A Novel Pathway for Biosynthesis of the Herbicidal Phosphonate Natural Product Phosphonothrixin Is Widespread in Actinobacteria

ABSTRACT

Phosphonothrixin is an herbicidal phosphonate natural product with an unusual, branched carbon skeleton. Bioinformatic analyses of the ftx gene cluster, which is responsible for synthesis of the compound, suggest that early steps of the biosynthetic pathway, up to production of the intermediate 2,3-dihydroxypropylphosphonic acid (DHPPA) are identical to those of the unrelated phosphonate natural product valinophos. This conclusion was strongly supported by the observation of biosynthetic intermediates from the shared pathway in spent media from two phosphonothrixin producing strains. Biochemical characterization of ftx-encoded proteins confirmed these early steps, as well as subsequent steps involving the oxidation of DHPPA to 3-hydroxy-2-oxopropylphosphonate and its conversion to phosphonothrixin by the combined action of an unusual heterodimeric, thiamine-pyrophosphate (TPP)-dependent ketotransferase and a TPP-dependent acetolactate synthase. The frequent observation of ftx-like gene clusters within actinobacteria suggests that production of compounds related to phosphonothrixin is common within these bacteria.

IMPORTANCE Phosphonic acid natural products, such as phosphonothrixin, have great potential for biomedical and agricultural applications; however, discovery and development of these compounds requires detailed knowledge of the metabolism involved in their biosynthesis. The studies reported here reveal the biochemical pathway phosphonothrixin production, which enhances our ability to design strains that overproduce this potentially useful herbicide. This knowledge also improves our ability to predict the products of related biosynthetic gene clusters and the functions of homologous enzymes.

INTRODUCTION

Herbicide resistant weeds have become a major challenge in modern agriculture, creating an urgent need for new phytotoxic molecules with novel biological targets that would evade existing resistance mechanisms (1). The 30-year gap since discovery of the last herbicide with a new target further emphasizes the need to develop strategies and sources for discovery of these compounds (2). It has been suggested that microbial natural products represent a promising and underexplored source of herbicides (3). Among these is phosphonothrixin, a herbicidal natural product made by the actinobacterium Saccharothrix sp. ST-888, which was isolated from soil near Iwaki-city, Japan in the mid-1990s (4). This molecule inhibits seed germination and induces chlorosis in gramineous and broadleaf weeds via a mechanism that does not involve the inhibition of chlorophyll biosynthesis, suggesting it may act on a different molecular target than known herbicides. Moreover, phosphonothrixin is inactive against bacteria and fungi, suggesting that it may have minimal impacts on the soil microbiome.

Structural characterization of phosphonothrixin revealed it to be a phosphonic acid (5), an emerging class of natural products with a striking propensity for useful bioactivities defined by their stable carbon-phosphorus (C-P) bonds (6). The diverse bioactivities of these molecules stem from their structural similarity to phosphate esters, carboxylic acids, and the catalytic intermediates formed by enzymes that act on these substrates. Accordingly, phosphonate isosteres of common metabolites often display highly specific binding to essential metabolic enzymes. However, due to the inert nature of the C-P bond, these molecules typically act as competitive inhibitors, rather than substrates. In other cases, appropriately placed labile moieties within phosphonate inhibitors react with active site residues, resulting in covalent inactivation of the enzyme (7, 8). Due to the prevalence of phosphate esters and carboxylic acids in essential metabolic pathways across all forms of life, the range of potential targets, including novel ones, for phosphonate inhibitors is vast, which may account for the higher rate of phosphonate commercialization relative to other natural product classes (6).

Phosphonothrixin is one of only two phosphonate natural products with a branched carbon skeleton, the other being the recently discovered herbicide pantaphos (9). This suggests novel biochemistry will be involved in its biosynthesis, an idea that is supported by analysis of the recently identified phosphonothrixin biosynthetic gene cluster (BGC) from Saccharothrix sp. ST-888 (10). This gene cluster is comprised of a putative 14-gene operon that confers production of phosphonothrixin when expressed in heterologous hosts. Like other characterized phosphonate BGCs, this operon encodes the enzyme phosphoenolpyruvate phosphonomutase (PEP mutase), which catalyzes the first step in all characterized phosphonate biosynthetic pathways. However, it lacks genes encoding known enzymes that catalyze the thermodynamically favorable subsequent steps needed to drive the unfavorable PEP mutase reaction (6). Although the decarboxylation-driven condensation of pyruvate and phosphonopyruvate (PnPy) was originally proposed to serve this purpose (10), recent data suggest that the biosynthetic pathway is actually more similar to that involved in the synthesis of the unrelated phosphonate natural product valinophos (11, 12). In this revised pathway, the thermodynamic driving force needed for the PEP mutase reaction is provided by the exergonic reduction of PnPy to phosphonolactate, followed by the multistep conversion to 2,3-dihydroxypropylphosphonic acid (DHPPA), the last common intermediate in the valinophos and phosphonothrixin pathways (Fig. 1). The remaining steps in the pathway involve oxidation of DHPPA to hydroxyoxopropylphosphonate (HOPPA) and subsequent conversion to phosphonothrixin by the combined action of two thiamine pyrophosphate (TPP)-dependent enzymes.

FIG 1.

