Discovery of the Biosynthetic Machinery for Stravidins, Biotin Antimetabolites

Abstract

Stravidins are peptide antibiotics produced by Streptomyces spp. Their antibacterial activity derives from an unusual amiclenomycin monomer, the warhead that inhibits biotin biosynthesis. Despite being discovered over five decades ago, stravidin biosynthesis has remained a mystery. Using our “metabologenomics” platform, we discover new stravidin analogues and identify the novel biosynthetic machinery responsible for their production. Analysis of the newly identified biosynthetic gene cluster (BGC) indicates the unusual amiclenomycin warhead is derived from chorismic acid, with initial steps similar to those involved in p-amino phenylalanine biosynthesis. However, a distinctive decarboxylation retains the nonaromatic character of a key ring and precedes a one-carbon extension to afford the warhead in its bioactive, untriggered state. Strikingly, we also identified two streptavidin genes flanking the new stravidin BGC reported here. This aligns with the known synergistic activity between the biotin-binding activity of streptavidin and the stravidins to antagonize both biotin biogenesis and bacterial growth.

Graphical Abstract

Biotin is a vital cofactor for key central metabolic enzymes in all domains of life.¹ De novo biotin biosynthesis is highly conserved among producing organisms, particularly for the enzymes catalyzing the bicyclic ring assembly (Figure 1A, (1)).² Within this pathway, the bioA gene, encoding for 7,8-diaminopelargonic acid (DAPA) aminotransferase (BioA), is essential for survival and persistence of mycobacteria,³ and therefore, BioA is a validated antituberculosis drug target.⁴

Figure 1.

BioA inhibition by amiclenomycin (Acm). (A) Acm (4) (red) inhibits BioA (green), a PLP-dependent transaminase, by forming an aromatized adduct with PLP (blue), which binds tightly to the enzyme active site.³¹ (B) Structures of known Acm-peptides and the new acetylated stravidin analogues (5 and 6) identified in this study (stereochemistry is displayed according to previously published structure³²). (KAPA, 7-keto-8-aminopelargomic acid; DAPA, 7,8-diaminopelargonic acid; PLP, pyridoxal-5′-phosphate; SAM, S-adenosyl-l-methionine; AMTB, S-adenosyl-2-oxo-4-thiomethylbutyrate; Ac Acm, N′-acetyl-amiclenomycin.)

During the “golden era” of antibiotic discovery decades ago, the biotin binding protein streptavidin and a mixture of the antibiotin dipeptides stravidin S2 (MSD-235 S2) (2) and stravidin S3 (3) (Figure 1B) were discovered from Streptomyces avidinii as the components of an intriguing antibiotic complex.^5,6 These components act synergistically to achieve biotin deficiency in the target microbe causing growth inhibition. Among the two stravidins, 3 was the major dipeptide.⁶ Stravidins contain the unique amino acid monomer amiclenomycin (Acm) (4) (Figure 1A) which was later individually isolated from Streptomyces lavendulae.⁷ Additional Acm-containing di- and tripeptides were also reported from Streptomyces venezuelae (Figure 1B).⁸

The bioactivity of the stravidins results from their Acm monomer, which antagonizes biotin biogenesis through BioA enzyme inhibition. This warhead binds to the substrate binding site of BioA and forms an irreversible adduct with the cofactor pyridoxal 5′-phosphate (PLP) through its characteristic amino-substituted 1,4 cyclohexadiene functionality, halting enzyme activity (Figure 1A).⁹ The bioactivity of 4 appeared to be specific to mycobacteria.⁷ However, Poetsch et al. showed that Acm-containing peptides are also active against Gram-negative bacteria,¹⁰ where they (but not the free Acm amino acid) are efficiently transported by peptide permeases. This denotes a clever prodrug strategy to target additional strains using this warhead, poised to become aromatic upon finding its target (Figure 1A). Despite these interesting findings, a clear understanding of the biosynthesis of amiclenomycin and stravidin in Streptomyces remains unclear. Recently, we reported a new approach combining bacterial metabolomics and genomics for the large-scale discovery of secondary metabolites and their associated biosynthetic gene clusters (BGCs).¹¹ Here, we utilize this “metabologenomics” platform to provide the first insight into the biosynthetic machinery for the production of 4 and discover new stravidin analogues.

