The diversity and evolution of microbial dissimilatory phosphite oxidation

Significance

Geochemical models of the phosphorus (P) cycle uniquely ignore microbial redox transformations. Yet phosphite is a reduced P source that has been detected in several environments at concentrations that suggest a contemporary P redox cycle. Microbial dissimilatory phosphite oxidation (DPO) converts soluble phosphite into phosphate, and a false notion of rarity has limited our understanding of its diversity and environmental distribution. Here we demonstrate that DPO is an ancient energy metabolism hosted by taxonomically diverse, autotrophic bacteria that exist globally throughout anoxic environments. DPO microorganisms are therefore likely to have provided bioavailable phosphate and fixed carbon to anoxic ecosystems throughout Earth’s history and continue to do so in contemporary environments.

Abstract

Phosphite is the most energetically favorable chemotrophic electron donor known, with a half-cell potential (E^o′) of −650 mV for the PO₄³⁻/PO₃³⁻ couple. Since the discovery of microbial dissimilatory phosphite oxidation (DPO) in 2000, the environmental distribution, evolution, and diversity of DPO microorganisms (DPOMs) have remained enigmatic, as only two species have been identified. Here, metagenomic sequencing of phosphite-enriched microbial communities enabled the genome reconstruction and metabolic characterization of 21 additional DPOMs. These DPOMs spanned six classes of bacteria, including the Negativicutes, Desulfotomaculia, Synergistia, Syntrophia, Desulfobacteria, and Desulfomonilia_A. Comparing the DPO genes from the genomes of enriched organisms with over 17,000 publicly available metagenomes revealed the global existence of this metabolism in diverse anoxic environments, including wastewaters, sediments, and subsurface aquifers. Despite their newfound environmental and taxonomic diversity, metagenomic analyses suggested that the typical DPOM is a chemolithoautotroph that occupies low-oxygen environments and specializes in phosphite oxidation coupled to CO₂ reduction. Phylogenetic analyses indicated that the DPO genes form a highly conserved cluster that likely has ancient origins predating the split of monoderm and diderm bacteria. By coupling microbial cultivation strategies with metagenomics, these studies highlighted the unsampled metabolic versatility latent in microbial communities. We have uncovered the unexpected prevalence, diversity, biochemical specialization, and ancient origins of a unique metabolism central to the redox cycling of phosphorus, a primary nutrient on Earth.

Phosphite (PO₃³⁻) is a highly soluble, reduced compound that can account for over 30% of the total dissolved inorganic phosphorus in diverse environments (1, 2). Evidence suggests that meteorite impacts deposited substantial phosphite quantities on early Earth, but its abiotic oxidation to phosphate after the great oxidation event (∼2.5 billion y ago [Gya]) is assumed to have rendered phosphite negligible in neoteric environments (3). Surprisingly, phosphite has been detected in diverse reducing environments, and up to 1 µM was observed in some surface waters, suggesting contemporary neogenesis (1, 3). Geothermal and hydrothermal systems may generate phosphite through metal phosphide corrosion and iron-mediated phosphate reduction, and some phosphite may be derived from biological phosphonate degradation or anomalous phosphate reduction (1, 4, 5). Meanwhile, some phosphite accumulation is likely attributable to anthropogenic activity because comparatively higher concentrations of phosphite have been identified in contaminated environments and industrial wastewaters (1, 2, 6, 7).

Despite its enigmatic distribution, functional gene presence in the Joint Genome Institute (JGI) Integrated Microbial Genomes and Metagenomes (IMG/M) database predicts that phosphite is assimilated as a phosphorus source by ∼1.5% of sequenced microorganisms (2, 8–10). However, the PO₄³⁻/PO₃³⁻ redox couple also has an extremely low potential (E^o′ = −650 mV), and microorganisms can alternatively use phosphite as a sole electron donor and energy source, excreting biogenic phosphate from cells (11). With the low potential of the PO₄³⁻/PO₃³⁻ redox couple, phosphite represents the most energetically favorable chemotrophic microbial electron donor described (12), yet only two dissimilatory phosphite-oxidizing microorganisms (DPOMs) have been cultured, and only one has been isolated.

