Discovery of Antimicrobial Phosphonopeptide Natural Products from Bacillus velezensis by Genome Mining

ABSTRACT

Phosphonate natural products are renowned for inhibitory activities which underly their development as antibiotics and pesticides. Although most phosphonate natural products have been isolated from Streptomyces, bioinformatic surveys suggest that many other bacterial genera are replete with similar biosynthetic potential. While mining actinobacterial genomes, we encountered a contaminated Mycobacteroides data set which included a biosynthetic gene cluster predicted to produce novel phosphonate compounds. Sequence deconvolution revealed that the contig containing this cluster, as well as many others, belonged to a contaminating Bacillus and is broadly conserved among multiple species, including the epiphyte Bacillus velezensis. Isolation and structure elucidation revealed a new di- and tripeptide composed of l-alanine and a C-terminal l-phosphonoalanine which we name phosphonoalamides E and F. These compounds exhibit broad-spectrum antibacterial activity, including strong inhibition against the agricultural pests responsible for vegetable soft rot (Erwinia rhapontici), onion rot (Pantoea ananatis), and American foulbrood (Paenibacillus larvae). This work expands our knowledge of phosphonate metabolism and underscores the importance of including underexplored microbial taxa in natural product discovery.

IMPORTANCE Phosphonate natural products produced by bacteria have been a rich source of clinical antibiotics and commercial pesticides. Here, we describe the discovery of two new phosphonopeptides produced by B. velezensis with antibacterial activity against human and plant pathogens, including those responsible for widespread soft rot in crops and American foulbrood. Our results provide new insight on the natural chemical diversity of phosphonates and suggest that these compounds could be developed as effective antibiotics for use in medicine or agriculture.

INTRODUCTION

Phosphonate and phosphinate (collectively Pns) natural products (NPs) are a class of compounds clinically and industrially used due to their potent bioactivities. Their defining carbon-to-phosphorus (C-P) bond provides resistance to degradation, while isosterism of the Pn moiety to phosphate and carboxylate functional groups enables highly specific competitive or suicide inhibition of essential metabolic enzymes. Primary metabolites containing carboxylate, phosphate ester, and phosphoanhydride groups can be mimicked by Pn inhibitors. Indeed, several Pn NPs have been made commercially available as essential research reagents, antibiotics, and herbicides. Fosmidomycin and FR-900098, isolated from Streptomyces lavendulae and Streptomyces rubellomurinus, respectively, both inhibit 1-deoxy-d-xylulose 5-phosphate reductoisomerase (DXR) to block the Rohmer pathway of isoprenoid biosynthesis (1–4). While this pathway is essential in plant chloroplasts, malarial parasites, and many bacterial pathogens, its absence in humans has prompted exploration of both compounds as potential clinical therapies (5, 6). Fosfomycin was originally discovered from several strains of Streptomyces and later again in Pseudomonas (7, 8). Prescribed for bacterial cystitis, fosfomycin forms a covalent adduct with UDP-N-acetylglucosamine enolpyruvyl transferase to block peptidoglycan synthesis (9). Lastly, phosphinothricin (PT) and its peptide conjugants (both originally isolated from Streptomyces) are potent inhibitors of glutamine synthetase and are the active ingredients within several lines of commercial and industrial herbicides produced by BASF (10). These examples demonstrate the broad importance of Pn NPs in medicine, agriculture, and biotechnology.

Molecular studies have elucidated the biosynthetic pathways for many Pn NPs, revealing a wealth of unusual enzymology and new metabolic paradigms (11, 12). Collectively, these have provided genetic and biochemical markers to identify strains by their Pn production pathways and bioinformatically predict intermediates and end products. Perhaps most importantly, all genetically characterized pathways for Pns initiate with the isomerization of phosphoenolpyruvate (PEP) to phosphonopyruvate (PnPy) by phosphoenolpyruvate mutase (PepM) (13–15). Based on this universally conserved first step, 6% of microbes possess biosynthetic gene clusters (BGCs) containing pepM, of which only a minuscule fraction have been characterized (16). The wealth of undiscovered Pn NPs suggested by these surveys, combined with their historically high rate of commercialization and efficacy (17), have renewed attention on these molecules as a source of novel pharmacophores for drug and herbicide development.

Molecular, genomic, bioinformatic, and analytical methods have been developed to empower access to new Pn NPs by genome mining (16, 18, 19). Targeted discovery of Pns from cryptic (uncharacterized) BGCs has uncovered new herbicidal and antimicrobial compounds, previously undescribed NP scaffolds, and the pathways and enzymes for their biosynthesis (18–23). These studies have also revised our understanding of the origin and function for many Pns, uncovering new roles for compounds previously thought to exist solely as components of macromolecules, metabolic side products, or chemical synthons. Aminomethylphosphonate (AMPn), a component of synthetic drugs, commercial herbicides, and industrial antiscaling agents, was recently found as a component of argolaphos A and B, broad-spectrum antibacterial phosphonopeptides produced by multiple strains of Streptomyces (18). The presence of the AMPn biosynthesis genes among diverse microbes and BGCs suggests its incorporation in many other inhibitory NPs (22). 2-Aminoethylphosphonate (2AEPn) has been estimated to be the most abundant Pn NP in nature given its ubiquity as a headgroup of lipids, glycans, and glycoproteins within bacteria, microbial eukaryotes, and mammals (24). With the recent discoveries of the phosphonocystoximate NP products (18) and fosmidomycin biosynthetic pathway (25), there is growing recognition that 2AEPn also forms the important core scaffold of inhibitory small-molecule Pns.

Genome mining has also revised our understanding of phosphonoalanine (PnAla), a noncanonical phosphonate amino acid best known as a potent agonist of glutamate receptors (26) and a common component of macromolecular lipids in microbial eukaryotes (27). While microbial degradation of PnAla has been well characterized (28, 29), its biosynthesis in bacteria has only recently come to light. The bacterial pathway for PnAla was discovered by targeting Pn BGCs that lacked genes encoding established PepM-coupling enzymes within Streptomyces. Isolation of the Pns from one of these groups resulted in the discovery of the phosphonoalamides, a series of antibacterial peptides containing an N-terminal PnAla residue (20). Heterologous expression established that only genes encoding PepM and a homolog of aspartate transaminase (putative PnPy-transaminase) were required to transform PEP into PnAla (via PnPy). Both genes were present within genetically diverse BGCs, suggesting that PnAla biosynthesis is a new branchpoint in Pn metabolism and a common strategy toward inhibitory Pn NPs (20).

