivated 'Entotheonella' symbiont, further supporting the relevance of these bacteria as sources of bioactive polyketides and peptides in sponges. Insights into misakinolide biosynthesis permitted targeted genome mining for other members, providing a cyanobacterial luminaolide producer as the first cultivated source for this dimeric compound family. The data indicate that this polyketide family is bacteria-derived and that the unusual macrolide diversity is the result of combinatorial pathway modularity for some compounds and of convergent evolution for others.

The intriguing structural and pharmacological features of actin-targeting macrolides¹ (Fig. 1) have attracted much attention among chemists and pharmacologists. Although chemically diverse, many of these cytotoxic compounds, such as swinholide A (1), misakinolide A (= bistheonellide A) (2), lobophorolide (3), tolytoxin (4), rhizopodin (6), and kabiramide C (7), exhibit similar actin-binding characteristics. Binding occurs at the same site of the barbed end of F-actin (the filamentous, polymerized form) and the corresponding region of G-actin (the globular monomer). While the cyclic polyketide core moieties bind to a hydrophobic patch on the actin surface, an exocyclic “tail” region intercalates into a narrow cleft that normally accepts the DNase I-binding loop of the neighbouring actin monomer during filament growth. This insertion results in a steric clash that causes filament disruption, capping, and/or sequestration of G-actin, depending on the toxin.

Figure 1.

Selected structures of structurally related macrolides and their biological sources. With the exception of luminaolide, all compounds have been shown to target actin. The colored regions denote shared structural features.

A remarkable feature of these compounds is their complex structural relationship, in which various large moieties are shared by substances (colored regions in Fig. 1), resulting in a diverse spectrum of molecular scaffolds with retention of overall activity. For example, the cryptic relationship between the completely dissimilar 1 and 7 becomes only obvious when compared with 2-6. Such interconnections are to our knowledge unparalleled among reported complex polyketides and evoke structural libraries generated by combinatorial synthesis. The relationship is particularly intriguing in light of the highly diverse organisms that are sources of these compounds, i.e., various bacteria^2–4 and a broad array of marine organisms, including sponges,^5–7 various mollusks,^8–10 a coral,¹¹ and brown¹² and red algae.¹³ It has therefore been speculated that the macroorganisms might not be the actual producers.¹⁴

We have previously provided evidence that almost all of the many bioactive polyketides and modified peptides in the marine sponge Theonella swinhoei (chemotype Y) are of bacterial origin^{15, 16} and produced by a single as-yet uncultured member of the diverse symbiont community, 'Candidatus Entotheonella factor' TSY1.¹⁷ 'E. factor' belongs to the new candidate phylum 'Tectomicrobia', members of which are widespread in sponges.¹⁷ In T. swinhoei Y, genomes of two closely related 'Entotheonella' phylotypes exhibit large differences in their biosynthetic gene repertoire. These data, along with previous compound localization studies on a Palauan T. swinhoei chemotype,^{18, 19} suggest that 'Entotheonella' members are metabolically highly variable and could be of more general biosynthetic importance in sponges. To investigate origins of other sponge-derived compounds, we selected T. swinhoei chemotype WA with a natural product profile that is distinct from that of the previously studied sponge. This sponge occurs at the same site as T. swinhoei Y and also harbors 'Entotheonella', but exhibits a distinct metabolic profile by containing the actin inhibitor misakinolide A (2) as the only known polyketide.

Here we provide insights into the biosynthesis of 2 by a bacterium that several lines of evidence suggest to be a member of 'Entotheonella'. We then used knowledge on the misakinolide polyketide synthase to mine cyanobacterial genomes for two additional biosynthetic routes of actin inhibitors, resulting in the discovery of a free-living bacterial source of macroalga-derived luminaolides. Finally, the combined data permitted an analysis of evolutionary mechanisms that underlie the unusual diversity of actin-targeting polyketides.

Results

Isolation of the putative misakinolide gene cluster

Complex bacterial polyketides are known to be products of two alternative variants of modular polyketide synthases (PKSs), cis- and trans-acyltransferase (AT) PKSs.^{20, 21} Both types comprise giant, multidomain enzymes that either contain (cis-AT) or lack (trans-AT) integrated AT domains. Other characteristic features of trans-AT systems are a much higher architectural diversity than found for cis-AT systems,²¹ a propensity to evolve in a mosaic-like fashion by extensive horizontal gene transfer,²² and their widespread presence in symbiotic bacteria.^{15, 23–27} The latter feature suggested a trans-AT PKS as candidate for the misakinolide pathway.

