A bifunctional nitrone conjugated secondary metabolite targeting the ribosome

Abstract

Many microorganisms possess the capacity for producing multiple antibiotic secondary metabolites. In a few notable cases, combinations of secondary metabolites produced by the same organism are used in important combination therapies for treatment of drug resistant bacterial infections. However, examples of conjoined roles of bioactive metabolites produced by the same organism remain uncommon. During our genetic functional analysis of oxidase encoding genes in the everninomicin producer Micromonospora carbonacea var. aurantiaca, we discovered previously uncharacterized antibiotics everninomicin N and O, comprising of an everninomicin fragment conjugated to the macrolide rosamicin via a rare nitrone moiety. These metabolites were determined to be hydrolysis products of everninomicin P, a nitrone linked conjugate likely the result of nonenzymatic condensation of the rosamicin aldehyde and the octasaccharide everninomicin F, possessing a hydroxylamino sugar moiety. Rosamicin binds the erythromycin macrolide binding site approximately 60 Å from the orthosomycin binding site of everninomicins. However, while individual ribosomal binding sites for each functional half of everninomicin P are too distant for bidentate binding, ligand displacement studies demonstrated that everninomicin P competes with rosamicin for ribosomal binding. Chemical protection studies and structural analysis of everninomicin P revealed that everninomicin P occupies both the macrolide- and orthosomycin-binding sites on the 70S ribosome. Moreover, resistance mutations within each binding site were overcome by the inhibition of the opposite functional antibiotic moiety binding site. These data together demonstrate a strategy for coupling orthogonal antibiotic pharmacophores, a surprising tolerance for substantial covalent modification of each antibiotic, and a potential beneficial strategy to combat antibiotic resistance.

Graphical Abstract

INTRODUCTION

Individual microbial genomes typically encode for the biosynthesis of several secondary metabolite families.¹ In rare cases however, has it been experimentally demonstrated that two secondary metabolites produced by the same organism function in tandem to impart their mode of action.² For example, pristinamycin is a two-component antibiotic biosynthesized by Streptomyces pristinaespiralis. Pristinamycin 1A is a macrolide component that binds the 23S portion of the 50S ribosome enhancing by 100-fold the binding affinity of the pristinamycin IIa, a depsipeptide that binds in a nearby site.³ Streptomyces clavuligerus is a producer of the β–lactam cephamycin C, a cephalosporin inhibitor of cell wall biosynthesis. Additionally, a separate biosynthetic gene cluster in S. clavuligerus produces the oxazolidine β–lactam clavulanic acid, a mechanism-based inhibitor of β–lactamase.⁴ In both cases, the combination of multiple antibiotic pharmacophores results in powerful clinical tools to combat antibiotic resistance. However, while resistance is replete in the environment from which antibiotic producing microorganisms have been isolated, examples of combination antibiosis are quite uncommon, these examples currently being among the exceptions to this rule.

Micromonospora carbonacea var. aurantiaca is known to produce the everninomicin (Evn) family of antibiotics (Evn D-G). Everninomicins are orthosomycin oligosaccharide antibiotics characterized by the oxidation of multiple glycosidic linkages into spirocyclic ortho-δ-lactone (orthoester) moieties.⁵ The heptasaccharide avilamycin orthosomycins are used clinically as antibiotics in animals, and the related octasaccharide Evn A (Ziracin) advanced to phase III chemical trials for the treatment of antibiotic resistant infections before being withdrawn for unstated pharmacological concerns. Subsequent structural studies have revealed a novel binding site and provided new insight into mechanism, binding interactions, and resistance.^6–7 In the course of functional genetic studies aimed at generating improved orthosomycin analogs, we discovered and characterized a previously unreported antibiotic Evn P, comprised of a deoxysugar nitrone linked fusion of the octasaccharide Evn F to an unrelated glycosylated polyketide macrolide antibiotic, rosamicin (Rosa). Using a combined genetic, biochemical, and cryo-electron microscopy (cryo-EM) structural approach, we demonstrate the productive binding locations and geometries of each pharmacophore in two distant sites within the 23S ribosome and investigate the consequences of antibiotic bifunctionality on antibiotic activity and resistance.