Here, we report the discovery of new PnAla-containing NPs produced by bacteria. Guided by genome mining, our survey of publicly available genomes suggested additional pathways for PnAla natural products from actinobacteria, including mycobacteria. Upon further analysis, we determined that this Pn BGC was not of mycobacterial origin but rather came from Bacillus velezensis. Purification and structure elucidation revealed a new series of phosphonopeptides, defined by PnAla at the carboxyl-terminus. These compounds exhibited strong inhibitory activity against plant and animal bacterial pathogens. The conservation of this BGC within numerous strains and species of plant-associated Bacillus suggests that these compounds may have a role in modulating microbial community composition within their natural environments.

RESULTS

Identification of a Bacillus PnAla BGC from a contaminated actinobacterial data set.

We focused our initial analyses on actinobacteria, which contain the greatest overall diversity of unstudied Pn NP BGCs among bacteria (16). First, we identified all actinobacterial genomes within the NCBI database that contained pepM. The 20-kb neighborhoods surrounding pepM were then screened for the presence of the putative PnAla aminotransferase (20) and absence of phosphonopyruvate decarboxylase (PPD) (30, 31), phosphonomethylmalate synthase (PMS) (32), and phosphonopyruvate reductase (PPR) (23). These four enzymes serve to define the early branchpoints in Pn biosynthesis and provide the thermodynamic driving force necessary to overcome the unfavorable energetics of PnPy formation. These analyses revealed only 13 of the 592 actinobacterial Pn BGCs were suggested to proceed through PnAla.

Based on a PepM sequence identity of 80% or greater, a threshold that effectively delineates both similarity of Pn BGCs and their resulting NPs (16), the 13 PnAla-BGCs were divided into 3 gene cluster families (GCFs). Setting aside the 11 strains previously established to produce phosphonoalamides A to D (20), the remaining 2 were separated into their own individual groups. Extensive genome fragmentation precluded meaningful neighborhood analysis of the first singleton pepM BGC (Streptomyces sp. strain SID7982). Interestingly, the remaining cryptic BGC originated from the genomic assembly of Mycobacteroides abscessus (formerly Mycobacterium abscessus) subsp. massiliense strain aerosol_aerosol_3 (henceforth designated AA3). While only a limited number of NPs have been discovered from mycobacteria, all have proven to be essential to their physiology and pathogenesis. These include siderophores, such as the mycobactins, carboxymycobactins, and exochelins, and the essential cofactor mycothiol (33, 34). Based on this reasoning, we hypothesized that the Pn NP produced from this BGC may also provide similar advantages in Mycobacteroides pathogenesis, including antagonistic properties useful for its survival in human or animal hosts.

We first analyzed the AA3 assembly to verify the source of the BGC and identify strains which could be readily obtained to characterize the potential Pn NPs. To our surprise, BLAST analysis of the pepM from AA3 matched not to mycobacteria but to strains of Bacillus (ranging from 69 to 99% identity). As Mycobacteroides and Bacillus belong to taxonomically distinct phyla, we hypothesized that the Pn BGC may have been derived from horizontal gene transfer or indicate contamination of either the Mycobacteroides isolate or the sequencing process. To understand the nature of this result, we first compared the AA3 data set to those for other closely related strains. Whereas Mycobacterium and Mycobacteroides genomes average 5.6 ± 0.9 Mbp and 5.1 ± 0.3 Mbp, respectively (see Tables S1 and S2 in the supplemental material) (35, 36), the AA3 assembly was a clear outlier at 10.7 Mbp. Our analysis of the 3,914 contigs binned them to six genera of bacteria. The overwhelming majority (64.1%; 2,510 contigs) were assigned to Bacillus, and the remainder of the contigs were distributed among Mycobacteroides (32.4%; 1,268 contigs), Staphylococcus (1.4%; 54 contigs), Rhodococcus (0.05%, 2 contigs), Rhodoferax (0.02%; 1 contig), and Streptomyces (0.02%; 1 contig) (Fig. 1A). Forty-two contigs (1.2%) did not match to any known organisms. Considering total nucleotide content, 80.6% (8.6 Mbp) was attributed to Bacillus and 17.5% (1.9 Mbp) to Mycobacteroides, with all remaining genera each representing <1% of the total data set. Three 16S rRNA genes were detected, matching to Bacillus velezensis Pilsner 2-2 (99.9% identity), Staphylococcus epidermidis NRC 113846 (100%), and Mycobacteroides abscessus T84b (100%). These results clearly showed the AA3 assembly to be a mixed data set with DNA from multiple strains.

FIG 1.

A contaminated genome of Mycobacteroides abscessus subsp. massiliense contains a phosphonoalamide BGC from Bacillus velezensis. (A) Composition of the contaminated genome by genera. The inner pie chart shows the proportion of contigs attributed to each genus, while the outer pie chart shows the proportion of bases attributed to each genus. (B) B. velezensis phosphonoalamide BGC. Dashed lines indicate proposed cluster boundaries. Proteins encoded by M. abscessus are at least 97% identical to those encoded by B. velezensis.

To identify the most likely origin of the Pn BGC in the AA3 assembly, we focused our attention on the contig containing pepM. The gene was carried on a 155-kbp contig (accession no. FVSM01000012) with a mean GC content of 46.7%. This is within the expected average GC content of B. velezensis (45.2 to 47.0%) (Table S3) but lower than for Mycobacterium (66.7%) and Mycobacteroides (64.3%) (Tables S1 and S2). Indeed, the sequence surrounding pepM was almost completely syntenic to the corresponding 100-kb region from Bacillus velezensis NRRL B-41580 (Fig. S1). Furthermore, the predicted proteins within the immediate pepM neighborhood of AA3 shared ≥97% sequence identity with their corresponding homologs in B. velezensis NRRL B-41580 (Fig. 1B; Table 1). Altogether, these results indicate Bacillus as the actual source of the Pn BGC rather than horizontal gene transfer into a member of Mycobacterium or Mycobacteroides. In fact, our reciprocal analysis of publicly available Mycobacterium and Mycobacteroides genomes indicated the complete absence of Pn BGCs from these genera except for those with contaminated data sets (e.g., AA3).