From misakinolide-containing T. swinhoei WA collected at Hachijo Island, Japan, we used a previously prepared enriched fraction of filamentous bacteria exhibiting the characteristic morphology of 'Entotheonella'.¹⁷ With degenerate PCR primers targeting PKS genes, this fraction consistently yields gene fragments that phylogenetically cluster with the trans-AT PKS group. In contrast, similar analyses with the total sponge DNA provides a large number of amplicons belonging to other PKS types, suggesting that the trans-AT PKS genes are associated with the filamentous fraction. To obtain information about the biosynthetic product of the trans-AT PKS(s), we initially attempted to construct a fosmid library from the enriched filamentous fraction. However, due to significant DNA instability problems, we instead prepared a 350,000 member metagenomic fosmid library using total sponge DNA, which we iteratively screened²⁸ to isolate a 99 kb region containing a large trans-AT PKS cluster, termed mis cluster, of ca. 90 kb length (Fig. 2, Supplementary Fig. 1, and Supplementary Table 1). Outside of the cluster, the region contained several transposase genes and gene fragments as well as other non-PKS gene candidates. Notably, on both sides of the mis locus we identified extended regions with genes that exhibited >80% nucleotide identity to the previously sequenced genomes¹⁷ of the 'Entotheonella' variants TSY1 and TSY2 (Supplementary Fig. 1 and Supplementary Table 1). In agreement, testing the filamentous fraction with 16S rRNA gene primers generated an amplicon that exhibited 98.3% identity to the 16S rRNA gene of 'E. factor TSY1' (Supplementary Fig. 2a). This sequence accounted for ca. 70% of the 16S reads obtained by pyrosequencing, suggesting the dominant presence of an 'Entotheonella' phylotype, designated here as TSWA1. For more conclusive data we subjected individual filaments isolated by fluorescence-activated cell sorting (FACS) to multiple displacement amplification (MDA).²⁹ PCR analysis of the amplified DNA revealed the presence of mis genes and the TSWA1 16S rRNA gene in several single-filament preparations (Supplementary Fig. 2b). Independent sequencing of the amplified DNA from two filaments (G6 and H6) generated in both cases a dataset that covered almost the entire isolated mis locus (Supplementary Fig. 2c). Both datasets contained a single 16S rRNA gene that was >99% identical to the previously amplified TSWA1 gene. For the filament G6, analysis of 139 genes that occur as single copy in 90% of bacterial genomes showed that only 6 of these were present in more than one copy in the MDA data (Supplementary Table 2). These numbers suggest that the sample contained little, if any, DNA from organisms other than 'Entotheonella'. For more insights into the taxonomy of TSWA1, we performed average nucleotide identity (ANI)³⁰ analyses using genomic data of TSY1, TSY2, and the two TSWA1 filaments. Considering a species boundary at 95-96%,³⁰ the returned values (Supplementary Table 3) suggest that TSY1, TSY2, and TSWA1 belong to three different 'Entotheonella' candidate species. The TSWA1 16S rRNA gene also exhibited 96% identity to that of 'Entotheonella palauensis', a previously reported symbiont of another T. swinhoei chemotype without known genome sequence.¹⁷ In light of these data, we propose to maintain 'E. factor' for TSY1 and to assign the new names 'Entotheonella gemina' to TSY2 and 'Entotheonella serta' to TSWA1.

Figure 2.

Unifying model for the biosynthesis of the actin-inhibitors misakinolide A, tolytoxin, and luminaolides. a, Misakinolide biosynthetic gene cluster. b, Scytophycin and tolytoxin gene cluster. c, Luminaolide B gene cluster. The 10 kb scale bar refers to all three clusters. d, Architecture of the PKS systems and proposed macrolide biosynthesis. For luminaolides, the stereochemistry is uncertain, and the PKS is only known for luminaolide B. The first ring closure in misakinolide biosynthesis is hypothetical and might alternatively occur after polyketide assembly. Circles represent the enzymatic domains. A, adenylation domain; AT, acyltransferase; DH, dehydratase; DH*, dehydratase variant with aberrant active site motif; ER, enoylreductase; FT, formyltransferase; KR, ketoreductase; KS, ketosynthase; KS⁰, nonelongating KS; MT, C-methyltransferase; OMT, O-methyltransferase; PS, pyran synthase; TE, thioesterase. Solid black circled represent acyl carrier protein domains. The grey AT domain is predicted to be non-functional. Gaps between domains denote protein boundaries, dashed lines connect domains within the same protein. Black triangles point to split modules belonging to two distinct proteins for the PKS variants indicated by the labels.

We also investigated the location of misakinolide A and 'Entotheonella' cells by matrix-assisted laser desorption/ionization imaging mass spectrometry (MALDI-IMS). MALDI-IMS is a mass spectrometry technique that has recently been applied to the study of bacterial chemical interactions³¹ and allows one to spatially localize (resolution ca. 50 µm) molecules of interest. We conducted IMS at high mass accuracy (60,000) to thin sections of the misakinolide-containing T. swinhoei. This analysis unexpectedly revealed a highly localized distribution of the polyketide in thin tissue zones lining pores, chambers, and the exterior (Fig. 3), while control IMS experiments on the sponge chemotype T. swinhoei Y that also harbours 'Entotheonella' but lacks mis genes¹⁷ did not result in detection of misakinolide (Supplementary Fig. 3). For the misakinolide chemotype WA, fluorescence in situ hybridization (CARD-FISH) using 'Entotheonella'-specific probes showed a distribution of the filaments at the same confined locations. Collectively, these data point to 'Entotheonella serta' TSWA1 as the origin of the mis locus.

Figure 3.

Co-localization of 'Entotheonella' and misakinolide A using a combination of CARD-FISH and high resolution imaging mass spectrometry (HR-IMS). a, localization of 'Entotheonella' by CARD-FISH. Brightfield image showing representative pores from a thin section of T. swinhoei WA (left). CARD-FISH results showing 'Entotheonella' to be localized around pores of the sponge tissue (middle). Overlay of the brightfield image with the fluorescent image obtained from CARD-FISH labeling of 'Entotheonella' filaments (right). b, localization of misakinolide in the sponge tissue. Representative optical image of a T. swinhoei WA thin section used for MALDI-IMS prior to matrix application (left). False color heat map representation of the spatial distribution of misakinolide A within the sponge tissue. (middle). Overlay of the optical image and the spatial distribution of misakinolide A (right). c, Magnified section of the north-eastern part of the thin section in b. Magnified section of the north eastern part of the optical image (left). Magnified section of the north eastern part of the false color heat map representation of the spatial distribution of misakinolide A (middle). The image shows misakinolide A to be located around pores of the sponge tissue (overlapping with the observations in a). Overlay representation of the magnified optical image of the sponge tissue section and the spatial localization of misakinolide A as shown by MALDI-IMS (right).