DPO was first identified in Desulfotignum phosphitoxidans FiPS-3, an autotrophic homoacetogenic facultative sulfate-reducing bacterium, isolated from Venetian brackish sediments (13). DPO in FiPS-3 is attributed to the ptx-ptd gene cluster (ptxDE-ptdCFGHI), which FiPS-3 likely acquired through horizontal gene transfer (HGT) (14–16). FiPS-3’s most closely related cultured isolate is incapable of DPO although the organisms share 99% 16S ribosomal RNA (rRNA) gene identity (17). The second known DPOM, Candidatus (Ca.) Phosphitivorax anaerolimi Phox-21, was enriched from wastewater collected in Oakland, California, and recently another Phosphitivorax strain (Ca. P. anaerolimi F81) was identified in Danish wastewater (18, 19). Phox-21 grows chemolithoautotrophically with phosphite and carbon dioxide (CO₂) as the sole electron donor and acceptor, respectively, and is the first naturally occurring species proposed to fix CO₂ via the reductive glycine pathway (18, 20, 21). The reductive glycine pathway has since been confirmed to naturally fix CO₂ in wild-type Desulfovibrio desulfuricans (22). Phox-21 harbors all ptx-ptd genes but, unlike FiPS-3, lacks ptdG (which encodes a putative transcriptional regulator) and shows no evidence of horizontal acquisition of the ptx-ptd cluster (14, 18). Understanding the evolutionary history of DPO metabolism is consequently limited by the existence of only two characterized DPOMs whose ptx-ptd clusters exhibit deviating patterns of composition and inheritance.

Scarce representation also limits our understanding of the genes, organisms, and environments that support DPO. It is difficult to predict the range of DPO taxa because D. phosphitoxidans FiPS-3 and Ca. P. anaerolimi represent distinct taxonomic classes (Desulfobacteria and Desulfomonilia_A), and their closest relatives are either uncultured or unable to catalyze DPO (17, 18). The environmental context of DPO remains ambiguous since DPOMs have only been identified in three distinct locations globally (17–19). Furthermore, the ptx-ptd cluster has unresolved genetic diversity. D. phosphitoxidans FiPS-3 and Ca. P. anaerolimi species have ptx-ptd clusters with alternative synteny and gene composition, and the PtxD proteins from FiPS-3 and Phox-21 share only 55% amino acid sequence similarity (18). Recognizing the breadth of hosts and environments supporting this metabolism and characterizing the underlying biochemistry and genetics would facilitate understanding of how DPOMs impact the phosphorus cycle.

Here we present the selective enrichment of diverse DPOMs in wastewater digester sludge from facilities around the San Francisco Bay area. Metagenome-assembled genomes (MAGs) uncovered 21 DPOMs spanning three disparate phyla. Comparative genomics revealed conservation of energy generation and carbon utilization pathways among DPOM genomes, despite taxonomic diversity. We also identified DPO genes throughout global metagenome databases and described the diversity of the ptx-ptd cluster. The phylogeny of ptx-ptd genes suggests that DPO metabolism is vertically inherited as a conserved unit since before the split of monoderm (gram-positive) and diderm (gram-negative) bacteria. Collectively, our results show that DPO is widespread across diverse environments and bacterial taxa, and likely represents a vestige of ancient microbial life.

Results

Selective Enrichment.

We hypothesized that DPOMs are cultivatable from wastewater sludge because phosphite can represent up to 2.27% of total dissolved wastewater phosphorus (23) and because both strains of Ca. P. anaerolimi were identified in wastewater digester sludge (18, 19). Accordingly, sludge from six San Francisco Bay area facilities was used to inoculate 30 enrichment cultures (Dataset S1). All cultures were grown in bicarbonate-buffered basal medium amended with 10 mM phosphite and multivariate exogenous electron acceptors (CO₂-only, CO₂ + SO₄²⁻, or CO₂ + NO₃⁻) (Dataset S1). Rumen fluid (5% by volume) was added to stimulate DPOM growth (18).