RESULTS

Generation and analysis of targeted gene replacement mutants.

Functional analysis of the Evn gene cluster (Supplementary Figure 1A) and Evn analog biosynthesis were performed by generating targeted gene knock outs of two oxidase genes, evdN1 and evdO1, from the Evn gene cluster (evd) encoded pathway using a two-step lambda-RED mediated gene replacement method.^8–9 The genes of interest were individually replaced on a cosmid containing most of the evd gene cluster in E. coli. The modified cosmids were then transformed into M. carbonacea var. aurantiaca using microporous conjugation, wherein two rounds of homologous recombination resulted in precise replacement of the target gene with the apramycin resistance cassette. Double crossover mutants were confirmed by PCR and Southern hybridization analysis and polar effects were ascertained by complementing mutations with wild-type genes in an integrative vector (Supplementary Figure 1B). Extracts from production cultures of each gene-replacement mutant were generated via solid phase extraction, silica gel fractionation, and analyzed by LC/MS.

Analysis of products of gene replacement mutants.

As shown in Figure 1A, analysis of extracts from the A-ring nitrososynthase gene replacement mutant ΔevdN1::aac(3)IV revealed loss of production of full-length Evn D–G and accumulation of a compound we termed Evn M. Genetic complementation with evdN1 did not result in restored production suggesting that, although the replacement was precise as indicated by PCR and Southern hybridization analysis (Supplementary Figure 2), polar effects introduced from the genetic replacement construct resulted in loss of activity of downstream genes (Supplementary Figure 1A). We previously reported successful replacement of the oxygenase evdO1 with the apramycin cassette resulting in the loss of production of all everninomicins (ΔevdO1::aac(3)IV).⁵ However we were unable to successfully perform genetic complementation due to inefficiency of the original transformation protocol and lack of an appropriate gene complementation vector. With our optimized vector, evdO1 genetic complementation resulted in the detection of two new truncated molecular weight halogenated metabolites termed Evn N and O (Figure 1A).

Figure 1.

Analysis of genetic replacements of evdN1 and evdO1. (A) LC/MS analysis of wildtype M. carbonacea var. aurantiaca, gene replacements evdN1 (ΔevdN1::ac(3)IV), evdO1 (ΔevdO1::aac(3)IV), and genetic complementation of evdO1 gene replacement (ΔevdO1::aac(3)IV GC). The chromatogram shows summed ion intensities in negative mode (orange) and positive mode (black) for Evn D–G and new metabolites. Chromatograms are extracted ion currents for positive (M+H) and negative (M−H) ions for the indicated compounds. Masses of all everninomicins in this study are found in Supplementary Table 1. (B) Structures of wildtype metabolites Evn A–G and deletion metabolite Evn H.

Structure elucidation of Evn M, product of the nitrososynthase gene replacement mutant.

Fragmentation analysis of accumulated intermediate Evn M (Figure 1A and Supplementary Figure 4) demonstrated that it is identical with the previously reported analog previously referred to as ‘everninomicin-2’ (Figure 2A), a minor by-product of large scale fermentation that lacks the A-ring nitrosugar and is not present in the wild-type strains under our production conditions.¹⁰

Figure 2.

evdO1 reveals nitrone linked macrolide congeners. (A) Structure of Evn M from evdN1 gene replacement strain. (B) Formation of nitrone moiety from the hydroxylamine of Evn F and aldehyde of Rosa. (C) Antibiotics Evn N and Evn O, identified in extracts of genetic complementation mutants of ΔevdO1 are likely hydrolysis products of Evn P, subsequently isolated from the wildtype M. carbonacea var. aurantiaca.