The ftx operon of Kitasatospora (formerly Saccharothrix) sp. ST-888 and predicted phosphonothrixin biosynthetic pathway. The 14 gene ftx operon is depicted at the top with predicted and/or experimentally validated functions of the encoded proteins shown above each gene. The proposed phosphonthrixin biosynthetic pathway is shown below with the enzyme catalyzing each reaction indicated. Genes without a role in the proposed pathway are color coded in gray. Abbreviations: phosphoenolpyruvate (PEP), phosphonopyruvate (PnPy), phosphonolactate (PnLac), phosphonolactate-phosphate (PnLac-Pi), phosphonolactaldehyde (PnLald), 2,3-dihydroxypropylphosphonic acid (DHPPA), 3-hydroxy-2-oxopropylphosphonate (HOPPA).

Here, we use a combination of in vitro and in vivo approaches to show that biosynthesis of phosphonothrixin via the common valinophos/phosphonothrixin pathway is widespread in actinobacteria. We also provide additional biochemical evidence for the proposed pathway and the function of two unusual TPP-dependent enzymes which combine to produce phosphonothrixin. The revised pathway uses most, but not all, of the genes in the phosphonothrixin BGC, suggesting that phosphonothrixin may be an intermediate rather than the final product of biosynthesis.

RESULTS

Bioinformatics analyses of the phosphonothrixin biosynthetic gene cluster.

The phosphonothrixin BGC from Saccharothrix sp. ST-888 is comprised of 14 genes arranged in a putative operon (Fig. 1), which we refer to as ftxA through ftxN to conform with standard bacterial nomenclature (13). Bioinformatic analyses of the encoded proteins strongly support the idea that phosphonothrixin is made by a pathway that shares early steps with the recently characterized valinophos pathway. Valinophos is synthesized via the intermediate (R)-2,3-dihydroxypropylphosphonic acid (DHPPA), which is made from PnPy in four steps catalyzed by the enzymes VlpB, VlpD, VlpC, and VlpE (12). Homologs of each of these enzymes are encoded by the ftx gene cluster, including FtxM (53% identity with VlpD), FtxL (64% identity with VlpC), and FtxN (54% identity with VlpE). The fourth enzyme FtxJ is more complex, but still consistent with the functions required for early steps in the valinophos pathway. FtxJ is a putative multifunctional protein comprised of an N-terminal domain with homology to the adenylation domain of nonribosomal peptide synthetases, and a C-terminal dehydrogenase domain with 57% identity with VlpB. Generic roles can be proposed for each of the remaining Ftx proteins, with data presented below supporting several of these assignments. FtxK is a member of the medium-chain dehydrogenases/reductase (MDR) alcohol dehydrogenase family, which are known to oxidize secondary alcohols to ketones (14). FtxEF and FtxG are both predicted as thiamine-pyrophosphate (TPP)-dependent enzymes. FtxEF is a heterodimeric member of the transketolase superfamily, whereas FtxG is a member of the decarboxylase superfamily (15), with the closest homolog in the Swiss-Prot database being acetolactate synthase from Lactococcus lactis (16). FtxI is a member of the 2-oxoglutarate-dependent dioxygenase family, which usually catalyze hydroxylation reactions, although a variety of other transformations, including desaturation, ring expansion and halogenation are known within this group (17). FtxA, FtxH, and the N-terminal domain of FtxJ are likely to comprise a multisubunit nonribosomal peptide synthetase (18); whereas FtxB, a member of the ATP-Grasp family, may also catalyze peptide biosynthesis (19). Finally, FtxC is a member of the major facilitator superfamily and likely to export the final product (20).

Phylogeny and genetic conservation of organisms encoding homologs of the ftx BGC.

We previously identified 13 actinobacteria that encode the ftx-like BGCs, including Saccharothrix sp. ST-888 (13), the strain in which phosphonothrixin was first identified (4). In the course of this work, we identified an additional 10 actinobacterial strains carrying homologous gene clusters. Similar BGCs were not observed in any taxa outside actinobacteria. To provide insight into the distribution and conservation of ftx-like BGCs, we examined the phylogeny of the twenty-two strains and genetic conservation within their homologous BGCs (Fig. 2). S. albus subsp. Albus B-2513 also encodes an ftx-like BGC; however, due to the poor quality of the draft genome sequence we did not include it these analyses. Organisms encoding ftx-like BGCs have global distribution with individual strains having been isolated in Japan, Europe, and North America (Table S1). Based on a larger phylogenetic analysis, including rpoB sequences from all currently available actinobacterial genomes in the NCBI RefSeq database (Data set S1), the 23 strains include a wide diversity of actinobacteria. Eleven strains fall within the genus Kitasatospora, including several strains that were originally classified as species of Streptomyces or Saccharothrix. This includes the original producing strain, here designated Kitasatospora sp. ST-888. In this regard, it should be noted that older taxonomic assignments often differ from the more accurate taxonomies produced by modern molecular methods (21). Of the remaining strains, four are members of the genus Streptomyces, three Saccharopolyspora, and one each in Saccharothrix, Cellulomonas, and Nocardia.

FIG 2.