Phosphite oxidation was observed in 26 of 30 enrichments and across all six wastewater facilities (Fig. 1 A and B and Dataset S1). When stationary-phase enrichments were respiked with phosphite, DPO activity resumed. No phosphite oxidation occurred in autoclaved controls (Fig. 1A). Based on prior experience (18), the high percentage of active DPO enrichments was unpredicted, indicating a greater prevalence of DPOMs than previously assumed.

Fig. 1.

DPO enrichment activity. (A) Representative phosphite oxidation by the SM1 community. Temporal ion concentrations are shown for live (solid lines) or autoclaved (dashed lines) inoculum. Enrichments were amended with 10 mM phosphite at the spike point. (B) Percent change of measured ions for each enrichment community. Each row represents one community; each column displays the percent accumulation or consumption of each titled ion. Row labels are colored according to the added electron acceptor (black, CO₂ only; blue, CO₂ + SO₄²⁻; green, CO₂ + NO₃⁻). A white dotted line denotes 50% consumption of PO₃³⁻. All percentages were calculated from concentration values prior to the first spike point. (C) Duration of ion depletion. Horizontal bars show the time frame for the metabolic activity of each measured ion. Colors correspond to B (red, PO₃³⁻; blue, SO₄²⁻; green, NO₃⁻).

CO₂ Preference.

DPO was impacted by the amended electron acceptor. Active enrichments with only CO₂ supported the highest average phosphite oxidation rate (0.64 ± 0.17 mM PO₃³⁻/d for CO₂ versus 0.56 ± 0.10 and 0.50 ± 0.20 mM PO₃³⁻/d for NO₃⁻ and SO₄²⁻, respectively). CO₂ also supported DPO from all six sample sites. Despite the availability of nitrate and sulfate, neither electron acceptor was definitively coupled to phosphite oxidation (Fig. 1C). While all amended cultures consumed nitrate, it was metabolized before phosphite oxidation was complete, suggesting utilization independent of DPO. In fact, when compared with other cultures with the same inoculum, nitrate delayed or even excluded DPO (Fig. 1C). Meanwhile, although sulfate was consistently consumed at the expected ratio, if reduced to sulfide coupled to phosphite oxidation (1 mol sulfate/4 mol phosphite), the timing of sulfate consumption was variable and frequently offset from DPO (Fig. 1C). This suggests that sulfate reducers may be utilizing a reduced metabolite from DPO activity. Consistent with this, both of the characterized DPOMs either grow preferentially (FiPS-3) or exclusively (Phox-21) by autotrophy and utilize CO₂ as an electron acceptor. In the case of FiPS-3, the reduced carbon end product is acetate (17), which is readily utilized by sulfate reducers. Our results support a DPOM preference for CO₂ and indicate that alternative electron acceptors may inhibit DPO activity.

DPOM Identification.

To characterize the active DPOMs, we recovered MAGs from CO₂-only enrichments. To identify candidate DPOMs, we searched all MAGs using custom-built profile hidden Markov models (pHMMs) (SI Appendix, Files S1–S7) for each of the seven ptx-ptd genes (14, 16). In total, 21 genomes had at least one gene from the ptx-ptd cluster (DPO MAGs) and, of these, 19 were of high quality (>90% complete; <5% redundant) (Dataset S3) (24). DPO MAGs were enriched in all phosphite-amended communities (compared with no-phosphite controls) (Fig. 2A) and were dominant in all but one community (SL1) (Fig. 2B). Furthermore, every sequenced community had at least one DPO MAG (Fig. 2B). These results confirmed that DPO activity in phosphite-amended enrichments was dependent on the ptx-ptd genes and further indicate that these genes serve as effective probes for DPOMs.

Fig. 2.