The mis cluster contains four large core PKS genes misC-F and the AT gene misG as candidates for polyketide biosynthesis (Fig. 2a, Supplementary Table 1). An additional type III PKS gene adjacent to this locus likely belongs to a different biosynthetic gene cluster, since almost identical genes are also present in 'Entotheonella' phylotypes TSY1 and TSY2, which lack the mis genes. We have previously reported a method to predict biosynthetic products of trans-AT PKSs by examining the phylogeny of ketosynthase (KS) domains,²² which catalyze polyketide chain elongation. This approach, validated for diverse polyketides,^{26, 32–35} is based on a on correlation of KS phylogeny with the structure of the incoming substrates close to the thioester terminus.^{22, 32} On this basis, prediction of the mis PKS product returned an almost perfect match for swinholide/misakinolide-type compounds (Fig. 2d, Supplementary Fig. 4-7, Supplementary Table 4). Likewise, an analysis of a diagnostic motif present in ketoreductase (KR) domains, which allows for the prediction of OH-bearing stereocenters and double bond geometries,^36–38 was in perfect agreement with the stereochemistry of these compounds (Supplementary Table 5). Additional diagnostic features for swinholides/misakinolides include two O-methyltransferase (OMT) domains corresponding to the two methoxy groups of the polyketide monomers and a predicted pyran synthase (PS) domain³⁹ predicted to form the dihydropyran ring. Thus, the in silico data were consistent with a role in a swinholide-type pathway. An unexpected deviation from PKS colinearity was the presence of a terminal PKS module in MisF that corresponded to the extended swinholide structure, although the sponge was only known to contain misakinolide A lacking this extension. We re-analyzed extracts of the sponge and the filamentous fraction by liquid chromatography-high-resolution mass spectrometry (LC-HRMS), but could detect only misakinolide A and closely related, non-expanded congeners (Supplementary Fig. 8). It is therefore likely that the last module is skipped in biosynthesis.

The mis PS domain catalyzes pyran ring formation

To obtain functional insights into the identity of the mis cluster, we expressed in E. coli the region of the PKS gene misF covering the putative PS domain. After heterologous production of the mis domain as a His-tagged protein, we incubated the purified enzyme with the synthetic test substrate 8 (Fig. 4). Extraction of the reaction mixture and HPLC-MS analysis revealed the formation of two new compounds, 9 and 10. Comparison with synthetic standards and NMR analysis of enzymatic products (Supplementary Fig. 9) identified them as syn- and anti-configured tetrahydropyrans. In control mixtures containing the substrate alone or in combination with boiled enzyme, these products were only present in trace amounts, likely due to spontaneous cyclization. The observed in vitro formation of the cyclic ether system by the PS suggested an analogous function within the mis PKS module, which is located at a position that matches that of the dihydropyran ring in misakinolide.

Figure 4.

HPLC profiles of test reactions investigating PS-catalyzed tetrahydropyran formation. i, Synthetic susbtrate 8. ii, Synthetic standard 9. iii, Assay mixture containing 8 and the expressed mis PS. iv, Assay mixture containing 8 and the boiled PS. Test substrates were racemic.

Identification of a cyanobacterial tolytoxin pathway

After identification of the mis PKS, we detected in our database of KS sequences a series of entries that showed high similarity to KSs of the mis PKS and all originate from the same cyanobacterium, Scytonema sp. PCC 10023 isolated from a subaerial habitat in Bermuda (raw genomic data of an ongoing genome sequencing study). The close relationship between PKSs from such diverse sources warranted a search for the polyketide product. Filling of sequence gaps resulted in two non-clustered loci harbouring trans-AT PKS genes (Fig. 2b, Supplementary Table 1). The deduced architecture of the cyanobacterial (tto) PKS showed considerable similarity to the mis assembly line. With the exception of the initial 4-5 modules, both PKSs exhibited a virtually indistinguishable domain organization and identical predicted KS specificity (Fig. 2d, Supplementary Fig. 4-7, Supplementary Table 1), indicating that a large portion of the tto polyketide product is closely related to misakinolides/swinholides. A KS specificity analysis²² (Supplementary Table 4) predicted the compound to belong to the scytophycin/tolytoxin group of cyanobacterial actin inhibitors. Since these compounds had previously not been reported from PCC 10023, we analysed cultures to search for scytophycin-type compounds. LC-HRMS analysis of crude extracts indeed revealed the presence of a compound with MS properties matching to tolytoxin (m/z 850.5264, [M+H]⁺, Δ -4.7 mmu) (Supplementary Fig. 10). Purification of this substance and comparison of its NMR spectra with the published data of tolytoxin⁴⁰ confirmed their identity (Supplementary Fig. 11-12). More detailed MS analysis also suggested the additional presence of low amounts of the congeners 6-hydroxy scytophycin B (m/z 818.5032, [M+H-H₂O]⁺, Δ −1.7 mmu, Supplementary Fig. 13) and 6-hydroxy-7-O-methyl scytophycin E (m/z 834.5338, [M+H-H₂O]⁺, Δ −2.4 mmu, Supplementary Fig. 14).

Identification of a cyanobacterial luminaolide producer

Having obtained more detailed knowledge on two pathways, we searched for further producers of related compounds. In an ongoing genome study of another cyanobacterium, Planktothrix paucivesiculata PCC 9631, isolated from a river in Paris, we detected several PKS gene contigs that phylogenetically clustered with the mis and tto systems. Gap filling yielded a ca. 99 kb cluster, termed lum (Fig. 2c). The architecture of the new PKS (Fig. 2d) closely resembled that of the scytophycin/tolytoxin PKS with the exception of two terminal modules that differed from a three-module tto region. In agreement with this architecture, a structural prediction resulted in a scytophycin-like polyketide that carries a terminal 3,5-diol moiety linked to the dihydropyran system instead of the 2,4-dien-7-ol present in scytophycins. Using this information in a substructure search of the AntiMarin database, the dimeric macrolide luminaolide was found as only hit (5, Fig. 2d). This result was intriguing, since 5 is only known from a marine coralline alga as a metamorphosis enhancer of coral larvae.¹³ To identify the polyketide, we subjected small culture extracts of PCC 9631 to LC-HRMS analysis, which revealed a compound with a similar HPLC retention time to and identical mass as an authentic luminaolide standard (Supplementary Fig. 15). For more conclusive structural information, we established a larger 8 L culture of the slow-growing strain, which permitted the isolation of 0.3 mg of the metabolite (11). Comparison of its ¹H NMR spectrum with that of authentic luminaolide suggested that the compounds were not identical but very close structural isomers (Supplementary Fig. 16). After more detailed analysis of 2D NMR data (Supplementary Fig. 17-20, Supplementary Table 6), we mapped the difference to an alternately positioned methyl group in each monomer of 11, being located at C-18 rather than the C-21 oxygen position of luminaolide (R₂ and R₁, respectively, in Fig. 2d). The compound was named luminaolide B.