TABLE 1.

Annotation and comparison of putative proteins encoded within the Pn BGCs of M. abscessus subsp. massiliense AA3 and B. velezensis NRRL B-41850

ORFa	M. abscessus subsp. massiliense AA3		% identity	B. velezensis NRRL B-41850
	Accession no.	NCBI annotation		Accession no.	Pfam annotation
orf1	SLC11717	Putative secreted protein	99.2	WP_053285257	Protein of unknown function DUF58
orf2	SLC11728	Uncharacterized protein involved in cytokinesis	99.9	WP_053285256	Transglutaminase-like superfamily
orf3	SLC11771	GMP synthase	100	WP_014417088	Glutamine amidotransferase class I
orf4	SLC11800	Biotin carboxylase	99.5	WP_053285255	ATP-Grasp domain
orf5	SLC11817	Biotin carboxylase	99.5	WP_053285254	ATP-Grasp domain
orf6	SLC11840	Isocitrate lyase family protein	99.0	WP_053285253	Phosphoenolpyruvate phosphomutase
orf7	SLC11886	Aspartate/tyrosine/aromatic aminotransferase	99.5	WP_053285252	Aminotransferase classes I and II
orf8	SLC11928	ABC transporter permease	97.1	WP_053285251	Major facilitator superfamily
orf9	SLC11953	RNA polymerase sigma factor	98.9	WP_014417094	Sigma 70, region 4
orf10	SLC11990	Uncharacterized protein	99.2	WP_053285250	Domain of unknown function
orf11	SLC12013	Putative hypoxanthine/guanine permease	99.8	WP_007408891	Permease family
orf12	SLC12048	Uncharacterized protein	98.9	WP_053285249

^aORF, open reading frame.

To gain insight into the potential structure and function of the potential Pn NPs, we analyzed the genes within the pepM neighborhood. Intriguingly, homologs for orf4 to orf8 (encoding ATP-Grasp ligases, PepM, aminotransferase, and major facilitator superfamily [MFS] transporter) were also present within the phosphonoalamide BGC of Streptomyces sp. strain NRRL B-2790 (Fig. S2). Based on these observations, we hypothesized that strain NRRL B-41580 may be a new source of PnAla-containing phosphonopeptides. However, differences in putative modification enzymes surrounding these five core genes and low sequence identities between the ATP-Grasp ligases suggested that the Pn NPs from B. velezensis would be distinct from those produced by Streptomyces.

Production screening and purification.

Having established the origin of the biosynthetic genes, we grew NRRL B-41580 and three additional strains possessing the same BGC (B. velezensis NRRL B-4257, Bacillus swezeyi NRRL B-41282, and B. swezeyi NRRL B-41294) in several media to identify the best producer strain and growth conditions for Pn production (Fig. S3). As NPs are often produced during stationary phase, we grew the strains for 3 days before harvesting their extracts and analyzing them by ³¹P nuclear magnetic resonance (NMR). Putative Pn NPs, indicated by signals with chemical shifts of 8 ppm or greater, were readily detected from all cultures, except when strains were grown in rhizocticin medium (RM) (Fig. S3). Among these, NRRL B-41580 grown in tryptic soy broth (TSB) demonstrated the greatest intensity of target signals. A subsequent production time course experiment showed maximal accumulation of the putative Pns at 7 days (Fig. S4).

Based on these results, 13 L of NRRL B-41580 TSB culture was prepared for ³¹P NMR-guided purification of the target compounds. We developed a 13-step procedure using differential precipitation, batch sorption, weak-anion exchange, strong-cation exchange, size exclusion, semipreparative flash chromatography, and normal- and reverse-phase high-performance liquid chromatography (HPLC) to purify three Pn NPs to homogeneity (Fig. 2A).

FIG 2.

Purification and structure elucidation for phosphonoalanine (compound 1), phosphonoalamide E (compound 2), and phosphonoalamide F (compound 3). (A) Methods used in the isolation of phosphonate natural products, from beginning (top) to end (bottom). The complete procedure is described in Materials and Methods. (B) Structures of purified compounds. Red arrows indicate ¹H-¹⁵N HMBC correlations. (C and D) ¹H-¹⁵N HMBC spectra of phosphonoalamide E (C) and phosphonoalamide F (D) show that the methylene protons of phosphonoalanine (δ_H ≈ 2 ppm) are correlated with an amide nitrogen (δ_N ≈ 125 ppm), demonstrating that phosphonoalanine is at the carboxy termini of both peptides. Table S7 summarizes all 2D NMR correlations observed from homo- and heteronuclear experiments.

Structure elucidation.

Compound 1 was isolated as a white amorphous solid. Its molecular formula was deduced as C₃H₉NO₅P⁺ from high-resolution mass spectrometry (HRMS) analysis ([M+H]⁺ calculated m/z 170.0213; observed m/z 170.0212; Δppm = −0.6). These values and the ¹H and ³¹P NMR data were identical to literature values of PnAla (Fig. S5 and S6; Table S4) (20). Thus, compound 1 was identified as l-phosphonoalanine (Fig. 2B).

Compound 2 was obtained as a white amorphous solid. Its molecular formula was deduced as C₆H₁₄N₂O₆P⁺ from HRMS analysis ([M+H]⁺ calculated m/z 241.0584; observed m/z 241.0583; Δppm = −0.4 [Fig. S7]). A series of one-dimensional (1D) and 2D homonuclear and 2D heteronuclear NMR experiments was performed to elucidate the chemical structure of this compound (Fig. S8 to S20). The ³¹P NMR spectrum showed a single resonance at 19.54 ppm (Fig. S8). The ¹H spectrum showed resonances for 3 methyl protons (H-3′; δ_H 1.46), 2 methylene protons (H-1; δ_H 1.93, 2.11), and 2 methine protons (H-2’, H-2; δ_H 4.04, 4.35) (Fig. S9). These were corroborated by ¹³C distortionless enhanced polarization transfer (DEPT) 135 and multiplicity-edited ¹H-¹³C heteronuclear single quantum coherence (HSQC) (Fig. S11 and S17) experiments. Analysis in 90% H₂O–10% D₂O revealed one additional amide proton undergoing solvent exchange (H-b; δ_H 8.27) (Fig. S12). ¹H-¹⁵N HSQC and ¹H-¹⁵N heteronuclear multiple bond correlation (HMBC) experiments confirmed that H-b was indeed directly attached to an amide nitrogen atom (N-b; δ_N 124.96) (Fig. S19 and S20).