Within a typical metabologenomics data set from many hundreds of Streptomyces strains, we identified an interesting case of a natural product/BGC pairing exhibiting a strong correlation (i.e., the same metabolite was detected by LC-MS above the set threshold levels¹¹ in 9 of 11 strains harboring the BGC). One specific ion detected in LC-MS data from extracts of Streptomyces sp. NRRL S-98 had a m/z value of 366.2383 (C₁₉H₃₂O₄N₃) and correlated to an orphan BGC annotated simply as “BGC_Others_61”.¹² Tandem MS (MS/MS) analysis of this species (Figure S1A) showed fragment ions characteristic of the stravidins, including one at m/z 307.2012 (C₁₇H₂₇O₃N₂, −0.4 ppm error, [M – H₂ – NHCOCH₃ + H]⁺). Stable isotope experiments using labeled amino acids were consistent with the presence of N-Me Ile in this molecule (Figure S2), which was supported by the presence of a characteristic and abundant signal for the immonium ion of N-Me Ile at m/z 100.1117 in the MS/MS spectrum (Figure S1A). Because of the novelty of the parent ion mass as well as the highly correlated BGC, we isolated this molecule for NMR analysis from another producing strain (Streptomyces sp. XY533) which showed higher metabolite expression.

Using NMR data to elucidate and verify the structure of this new stravidin, we found agreement among the MS results from stable isotope labeling, MS/MS analyses, and the 1D and 2D NMR data in DMSO-d₆ (Table S1), all confirming the presence of an N-Me Ile monomer. Notably, the secondary amine proton signal in this amino acid was absent and the N-Me group appeared as a singlet, but the presence of an additional N-substitution was excluded based on MS and NMR analysis. Further structural information provided by NMR spectra indicated the presence of two highly equivalent double bonds (δ_H ~ 5–6 ppm, δ_C ~ 125–130 ppm), suggesting the presence of a substituted 1,4-cyclohexadiene ring. Indeed, the COSY spectrum showed a strong correlation between the two protons at δ_H 2.62 and 4.74 ppm (positions 1 and 4 in the cyclohexadiene ring) (Figure S5), arising from the large long-range (⁵J) homoallylic coupling which is commonly seen for 1,4-cyclohexadienes.¹³ Additionally, an N-acetyl functionality was clearly observed within this monomer. Finally, a spin system was constructed where a 4-acetylamino-2,5-cyclohexadienyl moiety is separated from the α-proton at δ_H 3.88 ppm by two methylene groups, designating the presence of the unusual amino acid monomer N′-acetyl Acm. To satisfy all constraints of atom composition from the MS data, compound 5 was assigned as stravidin S5; an N′-acetyl analogue of 3 (Figure 1B).

Because of the co-occurrence of 2 and 3 in Streptomyces spp., we examined further the producing strains here and the metabologenomics correlations for a similar analogue of 2. Indeed, an ion at m/z value of 352.2226 (C₁₈H₃₀O₄N₃, −0.5 ppm error, [M + H]⁺), corresponding to the N′-acetyl stravidin S2 analogue, was present among the top five correlations to BGC_Others_61. MS/MS spectra (Figure S1A) showed the same fragmentation pattern as with 5, where two abundant fragment ions are an ion with a mass loss of 59 Da at m/z 293.1849 (C₁₆H₂₅O₃N₂, −0.4 ppm error, [M – H₂ – NHCOCH₃ + H]⁺), representing aromatized fragment ion of stravidin S2 and an abundant N-Me Val immonium ion at m/z 86.0964. The presence of an N-Me Val and not Ile in this molecule was confirmed by stable isotope labeling using labeled Val (Figure S2). Accordingly, this molecule was assigned as the new analogue stravidin S4 (6) (Figure 1B).

The evidence of the correlation between these new analogues and the gene cluster “BGC_Others_61” is strong by virtue of their co-occurrence in nine strains within our integrated metabologenomics data set. Beyond these strains, we searched for this BGC in additional members of this gene cluster family (GCF)¹² using BlastP of specific proteins and individual genome searches through EzBiocloud;¹⁴ such analysis showed that BGC_Others_61 is conserved in a total of 33 Streptomyces strains (Table S3). With the assignment of their BGC, we propose a biosynthetic pathway for 4–6 based on homology of the encoded gene products with proteins of known function (Figure 2, Table S2).

