Diversity and abundance of phosphonate biosynthetic genes in nature

Significance

Phosphonic acids are organophosphorus molecules containing direct carbon–phosphorus bonds that are often perceived as biological rarities. The data presented here show that the ability to synthesize diverse phosphonates is both widespread and relatively common among microbes. These findings are consistent with recent evidence suggesting that phosphonates are important intermediates in the global phosphorus cycle. Moreover, they support the idea that these molecules play a significant role in the biology of producing organisms, including a mechanism to sequester scarce phosphorus resources and to compete via chemical warfare using toxic phosphonate mimics of common metabolic intermediates.

Abstract

Phosphonates, molecules containing direct carbon–phosphorus bonds, compose a structurally diverse class of natural products with interesting and useful biological properties. Although their synthesis in protozoa was discovered more than 50 y ago, the extent and diversity of phosphonate production in nature remains poorly characterized. The rearrangement of phosphoenolpyruvate (PEP) to phosphonopyruvate, catalyzed by the enzyme PEP mutase (PepM), is shared by the vast majority of known phosphonate biosynthetic pathways. Thus, the pepM gene can be used as a molecular marker to examine the occurrence and abundance of phosphonate-producing organisms. Based on the presence of this gene, phosphonate biosynthesis is common in microbes, with ∼5% of sequenced bacterial genomes and 7% of genome equivalents in metagenomic datasets carrying pepM homologs. Similarly, we detected the pepM gene in ∼5% of random actinomycete isolates. The pepM-containing gene neighborhoods from 25 of these isolates were cloned, sequenced, and compared with those found in sequenced genomes. PEP mutase sequence conservation is strongly correlated with conservation of other nearby genes, suggesting that the diversity of phosphonate biosynthetic pathways can be predicted by examining PEP mutase diversity. We used this approach to estimate the range of phosphonate biosynthetic pathways in nature, revealing dozens of discrete groups in pepM amplicons from local soils, whereas hundreds were observed in metagenomic datasets. Collectively, our analyses show that phosphonate biosynthesis is both diverse and relatively common in nature, suggesting that the role of phosphonate molecules in the biosphere may be more important than is often recognized.

Phosphorus is an essential nutrient for all living organisms, required for the synthesis of nucleic acids, phospholipids, phosphorylated exopolysaccharides, and numerous metabolites. In most organisms, the preferred source of phosphorus is inorganic phosphate. However, because the majority of phosphate salts are highly insoluble, this ion is rarely available in concentrations that support unbridled growth. Thus, despite the fact that phosphorus is the 11th most abundant element in the Earth’s crust, it is a limiting nutrient in many ecosystems (1, 2). As a result, nature has evolved highly efficient phosphate transport systems, as well as systems for acquisition of phosphorus from essentially all known biomolecules containing this element (3, 4). These biomolecules are composed mostly of assorted phosphate esters and anhydrides secreted by other organisms during growth or released by decomposition after their death. An impressive body of research has characterized the processes by which these compounds are made and subsequently used. However, recent reports suggest that other, less-well studied phosphorus compounds may also be important in the biosphere (5). Among these are the phosphonic and phosphinic acids, compounds characterized by the presence of highly stable carbon–phosphorus (C–P) bonds in place of the labile carbon-oxygen-phosphorus linkages found in more familiar phosphorus-containing biomolecules.

Synthetic phosphonates have been known for more than a century, but their role in biology was not suspected until 1959 when 2-aminoethylphosphonic acid (AEP) was found in the acid hydrolysate of rumen protozoa (6). Since then, C–P compounds have been identified in a variety of bacteria, archaea, and eukaryotes (7, 8). In many of these organisms, phosphonates are found as constituents of extracytoplasmic macromolecules, including phosphonolipids and phosphonoglycans (5). It is generally believed that the inert nature of the C–P bonds in these structural molecules imparts resistance to hydrolytic enzymes, thus providing an advantage in phosphorus-limited environments (9). Other organisms produce a diversity of low-molecular-weight, bioactive phosphonates, whose inhibitory properties stem from the structural similarity of stable phosphonates to labile phosphate esters and carboxylates (5). The range of biological activities exhibited by these molecules is impressive, and examples with antibacterial, antiviral, antiparasitic, and herbicidal properties are known (5, 10). Thus, it seems likely that phosphonate biosynthesis has evolved in these organisms to provide a competitive advantage through chemical warfare.