Identification of a nitrone linked macrolide congener.

Interpretation of the mass fragmentation pattern of Evn N (Supplementary Figure 3), the remaining uncharacterized metabolite observed in ΔevdO1 genetic complementation strain, suggested that it possessed an A₁-A-B ring substructure, but remaining fragments were inconsistent with orthosomycins described to date. Extracts from scale-up fermentation of the wildtype strain were purified by Sephadex LH-20 fractionation, followed by preparative HPLC isolation of metabolites putatively containing the A₁-ring chromophore of 297 nm.

Two closely eluting abundant species were identified, Evn N and Evn O, and structures were determined via 2D-NMR (Figure 2C and Supplementary Figures 5–17, Supplementary Tables 2 and 3). Heteronuclear single quantum coherence (HSQC) spectra reveal only three anomeric centers, two of them consistent with the A and B-ring sugars via COSY and HMBC analysis. Analysis of the spin system connected to the B-ring via HMBC via C-10 revealed the gluconate congener of d-olivose. The third anomeric center was determined to be part of a spin system resolving into a desosamine sugar and connected to a 10-proton spin network flanked by an α–β unsaturated ketone and apparent macrocyclic polyketide ester. The alkoxyl methine on the other side of the putative macrolide ester is part of a spin system containing an ethyl side chain and epoxy group, evidenced to be adjacent to the α–β unsaturated ketone via HMBC. Sub-structural database search of this fragment revealed it to be part of a likely Rosa type macrolide. This was consistent with reported NMR data with the exception of altered shift of the aldehyde carbonyl to 147 ppm and shift of C-24 of the A-ring branched amino sugar in the trisaccharide. The mass of Evn N and O require a conjugating nitrone moiety, the product of reaction of a hydroxylamine (e.g. Evn F) with the aldehyde of Rosa (Figure 2B). This is consistent with the chemical shift of C-47, perturbation of C-24 and C-46, and NOESY data. Due to the anisotropic effect of the nitrone linkage, the E configuration of the nitrone is expected to be downfield of the Z configuration.¹¹ As the nitrone H-47 is 7.42 ppm in Evn O and 7.36 in Evn N, the configuration of Evn O and N are likely E and Z, respectively.

Speculating that Evn N and O were the likely orthoester hydrolysis products of a larger conjugate, extracts of wildtype M. carbonacea var. aurantiaca were searched for masses consistent with full-length Evn conjugates to Rosa (Figure 2C). Correspondingly, in addition to detecting m/z for Rosa and Evn F, a metabolite with m/z = 2084.5 [M-H]⁻, named Evn P, a putative precursor to the hydrolyzed saccharide complex Evn N/O was identified (Supplementary Figure S26). Indeed, after incubation of Evn P in 90% H₂O/10% DMSO for 48 hours it spontaneously hydrolyzed to Evn N when exposed to normal culture conditions (Supplementary Figure 22). Preparative isolation of Evn P resulted in a single stereoisomer, and NMR analysis (Figures S18–21) confirmed a single diastereomer with a structure of the conjugate of Evn F and Rosa. Based on chemical shift, NOE, and bioactivity (see below), the nitrone in Evn P is likely in the Z configuration, (Supplementary Figures S18 – 22, Supplementary Tables 2–4).

Prompted by the observation of the Evn N/O/P Rosa conjugates, we determined the genomic sequence of M. carbonacea var. auriantiaca via a hybrid assembly strategy using Oxford Nanopore MinION and Illumina reads. The final assembly consisted of a single 7,528,335 bp circular chromosome, in contrast to previous reports of linear chromosomes in Micromonospora sp. Biosynthetic gene clusters were analyzed using antiSMASH,¹² identifying the Ever and Rosa gene clusters. The oligosaccharide and macrolide biosynthetic gene clusters are spaced 414 kbp apart. Similarly, we determined the sequence of M. carbonacea var. africans, in which Evn nitrones are not detect, and we determined that this genome contains no Rosa gene cluster.