¹H-³¹P HMBC demonstrated correlation of methylene H-1 and methine H-2 protons with the phosphorus atom (δ_P 19.54) in a pattern consistent with PnAla (Fig. S14). ¹H-¹³C HSQC and ¹H-¹³C HMBC data then established direct attachment of H-1 with secondary carbon C-1 (δ_C 29.71), H-2 with tertiary carbon C-2 (δ_C 51.14), and C-2 that is connected to quaternary carboxylate carbon C-3 (δ_C 177.18) (Fig. S17 and S18). ¹H-¹H correlation spectroscopy (COSY) and total correlation spectroscopy (TOCSY) further verified that H1, H-2, and H-b were in the same spin system and indicated the amide moiety connected to C-2 (Fig. S15 and S16). Lastly, diagnostic splitting and J_CP coupling constants observed in the ¹³C spectrum confirmed the C-P bond (Fig. S10; Table S5). These data indicate that PnAla was a component of compound 2.

The second isolated spin system in the ¹H-¹H COSY and ¹H-¹H TOCSY spectra was consistent with alanine (H-2′, δ_H 4.03; H-3′, δ_H 1.46) (Fig. S15 and S16). This was corroborated by ¹³C DEPT 135, multiplicity-edited ¹H-¹³C HSQC, and ¹H-¹³C HMBC spectra (Fig. S11, S17, and S18). Observation of the free amine nitrogen (N-c; δ_N 40.60) of alanine in ¹H-¹⁵N HMBC spectra (Fig. 2C; Fig. S20), in conjunction with the above-described data, confirmed its direct attachment to the α carbon (C-2′, δ_C 49.34). The correlation between methyl protons of the alanyl side chain (H-3′; δH 1.48) and N-c further supported this as the free amine. Thus, the second half of compound 2 was composed of alanine.

Connectivity between the PnAla and Ala was established with the ¹H-¹³C HMBC data set (Fig. S18). Specifically, a correlation between the methine proton of PnAla (H-2, δ_H 4.33) and the carbonyl carbon of alanine (C-2′, δ_C 49.34) placed the amide moiety of the dipeptide. The absolute configuration of both amino acids was determined as l by Marfey’s analysis (Fig. S21 and S22). Thus, compound 2 is a dipeptide composed of l-Ala–l-PnAla. All NMR data are consistent with this structure (Fig. S8 to S20; Table S5). We named this compound phosphonoalamide E (Fig. 2B).

Compound 3 was obtained as a white amorphous solid. Its molecular formula was deduced as C₉H₁₉N₃O₇P⁺ from HRMS analysis ([M+H]⁺ calculated m/z 312.0955; observed m/z 312.0952; Δppm = −0.9 [Fig. S23]). As described above, 1D and 2D homonuclear and 2D heteronuclear NMR experiments were used to elucidate its chemical structure (Fig. S24 to S36). The ¹H spectrum showed resonances for 13 protons (Fig. S24). Multiplicity-edited ¹H-¹³C HSQC verified them as 6 methyl (H-3′, H-3″; δ_H 1.33, 1.46), 2 methylene (H-1; δ_H 1.98, 2.09), and 3 methine (H-2″, H-2′, H-2; δ_H 4.00, 4.29, 4.37) protons (Fig. S33). The final two amide protons (H-b, H-c; δ_H 8.23, 8.57) were observed to undergo solvent exchange (Fig. S28). ¹H-¹⁵N HSQC confirmed their direct connection to nitrogen atoms (N-b, N-c; δ_N 122.00, 123.49) (Fig. S35). As with compound 2, ¹H and ¹H-³¹P HMBC, ¹H-¹H COSY, ¹H-¹H TOCSY, ¹H-¹³C HSQC, and ¹H-¹³C HMBC data indicated that compound 3 contained PnAla, but with additional constituents attached at N-b (Fig. S24 and S30 to S34).

The observation of two amide protons (H-b and H-c; δ_H 8.23, 8.57) and two carbonyl carbons (C-1′ and C-1″; δ_C 174.17, 170.69) indicated the potential ligation of a dipeptide to PnAla (Fig. S25 and S28). Two additional spin systems within ¹H-¹H COSY and ¹H-¹H TOCSY data sets showed that compound 3 was composed of two alanine residues (Fig. S31 and S32). Placement of the Ala-Ala amide bond was confirmed by correlation between methine proton H-2′ and carbonyl carbon C-1″ (δ_C 170.69) in the ¹H-¹³C HMBC data set (Fig. S34). Correlation between the methyl group of the terminal alanine (H-3″; δ_H 1.46) and the amine of the terminal alanine (N-d; δ_N 40.22) in the ¹H-¹⁵N HMBC data set established alanine at the N-terminus of compound 3 (Fig. 2D; Fig. S36). The absolute configuration of all three amino acids was determined as l by Marfey’s analysis (Fig. S22 and S37). All NMR data are consistent with the structure of compound 3 (Fig. S21 to S34; Table S6) as the tripeptide l-Ala–l-Ala–l-PnAla. We named this compound phosphonoalamide F (Fig. 2B).

High-resolution tandem MS data provided rich fragmentation that further corroborated the structures of compounds 1 to 3 (Fig. 3; Fig. S38 and S39). Notably, ions for PnAla and a diagnostic decarboxylated product (m/z 124.0159) were observed within the MS spectra of all three compounds. An ion corresponding to Ala-PnAla (m/z 241.0577) was present in the spectra of compounds 2 and 3. The structure of compound 3 was further verified by additional daughter ions generated from fragmentation at the Ala-PnAla peptide bond (m/z 143.0813), the Ala-Ala peptide bond (m/z 241.0578), and the Ala-Ala moiety (m/z 115.0868).

FIG 3.

Tandem MS data for phosphonoalanine (A), phosphonoalamide E (B), and phosphonoalamide F (C). For each compound, the tandem mass spectra, annotated to highlight conserved fragmentation patterns, are on the left; their structures with the corresponding fragment ions are on the right.

Having established the complete chemical structures of all three compounds, we returned to analyze the initial concentrated extracts of each strain. LC-HRMS showed that three compounds were indeed present, with compound 3 in the greatest abundance (Fig. S40). These data validate that the Pns were indeed naturally produced and not artifactual derivatives (i.e., chemically modified products) generated during purification.