Although the number of proven phosphonate producers is relatively small, it is increasingly clear that these molecules compose a significant fraction of the bioavailable phosphorus in many ecosystems (11). ³¹P NMR analyses conducted in the 1990s suggested that phosphonates can account for nearly 25% of the high-molecular-weight dissolved organic P in marine environments (12, 13), whereas newer methods show that 5–10% of the phosphorus pool across the size spectrum is composed of C–P compounds (14). Similar studies show that phosphonates occur in a variety of soils, as well as in freshwater lakes and streams (15–18). The abundance of phosphonate catabolism genes in genomic and metagenomic datasets shows that the trait is widespread and common among microbes, with 30–40% of strains encoding one or more of these pathways (19, 20). Other studies revealed that these catabolic genes are often highly expressed in situ (21–23). Taken together, these data indicate that phosphonates are likely to be an important source of phosphorus in many environments.

In contrast to data supporting a significant role for C–P compounds in the biological phosphorus cycle, evidence regarding the source of phosphonates in nature remains scarce. In agricultural settings, these compounds may arise from xenobiotic sources, as synthetic phosphonates are widely used as herbicides and pesticides (e.g., glyphosate and phosphinothricin); however, in more pristine settings, these compounds must have a biogenic source. The characterized producers described above are not always abundant where phosphonates are detected. Thus, the principal sources of phosphonates in the biosphere remain to be identified.

Early studies addressing the source of environmental phosphonates relied on chemical analysis of whole organisms or tissue samples, which are limited in scope simply due to the inability to sample a significant fraction of the organisms present in any given environment. In most cases, the presence of phosphonates was assessed by ³¹P NMR using the distinctive chemical shift of C–P molecules, which is observed in the +5 to +45 ppm range, as opposed to the +5 to −20 ppm range observed for phosphate and its esters and anhydrides (24). However, ³¹P NMR is relatively insensitive and cannot detect phosphorus compounds that are present in low abundance, as is often the case in phosphonate producers. Furthermore, a variety of phosphate esters have chemical shifts in the phosphonate range, including many known biological products, such as glycerol 1′,2′-cyclic phosphate and 2′,3′-cGMP (25–27). Hence, additional confirmatory chemical tests for the presence of phosphonic acids (e.g., 2D NMR, mass spectrometry, chemical and enzymatic reactivity) are highly desirable, but rarely provided.

The unique properties of phosphonic acids have evoked considerable interest in their biosynthesis, and the genes and enzymes involved in the synthesis of several bioactive phosphonates, as well as those needed for the synthesis of certain phosphonolipids and phosphonoglycans, have now been characterized (5, 28–30). With a single exception, all known phosphonates are derived from phosphoenolpyruvate (PEP) by isomerization to phosphonopyruvate in a reaction catalyzed by the enzyme PEP mutase (PepM) (5). Because this initial reaction is thermodynamically unfavorable, net synthesis of phosphonates requires a thermodynamically favorable step immediately following PepM. Initial studies identified phosphonopyruvate decarboxylase (Ppd) as the enzyme providing this driving force; however, recent data show that other driving reactions exist (31, 32). Subsequent biosynthetic steps diverge to create a wide range of C–P biomolecules. The genes encoding these downstream enzymes are usually clustered together with the pepM gene, a trait that has greatly simplified genetic and biochemical studies of phosphonate biosynthesis and that allows prediction of the types of C–P molecules produced by the metabolic pathways encoded by these linked genes (29, 33, 34).