Figure 2.

Gene cluster responsible for stravidin production and the proposed biosynthetic pathway. (A) ORFs in BGC_Others_61 producing 5 and 6 and their putative functions (color coordinated with the legend). Two streptavidin genes (yellow) are found in close proximity to the BGC. (B) Proposed biosynthetic pathway for 5 and 6, shown as enzymatic reactions catalyzed by putative proteins from BGC_Others_61 (color coordinated with ORFs shown in panel (A); 2-MCM, 2-(4-methyl cyclohexa-2,5-dien-1-amine) malate; PDX, prephenate decarboxylase).

BGC_Others_61 (~16 kb) encloses 14 open reading frames (ORFs) (svnA–svnN; Figure 2A). We hypothesize that Acm biosynthesis starts with SvnA, a putative 3-deoxy-D-arabino-heptulosonic acid 7-phosphate (DAHP) synthase, which is known to catalyze the first step in the shikimate pathway to provide chorismic acid (7) as a precursor for several primary and secondary metabolites (Figure 2B).^15,16 By virtue of its stronger sequence identity (Table S2), we assigned SvnA as a putative class II DAHP synthase. Class I DAHP synthases are more common in bacteria and involved in primary metabolism to produce aromatic amino acids; they are usually under a strict feedback regulation by those amino acids or pathway intermediates. However, class II DAHP synthases are more widespread in plants, activated by tryptophan and not inhibited by aromatic amino acids.¹⁶ Bacteria utilizing chorismate in secondary metabolism usually harbor a class II DAHP synthase within their chorismate-derived secondary metabolite BGCs. This strategy, apparently used by the stravidin producers reported here, avoids negative feedback regulation of aromatic amino acids and provides an efficient shunt for the activity of the shikimate pathway into secondary metabolism.^17,18

The two genes svnN and svnK putatively encode an aminodeoxychorismate synthase and a 4-amino-4-deoxychorismate mutase with 46% and 34% sequence identity to the chloramphenicol biosynthetic enzymes CmlB and CmlD in S. venezuelae,¹⁷ respectively (Table S2). These are known to catalyze two consecutive steps converting 7 to an amino-deoxyprephenic acid intermediate (9) (Figure 3). Although these steps are shared in the biosynthesis of p-amino phenylalanine, a chloramphenicol precursor, the unique structure of 4 must involve additional characteristic steps which discriminates this pathway from other established ones utilizing chorismate (Figure 3). Typically, in the biosynthesis of aromatic amino acids, chorismate is converted into prephenate which is then subjected to aromatic decarboxylation followed by transamination. A similar mechanism involving aromatization occurs in the biosynthesis of p-amino phenylalanine after chorismic acid is converted into 4-amino-4-deoxychorimsate. Since 4 retains the essential 1,4-cyclohexadiene functionality in the prephenate ring, an exceptional nonaromatic decarboxylation must be occurring instead of the canonical pathways. In fact, the rare nonaromatic decarboxylation of prephenate has been previously encountered by the characterization of prephenate decarboxylase enzymes (PDXs), which produce the unusual amino acids anticapsin (BacA),¹⁹ choi (AerD),²⁰ tetrahydro-phenyalanine (SalX),²⁰ (Figure 3) and dihydro-phenyalanine (Plu3043).²¹ All of these enzymes uniquely decarboxylate prephenate at C-4 and protonate C-6 to produce the endocyclic diene dihydrohydroxy-phenyl-pyruvate (en-H₂HPP (15)). This intermediate is then subjected to allylic isomerization, enzymatically or non-enzymatically, to produce the exocyclic conjugated diene (ex-H₂HPP (16)) (Figure 3) as the thermodynamic product.²² Given that the ring system of 4 does not appear to originate from this conjugated diene system, we propose the action of a novel PDX within the stravidin pathway that retains the 1,4-cyclohexadiene functionality via a noncanonical previously unknown mechanism (Figure 3, bottom). After annotation of proteins from BGC_Others_61 based on assigned functions of homologues (Table S2), SvnJ remains the only protein with an unknown function. Accordingly, svnJ appears to be the best candidate to encode the proposed novel PDX.

Figure 3.