Relative abundance of DPO MAGs. (A) Relative abundance of MAGs across samples. Each point represents one MAG. Color represents the presence (black) or absence (gray) of any ptx-ptd genes. (A, Top) Comparison of samples from phosphite-amended exponential phase (+Pe) with no-phosphite (−Ps) controls. (A, Bottom) Comparison of samples from phosphite-amended stationary phase (+Ps) with no-phosphite (−Ps) controls. (B) Relative abundance of MAGs across time. Each subplot represents one community, while each stacked bar represents the community composition of one sample. Colors indicate the dominant (maroon), second dominant (pink), and third dominant (yellow) DPO members, and all remaining community members (gray). Relative abundance was calculated by dividing the mean coverage of a single MAG by the sum of mean coverages for all MAGs in the respective sample.

DPOM Taxonomy.

DPOM taxonomy assignments were made using 1) reconstructed 16S rRNA gene fragments (25), 2) multigene alignments using the Genome Taxonomy Database (GTDB) (26), and 3) alignment of the ribosomal S8 proteins (rpS8) (27). Assignments were congruent in each instance and are visualized in Fig. 3. Prior to our study, DPOMs had been identified as belonging to only two taxonomic classes of the Desulfobacterota phylum. In contrast, DPOMs in our enrichments span the monoderm–diderm taxonomic boundaries and include three phyla (Desulfobacterota, Firmicutes, and Synergistota) and six classes (Negativicutes, Desulfotomaculia, Synergistia, Syntrophia, Desulfobacteria, and Desulfomonilia_A) (Fig. 3).

Fig. 3.

Phylogenetic trees of DPO MAGs. (A) A phylogenetic tree of bacterial genomes from the GTDB was visualized with AnnoTree (74). Nodes of the tree represent class-level taxonomy, and those nodes with DPO organisms are highlighted according to the key. (B–D) Phylogenetic trees of the rpS8 marker gene showing the relationship of DPO MAGs to their closest relatives. Panels depict DPO MAGs belonging to the same phyla: (B) Firmicutes, (C) Synergistota, and (D) Desulfobacterota. The DPO MAGs from this study are bolded. Colored squares represent their dominance rank from Fig. 2B. Each close relative is annotated with its species name, accession number, and genome-source type (isolate vs. MAG), as well as its percent identity to the most closely related DPO MAG from this study. Clades are colored and labeled by taxonomic class. Internal nodes with a bootstrap support of >90% are indicated by closed circles and those with a support of >70% are indicated by open circles. (Scale bars, 0.2 change per amino acid residue.)

Desulfomonilia_A was the most sampled class of DPOMs (Fig. 3), comprising 13 of 21 DPO MAGs. They were enriched from all six sample sites and present in nine communities. Furthermore, they were the most relatively abundant DPO MAG (representing >85%) in each of eight communities, indicating a possible advantage under our enrichment conditions (Figs. 2B and 3). Desulfomonilia_A is an uncultured class that has recently been distinguished from the Desulfomonilia (https://gtdb.ecogenomic.org/). Consistent with this, the Desulfomonilia are represented by Desulfomonile tiedjei, which shares just 49% rpS8 sequence identity with the most closely related DPO MAG of Desulfomonilia_A (28, 29) (Fig. 3). All DPO MAGs of the Desulfomonilia_A class belong to the uncultured order UBA1062 (previously denoted GW-28) (18), which includes Ca. Phosphitivorax (Dataset S3). The monophyletic separation of the Desulfomonilia_A DPOMs supports the hypothesis that Ca. Phosphitivorax species are part of a unique order, and possibly a unique class, for which DPO is a common metabolic feature (18).

Beyond the Desulfomonilia_A DPOMs, we recovered eight additional genomic representatives from four classes (Negativicutes, Desulfotomaculia, Synergistia, and Syntrophia) (Fig. 3). While most of these are minority DPOMs in their respective communities, at least three (Pelotomaculaceae SL1, Ca. Smithella SM1, and Ca. Smithella LM1) dominate their DPOM populations (>83%) (Figs. 2B and 3). The Ca. Negativicutes and Ca. Desulfotomaculia MAGs represent the first DPO genomes taxonomically assigned to the Firmicutes phylum, highlighting the broad evolutionary divergence of DPOMs (30–32).