Antimicrobial activity.

We determined the antimicrobial activities of compounds 2 and 3 against a broad panel of bacteria, yeasts, and filamentous fungi using broth microdilution and disk diffusion assays. While neither compound exhibited antifungal activity, several bacterial strains were inhibited with various degrees of efficacy (Table 2). Compound 3 exhibited potent activity against enteric bacteria, including Escherichia coli K-12 (MIC = 6.25 μM) and Salmonella enterica subsp. enterica serovar LT2 (MIC = 12.5 μM), but neither Salmonella enterica serovar Tennessee E2007000304 nor serovar Livingstone 1236H was inhibited. Compound 3 demonstrated selectivity against Pseudomonas aeruginosa, inhibiting strain K but not PAO1. Serratia marcescens NRRL B-2544 was also inhibited but required a 200 μM concentration of the tripeptide. Considering that B. velezensis is a common epiphyte and soil microbe (37), we hypothesized that our compounds may be effective against bacteria isolated from these environments. Neither peptide inhibited Curtobacterium sp. strain MMLR14_014, Pseudomonas fluorescens Pf5, Pseudomonas mendocina ATCC 25411, or any of the tested Pseudomonas syringae and Acidovorax strains (Table 2). However, strong activity (MIC ≤ 12.5 μM) was observed when Bacillus subtilis ATCC 6633, Bacillus megaterium ATCC 19213, Erwinia rhapontici KSJ3948, Pantoea ananatis LMG 20103, and Paenibacillus larvae NRRL B-2544 were challenged with compound 3. In general, compound 2 demonstrated reduced potency against the assayed strains compared to that of compound 3.

TABLE 2.

Antibacterial activities of phosphonoalamides E (compound 2) and F (compound 3) against strains grown in minimal media

Order and strain	MIC (μM)
	Compound 2	Compound 3
Enterobacterales
Citrobacter freundii ATCC 8090	>200	>200
Enterobacter aerogenes CDC1998-68	>200	>200
Erwinia rhapontici KSJ3948	>200	6.25
Escherichia coli K-12	>200	6.25
Pantoea ananatis LMG 20103	100	3.12
Salmonella enterica subsp. enterica serovar Typhimurium LT2	>200	12.5
Salmonella enterica subsp. enterica serovar Tennessee E2007000304	>200	>200
Salmonella enterica subsp. enterica serovar Livingstone 1236H	>200	>200
Serratia marcescens NRRL B-2544	>200	200
Actinomycetales
Rhodococcus jostii RHA1	>200	>200
Curtobacterium sp. MMLR14_014	>200	>200
Bacillales
Bacillus subtilis ATCC 6633	200	12.5
Bacillus megaterium ATCC 19213	12.5	6.25
Paenibacillus larvae NRRL B-2605	—a	12.5
Staphylococcus aureus ATCC 23055	>200	>200
Pseudomonadales
Acinetobacter calcoaceticus ATCC 23055	>200	>200
Pseudomonas aeruginosa K	>200	25
Pseudomonas aeruginosa PAO1	>200	>200
Pseudomonas fluorescens Pf5	>200	>200
Pseudomonas medocina Palleroni ATCC 25411	>200	>200
Pseudomonas syringae pv. syringae B278a	>200	>200
Pseudomonas syringae pv. tomato Max 14	>200	>200
Burkholderiales
Acidovorax avenae ATCC 19860	>200	>200
Acidovorax citrulli ATCC 29625	>200	>200
Acidovorax delafieldii ATCC 17505	>200	>200
Acidovorax konjacii ATCC 33996	>200	>200
Acidovorax temperans ATCC 49665	>200	>200

^a—, not determined.

Diversity of PnAla biosynthetic pathways in Firmicutes.

We analyzed all PnAla-encoding gene neighborhoods of Firmicutes (Bacillota) in the NCBI nonredundant database to understand the biosynthetic diversity and distribution of this pathway. We identified 95 putative Pn BGCs with colocalized genes for PepM and the putative PnPy-transaminase in the absence of ppd, pms, and ppr. The majority of these PnAla BGCs were contained within Bacillus (95%). The remaining were distributed among Paenibacillus (3%), Abyssisolibacter (1%), and Yanshouia (1%) (Fig. 4A; Table S8). Aside from the eight Bacillus strains with only genus-level identification, all were members of the B. subtilis species complex (B. velezensis, B. amyloliquefaciens, B. swezeyi, B. subtilis, and B. siamensis). While the majority of PnAla-encoding strains originated from soils (29%), plants (24%), and food (19%), isolates cultivated from freshwater (5%), marine (5%), animal (5%), and air (2%) sources also contained the pathway.

FIG 4.

Diversity of PnAla-encoding Firmicutes and their BGCs. (A) 16S rRNA gene phylogenetic tree with Clostridium kluyveri as the outgroup. B. velezensis NRRL B-41580 is labeled red. Colors of the inner ring indicate genera, while those of the outer ring reflect BGC type. (B) The same colors are used to indicate GCFs within the PepM sequence similarity network (80% identity cutoff). Their corresponding pepM neighborhoods, with BGC genes annotated and labeled, are shown. The complete list of strains is provided in Table S8.

We classified the PnAla BGCs into five GCFs based on sequence similarity analysis of their PepM proteins (Fig. 4B). Group 1 contained all PnAla BGCs from Bacillus. Their putative BGCs were nearly identical to those of B. velezensis NRRL B-41580, indicating that all are likely to produce phosphonopeptides with PnAla at the C-terminus. This was supported by the detection of PnAla, compound 2, and compound 3 from additional Bacillus strains within this GCF (Fig. S39).

Paenibacillus contained two different PnAla BGCs. Like those of Bacillus and Streptomyces, the BGC of Paenibacillus forsythiae T98 (group 2) encodes putative ATP-Grasp ligases and an MFS transporter. However, low sequence similarity between these homologs hints at yet another series of phosphonoalamide-like compounds, possibly composed of different amino acids. In contrast, the absence of Pn genes beyond pepM and putative PnPy-transaminase in Paenibacillus sp. strains 32352 and OAS669 (group 3) suggests that biosynthesis may terminate with PnAla. The BGC from Yanshouia hominis BX1 (group 4) encodes an additional amino acid racemase immediately downstream of pepM, which may invert the stereochemistry of PnAla. Lastly, Abyssisolibacter fermentans MCWD3 (group 5) encodes several additional enzymes with the potential to ligate amino acids to PnAla. This may be representative of a pathway toward larger Pn tetrapeptides.