Phylogeny and conserved genes of organisms encoding homologs of the ftx BGC. A phylogenetic tree of the organisms encoding ftx-like BGCs based on their respective rpoB genes is shown on the left. Bootstrap support for the tree structure is indicated at each node. Based on a much larger phylogenetic analysis encompassing 3116 actinobacteria (Data set S1), those whose names are colored green fall into the genus Kitasatospora (despite their names); those in red, the genus Streptomyces; those in orange, Nocardia; and those in brown, Saccharopolyspora. Both Saccharothrix algeriensis DSM 44581and Cellulomonas sp. zg-B12 are correctly classified by this method. Strains marked with an asterisk have previously been shown to produce phosphonothrixin. The structures of the homologous BGCs are shown on the right with conserved genes labeled with the homologous ftx gene and matching colors.

As might be expected, conservation of genes within the ftx-like BGCs is consistent with the phylogeny of the encoding organisms (Fig. 2). Thus, all 11 Kitasatospora strains carry an ftx operon that is essentially identical to that found in K. sp. ST-888. The ftx operon in the four Streptomyces is also nearly identical, lacking only the ftxB gene. Accordingly, we expect the ftx-like BGCs in both the Kitasatospora and Streptomyces clades to direct biosynthesis of phosphonothrixin (or a highly similar molecule). Indeed, several of the strains in both clades have been shown to produce phosphonothrixin based on high resolution mass spectrometry (HR-MS) (22). Although the genetic organization differs, the ftx-like BGC of the Saccharopolyspora strains is also similar to the K. sp. ST-888 ftx operon, encoding homologs of 13 of the 14 genes, suggesting that these strains also have the capacity to produce phosphonothrixin congeners. The remaining ftx-like BGCs found in Nocardia, Cellulomonas and true Saccharothrix species are more divergent; however, all encode homologs of ftxD, ftxE, and ftxF and a highly conserved ftxJKLMN cassette in which ftxJ encompasses only the dehydrogenase domain.

Production of phosphonates by organisms encoding ftx-like BGCs.

To validate phosphonothrixin production and provide support for the proposed biosynthetic pathway, we grew five strains with ftx-like gene clusters in a variety of media. The presence and identity of phosphonates in the spent media were then assessed by NMR and coupled liquid chromatography/high-resolution mass spectrometric (LC-HRMS) analyses. K. sp. ST-888 and Kitasatospora (formerly Streptomyces) sp. NRRL F-6131 produced multiple phosphonates, two of which were definitively assigned as phosphonothrixin and the proposed pathway intermediate DHPPA (Fig. 3). A third compound was identified as 3-hydroxy-2-oxopropyl phosphonate (HOPPA, see below). S. sp. WM4235, S. roseoverticillatus NRRL B-3500 and Saccharopolyspora spinosa NRRL 18395 did not produce phosphonates under the conditions examined here, although S. roseoverticillatus NRRL B-3500 has previously been shown to make phosphonothrixin (22).

FIG 3.

Production of phosphonothrixin and pathway intermediates by Kitasatospora sp. ST-888 and NRRL F-6131. A) ³¹P NMR spectrum of concentrated spent media from K. sp ST-888 is shown at the top, the ¹H-³¹P HMBC NMR signals that correspond to each ³¹P signal are shown below. B) ³¹P NMR spectrum of concentrated spent media from K. sp NRRL F-6131 is shown at the top, the ¹H-³¹P HMBC NMR signals that correspond to each ³¹P signal are shown below. C) LC-HRMS analysis of the same extracts. Extracted ion chromatograms of DHPPA (m/z 155.0115 [M-H]-), HOPPA (m/z 152.9958 [M-H]-), and phosphonothrixin (FTX) m/z (197.0220 [M-H]-) are shown. S. durhamensis was used as a positive control for DHPPA and negative control for HOPPA and FTX. Other intermediates within the pathway were not detected.

FtxJ encodes a multidomain protein with phosphonopyruvate reductase activity.

A significant difference between the early steps of the valinophos and phosphonothrixin pathways lies in the putative bifunctional FtxJ protein versus mono-functional VlpB protein. To test whether FtxJ catalyzes the same reaction as VlpB, namely, reduction of PnPy to phosphonolactate (PnLac), we purified FtxJ after overexpression in E. coli and tested its activity in a coupled reaction with PEP mutase (Fig. 4A and Fig. S2 to 7). Because the PEP mutase (FtxD) from K. sp. ST-888 was poorly expressed in E. coli, we used the well-characterized rhiH-encoded PEP mutase from B. subtilus to produce PnPy in these experiments (23). Analysis of overnight reactions via 1D and 2D NMR experiments showed the production of PnLac in assays containing as purified FtxJ, but not when the enzyme was heat killed. Thus, the C-terminal domain of FtxJ performs the same function as VlpB. Whether the enzyme also catalyzes the adenylation reaction of a nonribosomal peptide synthase (NRPS) was not examined due to the large number of possible substrates for such a reaction and the possible requirement for other proteins within the putative multisubunit NRPS complex.

FIG 4.

