Food‐Poisoning Bacteria Employ a Citrate Synthase and a Type II NRPS To Synthesize Bolaamphiphilic Lipopeptide Antibiotics

Abstract

Mining the genome of the food‐spoiling bacterium Burkholderia gladioli pv. cocovenenans revealed five nonribosomal peptide synthetase (NRPS) gene clusters, including an orphan gene locus (bol). Gene inactivation and metabolic profiling linked the bol gene cluster to novel bolaamphiphilic lipopeptides with antimycobacterial activity. A combination of chemical analysis and bioinformatics elucidated the structures of bolagladin A and B, lipocyclopeptides featuring an unusual dehydro‐β‐alanine enamide linker fused to an unprecedented tricarboxylic fatty acid tail. Through a series of targeted gene deletions, we proved the involvement of a designated citrate synthase (CS), priming ketosynthases III (KS III), a type II NRPS, including a novel desaturase for enamide formation, and a multimodular NRPS in generating the cyclopeptide. Network analyses revealed the evolutionary origin of the CS and identified cryptic CS/NRPS gene loci in various bacterial genomes.

When life gives you citrate… Genome mining of food‐spoiling bacteria revealed unprecedented lipocyclopeptides (bolagladins) with antimycobacterial activity. A specialized citrate synthase and nonribosomal peptide synthetases (NRPSs) play key roles in the formation and loading of their unusual tricarboxylic fatty acid tail, thus generating a bolaamphiphilic natural product (see scheme).

Among the many food‐spoiling bacteria, one species stands out, as it is responsible for the governmental ban of a national dish: Burkholderia gladioli pv. cocovenenans (literally meaning making coconut poisonous). These bacteria poison the Indonesian specialty tempeh bongkrek, a coconut press cake produced by fermentation using a harmless mold fungus (Rhizopus). Yet, frequent contaminations of the Rhizopus cultures with bacterial toxin producers have been responsible for the deaths of thousands that have consumed the delicacy. ^[1] The causative agent of these intoxications has been identified to be the respiratory toxin bongkrekic acid. ^[1] To gain insight into its biosynthesis, we have sequenced the genome of B. gladioli pv. cocovenenans. In addition to the molecular basis of bongkrekic acid biosynthesis, we discovered various other biosynthesis gene clusters. ^[2] Yet, surprisingly little is known about the biosynthesis potential of the food‐poisoning bacteria, in particular with respect to peptidic natural products. Here, we deduce orphan nonribosomal peptide synthetase (NRPS) ^[3] gene clusters of B. gladioli pv. cocovenenans, and report the surprising discovery of a new family of antibiotics, noncanonical lipopeptides equipped with an unprecedented tricarboxylic acid tail that results from the recruitment of a citrate synthase.

Genome mining ^[4] of B. gladioli pv. cocovenenans revealed the presence of five orphan NRPS gene clusters. Bioinformatics predictions and BLAST homology search indicated that four of these gene clusters code for the biosynthesis of characterized lipopeptides, namely haereogladins (promoting biofilm formation), burriogladins (facilitating swarming), and icosalides (surfactant and swarming inhibitor) (Figure 1 A). ^[5] Furthermore, one NRPS gene cluster codes for the rare diazeniumdiolate siderophore gladiobactin. ^[6] We verified the predicted functions by HPLC‐HR‐MS of culture extracts and MS/MS fragmentation. ^[6] The fifth cryptic biosynthetic gene cluster (named bol) consists of a central 13 kb NRPS gene (bolH) that is flanked by a number of putative accessory genes (bolB‐U) (Figure 1 B and Table S7 and S8). By means of database searches we noted that genomes of numerous other B. gladioli isolates from highly diverse niches harbor orthologous, yet cryptic, gene clusters (Supporting Information). ^[7]

Figure 1.

Genome mining of the bacterial contaminant of tempeh bongkrek. A) Biosynthetic potential encoded in the chromosome. Gene loci for known compounds: ena, enacyloxin; bon, bongkrekic acid; cay, caryoynencin; tox, toxoflavin; NRPS: bgd, burriogladin; hgd, haereogladin; glb, gladiobactin; ico, icosalide; bol, cryptic NRPS gene locus, marked in red. B) Organization of the bol gene cluster. C) Encoded assembly line and predicted peptide backbone. FAL, fatty acid ligase; A, adenylation domain; T, thiolation domain or peptide carrier protein; C_S, starter condensation domain; ^LC_L, C domain fusing two l‐amino acids; C_Dual, C domain also promoting epimerization; TE, thioesterase domain. D) Mutant construction using homologous recombination, and PCR‐based verification of ΔbolH mutant. E) Production of 1 and 2 in B. gladioli pv. cocovenenans (WT), and metabolic profile of deletion mutant ΔbolH.