Evolution of actin-inhibiting polyketides

The highly unusual combinatorial-like structural properties of misakinolide, tolytoxin, luminaolide and other actin inhibitors raised the question of how this substance family had emerged. The isolation of three pathways in this study provided an opportunity to address this question. In addition, published biosynthetic genes exist for another member, the myxobacterial rhizopodin (compound 7, Fig. 1),³⁴ which shares part of the “tail” portion with tolytoxin and luminaolide, but features a different macrocycle. To test whether diversity resulted from exchanges of large PKS-encoding DNA regions, we inferred individual phylogenetic trees for DH, KR, MT, OMT, and TE domains in addition to the KS trees, comprising in total 135 domains from the four PKSs (Supplementary Fig. 21-26, summarization in Fig. 5). The data showed that for almost all examined domains of the tto system, the next phylogenetic neighbors were lum domains at analogous positions in the PKS. The exception was a region close to the tto PKS terminus (Fig. 5, region B) that is absent in the lum assembly line and corresponds to a missing portion of the luminaolide structure. For misakinolide, which exhibits a tolytoxin- and luminaolide-like macrocyclic portion but differs in the exocyclic region, the domain series matching to the shared macrolide moiety featured a similar neighboring pattern. In contrast, the remaining domains involved in tail biosynthesis appeared at tree positions distant from those of the tto and lum PKSs. These data are consistent with a scenario in which the misakinolide, scytophycin/tolytoxin, and luminaolide pathways arose from a common ancestor and diversified by deletion or acquisition of DNA regions encoding either a large upstream PKS portion (Fig. 5, region A, distinguishing mis from tto/lum) or a region close to the PKS terminus (region B, distinguishing mis/tto from lum). Comparison of the domain trees localized the recombination points at or shortly before the ER-KS boundary for the misakinolide/tolytoxin switch (region A) and at KS-DH and DH-KR boundaries for tolytoxin/luminaolide (region B). Within the subclades comprising exclusively mis, tto, and lum domains, the mis orthologs diverged at more basal positions in 19 of 23 cases. This topology suggests that the common macrolide core of misakinolide A, swinholide A, and tolytoxin/scytophycins is an ancestral feature and that the most recent event was the evolution of luminaolide biosynthesis from a scytophycin-like route by loss of five contiguous domains in PKS region B.

Figure 5.

Evolutive relationships of the rhizopodin (riz), luminaolide (lum), scytophycin/tolytoxin (tto), and misakinolide (mis) PKS systems. The figure summarizes phylogenetic data of Supplementary Fig. 22-27. Numbers refer to modules carrying KSs and the corresponding domain numbering in the phylogenetic analysis (e.g., riz DH5 is the DH domain in module 5). Colors symbolize the relationship between PKS components: Blue PKS portions are closely related and likely have a common evolutionary origin. Thick green lines connect direct, late-branching neighbors in phylogenetic trees, thinner blue lines connect early-branching neighbors, black asterisks denote domains at isolated tree positions within the same clade, red asterisks label singleton KS domains in separate specificity clades. Unlabeled domains were not included in the analysis. Grey dashed lines denoted possible insertion/deletion events. Regions A and B in the mis PKS and corresponding regions in the tto and lum PKSs were likely modified by recombination at the sites marked by the red triangles.

Unexpectedly, the rhizopodin PKS exhibits an entirely different phylogenetic pattern although most of the polyketide tail moiety is identical to those of scytophycins, tolytoxin, and luminaolides. All 31 examined riz domains were well-separated from the mis, tto, and lum homologs. An exception is riz KS1 from a specificity clade that contains only six members. In all cases, connecting nodes within trees are deeply positioned close to the root of major clades, with two of the six KSs belonging to shared polyketide structures (KS2 and 5) even being placed as singletons in specificity clades (Supplementary Fig. 5 and 6). Thus, in spite of great similarities at the polyketide level, the rhizopodin enzymes are not closely related to the other PKSs.

Discussion

Barbed-end actin binders exhibit several features that are to date unique among complex polyketides. They occur in many unrelated organisms and exhibit a degree of structural modularity unmatched by other known polyketides, resulting in more than 80 different compounds belonging to 8 skeletal types (swinholides, misakinolide A, scytophycins, luminaolide, aplyronines, reidispongiolides, rhizopodin, and kabiramide-like trisoxazoles). To contribute to an understanding of this intriguing metabolic puzzle, we first examined the true source of an animal-derived compound, misakinolide A from the microbially highly complex sponge T. swinhoei WA. All lines of evidence provided by metagenomic gene isolation, bioinformatic and biochemical PKS analysis, single-filament sequencing, and MS imaging suggest that this macrolide is produced by a member of the uncultivated candidate genus 'Entotheonella'. Interestingly, the misakinolide-like polyketides swinholide A (1) and congeners are also known from field collections of marine cyanobacteria.⁴ Moreover, a previous extraction-based study demonstrated that swinholide A titers in a Palauan 'Entotheonella'-containing sponge were highest in a microbial fraction comprising unicellular bacteria, while levels were low in the 'Entotheonella', cyanobacterial, and sponge cell fractions.¹⁸ These seemingly contradicting observations could be the result of horizontal gene transfer, of the presence of other producers in bacterial bulk samples, or in the case of the sponge study to polyketide absorption or sequestration by other cell types.

The mis PKS belongs to the trans-AT PKS family, which has recently been recognized as one of the major biosynthetic systems in bacterial specialized metabolism.^{21, 41} Architecturally, it closely follows the trans-AT colinearity rules²² with the exception of an apparently skipped last module and the as-yet unknown formation of the tail-associated THP ring. Biosynthesis of this THP unit is still unclear, since unlike for the dihydropyran system in the macrolide portion, the corresponding PKS module lacks a PS domain.³⁹ The module harbors an aberrant DH domain that might be involved in cyclization.⁴² An alternative possibility is ring formation by accessory enzymes, such as the putative kinase MisA that could activate a hydroxyl function to facilitate elimination or substitution. Misakinolide is among the few sponge polyketides,^{15, 26, 43} for which biosynthetic genes have been isolated. The insight that bacteria produce this and many other compounds isolated from sponges provide interesting opportunities to create renewable production systems.