³¹P NMR analysis of biochemical enzyme reactions using purified Ftx enzymes. A) Conversion of PEP to PnLac in a coupled reaction with FtxJ and PEP mutase. NADH used in the reaction was continuously regenerated via NAD⁺-dependent phosphite oxidation using the enzyme PtxD. The teal spectrum shows the reaction using heat-inactivated FtxJ. The red spectrum shows the reaction using as purified FtxJ. Additional NMR analyses of the active reaction are shown in Fig. S2-S7. B) Conversion of DHPPA to HOPPA by FtxK (forward reaction). NAD⁺ used in the reaction was continuously regenerated via NADH-dependent reduction of pyruvate using the enzyme lactate dehydrogenase. The teal spectrum shows the reaction using heat-inactivated FtxK. The red spectrum shows the reaction using as purified FtxK. Additional NMR analyses are shown in Fig. S8 to 13. C) Conversion of HOPPA to DHPPA by FtxK (reverse reaction). NADH used in the reaction was continuously regenerated via NAD⁺-dependent phosphite oxidation using the enzyme PtxD. The blue spectrum shows the reaction using heat-inactivated FtxK. The green spectrum shows the reaction using as purified FtxK. The red spectrum shows the green reaction after spiking with authentic DHPPA. Additional NMR analyses are shown in Fig. S15-S20. (D) Conversion of HOPPA and pyruvate to phosphonthrixin catalyzed by the combined action FtxEF and FtxG. The teal spectrum shows the reaction using heat-inactivated FtxEF. The red spectrum shows the reaction using as purified FtxEF and FtxG.

FtxK is a DHPPA dehydrogenase.

The data presented above show that the valinophos and phosphonothrixin biosynthetic pathways share early steps up to the production of DHPPA. Thus, DHPPA should be a substrate for one of the uncharacterized ftx-encoded enzymes. To test this, we purified authentic DHPPA from spent medium of the valinophos producer Streptomyces durhamensis for use as a substrate in assays with enzymes encoded in the ftx operon. Based on our bioinformatics analyses, FtxK is expected to catalyze the NAD(P)-dependent oxidation of an alcohol and, thus, is an obvious candidate for the enzyme that would metabolize DHPPA. Indeed, when recombinant FtxK was incubated with NAD⁺ and DHPPA a new product was produced as indicated by ³¹P-NMR spectroscopy (Fig. 4B). Based on NMR analyses, this product was shown to be identical to synthetic HOPPA (Fig. S8 to 13), which was also observed in spent media from two phosphonothrixin producers (see above). It should be noted that the oxidation of alcohols to ketones with NAD⁺ is highly unfavorable. Thus, these reactions were performed using pyruvate and lactate dehydrogenase as a cofactor regeneration system to drive the reaction in the forward direction. We also examined the reverse reaction, reduction of HOPPA to DHPPA with NADH, which is expected to be thermodynamically favorable. This reaction, which used synthetic HOPPA, produced a new peak in ³¹P NMR experiments, which was shown to be DHPPA by co-spiking with authentic material purified from S. durhamensis (Fig. 4C and S14 to 19). Thus, FtxK is an NAD-dependent DHPPA dehydrogenase.

The roles of FtxEF and FtxG in phosphonothrixin synthesis.

Transfer of an acetyl-moiety to the C-2 position of HOPPA would afford phosphonothrixin. Similar ketotransferase reactions are often catalyzed by TPP-dependent enzymes, thus we suspected that either FtxG or FtxEF would catalyze the next step in the phosphonothrixin pathway. To test this idea, we tried to purify K. sp ST-888 FtxEF after expression of a His-tagged allele in E. coli; however, poor expression and solubility led to very low yields. Therefore, we turned to expression of the homologous FtxEF proteins (89.6% and 90.3% identity, respectively) from the close relative K. sp. NRRL F-6131, which, as shown above, also produces phosphonothrixin. Coexpression of His-tagged FtxE with untagged FtxF allowed purification of both proteins as a heterodimeric complex, which has been observed in other TPP-dependent enzymes such as the apulose-4-phosphate transketolases and phosphonopyruvate decarboxylase (23, 24). The most common acetyl-donor for TPP dependent enzymes is pyruvate, which is decarboxylated during the ketotransferase reaction (25); however, we did not observe any new products by ³¹P NMR when HOPPA and pyruvate were incubated with purified FtxEF. Similar attempts with other known TPP-dependent acetyl-donors, including acetoin, d-xylulose 5-phosphate and d-fructose 6-phosphate, also failed to generate new products.

The presence of FtxG, a putative acetolactate synthase, prompted us to consider whether acetolactate could be the acetyl-donor for the FtxEF reaction. Initially, we tested acetolactate synthase activity of K. sp. NRRL-6131 FtxG purified after overexpression of a His-tagged allele in E. coli. As a control we also assayed the activity of the previously characterized catabolic acetolactate synthase (AlsS) from Bacillus subtilis (26). Both enzymes consumed pyruvate and produced acetolactate based on NMR analyses (Fig. S20), confirming FtxG as an acetolactate synthase. Because acetolactate is chemically unstable and not commercially available, we tested whether FtxEF could use acetolactate as an acetyl donor for phosphonothrixin synthesis in a coupled reaction with FtxEF, FtxG, pyruvate and HOPPA. In these reactions, roughly 50% of the HOPPA was converted to a new phosphonate signal at δ 16.1 ppm (Fig. 4D). Additional NMR and HRMS analyses of large-scale reactions, using HOPPA produced in situ from DHPPA with FtxK, confirmed this molecule as phosphonothrixin (Fig. S21-S26). Surprisingly, very little phosphonothrixin was made when acetolactate was produced in a separate reaction, with removal of FtxG by ultrafiltration prior to addition of FtxEF (Fig. S27). This result suggests that complex formation between FtxEF and FtxG is required for efficient conversion of pyruvate and HOPPA to phosphonothrixin.