Comparison of different pathways for the production of amino acid monomers from chorismate. Chorismic acid is converted to prephenic acid (upper pathway) by the action of chorismate mutases (e.g: AroQ, PheA, TyrA). The canonical aromatic decarboxylation produces aromatic amino acids (prephenate dehydratase (Pht) for Phe and prephenate dehydrogenase (Phd) for Tyr). An unusual nonaromatic decarboxylation of prephenate is catalyzed by a group of prephenate decarboxylases (PDXs) which include BacA (anticapsin in bacilysin), SalX (tetrahydro-phenylalanine (H₄Phe) in salinisporamide) and AerD (Choi in aeruginoside 126A). In secondary metabolism, chorismic acid is converted to 8 by a 4-amino-4-deoxychorismate synthase (such as CmlB (chloramphenicol) and SvnN (stravidin)). Similar to the upper pathway, 8 rearranges to 9 by the action of a mutase (CmlD (chloramphenicol) and SvnK (stravidin)). In chloramphenicol biosynthesis, a typical aromatic decarboxylation is catalyzed by CmlC, followed by an aminotransferase to give the monomer p-amino phenyl alanine. However, for Acm (4), a unique nonaromatic decarboxylation retains the 1,4 cyclohexadiene functionality as in the prephenate skeletal structure, and we propose a novel PDX is catalyzing this step. Unusual nonaromatic decarboxylation pathways are highlighted.

By comparing the skeletal structure of intermediate 10 to 4 (Figure 2B), it is clear that the latter has a one-carbon-extended side chain. Remarkably, within this cluster, the three putative proteins SvnI, SvnE, and SvnF show homology to LueA, LueC, and LeuB from different strains, with 42, 45, and 42% sequence identity, respectively (Table S2). The latter enzymes are responsible for one carbon elongation in Leu biosynthesis, and a set of homologous genes (hphA, hphCD and hphB) are also known to extend the Phe skeletal structure by one carbon to produce homophenylalanine.²³ Hence, SvnI is assigned to condense 10 with acetyl-CoA (Figure 2B), affording 11, a C2 branched form of malate: [2-(4-methyl cyclohexa-2,5-dien-1-amine) malate], or 2-MCM for short. Isomerization of 11 to 3-MCM (12) occurs by the putative isomerase SvnF. Compound 12 is then oxidized by the putative dehydrogenase SvnE, and a spontaneous decarboxylation²³ produces 13. Lastly, the putative product of the aminotransferase gene svnD, with 32% sequence identity to the E. coli aminotransferase IlvE, introduces the amino acid functionality to yield the unusual monomer amiclenomycin (4) (Figure 2B). The new combination of enzymatic steps from primary metabolism with the specialized chemistry of secondary metabolism creates an amazing blend of transformations all encoded by this new BGC.

In order to produce the novel N′-acetylated analogues reported here, an additional N′-acetylation of 4 is needed. Indeed, BGC_Others_61 contains two candidate GNAT N-acetyltransferase genes, svnG and svnM, one of which would fulfill what we propose as the final step to produce monomeric N′-acetyl Acm. These two genes are conserved in all 33 strains harboring this gene cluster. Formally, the timing and role of this new N-acetylation in both the biosynthesis and bioactivity are not certain. However, N-acetylation is a protective modification which could both shield stravidin warhead and function as a resistance mechanism as in other antibiotic systems,^24,25 particularly since all 33 strains also contain the biotin biosynthetic cassette bioFABD. Furthermore, the putative transporter SvnB is anticipated to aid in stravidin resistance for producing strains, as it shares 30% identity with the lincomycin pump, LmrA. To this end, we were interested in evaluating whether the original S. avidinii strain,⁵ from which stravidins S2 (2) and S3 (3) were initially isolated and characterized, would also produce the N-acetylated analogues 5 and 6. Accordingly, we obtained this strain through ATCC for LC-MS/MS analysis of culture extracts. Strikingly, and in alignment with our observations, we detected both 5 and 6 as the major metabolites produced by S. avidinii (Figure S3), which reflects the importance of this modification for the producing strain.