To correlate the bol gene cluster with metabolites we deduced the core peptide structure from in silico analyses and performed comparative metabolic profiling of the wild type and a targeted null mutant. First, we gleaned the architecture of the cryptic peptide from bioinformatics analysis of the bol assembly line. In brief, the encoded type I NRPS (BolH) consists of four modules each composed of condensation (C), adenylation (A) and thiolation (T) domains. Furthermore, genes for a type II NRPS ^[8] comprising freestanding adenylation (A, BolO) and peptidyl carrier protein (PCP, BolC) domains, and a fatty acid ligase (FAL, BolB) suggest that an activated N‐acyl amino acid would be merged with a tetrapeptide, likely yielding a lipopeptide with five amino acid residues. The detailed analysis of the A and C domain specificities, ^[9] where possible, predicted that an N‐acyl‐l‐Ala‐derived building block would be linked to an l‐Ser‐d‐Val‐l‐X‐l‐Ser chain (X, prediction of the amino acid specificity not possible) (Figure 1 C).

To create a null producer, we improved the current protocol for the genetic manipulation of B. gladioli pv. cocovenenans. We noted that this strain is naturally competent to take up DNA and achieved the targeted gene knockout of the NRPS gene by means of homologous recombination and integration of a kanamycin resistance cassette into bolH, using a double‐selection plasmid containing the pheS gene. ^[10] By antibiotic selection and PCR analysis the desired mutation was confirmed (Figure 1 D). By comparison of the metabolic profiles of wild type and ΔbolH mutant grown under various conditions we detected two metabolites (1 and 2) in the wild‐type culture that are absent in the mutant strain. HPLC–HRMS elucidated two major congeners with m/z 822.4168 [M−H]⁻ (1) and m/z 836.4322 [M−H]⁻ (2) (Figure 1 D). The deduced molecular formulas and the MS/MS fragmentation patterns suggested that 1 and 2 represent the predicted lipopentapeptides.

To determine their structures, we optimized culture conditions and subjected the upscaled fermentation broth (3 L) to chromatographic separation by size‐exclusion chromatography and preparative HPLC, yielding pure 1 (0.8 mg) and 2 (1.6 mg). The structures of 1 and 2 were fully elucidated by ESI‐HR‐MS, MS/MS, and 1D and 2D NMR experiments (Figure 2 A). For 2 the molecular formula of C₄₀H₆₃N₅O₁₄ was deduced from ESI‐HR‐MS, and the number of carbon atoms was corroborated by ¹³C and HMBC NMR data. The presence of a fatty acid residue was confirmed by analysis of ¹H NMR and DEPT‐135 NMR data. From ¹H,¹H‐COSY couplings the spin systems of five amino acids were deduced (Figure 2 A). ¹H,¹³C HMBC couplings of the NH protons of each amino acid with the corresponding carbonyl carbon atoms and the respective Cα atoms of the neighboring amino acid allowed the assignment of the peptide chain. A coupling constant of J=14 Hz between the α and β protons of the dehydro‐β‐alanine residue indicates the E‐configuration of the double bond. An ¹H,¹³C‐HMBC coupling of the β‐proton of the dehydro‐β‐alanine unit with C1 of the fatty acid revealed the N‐acyl connection. Comparison of the deduced molecular composition (C₃₉H₆₁N₅O₁₄) and the MS/MS fragmentation of 1 with the ones determined for 2 indicated an Ile to Val replacement in 1. 1D and 2D NMR analyses confirmed this assignment. The absolute configuration of the amino acids in 1 and 2 was unambiguously determined by Marfey's analysis ^[11] (Supporting Information).

Figure 2.

Structures and bioactivities of bolagladins. A) Structure elucidation by 2D NMR couplings and with Marfey and Sanger reagents. B) Schematic representation of canonical lipopeptides (LPs) versus bolagladin. C) Droplet‐collapse assay showing no tensioactive property of 1; NC, negative control: water; PC, positive control, tenside producer; WT, wild type; ΔbolH, mutant. D) Swarming areas of WT mutant on semisolid agar.

The structure of the fatty acid moiety proved to be more complex than anticipated. MS/MS analyses revealed a fragment ion with m/z 450.2480, indicating an identical, yet highly substituted fatty acid residue (C₂₄H₃₆NO₇) in 1 and 2. ¹H, ¹H‐COSY couplings of the protons H‐2 to H‐4 and H‐9 to H‐14 and 1D NMR signals for additional methylene functions established the fatty acid chain. Coupling constants of J=11 Hz (between H‐2 and H‐3) and J=10 Hz (between H‐11 and H‐12) indicated the Z‐configuration of the double bonds. ¹H,¹³C‐HMBC couplings (Figure 2 A) finally unveiled the presence of an unusual methoxy dicarboxylic acid terminus.