DISCUSSION

The data presented here support the prediction of Zhang et al. that the biosynthetic pathways for phosphonothrixin and valinophos would be identical up to production of the intermediate DHPPA (12). Our results are also largely consistent with those of Zhu et al. (11), although we disagree with the interpretation of these data (see below). The two pathways then diverge, with phosphonothrixin being made via oxidation of DHPPA to HOPPA, and subsequent condensation with pyruvate through the concerted action of two novel TPP-dependent enzymes. The shared pathway for DHPPA synthesis differs significantly from other known phosphonate biosynthetic pathways by use of the thermodynamically favorable reduction of PnPy to PnLac to drive the unfavorable PEP mutase reaction. Although valinophos and phosphonothrixin are currently the only characterized natural products to use this biosynthetic strategy, the PnPy reductase gene was shown to be common in microbes, with at least four distinct classes of putative phosphonate biosynthetic pathways branching from the PnLac intermediate (12). The branch leading to phosphonothrixin can now be defined by the presence of FtxK, a novel NAD-dependent dehydrogenase that oxidizes DHPPA to HOPPA, and two unusual TPP-dependent enzymes (FtxG and the heterodimeric FtxEF) that combine to convert HOPPA and pyruvate to phosphonothrixin (Fig. 1).

While the evidence that coincubation of FtxEF and FtxG is required for phosphonthrixin biosynthesis is strong, the mechanism by which these TPP-dependent enzymes catalyze this reaction remains unclear. Because FtxEF produces small amounts of phosphonothrixin using the products of the FtxG reaction, it is likely that FtxG synthesizes an acetyl-donor which is used by FtxEF for condensation with HOPPA. However, the nature of this intermediate and the mechanism by which it is transferred to FtxEF cannot be definitively assigned based on the evidence in hand. Three possibilities exist. The simplest explanation is that FtxG produces acetolactate, which is then used as the acetyl donor by FtxEF. Accordingly, FtxG is a member of the decarboxylase (DC) superfamily, which includes the acetolactate synthases, based on comparisons to a well-curated database of TPP-dependent enzymes (27). Moreover, our data, and those of Zhu et al., show that FtxG converts pyruvate to acetolactate in vitro (11). However, this reaction is known to be a side activity of other members of the DC superfamily, including glyoxylate carboligase and pyruvate oxidase (28). Thus, it is unclear whether the acetolactate synthase activity of FtxG is relevant for phosphonthrixin synthesis. An alternative possibility derives from the finding that FtxG from K. sp. ST-888 also converts free TPP into free hydroxyethyl-TPP using pyruvate as a cosubstrate (11). Based on this result, Zhu et al. suggested that FtxG is a novel “HE-TPP releasing enzyme” and that the true substrate for FtxEF is free HE-TPP. It should be noted, however, that these reactions utilized extraordinarily high concentrations of TPP (2 to 5mM), more than 1000-fold higher than the mean Kd of 3 μM for characterized TPP-dependent enzymes (29). Thus, the physiological relevance of this activity is also questionable. A third possibility, consistent with the requirement for coincubation of the enzymes, is that FtxEF and FtxG form a complex in which an acetyl-donating intermediate made by FtxG from pyruvate is donated directly to FtxEF for condensation with HOPPA. We speculate that this intermediate is either acetolactate or a direct transfer of the hydroxyethyl moiety from the TPP cofactor of FtxG to the TPP cofactor of FtxEF. Further studies will be required to distinguish between these possibilities.

The pathway for biosynthesis of phosphonothrixin described herein uses only 10 of the 14 proteins encoded by the ftx operon. Moreover, the N-terminal adenylation domain of FtxJ is also not used in the pathway. Therefore, it seems likely that phosphonothrixin is an intermediate, rather than the final product of the ftx operon. As described above, the putative ATP-Grasp enzyme FtxB and nonribosomal peptide synthase comprised of FtxA, FtxH, and the N-terminal domain of FtxJ are likely to catalyze condensation of phosphonothrixin with amino acids. Peptide modifications are common in phosphonate natural products, including phosphinothricin tripeptide, dehydrophos, rhizocticin, and plumbemycin (6). In these natural products, the peptide modifications promote uptake of the bioactive phosphonate moiety, which is subsequently released by ubiquitous cytoplasmic peptidases (often referred to as the “Trojan Horse” strategy). It should be noted that phosphonothrixin does not have a free amine that would typically be used as the nucleophile in the amide bond forming reactions catalyzed by nonribosomal peptide synthetases and ATP-grasp enzymes. Thus, the nature of the linkages created by these remaining ftx-encoded enzymes is speculative and likely involves novel biochemistry. The final gene in the ftx operon without a defined role in the biosynthetic pathway is ftxI, which is predicted to encode a α-ketoglutarate dependent dioxygenase. Although other activities have been shown in this enzyme superfamily, the primary activity of α-ketoglutarate dependent dioxygenases is hydroxylation of their substrates. Thus, it seems likely that additional modifications, beyond peptide formation will be found on the final ftx product.