Our study also shows that knowledge on biosynthetic pathways can reveal culturable sources for rare compounds that are macroorganism-derived. For two misakinolide-like trans-AT PKS clusters detected in cyanobacterial genomes, structural prediction and targeted genome mining resulted in the isolation of their polyketide products: tolytoxin previously described from other cyanobacterial strains⁴⁰ and, more interestingly, luminaolide B, a close congener of a dimeric polyketide only reported from a coralline alga.¹³ Further supporting misakinolide as product of mis PKS, both compounds share large moieties with this polyketide, but exhibit different tail regions. These structural features are perfectly mirrored at the PKS level, where mainly the modules encoding tail biosynthesis exhibit a divergent architecture. Additional correlations exist regarding distinct macrolide regions (lack of domains in the luminaolide PKS), O-methylation (presence/absence of free-standing MT genes), and the oxygenation pattern (presence of the cytochrome P450 gene ttoG for tolytoxin biosynthesis). Since no further redox components were identified, TtoG might directly accept a methyl-branched precursor for epoxidation.

These data on three new pathways were a useful basis to investigate the evolutionary basis of structural chimerism in the macrolide family. Similar moieties can principally result from shared ancestry of biosynthetic genes or from convergent evolution. We conducted phylogenetic analysis on 543 domains from the mis, tto, lum, and other PKSs, which also included the PKS of the actin inhibitor rhizopodin with a scytophycin/luminaolide-like tail moiety. The tree topologies clearly supported a common evolutionary origin for the shared moieties of misakinolide, tolytoxin, and luminaolide. The data suggest an evolutionary sequence that involves a scytophycin/tolytoxin-like compound as progenitor of luminaolides, with the structural change in the macrolide region being due to loss of a contiguous DNA region (Fig. 5, region B) encoding five PKS domains. The results are also consistent with an earlier switch of almost the entire misakinolide- and tolytoxin-type tail regions, which resulted from an exchange of DNA covering 4-5 PKS modules (Fig. 5, region A). Although the directionality of this change cannot be inferred, the structure of lobophorolide¹² (3) is of note in this context. It contains a misakinolide-like tail moiety but is remarkably similar to tolytoxin regarding oxygenation, methylation, and cyclization pattern, suggesting it to represent a “missing link” between these compounds. Misakinolide A might therefore be an archetype member of this family and close to the starting point of a sequence connecting misakinolide-, lobophorolide-, tolytoxin-, and luminaolide-type polyketides. Since the data assign the misakinolide producer to the recently proposed candidate phylum 'Tectomicrobia', which is phylogenetically distant from Cyanobacteria, an interesting question is how gene clusters move between such divergent organisms. Another striking example of inter-phylum natural product gene transfer is the production of the related onnamides, nosperin, pederin, and diaphorin by members of 'Tectomicrobia',¹⁷ Cyanobacteria,⁴⁴ Gamma-²³ and Betaproteobacteria,⁴⁵ respectively. Such distributions show that the occurrence of related symbiont-derived compounds in diverse sources does not necessarily imply production by similar bacteria but can be the result of complex vertical and horizontal evolutionary events.

Cis-AT PKSs mainly evolve by module duplication and by horizontal and vertical acquisition of entire PKS assembly lines.⁴⁶ Chimeric polyketides resulting from major PKS recombinations are therefore rarely encountered for this PKS family. In contrast, trans-AT systems exhibit a pronounced tendency to recombine and form new gene clusters in a mosaic-like fashion.²² This property can result in major recombinatorial polyketide diversification, as exemplified for the mis, tto, and lum products or compounds of the pederin family.^{15, 23, 26, 44, 45} Artificial PKS engineering is an important goal, but has been almost exclusively performed with cis-AT PKSs, rarely yielding compounds with dramatically altered skeletons. The natural propensity of trans-AT PKSs to form functional hybrids suggests these enzymes as attractive candidates for synthetic biology. In this context, evolutionary studies on hybrid genes to identify recombination points can provide useful guidelines. An example is the five-domain deletion leading to luminaolides, which occurred behind KS17. Previous phylogenetic²² and biochemical^47–49 studies suggested significant selectivity for the α,β-moiety of substrates for many extending KSs, but lower selectivity for the nonextending KS⁰ domains. In agreement with these data, the observed deletion topology preserves KS17 as acceptor for the incoming substrate, while a KS⁰ is the acceptor of the modified chain in luminaolide biosynthesis. Similar cut and paste strategies using sites preferred by nature might also be practicable for trans-AT PKS engineering.

Unexpectedly, in the trees inferred from 15 different rhizopodin PKS domains corresponding to the tolytoxin/luminaolide-like tail, almost all riz homologs are well-separated from domains of the other pathways (Fig. 5, Supplementary Fig. 21-26). These topologies do not support recent evolution of the three architecturally related PKS parts from a common ancestor. One scenario that would be consistent with the data is that closely related tail moieties had independently emerged at least twice, leading to rhizopodin in one case and to scytophycins, tolytoxin, and luminaolides in the other. As an alternative to convergent evolution, similar trees would also be observed if the first six modules of the riz, tto, and lum PKS were conserved for extremely long times and had appeared early in the evolution of trans-AT PKSs. We disfavour this latter model, since it would likely require the exchange of at least KS2 and KS5 without concomitant modification of the polyketide chain and furthermore imply that many PKSs with unrelated architecture had evolved from these six modules. Either way, convergent evolution or extremely long retention would be conceivable in cases where a producer gained a significant advantage from a polyketide structure, thus promoting multiple emergence or long-term retention of the corresponding PKS architecture. Studies on actin structures and synthetic minimized analogs indeed show that the similar tail regions of rhizopodin (5), kabiramide C (6), and related compounds are crucial and sufficient for G-actin binding and filament severing.^{1, 50} The importance of this moiety for bioactivity might therefore be a key factor underlying its repeated occurrence in nature.

