Analysis of the Ambocidin Biosynthetic Gene Cluster

A recent genomic survey of actinomyces species for BGCs encoding calcium-dependent antibiotics uncovered multiple species that had the genetic capacity to produce previously undescribed compounds.¹³ Among these, a silent BGC found in the genome of Streptomyces ambofaciens ATCC 23877 was of particular interest as the order and predicted substrate specificity of adenylation domains in the BGC indicated that the product was likely to be broadly different from anything previously described.¹³ We conducted our own analysis of the genome of S. ambofaciens ATCC23877 using antiSMASH 6.0¹⁴ and identified a 75 kb BGC that contains the biosynthetic features consistent with the production of a calcium-dependent lipopeptide (Figure 1A, Supporting Information Table S7). Manual examination of the genes at this locus (which we named the amb BGC) identified at least 23 putative biosynthetic genes (ambA–W). The amb cluster contains three nonribosomal peptide synthetase (NRPS) genes, ambD, ambI, and ambS, with a total of 13 predicted modules–indicating a 13-residue non-ribosomal peptide. In addition to the core NRPS biosynthetic genes, the amb BGC contains genes for biosynthesis of non-canonical amino acids, as well as genes for fatty acid biosynthesis and attachment.

Discovery of Ambocidin Cyclic Lipodepsipeptides via Heterologous Expression

Despite extensive characterisation of S. ambofaciens ATCC 23877 secondary metabolites,¹⁵ no calcium-dependent antibiotic activity has ever been reported, suggesting the amb BGC is not expressed under commonly employed laboratory cultivation conditions. We cultivated S. ambofaciens using six media commonly found to elicit antibiotic production¹⁶ and did not observe any calcium-dependent antibiotic activity, indicating that the amb cluster is likely to be transcriptionally silent or non-functional in its native host (Supplementary Figure S34).

As transcriptionally silent gene clusters arise from intrinsic and often cryptic repression mechanisms in the native host,¹⁷ we sought to move the entire amb cluster into a more permissive heterologous host. Our initial attempts to clone the amb cluster using CRISPR-Cas9 targeted transformation-assisted recombination (TAR) in yeast were unsuccessful, as can be the case with large biosynthetic gene clusters. We then turned to the recently developed CAPTURE method of whole-BGC cloning, which uses Cas12a to excise the BGC of interest followed by in vitro assembly with two DNA adaptors using the 3’–5’ exonuclease activity of T7 DNA polymerase.^{11a, 18} The resulting linear product is transformed into E. coli cells expressing Cre recombinase, which catalyses the recombination between the LoxP site on the free ends of each DNA adaptor–forming a circular plasmid.^{11a, 18} The original protocol uses LacZ/β-galactosidase screening to identify successful Cre recombination events. To streamline the protocol we replaced the lacZ gene with pheS, the product of which catalyses the incorporation of the 4-chorophenylananine (4-CP) into proteins, resulting in misfolding and cell death. Since pheS is removed following successful Cre-catalysed circularisation, the supplementation of 4-CP in the growth medium selects for correctly assembled BGC-containing plasmids. The use of 4-CP is additionally advantageous for being far cheaper than the β-galactosidase required for LacZ blue/white screening.

Using our optimised CAPTURE method we cloned the complete amb BGC and verified the final construct by PCR and restriction enzyme digestion of the purified plasmid (Supplementary Figure S1). We then conjugated the amb cluster into four Streptomyces strains commonly used for the heterologous expression of BGCs: Streptomyces avermitilis SUKA 17,^{11a, 19} Streptomyces albus Del14,²⁰ Streptomyces coelicolor M1152²¹ and Streptomyces venezuelae ATCC 15439.²² Exconjugants were fermented for 10 days, followed by solid phase extraction of the culture supernatants. Comparison of HPLC traces of each amb cluster-containing heterologous host to its corresponding empty-vector control revealed the production of new metabolites in S. avermitilis SUKA 17::amb, S. albus Del14::amb, and S. coelicolor M1152::amb, but not S. venezuelae::amb (Figure 2A.).