The closest cultured relatives of DPO MAGs share 57 to 95% rpS8 amino acid sequence identity (Fig. 3), which surpasses the species threshold (<98.3%) (27). Furthermore, multigene classification by the GTDB designates these related isolates as belonging to at least different genera (Dataset S4), making predictions about DPOM physiology from taxonomy unreliable. Regardless, all characterized DPO MAG relatives, regardless of taxonomy, are obligately anaerobic chemoorganotrophs. Furthermore, the Desulfomonilia_A, Desulfotomaculia, and Syntrophia classes contain canonical representatives that are dependent on syntrophic associations (19, 31, 33–35). The phylogenetic relatedness of our DPOMs to notoriously fastidious syntrophic organisms could explain the difficulty in isolating DPOMs (18).

The 16S rRNA gene is the canonical taxonomic marker for resolving microbial speciation. While not present in all DPO MAGs, 86% (n = 18) contained the 16S rRNA gene, enabling refined taxonomic analyses (Dataset S4). To capture the novelty of enriched DPOMs, we used EMIRGE to reconstruct full-length 16S rRNA gene sequences that were BLAST-searched in the SILVA database (25, 36). We determined that the DPOMs represented 14 new strains, 6 new species, and 1 new genus based on standardized relatedness metrics (27). Proposed names and etymologies are provided in Dataset S4. The new genus, represented by Cosmobacter schinkii SL3 [named in recognition of Bernhard Schink, for his exemplary contributions to microbiology and discovery of the first DPOM (13, 17)], is the second characterized genus of the Desulfomonilia_A UBA1062 order, in addition to Ca. Phosphitivorax. Consequently, UBA1062 was expanded to include two genera and five species (Fig. 3).

Metabolic Traits.

The genomes of FiPS-3 and Phox-21 have been used to predict the mechanism for DPO energy conservation (14, 18). In the model (Fig. 4), the Ptx-Ptd protein cluster is hypothesized to oxidize phosphite and generate NADH and adenosine triphosphate (ATP) through substrate-level phosphorylation. Alternative reducing equivalents are likely produced via a Na⁺ motive force, ferredoxin, and an electron confurcation mechanism. The model proposes CO₂ to be fixed into biomass via the reductive glycine pathway, as was suggested for Phox-21 (18). In contrast, FiPS-3 utilizes the Wood–Ljungdahl pathway (14). By comparing the genomes of DPO MAGs with FiPS-3 and Phox-21, we found highly conserved metabolic traits beyond the ptx-ptd gene cluster, regardless of taxonomy.

Fig. 4.

Metabolic model of energy conservation by Desulfomonilia_A DPOMs [adapted from Figueroa et al. (18)]. Reactions are diagramed in their hypothesized locations in relation to the inner membrane (IM) and outer membrane (OM) of a bacterial cell. Dashed lines represent mechanisms that have not been biochemically confirmed. Balanced equations are provided for phosphite oxidation and CO₂ reduction to d-lactate. Dissimilatory phosphite oxidation proteins: 1) PtdC, phosphite-phosphate antiporter; 2) PtxDE-PtdFHI, putative phosphite dehydrogenase protein complex. CO₂ reduction (reductive glycine pathway) proteins: 3) FdhAB/FdoGHI, formate dehydrogenase; 4) Fhs, formate:tetrahydrofolate (THF) ligase; 5) FolD, methylene-THF dehydrogenase/methenyl-THF cyclohydrolase; 6) glycine cleavage system (GcvH, lipoyl-carrier protein; GcvPAB, glycine dehydrogenase; GcvT, aminomethyltransferase; Lpd, dihydrolipoyl dehydrogenase); 7) GlyA, serine hydroxymethyltransferase; 8) SdaA/IlvA, serine dehydratase/threonine dehydratase; 9) LdhA, d-lactate dehydrogenase. Energy conversion proteins: 10) ATP synthase complex; 11) Rnf, sodium-translocating ferredoxin:NAD oxidoreductase complex; 12) NfnAB, NAD-dependent ferredoxin:NADP oxidoreductase.