Biological evaluation of new orthosomycins.

The everninomicin analogs were purified and tested for activity against S. aureus subsp. aureus Rosenbach in a broth microdilution assay (Supplementary Figure 23). Evn A (Ziracin) was used as a benchmark and possessed an MIC of 1 μg/mL, in line with literature values.¹³ Both the truncated Evn N and full-length everninomicin-rosamicin Evn P conjugates showed comparable potent activity against S. aureus with MICs of 1 and 2 μg/mL respectively. Truncated conjugate Evn O showed an increased MIC of approximately 10 μg/mL. Evn M was not isolated in sufficient quantities for activity analysis but its activity has been reported elsewhere and is comparable to Evn N/O and Evn P.¹⁴

Interaction of Evn P with the bacterial ribosome.

Orthosomycins, such as Evn A and avilamycin (Avn), have been shown to bind to a distinct site on the large ribosomal subunit,^15–17 which is located far (60 Å) from the peptidyltransferase center (PTC) active site and ribosomal tunnel where the majority of large-subunit targeting antibiotics interact, such as the chloramphenicols and macrolides (Figure 3A). Although the bound structure of the 16-membered macrolide Rosa has not been determined on the ribosome, competition assays suggest that Rosa has an analogous binding site within the ribosomal tunnel as other 14- and 16-membered macrolides, such as erythromycin (Ery) and spiramycin.¹⁸ Thus, the antimicrobial activity of Evn P could result from binding to either the Evn or Rosa binding site, by binding to both sites, or to an entirely new site. However, given the large distance between the Evn and macrolide binding sites on the large subunit, coupled with the short nitrone linker between the Evn F and Rosa moieties in Evn P, it is not possible for one molecule of Evn P to interact simultaneously with both the Evn and macrolide binding sites.

Figure 3.

Mapping the binding site of Evn P on the bacterial ribosome. (A) Overview of the binding sites of macrolides, such as Rosa (orange), and orthosomycins, such as everninomicin A (Evn A, purple) on the large ribosomal 50S subunit (grey). For reference, the P-site tRNA (green) is shown and central protuberance (CP) and L1 stalk are indicated. (B) Filter binding assay monitoring bound radiolabeled Ery in the presence of increasing concentrations of Evn H (purple), Rosa (orange), cold erythromycin (green) and Evn P (salmon). The error bars show the standard deviation from the mean of three independent experiments. (C–E) 23S rRNA nucleotides protected from chemical modification by DMS in the presence of (C) Evn H, (D) Rosa and (E) Evn P. (F) Electron density (grey mesh) for Evn P bound within the exit tunnel of the ribosome, with model for Rosa (orange) fitted. (G) Comparison of binding position of Rosa/Evn P relative to Ery (green) with 23S rRNA nucleotides A2058 and A2059 (blue) shown for reference. (H) Electron density (grey mesh) for Evn P bound to orthosomycin binding site of the ribosome, with model for Evn P (salmon) fitted. (I) Comparison of binding position of Evn P relative Evn A (purple) with 23S rRNA helices H89 and H91 (blue) and ribosomal protein L16 (green) shown for reference.

To determine whether Evn P can bind within the ribosomal tunnel analogous to macrolides, we employed competition assays using commercially available radiolabeled Ery. As expected, cold Ery and Rosa competed with radiolabeled Ery for ribosome binding in a dose-dependent manner (Figure 3B). As a negative control, we observed no competition between Ery and a full length Evn octasaccharide Evn H, consistent with their non-overlapping binding sites. Evn H, the D-ring des-C-3 methyl analog of Evn D was chosen as it was isolated by us in a parallel study, is equipotent to Evn D, and Evn D-G are not commercially available. By contrast, Evn P competed with Ery, such that 50% of the radiolabeled Ery was displaced by 100 μM Evn P. This suggests that despite the presence of the large Evn moiety attached to Rosa, Evn P can still access its binding site within the ribosomal tunnel. However, we note that the binding appeared to be less efficient than observed for Ery and Rosa, since at 100 μM, radiolabeled Ery binding was completely abolished by Rosa (Figure 3B). Therefore, the presence of the Evn moiety may interfere with the ability of Evn P to access the Rosa binding site.