The ftx-like BGCs are common in actinobacteria. Each of these BGCs include homologs of ftxD (PEP mutase), as well as a conserved ftxJKLMN gene cassette, which encode the enzymes needed to produce HOPPA. Most also include homologs of ftxEF and ftxG, suggesting they could produce phosphonothrixin; however, the ftx-like BGCs of Saccharothrix, Nocardia, and Cellulomonas lack ftxG homologs. If these organisms indeed make phosphonothrixin, then their FtxEF enzymes must use an acetyl donor that is readily available in the cell, such as the central metabolite pyruvate or acetolactate, a required intermediate in branched-chain amino acid biosynthesis. Genes encoding the putative peptide forming enzymes are also poorly conserved in the ftx-like BGCs observed in other actinobacteria. This finding is consistent with variation of these putative modifications being used to promote targeting of specific organisms, as in the case of the antibacterial plumbemycin and antifungal rhizocticin, which share the same bioactive phosphonate warhead with different amino acid modifications directing uptake to bacteria and fungi, respectively (30, 31). Finally, it is interesting to note that, while ftx-like BGCs are common in actinobacteria, they have not yet been observed in any other bacterial, archaeal, or eukaryotic organisms. Whether this distribution reflects shared phylogeny, or an environmental adaptation specific to the actinobacteria that carry these genes, is unknown. Identification of the biological target of phosphonothrixin, which is also currently unknown, may help to clarify this issue.

MATERIALS AND METHODS

Bacterial strains, plasmids, and culture conditions.

All primers, strains and plasmids used in this study are listed in Table S1. Kitasatospora (formerly Saccharothrix) sp. ST-888 was obtained from the National Institute of Bioscience and Human Technology, Agency of Industrial Science & Technology, Japan. Actinobacterial strains were maintained on ATCC172 medium (22). Escherichia coli strains were grown in Luria-Bertani (LB) broth or on 1.5% agar LB plates containing appropriate antibiotics. Antibiotics were used at the following concentration: ampicillin 100 μg/mL, kanamycin 50 μg/mL, chloramphenicol 15 μg/mL. Unless otherwise stated, all reagents were purchased from Sigma-Aldrich or ThermoFisher Scientific.

Identification and phylogenetic analysis of organisms encoding ftx-like gene clusters.

Organisms encoding homologous ftx-like gene clusters were identified using a catenated version of the ftx cluster as a query against NCBI reference genomes and whole-genome shotgun contigs databases with a cutoff of 40% Query Coverage. Potential hits were then examined using MultiGeneBlast (32) to generate a list of strains containing ftx-like gene clusters. Additional homologous genes in the immediate vicinity of these hits were identified by BLAST comparison of individual proteins with those encoded by the Kitasatospora sp. ST-888 ftx genes. Phylogenetic analysis of these organisms was accomplished by retrieving the rpoB DNA sequences from each organism and trimming to match an internal fragment widely used for multilocus sequence typing (21). Sequences were then aligned using the translation alignment tool provided in the Geneious Prime software package (Biomatters, Aukland, New Zealand). The resulting alignment was then used as input for RaxML using default parameters with 100 bootstrap replicates. The larger tree used for phylogenetic placement of these organisms was created using FastTree (33) with the rpoB sequences of 3116 actinobacteria present in the NCBI RefSeq representative genomes database (Data set S1).

Phosphonate production screening.

Freshly revived actinobacterial strains were inoculated into 16 × 150 mm culture tubes containing 5 mL ATCC 172 and grown on a roller drum (100 rpm) at 30_°C for 7 days. These starter cultures were used to inoculate agar plates of ISP4, GUBC (22), R2AS (22), and V8 media (4) with 1% soy flour (200 μL per plate). After 7 days at 30_°C, plates were frozen overnight, thawed at room temperature, compressed to collect 30 mL of aqueous extract, and lyophilized. The dried material was reconstituted in 1 mL of sterile dI H₂O and centrifuged at 3,000 RPM to sediment insoluble material. Crude extracts for each strain (500 μL aliquots produced from growth on the 4 media) were combined, lyophilized, and rehydrated to 500 μL with sterile dI H₂O. Methanol (MeOH) was added (2 mL) and the samples incubated at −20°C overnight before centrifuging at 3,000 RPM to sediment insoluble material. MeOH was removed from the aqueous fraction by rotary evaporation, followed by lyophilization and reconstitution in 500 μL sterile, deionized H₂O.

NMR spectroscopy.

Experiments were performed at room temperature on an Agilent DD2 600 MHz spectrometer (operated at 600 MHz, 151 MHz, and 243 MHz for ¹H, ¹³C, and ³¹P, respectively) with a OneNMR probe (at University of Illinois at Urbana-Champaign), or on a Bruker Avance Neo 400 MHz spectrometer (400 MHz for ¹H and 162 MHz for ³¹P) equipped with a 5 mm Prodigy Cryoprobe (at The Ohio State University Chemistry NMR facility). Some HMBC and HSQC spectra were collected on a Bruker Avance III 500 MHz spectrometer equipped with a 5 mm cryoprobe BBFO at 23°C with Topspin 3.6.2 software (at University of Illinois at Urbana-Champaign). Chemical shifts are reported in δ(ppm), referenced to 85% H₃PO₄ as an external standard for ³¹P chemical shifts. D₂O (50 μL) was added at 10–20% as the locking solvent. High molecular weight material, including proteins were removed from enzymatic assays prior to NMR analyses by passing 500 μL reactions through an Amicon Ultra-0.5 Centrifugal filter with an appropriate molecular size cutoff (10 or 30 kDa). The flowthrough was then mixed with D₂O to give a final concentration of 20%. Quantitative ³¹P NMR was performed using an internal standard of 0.5 mM dimethylphosphinate and acquisition was performed using 5-times the experimentally determined T1 relaxation time for the sample. Peak areas were calculated using MestReNova v12.0.3 software and compared to the internal standard.