Energy Conservation.

Like Phox-21, all DPO MAGs were missing a canonical membrane-bound electron transport chain (ETC), as complexes II to IV were either absent or incomplete (Fig. 5). Sporomusaceae SM1 of the Negativicutes class had a complete NADH:quinone oxidoreductase (complex I), including the N, Q, and P modules for NADH dehydrogenase activity, quinone reduction, and proton translocation, respectively. However, all other DPO MAGs only contained N-module subunits (Fig. 5 and Dataset S5). The N module houses the FMN and FeS clusters for electron transport, as well as the NADH binding site. It also chimerically associates with other protein complexes, such as formate dehydrogenases, catalyzing reversible NADH-dependent formate production (37, 38). Poehlein et al. suggested that the FiPS-3 N module may directly transfer electrons from NADH to ferredoxin (14). However, direct NADH-dependent ferredoxin reduction is thermodynamically unfavorable (39). Furthermore, the N module of DPO MAGs is located in various genomic contexts, making it unclear whether the commonality is uniquely associated with DPO activity or with alternative cellular functions.

Fig. 5.

Carbon and energy metabolism of DPO MAGs. Each DPO MAG was subjected to metabolic analysis via DRAM (64, 75). Within this heatmap, each cell represents a metabolic pathway (rows) for each DPO genome (columns). The number of genes for a given pathway is described by percent completion ranging from 0% (white) to 100% (brown). Pathways are organized into modules related to carbon metabolism, electron transport chain complexes, and other enzymes referenced in the text. Organisms are annotated with their taxonomic class.

In Phox-21, ferredoxin reduction by NADH is attributed to a sodium-translocating ferrodoxin:NADH oxidoreductase (Rnf) driven by a Na⁺ motive force (18) (Fig. 4). Consistent with Phox-21, an Rnf complex was present in the Synergistia and nearly all Desulfomonilia_A DPO MAGs (Fig. 5). In contrast, the Rnf was absent from the Negativicutes, Desulfotomaculia, and Syntrophia DPO MAGs, suggesting that it is dispensable or replaceable for DPO activity. The ion motive force for Rnf activity in Phox-21 is likely provided by a cation-translocating F-type ATPase at the expense of ATP (Fig. 4). The F-type ATPase was present in every DPO MAG, except one (Synergistaceae SL3) which had the V type (Fig. 5). While two genomes (Syntrophales LM1 and Pelotomaculaceae LM1) were missing several ATPase subunits, these were only 61 and 69% complete (Dataset S3) (24). Given the universal absence of an ETC in DPO MAGs, the ATPases are likely involved in ATP hydrolysis with the concomitant generation of a cation motive force.

CO₂ as an Electron Acceptor.

No DPO MAGs harbored functional pathways for methanogenesis or common respiratory pathways (oxygen, nitrate, or sulfate), which is similar to Phox-21 and consistent with the absence of ETC complexes (SI Appendix, Fig. S2). Furthermore, CO₂ was the only exogenous electron acceptor available to DPOMs in sequenced cultures. Consistent with Phox-21 (18), a physiological survey of one of our enrichments showed that CO₂ was necessary and sufficient to support phosphite oxidation and growth (Fig. 6). As observed in Phox-21, comparative genomics of DPO MAGs revealed a notable absence of any canonical CO₂ reduction pathways (Fig. 5). While FiPS-3 can use CO₂ as an electron acceptor by reducing it to acetate via the Wood–Ljungdahl pathway (17), carbon reduction in Phox-21 was attributed to the reductive glycine pathway (18). The reductive glycine pathway is composed of the methyl branch of the Wood–Ljungdahl pathway, combined with the glycine cleavage system, serine hydroxymethyltransferase, and serine deaminase to produce pyruvate as an anabolic intermediate (18, 22) (Fig. 4). The Phox-21 final product of CO₂ reduction remains unknown, as the genes for pyruvate conversion to acetate (phosphotransacetylase and acetate kinase) are missing from the genome. Lactate is a possibility, as the genomes of Phox-21 and all other Desulfomonilia_A DPO MAGs contain d-lactate dehydrogenase, which converts pyruvate to lactate at the expense of NADH (Fig. 4). This is an energetically favorable reaction that accounts for all reducing equivalents produced via phosphite oxidation according to Fig. 4 and