The unavailability of radiolabeled Evn for competition assays led us to employ chemical probing using dimethylsulfate (DMS) to assess the binding of Evn P to the ribosome. DMS is an alkylating reagent that methylates exposed adenine and cytosine residues on the Watson-Crick interface. Instead of using the classic method of identifying rRNA nucleotides protected by ligands from DMS methylation using reverse transcriptase (RT), wherein the primer extension reaction is blocked, we employed DMS-MaPseq that allows the RT to proceed past DMS-modified nucleotides by incorporating mismatches at those methylated positions at a high frequency.¹⁹ Counting these mismatches by high-throughput sequencing produces a quantitative readout of the extent of chemical modification. Treatment of E. coli 70S ribosomes with control antibiotics, Rosa or Evn H, resulted in the expected dose-dependent protection of 23S rRNA nucleotides that comprise the Rosa (e.g. A2058, A2059 and A2062) and Evn (e.g. A2471, A2534, A2483 and A2478) binding sites, respectively (Figures 3C–D and S24). When E. coli 70S ribosomes were treated with Evn P, dose-dependent protection of 23S rRNA nucleotides that comprise both the Rosa (e.g. A2058, A2059 and A2062) and Evn H (e.g. A2471, A2534, A2483 and A2478) binding sites were present (Figures 3E and S24), indicating that Evn P can bind to both the orthosomycin and macrolide binding site.

The chemical protection results were also confirmed structurally using cryo-EM where 70S ribosomes were incubated with Evn P and additional density for Evn P was observed both in the macrolide binding site (Figure 3F) as well as in the orthosomycin binding site (Figure 3H). Unfortunately, there are no small molecule structures of Rosa on or off the ribosome and the resolution obtained (3.7 Å) was insufficient to de novo model the Rosa moiety in the ribosomal tunnel. Nevertheless, we generated a tentative model for Rosa based on the similarity with the 16-membered macrolide tylosin. Consistent with our model, density for the single amino-sugar was observed at the equivalent position to the desosamine sugar of Ery, which is critical for macrolide binding due to interaction with A2058 and A2059 of the 23S rRNA (Figure 3G). At low thresholds, some additional density could be observed for the attached Evn moiety of Evn P. However, this density was poorly defined, suggesting flexibility, and was therefore not modelled. In the orthosomycin binding site, the density for the Evn component of Evn P was well-resolved, allowing the structure of Evn H to be fitted (Figure 3H). While additional density could be observed in the vicinity of the nitrone linker to the Rosa moiety, it was poorly ordered, suggesting no defined interaction of the Rosa moiety in the orthosmycin binding site. Evn P, like Avn, lacks the additional H-ring moiety (ring I) of Evn A, which was also validated by the electron density (Figure 3H, I).

Mechanism of action of Evn P on the bacterial ribosome.

