Biosynthesis of Bacillus Phosphonoalamides Reveals Highly Specific Amino Acid Ligation

Abstract

Phosphonate natural products have a history of commercial success across numerous industries due to their potent inhibition of metabolic processes. Over the past decade, genome mining approaches have successfully led to the discovery of numerous bioactive phosphonates. However, continued success is dependent upon a greater understanding of phosphonate metabolism, which will enable the prioritization and prediction of biosynthetic gene clusters for targeted isolation. Here, we report the complete biosynthetic pathway for phosphonoalamides E and F, antimicrobial phosphonopeptides with a conserved C-terminal l-phosphonoalanine (PnAla) residue. These peptides, produced by Bacillus, are the direct result of PnAla biosynthesis and serial ligation by two ATP-grasp ligases. A critical step of this pathway was the reversible transamination of phosphonopyruvate to PnAla by a dedicated transaminase with preference for the forward reaction. The dipeptide ligase PnfA was shown to ligate alanine to PnAla to afford phosphonoalamide E, which was subsequently ligated to alanine by PnfB to form phosphonoalamide F. Specificity profiling of both ligases found each to be highly specific, although the limited acceptance of noncanonical substrates by PnfA allowed for in vitro formation of products incorporating alternative pharmacophores. Our findings further establish the transaminative branch of phosphonate metabolism, unveil insights into the specificity of ATP-grasp ligation, and highlight the biocatalytic potential of biosynthetic enzymes.

Graphical Abstract

1. INTRODUCTION

Phosphonate and phosphinate (Pn) compounds are a class of natural products (NPs) characterized by direct carbon−phosphorus bonds. This moiety, an isostere of phosphate esters and carboxylic acids, is responsible for the bioactivity of Pn NPs, the majority of which are potent metabolic inhibitors.¹ Notable examples include the antibacterial fosfomycin, the herbicide phosphinothricin, and the antimalarial fosmidomycin. Fosfomycin (Monurol) is a clinically approved antibiotic that covalently inhibits MurA by mimicking phosphoenolpyruvate, preventing peptidoglycan biosynthesis.² Fosmidomycin has shown promise in clinical trials against malaria, where it inhibits 1-deoxy-d-xylulose 5-phosphate reductoisomerase to block the nonmevalonate pathway of isoprenoid biosynthesis.³ Outside of medicine, phosphinothricin (glufosinate) and its peptide derivatives, which irreversibly inhibit glutamine synthetase, are the active ingredients of multiple herbicide lines produced by BASF.⁴

Pn NPs have found commercial success in multiple industries, inspiring new genomics-driven techniques for their discovery.⁵ In the past decade alone, genome mining and targeted isolation have yielded multiple new Pn scaffolds, with new chemotypes, previously unknown headgroups, and a wide array of bioactivities.^6–10 However, genomic data suggests that an even greater number of Pn NPs await discovery, as nearly 7% of all bacteria are predicted to harbor biosynthetic gene clusters (BGCs) for the compounds, very few of which encode for known products.¹¹ To tap into this wealth of compounds, a greater understanding of Pn biosynthesis is required to accurately predict and classify the pathways, products, and potential biological activities of these BGCs.

All known Pn biosynthesis begins with the conversion of phosphoenolpyruvate (PEP) to phosphonopyruvate (PnPy) by phosphoenolpyruvate mutase (PepM). By itself, this reaction is highly unfavorable, requiring a coupled enzymatic reaction to thermodynamically drive C−P bond formation.¹² Previously, we leveraged this principle to search for Pn BGCs lacking a known coupling enzyme. This resulted in the isolation of l-phosphonoalanine (PnAla) and four tripeptides with an N-terminal PnAla (phosphonoalamides A−D) from Streptomyces sp. NRRL B-2790.⁷ A subsequent search for additional phosphonoalamide BGCs led to the discovery of C-terminal PnAla peptides (Ala−PnAla and Ala−Ala−PnAla, phosphonoalamides E and F) produced by Bacillus velezensis NRRL B-41580.¹³

While our studies have revealed PnAla and its peptide derivatives to have broad-spectrum antibacterial activity,^7,13 the role that PnAla plays in Nature is still unknown, six decades after its first isolation from biological material.¹⁴ The bioactivity of phosphonoalamides E and F, along with their isolation from B. velezensis, a noted biocontrol agent, further supports a potential role in microbial chemical ecology. Strains of this species have demonstrated efficacy against numerous bacterial and fungal phytopathogens and are used in commercial products such as RhizoVital, Botrybel, Serenade, Kodiak, and Taegro.¹⁵ B. velezensis dedicates a substantial portion of its genome to secondary metabolite production. The B-41580 genome contains the BGCs for 11 known antimicrobial compounds in addition to the phosphonoalamide cluster,¹⁵ suggesting that PnAla is yet another warhead within the B. velezensis arsenal.