In all three hosts where new peaks were apparent in HPLC traces, new [M+H]⁺ ions 947.5197 and 933.5038 were also observed, and these ions co-eluted with their putative dehydroxylated congeners 931.5248, and 917.5091 (Supplementary Figure S4). Surprisingly, only the organic extract of S. avermitilis SUKA-17::amb exhibited calcium-dependent antibiotic activity against Bacillus subtilis (Supplementary Figure S6). Manual comparison of the organic extracts of the three producing heterologous hosts by HPLC-MS revealed the additional ions [M+H]⁺1481.6907 (ambocidin A) and 1467.6749 (ambocidin B)–along with the respective dehydroxylated congeners 1465.6958 (ambocidin A1) and 1451.6801 (ambocidin B1)–exclusively present in S. avermitilis SUKA-17::amb (Figure 3A, Supplementary Figure S4).

We hypothesised the larger ions we observed (ambocidins A/A1 and B/B1) correspond to the mature, bioactive product of the BGC, while the smaller ions (ambocidins C−F) detected in S. albus Del14 and S. coelicolor M1152 are degradation products. In support of this hypothesis, we found that ambocidin A is actually detectable in S. albus Del14::amb and S. coelicolor M1152::amb, but only up to three days post culture inoculation, after which it is no longer present (Figure 3D, Supplementary Figure S10). To determine if this degradation is enzymatic, we incubated purified ambocidin A/A1 with concentrated supernatants or cell lysates from S. albus Del14, S. coelicolor M1152 and S. avermitilis SUKA1 cultures. Only S. albus Del14 or S. coelicolor M1152 culture supernatants were able to degrade ambocidin A/A1 to ambocidin C/C1 (Figure 3C, Supplementary Figure S7). Boiling the culture supernatant (Figure 3C, Supplementary Figure S7) or adding a broad spectrum metalloprotease cocktail (Supplementary Figure S8), prevented ambocidin A/A1 degradation, clearly implicating a protease enzyme that was either excreted, or released due to cell lysis.

We isolated ambocidin A/A1/B/B1 from S. avermitilis SUKA-17::amb using calcium-induced precipitation and reverse phase chromatography. The amino acid composition and connectivity for each metabolite was elucidated by NMR spectroscopy and LC–MS/MS (Supplementary Methods S12, Supplementary Figure S2, Supplementary Figures S13–S33, Supporting Information Tables S2–S3). With the exception of β-carbon of the β-hydroxy aspartic acid residue and the penultimate carbon in the lipid substituent, stereochemistry of all chiral centres was determined by comparison to reference standards using C3-Marfey's analysis²³ (Supplementary Figure S5, Supporting Information Table S4). Ambocidin A/A1/B/B1 are cyclic lipodepsipeptides comprised of 13 amino acids with a central 28-membered non-homodetic ring (Figure 2B). Ambocidin A and B contain the rare amino acid l-N^ϵ-hydroxy-arginine at position seven, while ambocidin A1 and B1 simply have unmodified l-arginine. The ambocidins are lipidated via the amino group of their N-terminal glycine residue. In the case of ambocidin A and A1, the attached fatty acid is (2Z,4E)-8-methyldeca-2,4-dienoic acid. The fatty acid attached to ambocidin B and B1, (2Z,4E)-8-methylnona-2,4-dienoic acid, is one carbon shorter.

Having accurate masses for the products of the amb BGC allowed us to specifically search LC–MS/MS chromatograms of extracts from S. ambofaciens for evidence of ambocidin production. This analysis confirmed that production of ambocidins A–F is repressed in S. ambofaciens. Interestingly, integration of a second copy of the entire amb BGC into the chromosome of S. ambofaciens appears to partially bypass repression, leading to production of ambocidin A, albeit at levels approximately 40 fold lower than seen with heterologous expression (Supplementary Figure S35).