The structures of 1 and 2 are highly unusual. The rare dehydro‐β‐Ala linker is known from the Streptomyces‐derived enamidonins, yet their biosynthesis has remained enigmatic. ^[12] Furthermore, owing to the unprecedented tricarboxylic fatty acid tail, 1 and 2 constitute a novel family of lipopeptides. Regular lipopeptides consist of a hydrophilic head moiety (linear or cyclic peptide) and a hydrophobic fatty acid tail. ^[13] The resulting amphiphilic nature of lipopeptides is usually reflected in their bioactivities, which encompass surfactant properties that may influence biofilm formation and swarming of the producer, and may disintegrate membranes of recipients (Figure 2 B).[ ^13b , ¹⁴ ] Compounds 1 and 2 clearly deviate from the canonical lipopeptide architecture and polarity as they have two hydrophilic groups at both ends of a hydrophobic chain and thus are natural bolaamphiphiles. ^[15] Therefore, we named them bolagladin A (1) and B (2).

The markedly poor solubility of the bolagladins hampered the routine profiling of biological activities of the isolated compounds. Thus, we first compared typical lipopeptide activities of B. gladioli pv. cocovenenans wild type and the ΔbolH mutant. We found that the mutant does not show altered tensioactive properties in a droplet‐collapse assay, and when grown on low‐density agar its swarming behavior is similar to the wild type (Figure 2 C and D). Notably, the strains do not produce any of the known tensioactive lipopeptides under these cultivation conditions. Next, we evaluated potential metal‐binding properties of 2, since bolU could code for a TonB‐like receptor that is typically associated with siderophores. However, strong metallophore activities were not observed in a CAS agar assay. Furthermore, stable iron adducts of 2 were not detected by mass spectrometry. Finally, we subjected 2 to antimicrobial assays using a panel of bacterial and fungal strains. We observed moderate antifungal activities of 2 against Candida albicans and Penicillium notatum, moderate activities against Staphylococcus aureus (MRSA) and Enterococcus faecalis (VRE), and more pronounced activity against Mycobacterium vaccae (Supporting Information).

As to the most unusual feature of the bolagladins, their branched tricarboxylic fatty acid tail, we predicted that an encoded citrate synthase (BolR), an acyl‐CoA ligase (BolB), an O‐methyl transferase (BolS), and two encoded priming ketosynthases III (KS III, BolM, BolP) play key roles in its formation. A phylogenetic analysis of BolR and a selection of citrate synthases (CS) and related biocatalysts showed that BolR falls into the group of CSs and related enzymes that promote an aldol reaction at the Si‐face of 2‐oxoglutarate (Figure 3 A). ^[16] Close relatives of BolR include the CSs from B. gladioli primary metabolism (CS I and CS II) and CSs from the biosynthetic pathways of squalestatins ^[17] and maleidrides. ^[18]

Figure 3.

Phylogenetic and mutational analysis of key pathway enzymes. A) Cladogram of Si‐ and Re‐face‐specific citrate synthases (Si/Re CS, CS I/CS II) from primary metabolism, squalestatin CS, maleidrides CS, isopropylmalate and Re‐citramalate synthase (IPMS and CMS). Outgroup: malate synthase (MS). B–D) Metabolic profiling of targeted mutants. *1 and *2, isotopes of 1 and 2. E) Structures of 3 and 4 formed in ΔbolR mutant, most likely resulting from incorporation of succinate and malate in lieu of citrate, and structures of 5 and 6 formed in ΔbolQ mutant.

To verify the role of BolR, we created a null mutant by double crossover. Metabolic profiling of the resulting ΔbolR mutant culture by HPLC–HRMS revealed that the production of 1 and 2 was fully abrogated. Instead, we detected two new metabolites with m/z 748.4157 [M−H]⁻ (3) and 764.4120 [M−H]⁻ (4) that are absent in the wild‐type metabolome (Figure 3 B). Their deduced molecular formulas (3: C₃₇H₅₈N₅O₁₁, 4: C₃₇H₅₈N₅O₁₂), MS/MS analyses and NMR data of 3 revealed that, in comparison to 2, both compounds are devoid of the terminal carboxymethyl moiety. In addition, 3 lacks the methoxy group (Figure 3 E). A plausible explanation for the production of 3 and 4 in the ΔbolR mutant would be the incorporation of succinate and malate building blocks in lieu of citrate (Figure 3 E). Apparently, the O‐methyl transferase BolS does not recognize the alternative substrate. Mutational analyses showed that both encoded KS III (BolM and BolP), and the ligases (BolO) are essential for the bolagladin biosynthesis, too (Figure 3 C). Since none of the congeners could be detected in these mutants, we propose that the function of the ligases and KS III is the activation and elongation of citrate and the fatty acid, respectively. At this stage, however, it cannot be ruled out that oxaloacetate is first activated and elongated before it is modified by the CS.