Detailed knowledge regarding C–P compound biosynthesis opened the door to gene-based methods for assessing the abundance and identity of biological phosphonate producers. Villarreal-Chiu et al. quantified the occurrence of Ppd homologs to estimate the prevalence of phosphonate metabolism in microbes, finding the gene in ∼10% of sequenced geneomes and genome equivalents in metagenomic datasets (20). PepM homologs—commonly located adjacent to genes involved in the synthesis of methylphosphonic acid, a putative precursor of methane in the aerobic ocean (8)—were also shown to be abundant (∼16% of genome equivalents) in the Global Ocean Survey (GOS) data (35, 36). Although both studies suggest that phosphonate production is relatively common, the methods used have the potential to overestimate the occurrence of phosphonate production due to the difficulties of assigning protein function on the basis of homology alone. PepM is a member of the large isocitrate lyase superfamily (37), whereas Ppd is a member of a large family of thiamine-pyrophosphate–using enzymes (38). Accordingly, homology-based searches (such as BLASTP) using Ppd and PepM as queries can identify family members that are not involved in phosphonate biosynthesis, even when fairly stringent cutoff values are used.

In this report we revisit the use of PepM to estimate the distribution and abundance of phosphonate production in nature. By using a stringent filtering strategy that requires all counted hits to include catalytically important amino acid residues that distinguish PepM from other isocitrate lyase superfamily members, we avoid ambiguities caused by the evolutionary ancestry of the enzyme. Moreover, by analyzing the pepM gene neighborhoods, we provide insight into the nature of the phosphonate molecules being produced. Our results reveal a surprising abundance and diversity of phosphonate biosynthesis in nature, suggesting that their role in the biosphere has yet to be fully appreciated.

Results

Ability to Synthesize Phosphonates Is Relatively Common and Widespread in Nature.

We searched for pepM homologs in the Integrated Microbial Genomes (IMG) database (39), GOS marine metagenomes (35, 36), and the Integrated Microbial Genomes with Microbiome Samples (IMG/M) database (40), using the sequences of biochemically validated PepM sequences as queries (31, 33, 34). All potential hits were screened for the highly conserved catalytic motif EDKXXXXXNS, which distinguishes PEP mutase from other members of the isocitrate lyase superfamily (41).

Two hundred and forty-seven bacterial genomes (5.7%) in the IMG genomes database encode pepM homologs. PEP mutase-encoding bacteria include 9 of 33 bacterial phyla represented, predominantly from Proteobacteria, Bacteroidetes, Firmicutes, Actinobacteria, and Spirochaetes, but the ability to produce phosphonates is not phylogenetically coherent except among Burkholderia, Selenomonas, and Nitrosococcus (SI Appendix, Fig. S1 and Dataset S1). Seventy-seven of 82 sequenced Burkholderia genomes (accounting for 31% of the pepM-positive genomes) encode at least one pepM gene, with some strains carrying as many as four, whereas eight of nine sequenced Selenomonas genomes and all four sequenced Nitrosococcus genomes contain pepM. Two archaeal genomes (1.1%), Nitrosoarchaeum limnia SFB1 and Nitrosopumilus maritimus SCM1, encode pepM. Seven pepM homologs were discovered in six eukaryotic genomes (3.2%), including sea snail (Lottia gigantean), sea anemone (Nematostella vectensis), and several protozoa (Paramecium tetraurelia, Perkinsus marinus, Tetrahymena thermophile, and Trypanosoma cruzi).

Analysis of the GOS metagenomes, which encompass near-surface marine environments from around the world, revealed pepM homologs from 59 of 79 sampling sites (SI Appendix, Table S1 and Dataset S2). The IMG/M microbiomes (40), which include metagenomic data from a variety of ecosystems, contain pepM homologs in 558 of the 1,281 datasets in the collection (SI Appendix and Dataset S3). To quantify the relative occurrence of phosphonate biosynthetic genes in these datasets, we compared the abundance of pepM to that of typical single-copy genes as described (42). By this measure, pepM genes occur in ∼7.6% of microbial genomes in the GOS and IMG/M metagenomes (assuming only one copy of pepM per genome) (SI Appendix; Dataset S2 and Dataset S3). The abundance of pepM is essentially the same in all GOS sites (between 4.4 and 8.7%), with a mangrove site and a hypersaline site (both in Ecuador) having the highest (20.7%) and the lowest (0.3%) abundance, respectively (Fig. 1A). In contrast, there are significant differences among the various IMG/M metagenomes based on the large set of metagenomic data analyzed, with mammalian- and molluscan-associated microbiomes having the highest median pepM abundances (17.4 and 29.9%, respectively) (Fig. 1B).