To ascertain the mechanism of action of Evn P during translation, we performed toeprinting assays to monitor the position of the ribosome on the ErmBL mRNA using reverse transcription²⁰ (Figure 4). The reactions were performed in the absence of antibiotics, in the presence of the control antibiotic thiostrepton (Ths), as well as increasing concentrations of Evn H, Rosa, and Evn P. In the absence of antibiotic, ribosomes translate the ErmBL mRNA and become stuck with Lys-tRNA in the P-site and the Ile “catch”-codon in the A-site because the amino acid Ile is not present in the reaction mix. By contrast, the addition of Ths led to a loss of ribosomes stuck at the Lys codon and the appearance of a strong band corresponding to ribosomes trapped on the AUG start codon, as previously reported.^21–22 Similarly, in the presence of Evn H, we observed that the majority of ribosomes became trapped at the start codon, as observed previously.¹⁵ By contrast, in the presence of Rosa, the majority of ribosomes became stuck with the Asp codon located in the P-site. This is consistent with previous findings that ErmBL-dependent stalling occurs at Asp10 of ErmBL and can be evoked by a variety of different macrolides bearing (e.g Ery) and lacking the C5-cladinose (e.g. solithromycin and telithromycin).^{6, 23} The toeprinting results in the presence of Evn P resemble closely those observed for Evn. These data suggest that Evn P has a similar mechanism of action to Evn, allowing initiation complex formation, but preventing further elongation by blocking the accommodation of the incoming aminoacyl-tRNA.¹⁵ While we observe no stalling at Asp10 in the presence of Evn P, we cannot rule out that this is because ribosomes bound by Evn P are already inhibited during the initiation phase of translation and therefore do not have a chance to translate further. However, for this to be the case, we would expect that Evn H and Evn P are more efficient translation inhibitors than Rosa.

Figure 4.

Toeprinting assay monitoring translation inhibition by Evn H, Rosa and Evn P using the ErmBL mRNA. Translation reactions were performed using the ermBL mRNA in the absence of antibiotics (−), the presence of thiostrepton (Ths), as well as increasing concentrations of Evn H, Rosa, and Evn P. Reverse transcription stops indicating ribosomes trapped during initiation with the AUG start codon in the P-site, as well as ribosomes stalled at Asp codon or Lys codon, are indicated with arrows. Sequencing lanes C, U, A and G are shown for reference with relevant ErmBL sequence. Because the ribosome protects 12 and 15 nucleotides from the A-site and P-site codons to the 3′ end of the mRNA where the reverse transciptase stops, the position of the toeprints (Asp and Lys arrows) are shifted compared to the sequencing lanes.

Evn P action on antibiotic resistant ribosomes.

To monitor efficiency of translation inhibition by Evn H, Rosa, and Evn P, we employed an in vitro E. coli lysate-based translation assay to monitor luminescence resulting from expression of a firefly luciferase (Fluc) reporter protein in the presence of increasing concentrations of each antibiotic. As seen in Figure 5A, Rosa inhibits the in vitro translation of Fluc with a half-inhibitory concentration (IC₅₀) of 6 μM, whereas Evn H and Evn P were at least an order of magnitude more efficient, displaying an IC₅₀ of 0.1 μM. We were next interested in analyzing whether Evn P could inhibit translation on ribosomes bearing mutations that confer resistance to macrolides and orthosomycins. To do this we generated translation extracts from E. coli strains (SQ171) where all rRNA operons are deleted from the genome and the only source of rRNA is a single plasmid-encoded copy bearing the macrolide resistance mutation A2059G²⁴ or the orthosomycin resistance mutation A2471C.^{16, 25} As expected, translation of Fluc by ribosomes bearing the A2059G mutation was unaffected by Rosa, even at high concentrations (100 μM), whereas Evn H maintained full activity, displaying an IC₅₀ of 0.1 μM (Figure 5B). Consistent with our findings that Evn P binds to both the macrolide and orthosomycin binding sites, Evn P displayed excellent activity against the macrolide resistant ribosomes. Conversely, Rosa inhibited translation on ribosomes bearing Evn region A2471C mutations with an IC₅₀ of 5 μM, similar to that observed on wildtype ribosomes (Figure 5C). As expected, the inhibition by Evn H was dramatically reduced on A2471C ribosomes, with only ~15% inhibition observed at 50 μM (Figure 5C), compared to 100% inhibition observed at <0.5 μM on wildtype ribosomes (Figure 5A). Evn P also inhibited translation on A2471C ribosomes; however, the IC₅₀ of 11 μM was ~2-fold higher than observed for Rosa, which may suggest that the presence of the Evn moiety could interfere with the Rosa moiety of Evn P finding its binding site within the ribosomal tunnel. Finally, we also generated translation extracts from a strain bearing both the A2059C and A2471C mutations with the expectation that the ribosomes should now be resistant to Evn P. Indeed, as shown in Figure 5D, we observed no inhibition of translation by Evn P up to concentrations of 50 μM, whereas our positive control, chloramphenicol, inhibited translation with an IC₅₀ of ~2 μM. Thus, collectively, these results indicate that Evn P can indeed bind to both the macrolide and orthosomycin binding sites on the ribosome. Due to the distance between sites, it is not possible for both sites to be occupied by EvnP simultaneously and only by generating mutations within both binding sites could the ribosomes obtain resistance to Evn P.