Online Methods

General

NMR spectra were recorded on a Bruker Avance III HD spectrometer equipped with a cold probe at 600 MHz for ¹H NMR and 150 MHz for ¹³C NMR. Chemical shifts were referenced to the solvent peaks at δ_H 2.05 for acetone-d ₆ and δ_H 1.94 and δ_C 1.39 for acetonitrile-d ₃. LC-ESI mass spectrometry was performed on a Thermo Scientific Q Exactive mass spectrometer. Sequencing, bioinformatics, compound isolation and characterization is described in Supplementary Note 1.

Organisms, plasmids and culture conditions

Theonella swinhoei WA (misakinolide chemotype) was collected in 2002 and 2011 at Hachijo-Jima, Japan (33°13.77' N, 139 73.47' E). A part of each sponge was stored in RNAlater, the remainder was subjected to cell separation immediately after collection. E. coli EPI300-T1^R (Epicentre) was used as host for the fosmid library, E. coli XL1blue was the general cloning host. Bacteria were cultivated in LB medium supplemented with 1.5% agar for solid media and with 100 µg/mL ampicillin or 5 µg/mL chloramphenicol for selection. The strain Scytonema sp. PCC 10023 was originally isolated from terrestrial sampling of felt-like stratum on a wall, Somerset Bermuda, and first described as Scytonema javanicum B-77-Scy.⁵¹ The strain Planktothrix paucivesiculata PCC 9631, recently renamed from Oscillatoria sp. PCC 9631,⁵² was isolated from water samples collected in the Marne River, France in 1996. Biomass of these axenic strains for genetic and chemical analyses was obtained by growing several cultures of 1.25 L of BG11 with 10 mM of NaHCO₃ for the Planktothrix strain and similar medium without NaNO₃ ⁵³ for the Scytonema strain at 25 °C under continuous light provided by Osram Universal White fluorescent tubes (20 µmol quanta m^-2 s^-1) with agitation and constant bubbling of 1% CO₂. The cultures in late exponential to linear growth phase were centrifuged at 12,000 × g for 10 min at 20 °C. The pellets were washed twice with sterile distilled water and immediately frozen in liquid N₂ prior to be lyophilized and further used in all analyses. Both strains are available at the Pasteur Culture collection of Cyanobacteria (http://www.cyanobacteria.web.pasteur.fr).

Preparation of the enriched 'Entotheonella' fraction

A protocol adapted from Bewley, Holland, and Faulkner¹⁸ as described previously¹⁷ was used for cell separation. 500 g sponge tissue of freshly collected T. swinhoei was processed. The resulting 'Entotheonella' fraction was resuspended in 200 ml Ca/Mg-free artificial sea water⁵⁴ and stored at 4 °C.

Isolation of the misakinolide gene cluster

The total DNA was isolated from one gram of T. swinhoei WA sample as reported previously for another chemotype of T. swinhoei.¹⁵ The isolated DNA was separated on low-melting-point agarose gel, and DNA fragments in the size range of 35-40 kb were purified from the gel. Subsequently, the size-selected DNA fragments were end-repaired, blunt-ligated into the fosmid vector pCC1FOS (Epicentre), packaged into lambda phage particles, and transfected into E. coli EPI300-T1^R (Epicentre) as described previously.¹⁵ The resulting clones were grown in a three-dimensional format in semi-liquid medium according to a previously developed protocol²⁸ at a clone density of up to 4000 cfus per ml, resulting in a 3D metagenomic library containing ∼350,000 clones. The library was screened²⁸ using the specific primers EKS27-f (5'-CTT GGG CCG CGT TGG AAG AT-3') and EKS27-r (5'-CCG CTT CGC CAA GCA CAA AG -3'). A positive clone (pTSW-AU1) was isolated and end-sequenced. A primer walking approach was used to isolate the entire PKS gene cluster. The library was initially screened with the primer pair EKS27RP-f (5'-TAG AGC ATC CGG GAT TTT CCT G-3') / EKS27RP-r (5'-AGT TGT TCA CAC GCG GGT GTT C-3') designed based on the pTSW-AU1 end sequences. The isolated fosmid (pTSW-AU2) was digested into smaller fragments with PstI, which were subsequently subcloned into the PstI site of pBluescript II SK(–) for end sequencing. The library was subsequently screened again in several further consecutive stages with specific primers designed based on end sequences of previously isolated fosmids. The following primer pairs were used: EKS27FP2-f (5'-TGG AGG TCA ACG CCC TTA ACG AAG TCT-3') / EKS27FP2-r (5'-GGA CCG CAC GTG GTT CTT TTC TGT GAA-3'), AU5KSEnd-f1 (5'-TCG TTT GGT GCA ACC GGC TC-3') / AU5KSEnd-r1 (5'-ACG TTG CTC TAC ACC CTC TGC G-3'), PEP-f (5'-GAT GAG TGG TCA AGC CTG TGT CGT GAA-3') / PEP-r (5'-CGC TAT GAA TTT GCT GCT GTA CAA CCA C-3'), respectively. In this way, three additional positive clones (pTSW-AU5, pTSW-AU10b, pTSW-AU11a) were isolated and entirely sequenced. The sequences of the five isolated clones were assembled using Geneious 5.4.6 (Biomatters). Remaining gaps between contigs were filled in by PCR, and the resulting continuous sequence was analyzed by BLAST homology searches and PKS domain analysis.