In this study, we establish the genetic and biochemical logic for phosphonoalamide biosynthesis in Bacillus. We defined the BGC using heterologous expression and biochemically demonstrated phosphonoalamide F to be the product of four enzymatic reactions. PnfD was shown to catalyze the interconversion of PnPy and PnAla, confirming it as an example of the transaminative early branch of Pn biosynthesis. Tripeptide formation was revealed to be a linear process catalyzed by two distinct ATP-grasp l-amino acid ligases, both of which displayed strict substrate specificity.

2. RESULTS

2.1. Delineation of the Phosphonoalamide Biosynthetic Gene Cluster.

During a previous search for PnAla-encoding BGCs within Actinobacteria, we identified a BGC attributed to Mycobacteroides abcessus subsp. massiliense strain aerosol_aerosol_3. We established that this genome was heavily contaminated and that the contig containing pepM belonged to a member of the Bacillus subtilis species complex. Fermentation of several strains encoding this BGC ultimately resulted in the isolation of two PnAla-containing peptides, phosphonoalamide E (Ala−PnAla) and phosphonoalamide F (Ala−Ala−PnAla) from B. velezensis NRRL B-41580.¹³

To establish the boundaries of this BGC, we first analyzed a 15-gene window (Table S1) centered on the pepM within the B-41580 genome (henceforth designated pnf C). This genomic neighborhood was compared to those of other Bacillus spp. encoding a PepM with ≥80% shared identity to PnfC, as BGCs meeting this similarity threshold generally produce similar Pn NPs.⁶ However, all genomes (90, Table S5) containing this BGC belonged to members of the B. subtilis group and displayed extensive conservation. As strains lacking pepM would be incapable of Pn biosynthesis, it would be highly unlikely that genes conserved between pepM⁺ and pepM⁻ strains would be involved in phosphonoalamide biosynthesis. Thus, we performed the reciprocal search of neighboring genes to delineate the cluster (Figure 1). Although orfs1−5 could not be confidently assigned to known functions, their conservation within closely related strains lacking pepM (B. subtilis 168 and B. amyloliquefaciens DSM7) led us to exclude them from the putative gene cluster. Genes downstream of pnf T were also excluded because their annotations as σ factor (orf11), yebDEG (orf15−17), and members of the pur operon (orfs18−29)) suggest functions unrelated to Pn metabolism. This resulted in our putative assignment of pnfABCDT as the biosynthetic genes. The B-41580 pnfABCDT were individually compared to those of all other pepM⁺Bacillus strains. The minimum percent identity observed was 81%, while the median percent identity across all comparisons was 99−100%.

Figure 1.

Synteny analysis of the B. velezensis pepM gene neighborhood. pnfABCDT were conserved between B. velezensis and B. swezeyi, which encode pepM (pnf C), but not between B. velezensis and closely related strains lacking pepM. Upstream genes were excluded based on their conservation within organisms lacking the ability to produce Pn. Downstream genes were excluded by the same basis but also by their identification as a σ factor, and members of the yeb and pur (purine biosynthesis) operons. Genes are numbered as space allows, with detailed annotation in Table S1.

This cluster assignment was further supported by similarities between the Bacillus and Streptomyces pepM gene neighborhoods: both encode a PepM, aminotransferase, two ATP-grasp ligases, and a transporter, and result in the biosynthesis of PnAla-containing tripeptides.^7,13 Thus, we reasoned that pnfAD would be essential for the production of phosphonoalamide F while pnf T may serve as a transporter. We proposed that PnfC would convert PEP to PnPy, PnfD would transaminate PnPy to PnAla, and the ATP-grasp ligases PnfA and PnfB would catalyze peptide bond formation to produce phosphonoalamides E and F.