Considering the known limitations of strain isolation, we speculate that the true diversity and abundance of PnAla BGCs in Firmicutes are much higher. Even with the small number of pathways observed in this study, their distribution raises questions about their chemical ecology, evolution, and potential effects on resident microbiome communities. Nonetheless, the antimicrobial activities of compounds 2 and 3 suggest that other NPs containing PnAla may also be effective against bacteria associated with disease.

DISCUSSION

The discovery of phosphonoalamides E and F from B. velezensis enriches our understanding of phosphonate biosynthesis and the chemical diversity of PnAla NPs. Although the BGCs from Streptomyces and Bacillus both include genes for putative PepM, PnPy-transaminase, and ATP-Grasp ligase enzymes, the strains yielded markedly different PnAla-containing phosphonopeptides. Streptomyces sp. B-2790 was previously shown to produce four Pn tripeptides with an N-terminal PnAla attached to combinations of alanine, valine, threonine, or isoleucine (20). In contrast, here we show that alanyl di- and tripeptides with PnAla at the C-terminus are the products of B. velezensis. Given the conservation of biosynthetic genes between these two strains, the diversity of their products most likely stems from differential specificity of their encoded ATP-Grasp ligases. These enzymes catalyze peptide bond formation by first activating amino acids into an acylphosphate intermediate using ATP. This then serves as the substrate for nucleophilic attack by the amine group of the adjoining amino acid (38). Indeed, ATP-Grasp ligases are frequently observed within Pn BGCs with functionally diverse activities. The BGCs for the rhizocticin and plumbemycin phosphonopeptides encode ATP-Grasp ligases implicated in the attachment of amino acids to the nonproteinogenic (Z)-l-2-amino-5-phosphono-3-pentenoic acid (APPA) warhead (19, 39). ATP-Grasp ligases in the valinophos BGC form a series of dipeptides that are attached to the terminal alcohol moiety of 2,3-dihydroxypropyl phosphonate via an unusual ester instead of a canonical amide bond (23). Finally, the role of these enzymes within argolaphos, pantophos, and O-phosphonoacetate serine biosynthesis remains elusive, as none are required for the biosynthesis of these Pn NPs despite their conservation within the BGCs (21, 22, 40). Understanding the catalytic diversity of the ATP-Grasp ligases and the molecular determinants of their activities will be required to improve bioinformatic predictions of end products during Pn genome mining.

The compositional differences between phosphonoalamide F (l-Ala–l-Ala–l-PnAla; from Bacillus) and A (l-PnAla–l-Ala–l-Val, from Streptomyces) resulted in different inhibitory activities against the same panel of bacteria. Phosphonoalamide F exhibited greater inhibitory activity (i.e., lower MICs) against E. coli K-12, S. enterica LT2, P. aeruginosa K, and S. marcescens B-2544 than was previously measured for phosphonoalamide A. This may reflect the natural specificities of different oligopeptide transporters in recognizing various phosphonopeptides and importing them across bacterial cell walls. This “Trojan horse” mechanism underlies the antimicrobial activities of many phosphonopeptides, including bialaphos, dehydrophos, and rhizocticin (41–43). Activity is dependent upon natural uptake and subsequent hydrolysis by cytosolic peptidases, releasing the active phosphonate warhead. This is best exemplified by rhizocticin, plumbemycin, and phosacetymicin, which display differential efficacy against bacteria and fungi depending on the composition of amino acids ligated to APPA (19, 42, 44). The activity profiles of the phosphonoalamides provide further evidence that phosphonopeptide potency and selectivity may be tunable via chemical modifications designed to facilitate cellular import.

Beyond pharmaceutical development, the Bacillus-derived phosphonoalamides have the potential to address the pressing need for commercial bactericides in agriculture. These phosphonoalamides inhibited E. rhapontici and P. ananatis, both of which are major pests of numerous cash crops. E. rhapontici is the causative agent of bacterial soft rot in multiple vegetables, including rhubarb, sugar beets, and tomato, as well as pink seed disease of many commercial grains (45). P. ananatis is a pathogen of maize, rice, tomato, and melons but is most burdensome in onion agriculture, where entire harvests can be lost to it (46, 47). Our compounds also inhibited S. marcescens and P. larvae, devastating bacterial pathogens of the Western honeybee, an essential pollinator used in 90% of commercial agriculture worldwide (48). Although S. marcescens is a natural member of the gut microbiome of honeybees, it is also an opportunistic pathogen which may contribute to their global population decline (49). In contrast, P. larvae is the causative agent of American foulbrood, a fatal honeybee disease that can lead to widespread outbreaks disrupting natural and commercial pollination. The bacterium invades and cannibalizes honeybee larvae, releasing spores that are spread from hive to hive (50). Given the need for effective treatments to mitigate losses from these diseases, the phosphonoalamides represent a potential opportunity for development as commercial pesticides. Understanding the molecular target of the phosphonoalamides and potential mechanisms of resistance will inform their suitability for commercial use and the creation of more potent analogs.

Finally, our results highlight the opportunity for Pn NP discovery in taxa outside actinobacteria. Although actinobacteria harbor the greatest diversity in their NP pathways and have a storied history of producing pharmaceutically and agriculturally useful compounds, Pn BGCs are abundant within other taxa, including the Firmicutes and Proteobacteria (16). Thus, the potential utility and physiological importance of phosphonates produced by these microbes should not be overlooked.

MATERIALS AND METHODS

Strains, media, and general culture conditions.

B. velezensis NRRL B-41580, B. velezensis NRRL B-4257, B. swezeyi NRRL B-41282, and B. swezeyi NRRL B-41294 were obtained from the USDA NRRL strain collection (Peoria, IL). In addition to those in Table 2, strains used in bioactivity assays included Mycobacterium smegmatis NRRL B-14616, Candida albicans ATCC MYA-2876, Debaryomyces hansenii CBS-767, Saccharomyces cerevisiae KSJ4150, Schizosaccharomyces pombe ATCC 24843, Kluyveromyces lactis Y-8279, Sporobolomyces roseus KSJ4237, Aspergillus niger ATCC 16404, Aspergillus nidulans FGSC A4, Neurospora crassa FGSC 4200, and Penicillium notatum ATCC 9478. E. coli, S. enterica, and M. smegmatis strains were grown at 37°C. All other strains were grown at 30°C.