$$6{\text{PO}}_3^{3–}+3{\text{CO}}_2+3{\text{H}}_2\text{O}\to 6{\text{PO}}_4^{3–}+{\text{C}}_3{\text{H}}_6{\text{O}}_3$$

$$\mathrm{\Delta }\mathit{\text{G}}°\prime ;=–29{\text{kJ/mol e}}^{–}\left(–348\text{kJ/mol lactate}\right).$$

Fig. 6.

CO₂-dependent DPO activity. Growth and phosphite concentrations were temporally monitored in the presence and absence of CO₂ for the SV3 community. Autoclaved controls showed no activity. Error bars represent SD of triplicate cultures.

CO₂ Fixation to Biomass.

In addition to serving as the electron acceptor for DPOMs, CO₂ is also fixed into biomass as the carbon source (18) (Fig. 6). While none of the DPO MAGs contained any canonical CO₂ fixation pathways (40), 12 in the Desulfomonilia_A, Negativicutes, and Syntrophia classes had all the genes necessary for CO₂ fixation to pyruvate via the reductive glycine pathway (22) (Figs. 4 and 5). Of the residual nine DPO MAGs whose reductive glycine pathway was incomplete, four were missing homologs of serine deaminases, preventing the final conversion of serine to pyruvate (Dataset S6). The remaining five DPO MAGs (ranging from 61.3 to 98.7% completion) were missing between one and four genes involved in formate and/or glycine transformations, severely impeding the overall pathway (Datasets S5 and S6). It is possible that homologous enzymes may perform the reactions of missing genes, as might be the case for one genome (Syntrophales LM1) which harbored a serine-glyoxylate transaminase as opposed to the standard serine deaminase (Dataset S6). Even if not a universal carbon fixation pathway in DPOMs, our analyses suggest the reductive glycine pathway might be an important autotrophic mechanism across diverse DPO taxa. Carbon-tracing studies will be necessary to understand how individual DPOMs use CO₂ to simultaneously generate biomass and capture energy from phosphite oxidation.

ptx-ptd Cluster Diversity.

DPO activity in FiPS-3 and Phox-21 was attributed to the ptx-ptd gene cluster, and only organisms with ptx-ptd genes were enriched here, positing this to be the dominant, or possibly sole, metabolic pathway underlying dissimilatory phosphite oxidation (14, 15, 18) (Fig. 2A). To determine the prevalence and diversity of DPOMs beyond our enrichments, we used the PtxD protein sequence from FiPS-3 as a marker gene to query the IMG/M protein sequence space (Dataset S7). We recovered 15 positive hits that were phylogenetically compared with the PtxD from our enriched DPO MAGs and the two previously known DPO species (Fig. 7).

Fig. 7.