To test this hypothesis, we cloned pnfABCD from B-41580 into pDG1730 for expression within a heterologous host. The resulting plasmid (pKSJ652) was transformed and integrated into the genome of B. subtilis168, and the process was repeated with pDG1730 to provide a negative control. Metabolites produced from the resulting strain, B. subtilis KSJ2140 (168 amyE::pKSJ652), were analyzed by ³¹P NMR and compared to the empty vector control strain, B. subtilis KSJ2139 (168 amyE::pDG1730). Significant accumulation of a Pn was observed within B. subtilis KSJ2140 extracts (Figure 2A) but absent from the control strain. This compound was identified as phosphonoalamide F by ³¹P NMR (δ_P 17.8), ¹H−³¹P HMBC (correlated δ_H 4.03, 1.87, and 1.69), and LC-HRMS analyses (m/z 312.0955), which were all consistent with literature values (Figure 2).¹³ These results are congruent with our synteny analyses and clearly demonstrate pnfABCD as the only genes required for phosphonoalamide F biosynthesis.

Figure 2.

Heterologous expression of pnfABCD. (A) ³¹P NMR spectra of the empty vector (pDG1730) and pKSJ652 integrant extracts. (B) ¹H−³¹P HMBC of the pKSJ652 integrant extract, demonstrating correlated proton resonances. (C) LC-HRMS showing EICs for [M + H]⁺ ions of PnAla (m/z 170.0213), phosphonoalamide E (m/z 241.0584), and phosphonoalamide F (m/z 312.0955).

2.2. PnAla Production by PnfC and PnfD.

After establishing the minimal BGC, we focused our attention on the biosynthesis of PnAla. Our previous heterologous expression experiments demonstrated that introduction of genes encoding PepM and a PLP-dependent aminotransferase from Streptomyces sp. NRRL S-515 were sufficient for PnAla production.⁷ As PnfD was similarly annotated as a PLP-dependent aspartate aminotransferase (AAT), we proposed that it would catalyze transamination of PnPy using aspartate as an amino donor (Figure 3A).

Figure 3.

PnAla production by PnfC and PnfD. (A) Biosynthetic scheme for l-PnAla formation. (B) ³¹P NMR spectrum of PnfCD reactions resulting in conversion of PEP (gray) to PnAla (green). (C) ³¹P NMR spectrum of PnfD reverse reaction resulting in conversion of PnAla to PnPy (yellow).

To test this hypothesis, we overexpressed and purified recombinant PnfC and PnfD from Escherichia coli for in vitro assays (Figure S2). Interestingly, purified recombinant PnfD was yellow in color, suggesting the presence of a bound chromophore. This was suspected to be an internal aldimine with PLP, and UV−vis spectroscopy did reveal an absorbance maximum at 439 nm, consistent with other PLP-dependent aminotransferases (Figure S3A).^16,17 We performed a series of spectrophotometric experiments to understand the nature of the protein. Upon treatment of PnfD with sodium borohydride, we observed a decrease in absorbance at 439 nm with a concomitant absorption increase at 344 nm (Figure S3B). These characteristics were consistent with reduction of the aldimine linkage between PLP and a lysine residue.¹⁶ Incubation of PnfC with l-cycloserine shifted the absorbance maxima to 381 nm (Figure S3C). This was consistent with formation of a stable adduct between l-cycloserine and PLP, resulting from the formation of a transient oxime intermediate.¹⁷ Having established PLP as the chromophore, we incubated PnfC with a stoichiometric amount of PLP to allow for full occupancy. By comparing the absorbance at 439 nm with and without exogenous PLP, the occupancy of as-purified recombinant PnfD was estimated to be 40% (Figure S3A).

We demonstrated the activity of PnfD as a PepM-coupling enzyme through in vitro reconstitution. Purified PnfD was incubated with PnfC, PEP, l-Asp (amino donor), and PLP. Over 24 h, this resulted in consumption of PEP (−1.0 ppm) and production of a new species (16.7 ppm, 27% yield) in ³¹P NMR spectra (Figure 3B). To determine if other amino donors would be more favorable, we substituted l-Asp with each of the other proteinogenic amino acids (Figure S4). While PnAla production was observed with l-Glu (19% yield), l-Asn (17% yield), and l-Cys (11% yield) as amino donors, the yield of PnAla remained highest with l-Asp (Figure 3B).