All chemicals and reagents were from Sigma-Aldrich (St. Louis, MO), Fisher Scientific (Waltham, MA), or VWR (Radnor, PA). Media used in this study included nutrient broth (NB; 3 g beef extract, 5 g peptone), GUBC (18), R2AS (18), tryptic soy broth (TSB; 17 g of tryptone, 3 g of soytone, 2.5 g of dextrose, 5 g of NaCl, 2.5 g of K₂HPO₄; pH to 7.3 prior to autoclaving), rhizocticin medium (RM) (42), M9 with 20 mM glucose, YPD (10 g of yeast extract, 20 g of peptone, 20 g of dextrose), yeast minimal medium (YMM; 6.7 g of yeast nitrogen base without amino acids, 20 g of dextrose), and malt extract medium (MEM; 20 g of malt extract, 20 g of dextrose, 6 g of peptone). M9 glucose medium was supplemented with 1 mM thiamine hydrochloride and 0.05 mM nicotinamide for Staphylococcus (SSM9PR) and solution B metals for Paenibacillus larvae (18). All components were dissolved in deionized water (dIH₂O). Sixteen grams of agar was added per liter of medium for plates.

Bioinformatics.

The genomic assembly for Mycobacteroides abscessus subsp. massiliense strain aerosol_aerosol_3 (Assembly accession no. GCA_900138605.1) was retrieved from NCBI and analyzed locally. The taxonomic origin of contigs was determined by analyzing the first 2 kb of each against the NCBI nonredundant nucleotide database using BLASTn and recording the genus of the top hit. Complete genomes of Mycobacterium, Mycobacteroides, and Bacillus velezensis within the NCBI Genomes database were retrieved and analyzed locally.

PepM sequences from actinobacteria were identified and retrieved from the NCBI database as previously described (20). The PepM (WP_239689795) and putative PnPy-transaminase (WP_053285252) from B. velezensis NRRL B-41580 were used as query sequences for cblaster (51) to identify putative gene neighborhoods within Firmicutes. GenBank files were downloaded from the NCBI database using the rentrez R package and manually analyzed to verify the absence of PnPy decarboxylase, PnPy reductase, and phosphonomethylmalate synthase. Open reading frames and synteny were analyzed using cblaster, clinker (52), or EasyFig (53). BLAST analyses were performed against the NCBI nonredundant (nr) protein database and Pfam (54). Sequence similarity networks were created using the Enzyme Function Initiative enzyme similarity tool (55) and visualized using Cytoscape.

16S rRNA gene sequences were identified from assemblies of PnAla-BGC-containing Firmicutes (Table S9). Wherever possible, partial 16S sequences within assemblies were combined with overlapping fragments to create more complete sequences. The 16S rRNA gene sequence from Clostridium kluyveri NBRC 12016 served as the outgroup. Sequences were aligned using the MAFFT (56) and phylogenic tree calculated using FastTree (57).

Production screening.

B. velezensis NRRL B-4257, B. velezensis NRRL B-41580, B. swezeyi NRRL B-41282, and B. swezeyi NRRL B-41294 were revived onto TSB plates from glycerol stocks and inoculated into 20- by 150-mm test tubes containing 5 mL of the same medium. Starter cultures were grown at 30°C for 24 h on a rotary shaker (220 rpm) and used to inoculate 125-mL baffled flasks containing 25 mL of NB, GUBC, R2AS, TSB, or RM (500 μL per flask). These were incubated on a rotary shaker (30°C, 220 rpm) for 3 days. Time course experiments were performed in 2.5-L Ultra-Yield flasks containing 1 L of production medium. These were inoculated with 10 mL of starter cultures and incubated on a rotary shaker (30°C, 220 rpm). Samples were withdrawn (25 mL) at 3, 5, and 7 days of growth.

Cultures were harvested by centrifugation at 18,592 × g and 4°C for 10 min. Clarified supernatants were lyophilized to dryness and reconstituted in 1 mL of dIH₂O. Concentrated extracts were then amended with 10% D₂O for ³¹P NMR analysis. Putative phosphonic acids were identified as ³¹P NMR resonances with chemical shifts of 8 ppm or greater.

Production scale-up.

A starter culture of B. velezensis NRRL B-41580 (as above) was used to inoculate 500-mL Fernbach flasks containing 125 mL of TSB medium. After 36 h of growth at 30°C on a rotary shaker (220 rpm), these seed cultures were used to inoculate 13 individual 2.5-L Ultra-Yield flasks each containing 1 L of TSB. These production cultures were grown for 7 days at 30°C on a rotary shaker (220 rpm). Cultures were harvested by centrifugation (18,592 × g and 4°C for 10 min). The clarified supernatant was removed and set aside. Cell pellets were resuspended in 500 mL of methanol (MeOH) and vigorously agitated by vortexing to extract residual metabolites. The aqueous and methanolic fractions combined to yield 13.5 L of starting material.

Purification of phosphonate natural products.

The above-described material was concentrated to 1 L by rotary evaporation. MeOH was added (75% [vol/vol]) to the crude extract and incubated at −20°C to remove undesired components through bulk precipitation. The soluble fraction was concentrated by rotary evaporation to 500 mL and incubated overnight with Amberlite XAD16 resin (200 g) at 16°C with gentle agitation. Metabolites were sequentially eluted with dIH₂O (750 mL) and MeOH (3 volumes of 750 mL). Pn-containing fractions were combined and subjected to another round of MeOH precipitation (80% [vol/vol]) and reconcentrated to 450 mL. Weak-anion-exchange chromatography (WAX) using iron-chelated Chelex-100 resin was performed, followed by reverse-phase medium-pressure flash chromatography, both as previously described (20). The resulting Pn-containing fractions were combined, concentrated to 20 mL, precipitated with MeOH (90% [vol/vol]), and reconstituted in 30 mL of dIH₂O.