MALDI-IMS for misakinolide A localization in the sponge

Fresh Theonella swinhoei WA sponge collected at Hachijo Island, Japan was immediately frozen upon collection and stored at −80 °C. Thin slices of the sponge were cut using a microtome (Microm Thermo Fisher). For this purpose, sponge tissue was immobilized on a stainless steel support with Tissue Tek^® O.C.T™ Compound and Cryomolds®. The sponge tissue consisting of a representative cross section of the sponge (containing outer and inner regions of the sponge) was cut into 50 µm slices. Tissue sections were either carefully transferred onto a stainless steel target plate (Thermo Scientific) suitable for MALDI-IMS experiments or onto a microscopy slide for subsequent CARD-FISH assays to detect 'Entotheonella'. MALDI-IMS was conducted on a MALDI LTQ-Orbitrap™ mass spectrometer (Thermo Scientific). Before matrix applications, stainless steel target plates were scanned using a HP ScanJet G4050 scanner. The Sunchrome SunCollect MALDI Spotter was used to uniformly cover the sponge tissue with universal MALDI matrix (1:1 mixture of HCCA and DHB; Sigma Aldrich). The matrix solution (30 mg/mL in methanol) was sprayed onto the sponge tissue at a distance of 40 mm from the sample surface and using a spray volume of 50 µL/min. In total, the sponge tissue was covered with 5-6 layers of matrix. The samples were measured in positive mode, a laser energy of 45 µJ, 2 microscans/step, and a resolution of 60.000. Data were acquired in the mass range between 500 and 1600 m/z. The spatial resolution was set to 60 µm which corresponds to the maximal spatial resolution of the instrument. Imaging data were analyzed with ImageQuest. Masses of interest were assigned to a false color or a false color head map scheme and the optical picture taken before sample application was superimposed onto the false color pictures. Control MALDI-IMS experiments were conducted using the same conditions and parameters, but the sponge chemotype T. swinhoei Y, which lacks misakinolide.

'Entotheonella'-specific FISH probe ESP219

The Entotheonella spp. specific FISH probe ESP219 (5'-CCG CAA GCY CAT CTC AGA CC-3') was designed from an alignment of currently available 'Entotheonella' 16S rRNA sequences from the SILVA rRNA database (http://www.arb-silva.de). Optimization of binding site and probe length were performed with mathFISH.⁵⁵ Monolabeled (5') probes with Cy5 were commercially synthesized by Microsynth (Switzerland). Hybridization conditions were optimized by Clone-FISH as previously described.⁵⁶ Briefly, the 16S rRNA of 'Entotheonella factor TSY1'¹⁷ and a construct with a single nucleotide mismatch (pSNPA225C) in the probe binding site were cloned into pBluescript and expressed in E.coli BL21 (DE3). Cells were fixed with 4% PFA in PBS (overnight, 4 °C) cells and stored at -20 °C until use. To optimize hybridization conditions, the relative amount of FISH-positive cells with increasing amounts of formamide (FA) in the standard hybridization buffer (0.9 M NaCl, 20 mM TRIS-Cl, pH 7.6, 0.01% SDS) ranging from 0-80% at a fixed temperature of 46 °C were measured using flow cytometry. The optimal hybridization condition was defined as the FA concentration where the signal for cells expressing the pSNPA225C construct fall below 1%. For CARD-FISH experiments, the optimized FA concentration of 35% at 46 °C was increased to 55% according to the formula of Wetmur⁵⁷ due to lower hybridization temperature of 35 °C.

CARD-FISH for localization of 'Entotheonella'

CARD-FISH analysis was carried out as previously described⁵⁸ with slight modifications. Slices were fixed in 4% PBS-buffered paraformaldehyde (4 °C, overnight), washed with 1×PBS (r.t.), dehydrated with 99% ethanol (5 min, r.t.) and subsequently air-dried (2 h, r.t.). After permeabilization with lysozyme (10 mg/mL) for 20 min at r.t., endogenous peroxidases were inactivated with 0.01 M HCl for 15 min at r.t. The slices were thoroughly washed with 1×PBS and ddH₂O, dehydrated in 99% ethanol for 1 min and air-dried. Whole cell in situ hybridization of slices was performed using a newly designed, 'Entotheonella'-specific oligonucleotide probe ESP219 and with the control probe NON338 (5'-ACT CCT ACG GGA GGC AGC -3'; Supplementary Fig. 3) coupled to horseradish peroxidase (HRP) (biomers, Germany). For hybridization, slices were covered with hybridization solution (0.9 M NaCl, 20 mM Tris-HCl pH 7.6, 10% wt/vol dextran sulfate, 0.05% wt/vol sodium dodecyl sulfate [SDS], 55% vol/vol formamide, 1% wt/vol nucleic acid blocking reagent [Roche], 0.5 mg/mL of herring sperm DNA [Sigma]) containing 0.5 ng/µL of HRP-labeled probe, placed in a moistening chamber and incubated for 2 h at 35 °C. Afterwards, slices were washed with prewarmed washing buffer (3 mM NaCl, 5 mM EDTA pH 8.0, 20 mM Tris-HCl pH 7.6, 0.05% wt/vol SDS) for 30 min and with PBST (0.01% vol/vol Triton X-100 [Sigma]) for 45 min at 37 °C. After equilibration of the probe-delivered HRP in 1 × PBS for 15 min at r.t., detection was performed by incubation with amplification buffer (1 × PBS pH 7.6, 2 M NaCl, 20% wt/vol dextran sulfate, 0.1% wt/vol nucleic acid blocking reagent [Roche], 0.0015% wt/vol H₂O₂) containing Alexa Flour® 633-labeled tyramides for 2 h at 37 °C in the dark. The slices were then washed twice with PBS-T for 15 min at RT, rinsed with ddH₂O, dehydrated with 99% ethanol for 1 min, air-dried and stored at 20 °C until microscopic analysis. Slices were covered with mountant (Cityflour, Ltd., UK) and observed under a Zeiss Axioskop 2 epifluorescence microscope equipped with a 75 watt xenon arc lamp (XBO 75), an appropriate filter set for Alexa Fluor® 633 and a 10x Plan Apochromat objective.