Figure 5.

Effect of Evn P on translation using wildtype, Evn, and macrolide resistant ribosomes. (A-D) in vitro translation using firefly luciferase as a reporter to monitor antibiotic inhibition on (A) wildtype E. coli 70S ribosomes, (B) E. coli A2059G macrolide-resistant ribosomes, (C) E. coli A2471C orthosomycin-resistant ribosomes, and (D) E. coli A2059G/A2471C macrolide and/or orthosomycin-resistant ribosomes. In (A-C) and (D), the error bars show the standard version from the mean for 2 and 3 independent experiments, respectively.

DISCUSSION

The initial motivation for this work was the functional analysis of two oxidase genes. EvdO1 has previously been established biochemically as an oxygenase with the potential to catalyze the formation of the orthoester linkages present in Evn.⁵ However the substrates of this oxidase remain unknown. Unexpectedly, no everninomicin-like metabolites were identified in the knockout strain ΔevdO1::aac(3)IV. However, complementation with evdO1 using a modified integrative vector restored production, as evidenced by the truncated everninomicin-rosamicin conjugates Evn O/P. It is likely that polar effects resulting from the apramycin-resistance cassette attenuated essential biosynthetic genes downstream of evdO1 in the complemented strain, resulting in the loss of biosynthesis of the eastern portion of everninomicin, or that the intermediates are not accumulated in sufficient quantities for detection. However, the oxidation state at C-1 of the hydrolyzed C-ring of the truncated product suggests that the orthoester linkage must have been formed prior to degradation. As complementation with evdO1 restored this metabolite, it is consistent that EvdO1 catalyzes the formation of the orthoester linkage between rings C and D. However, as the ordering and substrate dependence of the biosynthetic sequence is not yet fully determined, further experiments are needed to identify the substrate of EvdO1.

Evn E, the amino congener of Evn D occurs in fermentations along with every possible oxidation state of nitrogen (amino, hydroxylamino, nitroso, and nitro). Remarkably, the amino congeners of Evn’s are unique in the N-oxidation series for their activity against Gram negative pathogens, and we attempted to generate these herein via knockout of the biochemically characterized nitrososynthase gene evdN1.²⁶ Unexpectedly, targeted replacement of evdN1 resulted in the production of a metabolite lacking the entire terminal evernitrose (A-ring), and genetic complementation with evdN1 did not restore production of full-length everninomicins. Since the genes encoding the synthesis of evernitrose are clustered in an apparent TDP-(l)-evernitrose operon beginning with evdN1 (Supplementary Figure 1A), it appears that replacement of evdN1 resulted in functional loss of the entire operon and subsequent loss of the A-ring. Wild type everninomicins display all stages of N-oxidation indicating that TDP-evernosamine synthesis and/or glycosylation are not dependent upon complete N-oxidation. These data are consistent with the A-ring being added last in the biosynthesis of everninomicin.