Fig. 1.

pepM gene abundance in GOS metagenomes and IMG/M microbiomes. (A) Boxplot of prokaryotic genome equivalents for pepM occurrence by habitat type in GOS. Single black lines represent the median value for environments that were sampled only once. No significant difference was found in the relative pepM abundance across GOS habitats (P = 0.9328, Kruskal Wallis test applied to habitats with more than one sampling site). (B) Boxplot of percentage of prokaryotic genome equivalents for pepM occurrence by ecosystem type in IMG/M microbiomes. Relative pepM abundance across various ecosystems differed significantly (P < 3.3 × 10⁻⁵, Kruskal Wallis test applied to categories with more than one sampling site). In A and B, the number of sampling sites for each type is shown in parentheses. (C) Distribution of predicted prokaryotic phyla for pepM homologs identified in GOS metagenomes. (D) Distribution of predicted prokaryotic phyla for pepM homologs identified in IMG/M microbiomes. Phyla that account for <1% are grouped in “Others.”

To gain insight into which taxa are associated with phosphonate biosynthesis in the metagenomic data, we used PhymmBL, a bioinformatics tool that allows strong inference regarding the phylogeny of organisms from which individual sequence reads are derived (43). This analysis predicts that Proteobacteria, followed by Firmicutes, Cyanobacteria, and Spirochaetes, are the most abundant pepM-associated phyla in the marine environments represented in the GOS data, with 10 other phyla represented at lower abundances (Fig. 1C; SI Appendix; Dataset S2). At the genus level, 40% of pepM reads seem to be derived from Prochlorococcus (17.9%), Candidatus Pelagibacter (11.8%), and Borrelia (10.8%) whereas remaining reads were predicted to belong to the other 62 genera (SI Appendix; Dataset S2). In the IMG/M metagenomes, at least 28 different prokaryotic groups were predicted at the phylum level, primarily from Firmicutes, Proteobacteria, and Bacteroidetes (Fig. 1D; SI Appendix; Dataset S3), with the top five pepM-containing genera being Bacteroides (7.0%), Clostridium (6.8%), Bacillus (6.6%), Nitrosopumilus (6.5%), and Streptococcus (4.3%) (SI Appendix and Dataset S3). PhymBL analysis of a pepM PCR clone library from local soils (see below) revealed a similar distribution with 97% of the reads being assigned to Proteobacteria and only minor fractions assigned to Actinobacteria (1.5%), Nitrospirae (1.2%), and Acidobacteria (0.2%). The two most abundant subgroups in this library were the proteobacterial orders Burkholderiales (73.6%) and Rhizobiales (19.2%).

Differences in abundances observed between these various datasets are almost certainly due to the substantial differences between the sampled environments for metagenomic data and the extreme sampling bias toward pathogenic microbes for the genome sequence data.

Finally, we assessed the prevalence of phosphonate biosynthesis in cultivable bacteria by screening using degenerate PCR primers designed from conserved PepM amino acids motifs. Due to our interest in bioactive phosphonic acids, we focused our efforts on actinomycetes, which are known to be prolific producers of diverse natural products. We examined 1,649 strains isolated from local soils using protocols that are selective for spore-forming actinobacteria and an additional 973 strains from the US Department of Agriculture Agricultural Research Service actinobacteria collection: of these, 120 strains (4.6%) gave PCR products that were verified to be authentic pepM amplicons by DNA sequencing.

pepM Gene Neighborhoods Suggest Considerable Diversity in Phosphonate Biosynthetic Pathways.