Expression and analysis of the misakinolide PS domain

For PCR amplification of the gene region encoding the PS domain, the fosmid pTSW-AU5 was used, which contains a portion of the misakinolide biosynthesis gene cluster. PCR was performed using the primer pair MisPS-for (5'-CAA GTC ATA TGT CGC CTT GTT TTG CTG AAA CG-3') and MisPS-rev (5'-GAC TAT CTC GAG CGG ATT GTC TTG TGC AAG CAT-3'). After restriction with NdeI and XhoI, the PCR product was ligated into the same sites of pET29b to yield plasmid pPP09. The PS was heterologously produced in E. coli BL21 (DE3). Typically, TB medium (1 L) containing kanamycin (25 µg/mL) was inoculated with 0.01 volumes (10 mL) overnight culture and incubated to an OD₆₀₀ of 0.8-1.0 (250 rpm, 37 °C). Expression of the PS domain was induced with IPTG (1 mM), and cultures were shaken 14 hours (250 rpm, 16 °C). For purification, cells were harvested by centrifugation (3600 rpm, 4 °C, 40 min), resuspended (1 g/4 mL) in lysis buffer (20 mM Tris-HCl, 0.5 M NaCl, 10 mM imidazole, 10% glycerol, pH 8.0) and broken by sonication on ice (5×10 s pulses, 75% amplitude). After centrifugation (11,000 rpm, 4 °C, 30 min) the lysate was bound to 750 µL Ni-NTA resin (1 h, 4 °C) and loaded onto a PolyPrep® column. The column was washed with lysis buffer (2×2 column volumes) and lysis buffer containing 30 mM imidazole. The protein was eluted with lysis buffer containing 300 mM imidazole. Fractions were analysed by SDS-PAGE. Desalting was conducted by buffer exchange into storage buffer (25 mM HEPES, 100 mM NaCl, 10% glycerol, pH 7.5) with a GE Healthcare PD column. Finally the protein was concentrated with a Sartorius Vivaspin 500 column® (MWCO 10,000). Activity assays were conducted in a total volume of 200 µL containing 12 mM NAC thioester substrate (20 μL of 10L stock dissolved in H₂O containing 2% DMSO) and ca. 4 mg/mL of the PS dissolved in storage buffer (180 μL). Reactions were incubated at 37 °C for 16 h and extracted with EtOAc (2 × 200 µL). After evaporation of the solvent, the sample was resuspended in MeOH (750 µL) and analysed by HPLC and LCMS. For NMR measurement the product was resuspended in CDCl₃. The preparative assay was performed as described previously for DH domains,³⁷ using the PS instead of the DH and compound 8 ³⁹ as test substrate. Assay reactions were analyzed by analytical reversed phase HPLC (column: Knauer, Eurospher II 100-5, 5 µ, 250 × 4 mm; flow rate: 1 mL/min, mobile phase: 50% MeOH/H₂O; λ 230 nm). Preparative assay reactions were purified by semi-preparative reversed phase HPLC (column: Phenomenex, Luna, C18, 5μ, 250 × 10 mm; flow rate: 3 mL/min; mobile phase: 30% CH₃CN/H₂O) to give recovered alcohol 8(t_R 9.9 min), anti-10 (t_R 17.1 min) and syn-9 (t_R 22.3 min,). NMR data for products 9 and 10 were consistent with those previously reported.³⁷

anti-10: ¹H NMR (600 MHz, CDCl3) δ = 5.85 (brs, 1H), 4.32 – 4.27 (m, 1H), 3.94 (m, 1H), 3.50 – 3.37 (m, 2H), 3.09 – 2.98 (m, 2H), 3.00 (dd, J= 14.3, 8.6,1H), 2.62 (dd, J=14.3, 5.2, 1H), 1.96 (s, 3H), 1.74 – 1.59 (m, 4H), 1.40 – 1.35 (m, 1H), 1.33 – 1.27 (m,1H), 1.17 (d, J=6.5, 3H).

syn-9: ¹H NMR (600 MHz, CDCl3) δ = 5.80 (br s, 1H), 3.83 – 3.77 (m, 1H), 3.47 – 3.41 (m, 3H), 3.09 – 2.99 (m, 2H), 2.83 (ddd, J=14.8, 7.6, 1.5, 1H), 2.62 (ddd, J=14.8, 5.4, 1.3, 1H), 1.96 (d, J=1.2, 3H), 1.86 – 1.79 (m, 1H), 1.61 – 1.48 (m, 3H), 1.29 – 1.14 (m, 2H), 1.14 (dd, J=6.2, 1.4, 3H).

MS analysis of misakinolide A and congeners

Whole sponge samples of T. swinhoei WA and the 'Entotheonella'-enriched cell fractions were each extracted with EtOH and MeOH. The samples were subjected to UPLC HR HESI-MS/MS analysis, followed by molecular network analysis. HR HESI-MS Data were collected on a Thermo Q Exactive™ coupled to a Dionex Ultimate® 3000 UPLC system. A solvent gradient (A = H₂O + 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) was used on a Phenomenex® Kinetex ™ 2.6 µm C18 100 Å (150 × 4.6 mm) column at 27 °C. The MS was operated in positive ionization mode at a scan range of 200 to 1600 m/z or 1000-1600 m/z, respectively. The spray voltage was set to 3.7 kV and the capillary temperature to 320 °C. MS data were acquired in a data dependent manner. A parent ion scan was conducted with a resolution of 70,000. The top ten most abundant peaks from each parent ion scan were subsequently fragmented with a resolution of 17,500 and a normalized collision energy of 35 (NCE). MS² scans were conducted with an AGC target of 1×10^5 or a maximal injection time of 250 ms. The isolation width was set to 4 m/z, the dynamic exclusion to 10 seconds. Thermo raw files were converted into mzXML files using MSExport. MS Cluster was used to merge identical spectra to consensus spectra obtained from metabolites that were fragmented multiple times. Cosine values resulting from spectral alignments that were calculated for every possible pair of spectra were generated. A cosine value of one indicates identical spectra whereas a cosine value of zero indicates no spectral correlations. The cosine value threshold was set to 0.65. Spectra obtained from solvent controls were removed. The resulting network was visualized, color coded and analyzed using cytoscape 3 (www.cytoscape.org).⁵⁹

Accession codes

GenBank. Sequence data have been deposited under accession codes KR857270-KR857273 and LC054533 - LC054540.

Supplementary Material

Note: Supplementary information is available on the Nature Chemical Biology website.