Having established PnAla biosynthesis with PnfD, we examined its ability to catalyze the reverse reaction, using PnAla as an amino donor to convert oxaloacetate (OAA) to l-Asp. Incubating PnfD with PnAla, OAA, and PLP resulted in consumption of PnAla (16.6 ppm) and formation of a new species (9.9 ppm) in ³¹P NMR (Figure 3C). LC-HRMS confirmed this new species as PnPy (Figure S5). Intriguingly, the yield of PnPy was low, suggesting that the equilibrium constant of the PnfD reaction may favor PnAla. To further examine the effects of substrate concentration on this directionality, we monitored both forward reactions containing 1 mM PnPy and a large excess (10 mM) of amino donor (Asp) and reverse reactions containing 1 mM PnAla with a large excess (10 mM) of keto-acid acceptor (OAA) using ³¹P NMR over the course of 2 h (Figure S6). In both the forward and reverse PnfD reactions, conversion occurred rapidly, reaching apparent equilibrium within 10 min. As the majority of enzymatic transamination reactions are freely reversible, with equilibrium constants close to one,¹⁸ we expected both reactions to proceed toward completion given the large excess of supplied cosubstrate. However, the yield of PnAla in the forward reaction (96%) was far higher than the yield of PnPy in the reverse reaction (60%). This suggests that PnfD has an inherent preference for the transamination of PnPy over the transamination of PnAla, allowing it to drive PnAla biosynthesis.

Our combined results demonstrate PnfD as a reversible PLP-dependent transaminase that functions as a PepM-coupling enzyme by converting PnPy to PnAla. While an excess of amino donor could be used to drive the forward reaction to near completion, an equivalent excess of keto-acid acceptor was unable to drive the reverse reaction past 60% conversion, suggesting that reaction characteristics inherent to PnfD control the directionality of biosynthesis.

2.3. Phosphonoalamide Biosynthesis in Bacillus is a Linear Pathway.

Having established the synthesis of PnAla from PEP by PnfC and PnfD, we sought to reconstitute the reactions leading to peptide formation. The remaining genes in the BGC, pnfA and pnf B, encoded putative ATP-grasp ligases, are a family of enzymes that implicated formation of other phosphonopeptides including valinophos, rhizocticin, and plumbemycin pathways.^9,19–21 In canonical peptide bond formation by ATP-grasp ligases, one amino acid is first activated as a carboxylate, forming an acylphosphate intermediate upon ATP hydrolysis. The amine of the second amino acid is then primed for nucleophilic attack, forming an amide bond. The C-terminal position of PnAla in the Bacillus compounds suggested it would be a nucleophile in the biosynthesis of phosphonoalamide E and F. Based on this logic, there would be two possible routes to phosphonoalamide F (Figure 4A). In the convergent pathway, one ligase would produce Ala−Ala and the second would ligate PnAla to the alanyl dipeptide to form phosphonoalamide F. However, this would not provide a clear explanation for our isolation of phosphonoalamide E (Ala−PnAla). Alternatively, ligation of Ala to PnAla may occur in a linear manner, first forming phosphonoalamide E and then forming phosphonoalamide F, with one ligase catalyzing each reaction.

Figure 4.

Biosynthesis of phosphonoalamide F. (A) Potential biosynthetic routes to phosphonoalamide F. The convergent pathway (gray) was excluded based on the biochemical activity of the ATP-Grasp ligases. (B) Reactions with Ala and PnAla, highlighting the formation of Ala−PnAla (cyan) by PnfA. (C) Reactions with Ala and Ala−PnAla, highlighting the formation of Ala−Ala−PnAla (blue) by PnfB.

To test this hypothesis, recombinant PnfA and PnfB were expressed and purified from E. coli. As it was unclear which enzyme would catalyze the first ligation, we incubated Ala, PnAla, Mg²⁺, and ATP with either PnfA or PnfB and monitored the reactions by ³¹P NMR (Figure 4B). In the reaction with PnfA, PnAla (16.6 ppm) was completely consumed, with appearance of a new species (17.0 ppm). No product was observed when any component was omitted (Figure S7). LC-HRMS identified this product as phosphonoalamide E (Ala−PnAla, Figure S8A(i)), which was confirmed by fragmentation analysis (Figure S9B and Table S6). PnAla remained untransformed in the reaction with PnfB.