The peptide bond formation in the dipeptide bacilysin is catalyzed by the L-amino acid dipeptide ligase BacD, which catalyzes the condensation of L-Ala with the unusual monomer anticapsin.^27,28 In a similar fashion, we propose that the dipeptide feature in stravidins is introduced through a dipeptide ligase enzyme encoded by svnC. Indeed, Blast searches showed that SvnC is an ATP-dependent carboxylateamine ligase (glutamate-cysteine ligase) homologue with 54% sequence identity, which highlights this protein as a possible candidate for the proposed role. As with BacD, a relaxed substrate specificity for this putative ligase must be responsible for selecting either Val or Ile as the N-terminal substrate in stravidins (Figure 1B). Furthermore, svnH encodes a putative class I SAM-dependent methyl-transferase (Table S2), which explains the presence of the N-methylated Val or Ile in the stravidins.

Stravidins were initially discovered as part of a synergistic antibiotic complex with the biotin-binding protein streptavidin.^5,29 While studying this protein and its production by different Streptomyces strains, Bayer et al.³⁰ noted that streptavidin occurs only in Acm-producing strains. Moreover, they discovered two genes in those strains expressing two similar streptavidin proteins. Strikingly, we identified two streptavidin genes in close proximity to the proposed cluster (Figure 2A), which aligns with these previous observations and further supports this compound/BGC correlation. Notably, both streptavidin genes were conserved in all 33 strains harboring this cluster (Table S3).

For additional validation of the stravidin BGC assigned here, we heterologously expressed this BGC in S. lividans 66, which lacks this cluster and produces neither 4 nor stravidins. Thus, BGC_Other_61 from Streptomyces sp. NRRL S-98 was cloned using a newly developed CRISPR-Cas9 technology (Varigen Biosciences), and the recombinant DNA (SvnBAC) was subsequently conjugated into S. lividans 66 (Figure 4A). Culture extracts of the exconjugant S. lividans 66-SvnBAC were analyzed by UHPLC-HRMS/MS. Successfully, both compounds 5 and 6 were heterologously expressed and detected at a titer equivalent to the native producer (Figure 4B), with their identities also confirmed by MS/MS (Figure S1B).

Figure 4.

Heterologous expression of 5 and 6 in S. lividans 66. (A) Workflow for cloning BGC_Others_61 from Streptomyces sp. NRRL S-98 genome and heterologous expression in S. lividans 66. Extracted metabolites from the generated derivative S. lividans 66-SvnBAC were analyzed by LC-MS/MS for the presence of 5 and 6. Some graphic materials are provided by D. Mead. (B) S. lividans 66-SvnBAC produces both metabolites 5 and 6 with the same high abundance as the native producer Streptomyces sp. XY533. Retention times are shown in blue, while m/z values are shown in red. *The cloned gene cluster shows 100% similarity in the two strains Streptomyces sp. NRRL S-98 and Streptomyces sp. XY533.

As an initial evaluation of the impact of N-acetylation on stravidin bioactivity, we attempted to analyze the structures of stravidin S3 (3) and stravidin S5 (5) using the recently developed tool eNTRyway, which predicts the likelihood of a small-molecule to accumulate in Gram-negative bacteria based on molecular physicochemical properties.²⁶ As expected, by masking the primary amine functionality, the N-acetylated version 5 showed weaker probabilities of accumulation compared to the free-amine counterpart 3. However, further biological evaluation is essential for a conclusive comparison of the bioactivities of these analogues and to assess the possibility of in vivo deacetylation of 5.

The discovery of the stravidins ~60 years ago is inspiring for several reasons. First, it provided a unique structural scaffold produced by Streptomyces spp., complete with natural moieties serving a prodrug function. Second, the stravidin system also supplied an example antimetabolite strategy using a synergistic protein/small molecule complex to antagonize a specific metabolic pathway. Third and last, it allowed for the validation of BioA as a successful drug target for TB. The aforementioned points highlight the importance of this unique system to the biomedical and research communities. Here, we provided the first report of the molecular basis for the production of the warhead monomer (Acm), the dipeptide prodrugs (stravidins), and the antibiotic complex (stravidins with streptavidin). These findings fill an essential gap in the exciting investigation of the stravidins and enable synthetic biology approaches. Our findings could enable strain engineering to produce these compounds or analogues for the use as antibiotics or herbicides. Moreover, the new enzymology proposed for stravidin biosynthesis will enable new insights and extend options for chemoenzymatic conversions in the future.