l‐Asp‐PCP (BolC), would be generated by the freestanding A domain BolO. l‐Asp‐PCP would then be transformed into β‐Ala‐PCP by means of a pyridoxal phosphate‐dependent decarboxylase, BolN, in analogy to the destruxin and fluvirucin biosynthetic pathways. ^[19] The activated tricarboxylic fatty acid would then be transferred onto the type II NRPS to form the acyl‐β‐Ala subunit by the second KS III. A similar cooperation of two KS III and an ATP‐dependent ligase has been implicated in the biosynthesis of unusual fatty acids in mupirocin and thiomarinol biosynthesis. ^[20] Finally, a desaturase would introduce the enamide double bond. We confirmed this model by targeted gene inactivation of bolC, bolN and bolB (Figure 3 C). To pinpoint the desaturase in charge of the unusual peptide modification we generated and profiled a range of candidate mutant cultures (ΔbolL, ΔbolF, and ΔbolQ). While the ΔbolL mutant showed the wild‐type profile, in the ΔbolF and ΔbolQ mutants production of 1 and 2 was fully abrogated. In the ΔbolQ mutant we identified congeners with m/z 838.4484 [M−H]⁻ (5) and 824.4327 [M−H]⁻ (6) lacking the enamide double bond (Figure 3 E). Thus, BolQ seems to be a novel trans‐acting desaturase that generates the unusual enamide substructure online. Finally, the rare tricarboxylic‐acyl‐β‐dh‐Ala brick would be transferred onto the modular NRPS to assemble the full‐length peptide (Figure 4 A).

Figure 4.

Model of bolagladin biosynthesis, and network analysis for genome mining. A) Biosynthesis of the tricarboxylic fatty acid involving citrate synthase (CS) BolR; MT, methyltransferase; KS III, priming ketosyntase; FAS, fatty acid synthase; DC, decarboxylase; DS, desaturase. B) Sequence similarity network and genome neighborhood networks identify the evolutionary origin of BolR and enable the discovery of cryptic CS/NRPS gene clusters.

To gain insight into the evolution and distribution of similar biosynthetic pathways we created sequence similarity networks (SSNs) and genome neighborhood networks (GNNs) using CS I and BolR as the query sequences (Figure 4 B). ^[21] The CS I network forms a single cluster that places the CS gene exclusively among genes associated with primary metabolism in Proteobacteria. The BolR‐based network, in contrast, shows several clusters and revealed a high synteny of orthologous genes for citrate synthase (BolR), KS III (BolM and BolP), the A domain (BolO) and PCP (BolC) all putatively involved in the biosynthesis, activation and transfer of a citrate building block onto the assembly line. Our BolR network analysis led to the discovery of cryptic NRPS gene clusters that could code for the biosynthesis of related tricarboxylic fatty acid‐derived lipopeptides in Burkholderia cepacia and Burkholderia stabilis. Furthermore, we found NRPS‐independent biosynthetic pathways in genomes of Burkholderia, Pseudomonas, Collimonas, and Chromobacterium species.

Beyond genome mining for novel natural products, the SSNs and GNNs shed light on the origin of BolR. While regular CS and their surrounding genes are highly conserved in Proteobacteria, the bolR ancestor gene likely emerged from an ancient gene duplication event, followed by diverging evolution. Having been recruited by secondary metabolism alongside with accessory enzymes, it warrants metabolic fluxes that allow the incorporation of tricarboxylic fatty acids into lipopeptides, thus increasing structural diversity.

In conclusion, by genome mining of infamous food‐spoiling bacteria we discovered a new family of antibiotics that are an important addition to the metabolic inventory of diverse, ecology‐ and health‐relevant bacteria of the genus Burkholderia. ^[20] The bolagladins feature an unprecedented lipopeptide modification, a tricarboxylic fatty acid that creates a natural bolaamphiphile. We unveiled the key role of a citrate synthase in the bol pathway and provide an evolutionary model for the recruitment of a biocatalyst from the Krebs cycle. Using this designated citrate synthase as a bioinformatics handle led to the discovery of related cryptic pathways. Furthermore, the unusual lipopeptide assembly line, including the novel enamide‐forming desaturase, may provide novel tools for bioengineering approaches.

Conflict of interest

The authors declare no conflict of interest.

Supporting information

As a service to our authors and readers, this journal provides supporting information supplied by the authors. Such materials are peer reviewed and may be re‐organized for online delivery, but are not copy‐edited or typeset. Technical support issues arising from supporting information (other than missing files) should be addressed to the authors.

Supplementary

B. Dose, C. Ross, S. P. Niehs, K. Scherlach, J. P. Bauer, C. Hertweck, Angew. Chem. Int. Ed. 2020, 59, 21535.