Mass spectrometry.

Chelex-Fe weak anion-exchange chromatography was performed on the pooled extracts (50 μL) as previously described (12) to enrich phosphonates and remove ion-suppressing species. Processed samples were reconstituted in sterile dI H₂O (50 μL), of which 10 μL was combined with 90 μL of acetonitrile (ACN) 10 mM NH₄HCO₃ pH 9.2. These were vigorously vortexed, centrifuged for 10 min at 13,000 g, and the clarified supernatant was withdrawn for LC-HRMS analysis as previously described (12). High resolution mass spectrometry of phosphonothrixin was carried out at the Metabolomics Center, Roy J. Carver Biotechnology Center (UIUC) using a Dionex Ultimate 3000 series HPLC system (Thermo, Germering, Germany) with a Q-Exactive MS system (Thermo, Bremen, Germany). Xcalibur 4.1.31.9 software was used for data acquisition and analysis.

Synthesis of the 3-hydroxy-2-oxopropylphosphonic acid (HOPPA).

Diethyl (3-[methoxymethoxy]prop-1-yn-1-yl)phosphonate was prepared as described (Fig. S1, [34]). Freshly distilled diethyl phosphite (1 equiv., 700 mg) in 8 mL of anhydrous DMSO was added to a flame-dried round-bottom flask attached to a Schlenk line to maintain a dry environment. Freshly distilled 3-(methoxymethoxy)prop-1-yne (MOM-protected propargyl alcohol) (1.2 equiv.), copper iodide (0.1 equiv.) and diethyl amine (0.2 equiv.) were added. The reaction was then stirred at 55°C under dried air for 24 h. The product was purified by diluting with 50 mL of H₂O and extracted with 3 × 40 mL of ethyl acetate. The combined organics were then washed with 3 × 20 mL of H₂O, dried over magnesium sulfate, filtered, and concentrated. The resulting residue was purified using SiO₂ gel column chromatography eluting with 100% EtOAc to obtain the final product in 28% yield.

Diethyl-protected β-ketophosphonate was prepared according to (35) (Fig. S1). The alkynylphosphonate (1 equiv., 373 mg) was combined with methanol (1.56 mL), H₂O (156 μL), and AgNO₃ (0.17 equiv.). After 12–14 h at 120°C in a sealed tube, the reaction was allowed to cool to room temperature before rotary evaporation of solvent. The product was purified using SiO₂ gel column chromatograply eluting with ethyl acetate: methanol (9:1) to obtain purified product in 45% yield. Deprotection was carried out in a flame-dried round-bottom flask under dried nitrogen. The diethyl-protected beta-ketophosphonate (1 equiv., 80 mg) was combined with anhydrous DCM (3 mL). While the solution was stirring under nitrogen, trimethylsilyl bromide (TMSBr, 3 equiv.) was added dropwise and the reaction was allowed to stir under nitrogen at 45°C for 90 min. After cooling the contents to room temperature, solvent was removed via rotary evaporation. Then, under constant stirring, 10 mL of H₂O was slowly added before lyophilizing overnight. To ensure adequate removal of HBr and TMS, lyophilized material was redissolved in water and lyophilized a second or third time to obtain the final product in 60% yield. ¹³C NMR characterization confirmed formation of the desired product (Fig. S10).

Diethyl-protected β-ketophosphonate was prepared according to (35). Alkynylphosphonate (1 eq, 373 mg) was combined with methanol (1.56 mL), H₂O (156 μL), and AgNO₃ (0.17 eq). After 12 to14 h at 120°C in a sealed tube, the reaction was allowed to cool to room temperature before rotary evaporation of solvent. The product was purified with SiO₂ column (ethyl acetate: methanol [9:1]) to obtain purified product in a 45% yield. Deprotection was carried out in a flame-dried round-bottom flask under dried nitrogen, the diethyl-protected beta-ketophosphonate (1 eq, 80 mg) was combined with anhydrous DCM (3 mL). While this was stirring under nitrogen, TMS-Br (3 eq) was added dropwise and allowed to stir under nitrogen at 45°C for 90 min. After cooling the contents to room temperature, solvent was removed via rotary evaporation. Then, under constant stirring, 10 mL of H₂O was slowly added before lyophilizing overnight. To ensure adequate removal of HBr and TMS, lyophilized material was redissolved in water and lyophilized a second or third time to obtain final product in a 60% yield.

Protein expression and purification.

Plasmids used for protein expression are described in Table S1. The open reading frames of interest were PCR amplified from K. sp. ST-888 genomic DNA with the exceptions of the ftxE, ftxF, and ftxG genes which were amplified from S. sp. NRRL F-6131 genomic DNA. RhiH (PEP mutase) was PCR amplified from Bacillus subtilis ATCC 6633. Primers were designed with NdeI, XbaI, or HindIII restriction sites to allow screening of pET28B plasmid constructs by restriction analysis. FtxE (site 2, BamHI/HindIII) and ftxF (site 1, NdeI/KpnI) were cloned into coexpression plasmid pETDuet-1. PCR products were purified using a QIAquick PCR Cleanup kit (Qiagen Inc.). Linearized pET28b plasmid was combined with PCR products in Gibson assembly reactions, except for ftxJ, which was first cloned into the pGEM-T Easy expression plasmid, then excised by restriction enzyme digestion and cloned into the pET28b plasmid as per the manufacturer’s instructions. Gibson assembly reactions were used to transform E. coli Rosetta (DE3) pLysSRARE. Plasmid constructs were verified by restriction digestion and Sanger sequencing Roy J. Carver Biotechnology Sequencing Center, University of Illinois at Urbana-Champaign (UIUC) or at ACGT, Inc. (Wheeling, IL).