Because the genes encoding phosphonate biosynthetic pathways are typically clustered together with the pepM gene (29, 33, 34), examination of the pepM gene neighborhood provides insight into the diversity of their phosphonic acid products. Accordingly, we examined the pepM gene neighborhoods in all known phosphonate producers, as well as those encoded in sequenced microbial genomes and in 25 of the pepM-positive actinobacteria, which were cloned and sequenced as described (SI Appendix, Materials and Methods). Putative gene functions were assigned on the basis of genome annotations and homology of the protein products to enzymes of known function. Annotated diagrams of all gene clusters are shown in SI Appendix, Figs. S2–S26 and at http://file-server.igb.illinois.edu/˜xyu9/Dataset_S4._Phosphonate_gene_clusters.html.

Several large groups of nearly identical gene clusters were observed that appear to be involved in synthesis of phosphonolipids or phosphonoglycans (Fig. 2A and SI Appendix, Fig. S27). The largest of these, composed of 93 gene clusters found mostly in Burkholderia, consists of a seven-gene putative operon that we propose may be responsible for synthesis of phosphonolipids with 1-hydroxy-2-aminoethylphosphonate as the head group (Fig. 2A, group 1; SI Appendix, Fig. S27), similar to the lipids previously characterized in Bdellovibrio stolpii (44). A second large group is found solely in Burkholderia species, most of which also encode the group 1 gene cluster (Fig. 2A, group 2). We predict that these genes are responsible for synthesis of 2-hydroxy-phosphonoacetate, which may be used as an alternative phosphonolipid headgroup in these strains (SI Appendix, Fig. S27). A third group including 27 gene clusters, mainly from Bacteroides and Treponema, includes a locus previously implicated in the synthesis of capsular polysaccharide B by Bacteroides fragilis, which contains AEP in an ester linkage to a hydroxyl group of the carbohydrate (45) (Fig. 2A, group 3). Interestingly, the nearly identical cluster found in Bacteroides eggerthii DSM 20697 has replaced the gene needed to make AEP from phosphonoacetaldehyde with a gene that produces hydroxyethylphosphonate (HEP) from the same substrate (46). Thus, we conclude that some Bacteroides strains produce a similar capsular polysaccharide decorated with HEP instead of AEP. Group 4 and several additional groups with a few members each also contain genes related to lipid biosynthesis and, thus, are probably involved in the synthesis of phosphonolipids (Fig. 2A and SI Appendix, Fig. S27). Collectively, these groups account for approximately half of the identified pepM gene clusters. The remaining pepM gene neighborhoods are highly diverse and include the genes for synthesis of the known bioactive phosphonates, as well as those from all of the actinomycetes sequenced in this study (Fig. 2A, group 9 and group 10; SI Appendix, Fig. S27). We expect that the majority of these gene clusters will direct synthesis of low-molecular-weight C–P compounds. Interestingly, 10% of sequenced genomes analyzed in our dataset lack the gene encoding Ppd; thus Ppd-independent phosphonate biosynthesis is common in nature.

Fig. 2.

Analysis of PepM and phosphonate gene clusters from microbial genomes. (A) Maximum-likelihood tree of PEP mutase sequences from NCBI and 25 actinomycete strains from this study. The tree was calculated with the FastTree program (53) with default settings. The tree is rooted with 2-methylisocitrate lyase sequence (NP_286072). The branch is colored based on the source of pepM sequences: 25 actinomycete isolates from this study (red), NCBI archaeal genomes (purple), NCBI bacterial genomes (blue), and NCBI genomic fragments (green). Selected pepM groups are highlighted by color shading, with the number of sequences within a group shown in parentheses. Known phosphonate compounds are indicated and marked with an asterisk. The phosphonate biosynthetic loci for 25 actinomycete strains from this study are shown in SI Appendix, Fig. S2–S26. The phosphonate gene cluster for each organism is listed at http://file-server.igb.illinois.edu/˜xyu9/Dataset_S4._Phosphonate_gene_clusters.html. and is shown in the same order as in the tree. (B) Phosphonate gene cluster similarity as a function of PepM identity. Phosphonate gene cluster similarity was calculated using the fraction of homologous genes shared by two gene clusters. PepM identity was calculated using pairwise deletion of missing sites across the entire PepM alignment. Gene-cluster similarity measures were binned by PepM identity at intervals of 0.02 and plotted with a standard boxplot. The line shown is a linear regression over the PepM identity range 0.6–1.0, with equation ŷ = −0.86 + 1.6x; R-square = 0.74. The correlation has a P value of 2.2 × 10⁻¹⁶. The full dataset is plotted in SI Appendix, Fig. S28.