These two λ–RED-mediated polar effects highlight the need for improved gene replacement and inactivation methods in actinomycetes. To date we have attempted without success to adapt CRISPR-Cas9 based approaches to M. carbonacea var. aurantiaca. However, we were pleased that the functional analysis studies herein informed the discovery of the previously unreported bifunctional antibiotic Evn P. The nitrone functionality in natural products is extremely uncommon. Fungal indole alkaloid roquefortine L²⁷ and plant quinoline alkaloid cincholenines²⁸ are among the few natural products reported to date to possess this functional group and display this functionality in the form of oxidized dihydropyrrole and tetrahydropyridine, respectively. Recently, malleonitrone, a nitrone conjugated siderophore pair was isolated from mixed culture of the pathogen Pseudomonas aeruginosa and Burkhodheria thailandensis. The conjoined action of these intergeneric siderophores possessed antibiotic activity.²⁹ To our knowledge, Evn P is the first reported bridging nitrone functional antibiotic from a single stain. The identification of the bifunctional rosamicin-everninomicin conjugate was facilitated by evdO1 knockout complementation experiments, which accumulated metabolites within the typical scanning range of our mass spectrometer. The formation of Evn P may be a non-enzymatic process, as the condensation of a hydroxylamine and aldehyde occurs spontaneously under physiological conditions.¹¹ Thus, we cannot completely exclude the possibility that the bifunctional antibiotic arises artificially during the fermentation conditions, rather than as a metabolite produced adaptively by M. carbonacea var. aurantiaca.

Nevertheless, we could show that the bifunctional rosamicin-everninomicin conjugate Evn P maintains potent antimicrobial activity with an MIC of 1 – 2 μg/ml against S. aureus (Supplementary Figure 23). Using a combination of competition, chemical protection, and structural analysis, we demonstrate that Evn P can bind to both the macrolide and everninomicin binding sites on the ribosome (Figure 3). Moreover, using cell-free in vitro translation systems, we observed that Evn P maintains excellent inhibitory activity (IC₅₀ of 0.1 μM), analogous to the orthosomycin Evn H (Figure 5A). Toeprinting assays revealed that Evn P, like Evn H, traps ribosomes on the start codon, presumably by preventing the accommodation of the first amino-acyl-tRNA during the first elongation step.¹⁵ However, an advantage of Evn P over Evn H is that the A2471C mutation in the 23S rRNA that confers resistance to orthosomycins,^{15, 2} such as Evn H, does not confer resistance against Evn P. Moreover, we could demonstrate that this is because the Rosa component of Evn P can direct binding of the drug to the macrolide binding site on the ribosome and thereby maintain inhibitory activity. This was validated by showing that ribosomes bearing the A2471C mutation, plus an additional A2059G mutation that confers resistance to macrolides, led to high level resistance against Evn P (Figure 5D). Thus, we conclude that it will be more difficult for bacterial ribosomes to obtain resistance to Evn P since simultaneous alteration (mutations and/or modifications) of two distinct sites within the 23S rRNA will be required to obtain resistance. Indeed, producers of macrolides and orthosomycins often protect their own ribosomes via methyltransferases that modify nucleotides within the binding sites, such as methylation of A2058 (macrolides),³⁰ U2479. G2535 (avilamycin),^30–31 and G2470 (everninomicin).³². It will be interesting to ascertain how the producer of Evn P, Micromonospora carbonacea, obtains resistance to this bifunctional rosamicin-everninomicin conjugate.

CONCLUSION

Given the importance of identifying new methods to combat antibiotic resistance, dual pharmacophore targeting has been proposed as a synthetic strategy to improve pharmacological properties via the generation of antibiotic hybrids and antibiotic conjugates.³³ Dual targeting of orthogonal binding sites within the ribosome demonstrates how conjugation can potentially hinder the development of antibiotic resistance. The interception of hydroxylamino-functional Evn F via Rosa aldehyde in Micromonospora carbonacea fermentation results in the formation a potent stable bifunctional nitrone conjugated compound, generating a new type of hybrid ribosome engaging antibiotic exemplifying this useful strategy.