Bioactivity and Structure-Activity Relationship

Ambocidin A is active against the Gram-positive bacteria Bacillus subtilis, Staphylococcus aureus, methicillin-resistant Staphylococcus aureus (MRSA), and vancomycin-resistant Enterococcus faecium (VRE). The minimal inhibitory concentration (MIC) values for ambocidin A range from <0.031 to 1 μg/mL in broth dilution assays (Table 1). In contrast, ambocidin A1 exhibits a 8–16 fold increase in MIC against all tested Gram-positive bacteria except B. subtilis, clearly implicating the N^ϵ−OH group in Arg⁷ as a key feature for bioactivity. Ambocidin B has similar MIC values to ambocidin A1, while ambocidin B1 has greatly reduced potency compared to ambocidin A (>64 times less active against S. aureus and MRSA) Taken together, these data indicate that both fatty acid length and the hydroxylation state of Arg⁷ are important factors for ambocidin bioactivity, with ambocidin A being the most potent congener. None of the ambocidins are active against any of the Gram-negative bacteria tested, nor did any exhibit toxicity against HCT-116 human colon carcinoma cells (MIC>32 μg/mL). Ambocidin A was assessed for haemolytic activity at concentrations up to 200 μg/mL, and did not result in any detectable haemolysis (Supplementary Figure S36). The activity of ambocidin A was found to be calcium dependent when tested against a panel of Gram-positive pathogens at a range of calcium concentrations, with higher concentrations of calcium resulting in lower MIC values against all strains tested (Figure 4A, Supporting Information Table S9)

Determination of the Mechanism of Action of the Ambocidins

Cyclic lipopeptide antibiotics typically target components of the Gram-positive bacteria membrane such as lipid II,^11b undecaprenyl monophosphate (C55-P),²⁴ or undecaprenyl diphosphate (C55-PP).^24a To determine the mechanism of action of the ambocidins, we used HPLC-MS to monitor metabolic changes in S. aureus upon exposure. We observed the accumulation UDP-MurNAc-pentapeptide with ambocidin A treatment, indicating inhibition of cell wall biosynthesis (Figure 4A). Next we monitored membrane disruption by using SYTOX green dye and found that, unlike daptomycin, ambocidin A/A1 does not disrupt the MRSA cell membrane (Figure 4B). To determine which cell wall component ambocidin A targets, we supplemented purified lipid II, C55-P or C55-PP, into B. subtilis cultures, followed by challenging with ambocidin A/A1. The addition of lipid II at 5 : 1 molar ratio completely rescued B. subtilis from ambocidin-induced antibiosis. In contrast, C55-P and C55-PP at similar molar ratios had no effect on antibiosis (Figure 4D). The clinically used antibiotic vancomycin also targets lipid II, specifically binding to its d-ala-d-ala motif. Vancomycin-resistant Enterococcus faecium (VRE) is altered at this motif through a missense mutation, significantly attenuating vancomycin affinity. That the ambocidins all have potent activity against VRE indicates that they use a inhibition mechanism that does not bind the d-ala-d-ala. Furthermore, we found that a 200-fold molar excess of diacetyl-lys-d-ala-d-ala-OH (which mimics the d-ala- d-ala motif on lipid II) has no effect on ambocidin A bioactivity, whereas vancomycin is antagonised by a 12-fold molar excess (Supporting Information Table S5), providing further evidence that the ambocidins bind lipid II at a different location from that of Vancomycin. Taken together, our experiments demonstrate that the ambocidins inhibit cell wall biosynthesis–without causing membrane disruption–by antagonising lipid II at a site other than the well-established d-Ala-d-Ala motif targeted by vancomycin.