Protein expression was carried out as previously described (36) with the following modifications. To induce protein expression IPTG was added to a final concentration of 1 mM. Cells were lysed by three passages through a chilled French pressure cell. All buffers used for protein purification contained 300 mM NaCl, 50 mM K₂HPO₄ (pH 7.5) and 10% glycerol. Protein samples were concentrated using Amicon Ultra-15 Centrifugal filter columns with the appropriate molecular size cutoff (10 or 30 kDa) prior to desalting. Imidazole was removed from protein samples using PD-10 desalting columns per the manufacturer’s instructions (GE Healthcare).

Enzymatic assays.

All enzymatic assays were carried out in 1.5 mL Eppendorf tubes in a final volume of 500 μL and incubated overnight at room temperature unless otherwise stated.

(i) FtxJ assay. FtxJ was assayed in a coupled reaction in which PnPy was enzymatically generated from phosphoenolpyruvate (PEP) using the PEP mutase (RhiH) from B. subtilus (23). PnPy was reduced to phosphonolactate (PnLac) by FtxJ using NADH, which was continuously regenerated via phosphite dehydrogenase (PtxD) using phosphite as the electron donor (37). Inactive FtxJ was prepared by incubation at 100_°C for 5 min. Assays were conducted in 50 mM sodium phosphite buffer (pH 7.5) containing 10 mM phosphoenolpyruvate, 200 μM NADH, 1 mM MgCl₂, 4.5 μM purified FtxJ and 45 μM RhiH PEP mutase, and 32 μM PtxD. Phosphonate production and consumption were assayed by ³¹P NMR as indicated above.

(ii) FtxK assays. FtxK activity was assayed in the reverse direction in a reaction using continuous regeneration of NADH with PtxD and phosphite as described (37). Assays were conducted in 50 mM sodium phosphite buffer (pH 7.5) containing 2.5 μM purified FtxK, 5 mM MgCl₂, 2 mM HOPPA, 200 μM NADH, and 25 μM PtxD. FtxK activity was assayed in the forward direction in a coupled assay using lactate dehydrogenase (Sigma Chemical, St. Louis, MO) and pyruvate to regenerate NAD⁺. Assays contained 2 mM 2,3-dihydroxypropyl phosphonic acid (DHPPA), 4 μg/mL lactate dehydrogenase, 10 mM sodium pyruvate, and 250 μM NAD⁺. Phosphonate production and consumption were assayed by ³¹P NMR as indicated above.

(iii) FtxG assays. Acetolactate synthase activity was assessed by ¹H and ¹³C NMR following incubation of purified FtxG (2.5 μM) in 50 mM K₂HPO₄ buffer (pH 7.5) with 5 mM MgCl₂ and 20 mM sodium pyruvate. The products were compared to those of control reaction carried out under identical conditions using acetolactate synthase (AlsS) from Bacillus subtills produced by overexpression in E. coli using pMC119 (Table S1).

(iv) Coupled FtxEF assays. Small-scale reactions using synthetic HOPPA were coupled with FtxG and contained 2.5 μM the purified FtxE and FtxF, 2.5 μM purified FtxG enzyme, 5 mM MgCl₂, 50 mM K₂HPO₄, 200 μM TPP, 1 mM HOPPA and 20 mM sodium pyruvate. Large-scale FtxEF/FtxG reactions that were used for structural assignment of the product utilized HOPPA that was made in situ with FtxK as described above. These 5 mL reactions were performed in a 15 mL conical tubes containing 2.5 μM FtxEF, 2.5 μM FtxG, 2.5 μM FtxK, 4 μg/mL lactate dehydrogenase, 1 mM DHPPA, 5 mM MgCl₂, 50 mM K₂HPO₄, 200 μM TPP, 20 mM sodium pyruvate, and 250 μM NAD⁺. Protein was removed and the reaction products purified by HPLC as follows. Samples were dried, resuspended in 70% acetonitrile with 10 mM NH₄HCO₃ (pH 8.0), and then separated using an Atlantis HILIC Silica column (10 × 250 mm, 5 μm particle size) with a flow rate of 5 mL/min using 10 mM NH₄HCO₃ (pH 8.0) (Solvent A) and 90% acetonitrile 10 mM NH₄HCO₃ (pH 8.0) (Solvent B). The separation was performed by an initial step of 10 min at 90% solvent B, followed by a linear gradient to 50% solvent B over 50 min and an additional 13 min at 50% solvent B. Fractions were collected at 2-min intervals and screened for the presence of phosphonothrixin by ³¹P NMR. Fractions containing phosphonothrixin were combined, dried, reconstituted in 10% vol/vol D₂O, and analyzed by 1D and 2D ¹H, ³¹P and¹³C NMR experiments.