PEP Mutase Sequence Similarity Is Strongly Correlated with Similarity of PepM Gene Neighborhoods.

Due to the strongly conserved gene neighborhoods seen in many closely related strains, we were interested in whether PepM phylogeny was a reflection of the organismal phylogeny. To test this idea, we constructed a PepM phylogenetic tree using the sequences obtained from sequenced microbial genomes and from the 25 actinomycete gene clusters sequenced in this study (Fig. 2A). For this analysis, neighborhoods defined to encompass 6 genes upstream and downstream of pepM (13 genes in total). The results show scant correlation between the PEP mutase phylogeny and organismal phylogenies. This is strikingly illustrated by the group 1 and group 2 gene neighborhoods, which encode distantly related PEP mutase sequences, but which are often found in the same organism. Many other related PepM proteins are also found in unrelated organisms (e.g., the group 3 gene cluster PepM found in Treponema and Bacteroides and the group 1 gene cluster PepM found in Burkholderia and Cupriavidus).

In contrast, there is a strong correlation between PEP mutase phylogeny and the pepM gene neighborhood. Including those discussed above, we found 12 groups of nearly identical gene neighborhoods found in at least three different organisms (Fig. 2A). All of the PEP mutase sequences from each group are monophyletic, suggesting that similarity of PEP mutase is a good predictor of the other genes in the phosphonate biosynthetic pathway. To test this idea in a more rigorous fashion, we plotted the similarity of 342 pepM gene neighborhoods against the similarity of PEP mutase amino acid sequences in all pairwise combinations (58,311 comparisons). The data reveal a highly significant linear correlation for PEP mutase pairs having greater than 60% identity, with essentially no similarity in the pepM gene neighborhood at lower values (Fig. 2B and SI Appendix, Fig. S28).

Numerous Undiscovered Phosphonate Natural Products Are Likely to Exist.

The preceding analysis shows that, when two organisms have similar PepM sequences, they are much more likely to encode similar (if not identical) phosphonate metabolic pathways and therefore produce similar phosphonate molecules. The converse is also likely to be true. Therefore, we can estimate the diversity of phosphonate biosynthetic pathways in nature by examining the number of different PepM sequences in various datasets. To do this, we estimated the number of different PepM sequences in the microbial genomes, GOS metagenomes, and IMG/M metagenomes using rarefaction analysis (47, 48). We also constructed and sequenced a large library of pepM PCR amplicons using DNA isolated directly from local soils to provide a greater depth of coverage specific to the pepM gene.

Based on the resulting collector’s curves, it is clear that the diversity of PepM sequences in nature is very large (Fig. 3). Among the datasets examined, only the soil PCR library was sampled at sufficient depth to approach saturation. Using an 84% PepM identity level, which represents conservation of roughly half of the genes in the cluster, we project that there are at least 135, 248, 213, and 60 PepM groups in the microbial genomes, GOS metagenomes, IMG/M metagenomes, and soil clone libraries, respectively (SI Appendix, Table S2). Significantly, there is little overlap between the PepM groups found from different environments (e.g., distinct PepM sequences found in aquatic ecosystems, terrestrial ecosystems, or animal-associated microbiota) (SI Appendix, Fig. S29). Thus, a realistic estimate of phosphonate diversity in nature would be closer to the sum of the four richness estimates.

Fig. 3.

Rarefaction analysis of amino acid sequences of pepM identified from IMG microbial genomes, IMG/M microbiomes, GOS metagenomes, and soil pepM clone libraries. Rarefaction curves are shown for OTUs with differences not exceeding 16%.