We scaled up the PnfA reaction and purified Ala−PnAla for use in tripeptide ligation assays. Ala−PnAla was isolated from the reaction mixture by strong cation exchange chromatography and HILIC HPLC. The purified dipeptide was >95% pure as determined by ¹H NMR (Figure S10). Reactions containing Ala, Ala−PnAla, Mg²⁺, and ATP with either PnfA or PnfB were prepared and monitored by ³¹P NMR (Figure 4C). A new species (17.3 ppm) was observed in the PnfB reaction, which was identified as phosphonoalamide F by LC-HRMS (Figure S11Ai) and confirmed by fragmentation analysis (Figure S9A and Table S6). Ala−PnAla remained untransformed in the reaction with PnfA. These results demonstrated the phosphonoalamide biosynthesis proceeds through a linear pathway in Bacillus.

To further exclude the possibility the peptides may result by convergent biosynthesis from secondary activity of the ATP-Grasp ligases, we prepared additional control reactions using the recombinant enzymes. First, we incubated Ala−Ala, PnAla, ATP, and Mg²⁺, with either PnfA or PnfB. Surprisingly, a new species (17.1 ppm) was observed in the ³¹P NMR spectra of the PnfA reaction (Figure S12A). The NMR shift of the product was consistent with Ala−PnAla instead of Ala−Ala−PnAla (17.4 ppm). Indeed, the LC-HRMS revealed the signal as the Ala−PnAla dipeptide (Figure S12B). As free Ala was not added to the reaction, we suspected Ala contamination in our stock of Ala−Ala. However, this possibility was eliminated based on the absence of signals corresponding to free Ala in the ¹H NMR spectrum (Figure S13A). As our ligation reactions took place at alkaline pH, we hypothesized that Ala−Ala may be undergoing base-mediated hydrolysis to generate free l-Ala. Indeed, when Ala−Ala was incubated in reaction buffer and analyzed by LC-HRMS at sequential time points, free Ala steadily increased, revealing the source of Ala for Ala−PnAla formation (Figure S13B).

To further exclude the involvement of Ala−Ala in phosphonoalamide F biosynthesis, we incubated PnfA with PnfB, Ala, PnAla, ATP, and Mg²⁺. Analysis of the reaction by ³¹P NMR and LC-HRMS demonstrated the formation of Ala−PnAla and Ala−Ala−PnAla (Figure S14), but not Ala−Ala (Figure S15B). Ala−Ala was also absent from separate reactions containing Ala, ATP, and Mg²⁺ with only PnfA (Figure S15C) or PnfB (Figure S15D). These data demonstrate that the Ala−Ala dipeptide is not a product of the pathway.

Inspired by the success of the PnfAB reaction, we reconstituted the entire pathway in a one-pot reaction containing PnfA, PnfB, PnfC, PnfD, PEP, PLP, Asp, Ala, PnAla, ATP, and Mg²⁺. Conversion of PEP was low (8.3%), with only one product signal visible by ³¹P NMR (Figure S16A). Nonetheless, LC-HRMS revealed the formation of PnAla, Ala−PnAla, and Ala−Ala−PnAla, with the dipeptide as the major product (Figure S16B). In vivo biosynthesis heavily favored tripeptide production, as demonstrated by isolation from the native producer¹³ and heterologous expression, suggesting that our in vitro reaction conditions require optimization. While our independent PnfA (Figure S8A(i)) and PnfB (Figure S11A(ii)) reactions result in near-complete turnover of Pn substrates, the PnfAB reaction (Figure S14A) produces equivalent amounts of dipeptide and tripeptide, further highlighting the complexity behind coordinating biosynthesis.