The effectiveness of the clinically used lipopeptide antibiotic daptomycin is blunted in the presence of serum or pulmonary surfactant.²⁵ To determine if ambocidin A is similarly affected, we measured the MIC of ambocidin A against B. subtilis in the presence of each of these. The MIC only increased two-fold in the presence of 10 % human serum, and four-fold with 1 % pulmonary surfactant. In contrast, the daptomycin MIC increases 8-fold when incubated with 10 % human serum and more than 256-fold increase with 1 % pulmonary surfactant (Supporting Information Table S6). Ambocidin A is therefore only mildly affected by these clinically relevant physiological factors.

Assignment of AmbT as an Arginine N^ϵ-Hydroxylase

To understand the biosynthetic origin of the important bioactivity-enhancing N^ϵ-hydroxyarginine in the ambocidins, we performed a targeted gene deletion. Among the genes present in the amb gene cluster, we identified ambT as the most likely candidate for catalysing arginine hydroxylation. The AmbT protein was identified by the built-in pHMM scan functionality of antiSMASH 6.0 as belonging to the YqcI/YcgG family of heme-dependent oxygenases (e-value for hit to PF08892=9.1e–57), members of which have been shown to catalyse N^ϵ-hydroxylation of arginine in the biosynthesis of miharamycin²⁶ and argolaphos²⁷ (Figure 5A). To confirm the role of ambT in arginine hydroxylation we used a modified form of CAPTURE to disrupt ambT with a kanamycin-resistance cassette in-vitro, creating amb::ΔambT. Vexingly, despite repeated attempts to conjugate ambΔambT into S. avermitilis we were unable to obtain any exconjugants. As a workaround, we delivered amb::ΔambT into S. coelicolor M1152, which yielded viable exconjugants. HPLC-HRMS analysis of the resulting fermentation extracts showed that production of N^ϵ-hydroxyarginine-containing ambocidins (ambocidins C/D) was completely abolished and that the non-hydroxylated congeners (ambocidins C1/D1) are the new major products (Figure 5B, Supplementary Figure S9). The abolished production of N^ϵ-hydroxyarginine in the hosts containing amb::ΔambT clearly demonstrates that AmbT is the enzyme responsible for catalysing this important transformation in ambocidin biosynthesis.

YqcI/YcgG Family Arginine N^ϵ-Hydroxylases Markers for Structural Novelty and Bioactivity

N^ϵ-hydroxyarginine is a rare structural motif in microbial natural products. By searching the Dictionary of Natural products for this moiety, we were able to identify just four compound series’: the miharamycins,²⁶ asterobactins,²⁸ pentaminomycins²⁹ and argolaphos.²⁷ Addition of the ambocidins to this group brings the total number of reported instances of this motif to five, with three of these possessing antimicrobial activity. This result suggested to us that proteins in the YqcI/YcgG family might serve as a useful biosynthetic marker for both structural novelty, and clinically relevant bioactivity. To investigate this possibility further, we conducted a search of all sequenced bacterial genomes available in NCBI for BGCs encoding large non ribosomal peptides (>=8 Adenylation Domains) and at least one ambT homologue using the Boolean search functionality of the antiSMASH database. This search returned 124 BGCs, the majority of which were similar or identical to the BGCs siderophores plantaribactin, gramibactin and megapolibactin. Each of these BGCs encode compounds that contain the amino acid graminine, a moiety which is derived from arginine via an as yet unknown biochemical transformation.³⁰ By further refining our search to exclude siderophores, followed by manual removal of BGCs that appeared to be identical or near-identical to those for previously described compounds, we were left with ten unique BGCs (Figure 5C, Supporting Information Table S8). Among these ten BGCs, one appeared to encode a compound that was very similar, or perhaps identical, to the ambocidins (Supplementary Figure S12) and four further BGCs had a calcium dependent antibiotic as their closest match among all characterised BGCs in the MiBiG database (Supporting Information Table S8). We posit that the BGCs identified in this analysis, particularly those whose closest characterised relative encodes an antibiotic, might serve as a fruitful targets for future genome mining efforts.