Discussion

Biogenic C–P compounds were discovered more than 50 years ago and have been found in a variety of organisms, yet they are often portrayed as rare molecules with highly specialized functions (49). In contrast, our results show that the genetic capacity for phosphonic acid synthesis is common, widespread, and diverse, with more than 5% of sampled microbial cells encoding putative pepM genes. These microorganisms are found in every environment and include members of phylogenetically diverse groups. Moreover, these data probably underestimate the prevalence of phosphonate biosynthesis because larger eukaryotes such as mollusks, anemones, and corals, which are known to make phosphonates (7), are not well represented in either genomic or metagenomic datasets. Furthermore, numerous truncated homologs found in the metagenomic data were removed from our count because they lacked the diagnostic PepM catalytic motif. It is very likely that some of these would have contained the PepM motif if the full-length gene sequence had been available. This difference in counting method is also likely to explain the higher PepM abundances (∼16%) that we previously reported (8) because PepM homologs were not filtered for the EDKXXXXXNS motif in that study.

Our results also show that phosphonate biosynthesis is highly diverse. Our calculations predict that hundreds of unique phosphonate molecules remain to be discovered in nature. Based on the content of available sequenced gene clusters, it is probable that many of these molecules will contain similar phosphonate moieties (e.g., AEP and HEP) attached to different glycans or lipids. However, many of the gene neighborhoods, including those for all known bioactive phosphonates, lack obvious candidates for gene glycan or lipid synthesis, suggesting that numerous small-molecule phosphonates await discovery. Given the potent bioactivity of this class of molecule, it is likely that pharmaceutically useful compounds will be among this group. In this regard, most of the actinomycete strains examined in this study have either unique pepM gene or ones that are similar to known phosphonate antibiotic gene clusters, suggesting that this group represents a rich source to mine for bioactive phosphonate natural products (Fig. 2A, red branches).

Materials and Methods

Bacterial Strains, Plasmids, and Culture Conditions.

The bacterial strains and plasmids used in this study are listed in SI Appendix, Table S3. Growth conditions and molecular biology methods are described in SI Appendix, Materials and Methods.

Identification of PEP Mutase Gene Homologs in IMG-Sequenced Genomes, IMG/M Microbiomes, and GOS Metagenomes.

PEP mutase gene homologs were identified from all sequenced genomes in the IMG database (39), all microbiomes in the IMG/M database (40), and GOS metagenomes in the CAMERA database (50) as described in SI Appendix, Materials and Methods. Genome equivalents and the percentage of genome equivalents with pepM in IMG/M microbiomes and GOS metagenomes were determined using the method suggested by Howard et al. (42).

Soil pepM Gene Clone Library Construction, Sequencing, and Sequence Analysis.

Clone library construction, sequencing, and sequence analyses were performed as described in SI Appendix, Materials and Methods.

Strain Isolation, pepM Gene Screening, Sequencing, and Sequence Annotations of Phosphonate Biosynthetic Gene Clusters from Actinomycetes.

Twenty-five pepM-positive actinomycete strains were chosen for cloning and sequencing of their phosphonate gene clusters: 21 isolates were from the genus Streptomyces and one each from the genera Kitasatospora, Lechevalieria, Micromonospora, and Saccharothrix (SI Appendix, Table S3). Strain isolation, pepM screening, gene cluster sequencing, and sequence annotations are described in SI Appendix, Materials and Methods and Figs. S2–S26.

PEP Mutase Sequence Diversity Analysis.

PepM amino acid sequences from IMG-sequenced genomes, IMG/M microbiomes, GOS metagenomes, and soil clone libraries were aligned with ARB software (51), and distance matrices were used as input to Mothur software (52) for clustering sequences into operational taxonomic units (OTUs), generating rarefaction curves, and calculating richness estimates. The PEP mutase maximum-likelihood tree was constructed with FastTree (53) as described in SI Appendix, Materials and Methods.

Phosphonate Biosynthetic Gene Cluster Analysis.

Phosphonate gene clusters from the National Center for Biotechnology Information (NCBI) sequenced genomes and from 25 actinomycete isolates from this study were compared and analyzed as described in SI Appendix, Materials and Methods.

Nucleotide Accession Numbers.

Twenty-five actinomycete phosphonate biosynthetic gene cluster sequences have been deposited in GenBank under accession nos. KF386859–KF386883. Partial soil PEP mutase gene sequences have been deposited in GenBank under accession nos. KF386884–KF387474.