2.4. PnfA and PnfB Are Highly Specific Amino Acid Ligases.

In contrast to the diverse tripeptides isolated from Streptomyces,⁷ the only phosphonopeptides isolated from B-41580 were Ala−Ala−PnAla and its immediate precursor Ala−PnAla.¹³ This suggested that PnfA and PnfB may display strict substrate specificity. To test this hypothesis, we began by incubating PnfA with PnAla and each of the proteinogenic amino acids (Figure S8). Monitoring the reactions by ³¹P NMR and LC-HRMS identified Ser−PnAla to be the only other dipeptide produced by PnfA, the structure of which was established by fragmentation analysis (Figure S9C and Table S6). Overall turnover of PnAla was much lower with Ser (8%) than with Ala (100%), indicating that replacement of a methyl group hydrogen with a hydroxyl moiety reduced carboxylate acceptance. The methyl side chain was required for catalysis, as Gly (noralanine) was not ligated to PnAla by PnfA. The second ATP-Grasp ligase in the pathway, PnfB, also exhibited high substrate specificity for Ala. Reactions containing PnfB, Ala−PnAla, and each of proteinogenic amino acids yielded no products as detected by ³¹P NMR and LC-HRMS other than Ala−Ala−PnAla (Figure S11).

We determined whether PnfA could be used to synthesize other phosphonodipeptides using reactions containing Ala and other aminophosphonate substrates. These included 2-amino-ethylphosphonate (2AEP), 1-hydroxy-2AEP (1H2AEP), 3-aminopropylphosphonate (3-APPn), phosphinothricin (PT), l-2-amino-4-phosphonobutyrate (l-AP4), l-2-amino-5-phosphonopentanoate (l-AP5), aminomethylphosphonate (AmPn), (R)-1-aminoethylphosphonate (l-Ala(P)), (1R)-1-amino-2-methylpropylphosphonate (l-Val(P)), and 4-amino-phenylphosphonate (4-APhePn) (Figure S17A). These substrates were chosen to provide structural insights into nucleophile selectivity by PnfA, as they differed from PnAla with respect to side chain length (l-AP4 and l-AP5), presence of a carboxyl group (2AEP), use of a phosphinate moiety (PT), or a combination of the above and other modifications. Reactions with Ala and these alternative nucleophiles were monitored by LC-HRMS (Figure S17B). All signals corresponding to products were further characterized by fragmentation analysis to unambiguously establish their chemical identities. These revealed PnfA to synthesize additional phosphonate dipeptides including Ala-2AEP (9.4% yield), Ala-1H2AEP (6.4%), and Ala-AP4 (trace amounts; Figures S18 and S19 and Tables S7 and S9).

The activity profile of PnfA with the tested substrates allowed us to propose rudimentary rules for the nucleophile specificity. Using PnAla as the canonical substrate, acceptance of 2AEP revealed that a carboxyl group was not required, as 2AEP is equivalent to decarboxylated PnAla. Acceptance of l-AP4, but not l-AP5, demonstrated that the amino acid side chain could be extended by one carbon, but not two. However, these moieties could not be combined, as 3-APPn, which exhibits side-chain extension by one carbon and loss of the carboxylate moiety, was not accepted by PnfA. Likewise, PT, the phosphinate analogue of l-AP4, failed to ligate with PnAla. While PnfA accepted 1H2AEP as a substrate, which is decarboxylated and substituted at the β carbon, other compounds with multiple structural differences from PnAla (AmPn, Ala(P), Val(P), and 4-APhePn) were not ligated. Although PnfA formed dipeptides between Ala and PnAla, 2AEP, and 1H2AEP, PnAla was the only substrate among the three substrates to be successfully ligated to Ser (7.2% yield) (Figure S20B).

As all our accepted nucleophilic substrates contained a Pn group, we wondered if this moiety was strictly required by PnfA. Therefore, we incubated Asp, the proteinogenic isostere of PnAla, with Ala or Ser. LC-HRMS revealed product formation in both reactions (Figure S21A) and fragment analysis identified the products as Ala−Asp and Ser−Asp (Figure S21B and Table S8). Nonetheless, Pns remained the preferred substrate, with significantly greater amounts of unligated Ala and Ser remaining in reactions containing Asp versus PnAla.