At this point the sample was highly enriched in Pn compounds but contained significant amounts of interfering salts. To remove them, the sample was first fractionated over a Sephadex G25 size exclusion column (5 by 100 cm, 1.9-L bed; dIH₂O at 1 mL min⁻¹; 10-mL fractions). Pn-containing fractions were combined, concentrated, and applied to a Chromabond HLB column and eluted stepwise with MeOH (2.5 by 10 cm, 20 mL bed; 100 mL of 0, 25, 50, 75, and 100% MeOH at 1 mL min⁻¹; 20-mL fractions). Fractions with Pns were combined and concentrated to 30 mL. This was then successively purified over two additional size exclusion chromatography (SEC) columns. The first was a Sephadex LH-20 column (2.5 by 180 cm, 1.2 L bed; dIH₂O at 1 mL min⁻¹; 10-mL fractions), and the second a BioGel P2 column (2.5 by 100 cm, 600 mL bed; dIH₂O at 0.5 mL min⁻¹; 3-mL fractions).

Following fractionation, the sample was lyophilized and dissolved in dIH₂O for purification by HPLC. This was first separated using a Phenomenex (Torrance, CA) Fusion-RP column (10 by 250 cm) using a gradient program with dIH₂O and MeOH (0 to 10 min of 100% dIH₂O, 10- to 45-min linear gradient to 70% MeOH, 45- to 46-min linear gradient to 100% MeOH, 46 to 55 min of 100% MeOH, 55 to 75 min of 100% dIH₂O; 3 mL min⁻¹; 3-mL fractions). The Pns were not retained on the column, indicating a high degree of hydrophilicity. Nonetheless, this step further reduced chemical complexity by removing all remaining hydrophobic compounds. The Pns were then dissolved in 70% acetonitrile (MeCN) and separated on a Waters (Milford, MA) BEH amide column (10 by 250 mm) using a gradient program with dIH₂O and MeCN containing 0.1% formic acid (FA) (0 to 5 min of 70% MeCN, 5- to 35-min linear gradient to 40% MeCN, 35 to 45 min of 40% MeCN, 45- to 50-min linear gradient to 70% MeCN, 50 to 70 min of 70% MeCN; 4 mL min⁻¹; 4-mL fractions). The Pn-containing fractions (12 to 14 min) were combined, reconstituted in 75% MeCN, and separated using an isocratic method with the same column (75% MeCN with 0.1% FA; 4 mL min⁻¹; 4-mL fractions). This yielded 25 mg of purified compound 3, but compounds 1 and 2 remained unresolved from each other. Fractions containing the latter were lyophilized, dissolved in 2 mL of dIH₂O, acidified to pH 3 with FA, and separated by strong cation exchange (SCX) using Dowex 50W-X8 (1 by 10 cm, 5-mL bed; 2 mL min⁻¹; 2.5-mL fractions). The column was eluted first with dIH₂O (50 mL) and then with 0.1% FA (25 mL). Water fractions containing Pns were combined to yield 0.8 mg of pure compound 1. FA fractions containing Pns were combined to yield 0.8 mg of pure compound 2.

NMR analyses and mass spectrometry.

NMR experiments were performed at the Ohio State University Campus Chemical Instrument Center. All NMR spectra were acquired at 25°C on a Bruker Avance III HD Ascend 600-MHz spectrometer (600 MHz for ¹H, 150 MHz for ¹³C, and 243 MHz for ³¹P) equipped with a Bruker 5-mm Smart Broadband Observe solution probe (BBFO), a Bruker Avance Neo 400-MHz spectrometer (400 MHz for ¹H, 100 MHz for ¹³C, and 162 MHz for ³¹P) equipped with a 5-mm Prodigy cryoprobe, or a Bruker Avance III HD Ascend 700-MHz spectrometer (700 MHz for ¹H, 176 MHz for ¹³C, and 283 MHz for ³¹P) equipped with a 5-mm triple-resonance Observe (TXO) cryoprobe. Spectra were processed using MestReNova 12 software. HRMS and HRMS/MS were performed on an Agilent 6540 UHD Accurate-Mass quadrupole time of flight (Q-TOF) system equipped with an Agilent 1260 Infinity II HPLC as previously described (20). NMR and MS data were analyzed using MestReNova 12.

Marfey’s analysis.

Compounds 2 and 3 (0.2 mg) were dissolved in 0.5 mL of 6 N HCl and heated to 100°C in a sealed reaction vial for 16 h. Samples were dried at 40°C under a gentle stream of air to remove HCl. Hydrolysates were dissolved in 50 μL of water and transferred to microcentrifuge tubes, and 50 mM solutions (50 μL) of each amino acid standard were prepared. Each sample or standard was combined with 20 μL of 1 M NaHCO₃ and 100 μL of a 1% solution of N_α-(2,4-Dinitro-5-fluorophenyl)-l-alaninamide (FDAA) in acetone and incubated in a heating block at 40°C for 1 h. Samples were cooled to room temperature and neutralized with 20 μL of 1 M HCl. Derivatized amino acid standards and hydrolysates were diluted 50-fold into 10% MeCN with 0.1% FA. Derivatized PnAla solutions were diluted 50-fold into water with 0.1% FA. These were analyzed by LC-MS using a Phenomenex Fusion-RP column (2 by 100 mm, 4 μm) with a gradient of 0 to 100% MeCN containing 0.1% FA over 30 min.

Antimicrobial assays.

Compounds 2 and 3 were repeatedly lyophilized and exchanged with ultrapure dIH₂O to ensure that all residual modifier was removed prior to assays. Susceptibility testing was performed using the broth microdilution method in 96 round-well microtiter plates following general guidelines by the Clinical and Laboratory Standards Institute as previously described (20). Compounds were prepared as 50× stocks (10 mM) by dissolving in dIH₂O. All bacterial strains were grown and assayed in M9 glucose medium (with amendments added as required). Yeast strains were grown and assayed in YMM. Controls wells contained 200 μM kanamycin (for bacteria) or nystatin (for yeasts), no compound addition (vehicle only), and no cell addition (added media instead). Culture densities (optical densities at 600 nm [OD₆₀₀]) were recorded using a Bio-Rad xMark microplate spectrophotometer after 16 h for all strains except bacilli, which were recorded after 24 h. MIC was defined as the lowest concentration of compound that resulted in ≥90% growth inhibition. MIC values are from triplicate assays performed on separate days. Disk diffusion assays were used to assess antimicrobial activity of the purified compounds against filamentous fungi and against M. smegmatis NRRL B-14616 as previously described (20).