Finally, we determined whether the specificity of PnfA would include substrate stereochemistry. Dipeptide products were absent in the reactions containing d-Ala and l-PnAla (Figure S22A). However, when PnfA was supplied with l-Ala and dl-PnAla, roughly half of the dl-PnAla was converted to dipeptide (Figure S22B). The stoichiometry of this reaction immediately suggested that PnfA had consumed one enantiomer of PnAla while ignoring the other. Indeed, Marfey’s analysis of the reaction mixture revealed the unreacted PnAla to be overwhelmingly in d configuration, with little, if any, l-PnAla remaining (Figure S22C). Altogether, these data demonstrate that PnfA is an enantioselective dipeptide ligase with strict specificity for both carboxylate and nucleophilic substrates.

To apply our characterization of PnfA and PnfB toward the elucidation of phosphonoalamide biosynthesis in Streptomyces, we aligned PnfA and PnfB with the conserved ligases in Streptomyces (PnaB and PnaC) and two structurally characterized ATP-Grasp l-amino acid ligases: BacD,^22,23 which activates Ala as a carboxylate, and LdmS,²⁴ which activates Asp as a carboxylate (Figure S24). From the structure of phosphonoalamide A (Figure S25), one of the Streptomyces ligases can be inferred to activate PnAla, for which Asp is the proteinogenic isostere (such that LdmS may serve as a reference), and the other to activate Ala (such that BacD, PnfA, and PnfB may serve as references). The catalytic Arg was completely conserved among all ligases, while many of the residues implicated in Mg²⁺ and ADP/ATP binding demonstrate complete or near-complete conservation. One residue implicated in substrate binding, Glu273 of BacD, is conserved among all ligases shown to activate Ala (BacD, PnfA, and PnfB) as well as PnaC but not PnaB, suggesting that PnaC may also be capable of activating Ala as a carboxylate.

3. DISCUSSION

We have elucidated the complete pathway for the phosphonoalamide biosynthesis using a combination of comparative genomics, heterologous expression, and biochemical reconstitution experiments. Of the contiguous genes conserved between Bacillus strains, only four encoded enzymes were essential for biosynthesis. The pathway begins with the interconversion of PEP to PnPy by PepM (PnfC), which is coupled to the transamination of PnPy to PnAla by PnfD. Ligation of PnAla to Ala is catalyzed by PnfA, forming Ala-PnAla (phosphonoalamide E). Subsequent ligation of Ala−PnAla to Ala by PnfB results in production of Ala−Ala−PnAla (phosphonoalamide F, Figure 4A).

Similar to other aminotransferases, PnfD exhibited relaxed substrate specificity, accepting Asp, Glu, Asn, and Cys as amino donors for the transamination of PnPy into PnAla. While utilization of Asn and Cys as amino donors is unusual for AATs,²⁵ this ability may simply be overlooked, as recent studies have discovered other bacterial AATs capable of cysteine- and asparagine-oxo-acid transamination.^26,27 Transamination of PnPy to PnAla was a reversible reaction, but the limited degree to which substrate concentration affected the reverse reaction was unexpected. While a large excess of amino donor was able to drive the forward reaction to near completion (96% yield of PnAla), an equivalent excess of keto-acid acceptor was unable to drive the reverse reaction as far (60% yield of PnPy). Taken together, this suggests that PnfD has an inherent preference for the transamination of PnPy over the transamination of PnAla, allowing this reversible enzyme to dictate the directionality of Pn biosynthesis.

PnfA and PnfB provide further examples of ATP-grasp amino acid ligases involved in phosphonopeptide biosynthesis. Both ligases were highly specific, such that PnfB did not accept any alternate substrates provided and PnfA only tolerated small differences in carboxylate and nucleophile structure. The high selectivity of these enzymes provides an opportunity to understand structural differences between di- and tripeptide ligases, as they activate the same carboxylate substrate (Ala), are produced by the same organism, and act within the same biosynthetic pathway. Additionally, the specificity of PnfA for l-PnAla, even in the presence of stoichiometric d-PnAla, offers a method for separation of the two enantiomers. The commercial cost of pure d-PnAla is roughly 12-fold higher than that of l-PnAla and nearly 80-fold that of dl-PnAla, highlighting the difficulty of obtaining specific isomers.

The activation of PnAla as a nucleophile by PnfA, rather than as a carboxylate, results in the alternate connectivity of the Bacillus phosphonoalamides (C-terminal PnAla) as compared to the Streptomyces phosphonoalamides (N-terminal PnAla). To the best of our knowledge, this is the only example in NP biosynthesis where the same pharmacophore is differentially incorporated (carboxylate vs nucleophile) by ATP-grasp ligases. This is unusual for phosphonopeptides, as other groups with conserved Pn pharmacophores retain the same connectivity. The rhizocticins and plumbemycins contain the threonine synthase inhibitor (Z)-l-2-amino-5-phosphono-3-pentenoate (APPA) as a C-terminal residue,^21,28–30 while bialaphos, trialaphos, and phosalacine contain the glutamine synthetase inhibitor phosphinothricin at their N-terminus.⁴

Also unique to this pathway is the production of a single end product (phosphonoalamide F) rather than a mixture of tripeptides. In this respect, the Bacillus phosphonoalamides are more similar to phosphonopeptides produced by nonribosomal peptide synthetases (phosphinothricin (PT) tripeptide, phosalacine; Figure S25)^4,31 or tRNA-dependent GCN5-related N-acetyltransferases (argolaphos, dehydrophos, fosfazinomycin; Figure S25),^10,32–34 which result in specific, invariant natural products.

Outside of the Bacillus phosphonoalamides, ATP-grasp ligases appear to underlie a strategy for producing multiple phosphonopeptides from a single pathway. The rhizocticin, plumbemycin, valinophos, and Streptomyces phosphonoalamide pathways all use ATP-grasp ligases to produce multiple compounds with the same Pn warhead.^7,9,19,20,28 Moreover, the composition of proteinogenic amino acids within these antimicrobial phosphonopeptides has been shown to influence their selectivity.^29,35,36 Most strikingly, the rhizocticins and plumbemycins both contain the threonine synthase inhibitor APPA as a C-terminal residue. However, the rhizocticins are antifungals while the plumbemycins display antibacterial activity.^28–30 Phosphonoalamide A (PnAla−Ala−Val, from Streptomyces) and phosphonoalamide F (Ala-Ala-PnAla, from Bacillus) also exhibit different spectra of antimicrobial activity,^7,13 but it remains to be seen whether this is due to amino acid composition or position of the PnAla moiety.

The convergent alanyl derivatization of Pn moieties observed in bialaphos, trialaphos, and the Bacillus phosphonoalamides, which are produced by taxonomically distant organisms and use different mechanisms for amino acid ligation, suggests that alanylation may be advantageous. Indeed, bialaphos is a more potent antimicrobial than phosphinothricin,³¹ and each Ala residue ligated to PnAla results in lower MICs.^7,13 Ala appears to be a preferred constituent of antimicrobial phosphonopeptides, as roughly half of these natural products contain at least one Ala residue: phosphonoalamides A (PnAla−Ala−Val), C (PnAla−Ala−Ile), E (Ala−PnAla), and F (Ala−Ala−PnAla),^7,13 plumbemycins A (Ala−Asp−APPA) and B (Ala−Asn−APPA),²⁸ phosalacine (PT−Ala−Leu),⁴ bialaphos/phosphinothricin tripeptide (PT−Ala−Ala),³⁷ and trialaphos (PT−Ala−Ala−Ala).³⁸

B. velezensis also utilizes alanyl derivatization in the production of the antimicrobial dipeptide bacilysin. Coincidentally, bacilysin is also assembled by an ATP-grasp ligase and composed of an N-terminal Ala and a C-terminal l-anticapsin pharmacophore.³⁹ Likewise, bacilysin is a much more potent antibacterial than free anticapsin.⁴⁰ Thus, it stands to reason that alanyl incorporation may afford the broadest uptake of these peptides, obviating the need for peptide diversification. Alternatively, these single products may be ignored by the oligopeptide transporters of producing strains, offering a means of self-resistance. Further investigation is required to address these hypotheses, yielding insights that can be applied to the rational design of antimicrobial peptides.

The biosynthesis of phosphonoalamides E and F in Bacillus further establishes the transaminative branch of Pn NP metabolism. Peptide biosynthesis reactions revealed the strict specificity of ligases within this pathway, while demonstrating the importance of different functional groups for substrate acceptance. The alternative incorporation of PnAla (C- vs N-) within this group of phosphonopeptides suggest that further exploration of this biosynthetic branch will yield additional variations on this theme, all of which can be leveraged to decipher the molecular mechanisms underlying their bioactivity.