Phylogenetic trees of the phosphite dehydrogenase PtxD. (A) The PtxDs from IMG/M metagenomes and DPO MAGs were aligned with proteins from the 2-hydroxyacid dehydrogenase family (Pfam PF00389; set representative proteomes to 15%). Protein subfamilies were assigned based on Matelska et al. (41). An arrow indicates the location of PtxD proteins that are associated with DPO-PtdC but clade with assimilatory phosphite oxidation PtxD (APO). (Scale bar, 0.5 change per amino acid residue.) (B) Refined tree of all PtxDs within the DPO-PtxD clade. PtxDs from the IMG/M are in light black font and labeled with their source environment and scaffold ID. PtxDs from our enriched DPO MAGs are bolded and labeled with their bacterial host name. PtxDs that belong to a binned organism are highlighted based on their taxonomic class. Published organisms with validated DPO activity are in red font. Only genes adhering to the IMG/M data usage policy are shown. Internal nodes with a bootstrap support of >70% are indicated by closed circles and those with a support of >50% are indicated by open circles. (Scale bar, 0.3 change per amino acid residue.) (C) The presence (maroon) or absence (light pink) of ptx-ptd genes in each genome was determined using custom pHMMs. Genes that were absent from a DPO MAG but present in the assembly are in gray, where phylogeny, tanglegrams, and synteny were collectively used to predict the most likely host. (D) Horizontal gray bars display the size (bp) of the contig on which each PtxD was found and are in logarithmic scale to visualize the full range of contig lengths. The black dotted line indicates the minimum length for all seven ptx-ptd genes to be present, based on FiPS-3 sequences (7,137 bp). Asterisks signify contigs that were binned.

Our analysis revealed that the DPO-PtxD formed a monophyletic clade that included all validated DPOMs (i.e., FiPS-3, Phox-21, and our enriched DPOMs). The DPO-PtxD belonged to the glyoxylate/hydroxypyruvate reductase B (GHRB) protein subfamily of the d-2-hydroxyacid dehydrogenases (2HADHs). The closest relatives of DPO-PtxD are the sugar dehydrogenases and the PtxD homologs involved in phosphorus assimilation (41) (Fig. 7A and Dataset S8). The DPO-PtxD can be distinguished from closely related proteins based on the presence of nearby ptd genes (Fig. 7). The closest non-DPO homolog (Ga0209611_101991891) of the DPO-PtxD lacks the remaining ptx-ptd genes in the inclusion matrix (Fig. 7C), demonstrating the specificity of our custom pHMMs (SI Appendix, Files S1–S7).

The DPO-PtxD was found exclusively in anoxic environments (Fig. 7B). The predicted failure of DPOMs to occupy oxic environments, despite the thermodynamic favorability of DPO coupled to oxygen respiration (ΔG^o′ = −283 kJ⋅mol⁻¹ PO₃³⁻), suggests that metabolic proteins may be oxygen-sensitive. Alternatively, DPO metabolism may be dependent on the biochemical pathways of anaerobes. While DPOMs appear to be common members of diverse anoxic environments, further analyses will be required to describe their relative abundance in natural habitats.

Evolutionary History.

The DPO evolutionary history was ascertained using 1) genomic features, 2) comparative taxonomic clustering, and 3) syntenic conservation. Within the DPO-PtxD clade, proteins clustered based on host taxonomy, and the PtxD was distinguishable at the genus level (Fig. 7B). The only deviation from this pattern was Ca. Smithella phosphorovis LM1 of the Syntrophia class, which had a PtxD lineage consistent with the Ca. Phosphitivorax species of the Desulfomonilia_A class (Fig. 7B). The ptx-ptd cluster from Ca. S. phosphorovis LM1 occurred on a single contig (13,378 bp) that hosted an IS91 family transposase. This contig had a sequencing depth (64.7×) threefold that of the bin’s average coverage (19.4×), and the GC content (57.4%) was 3.5% higher than the host genome mean GC content (53.9%). Together, these findings suggest that, like FiPS-3 (14), Ca. S. phosphorovis LM1 likely acquired its ptx-ptd genes through HGT. Consistent with this conclusion, the LM1 community assembly did not include taxonomic marker genes for Ca. Phosphitivorax species, and the assembly graph supported the binning results, precluding a different bin assignment for this contig.

In contrast to FiPS-3 and Ca. S. phosphorovis LM1, most PtxDs clustered according to host taxonomy, indicating that most DPOMs likely acquired their PtxD via vertical inheritance (Fig. 7B). Similar taxonomic clustering occurred for PtdC and PtdF, further suggesting that the ptx-ptd genes are inherited as a metabolic unit (SI Appendix, Fig. S3). Tanglegram analyses facilitate a coarse approximation of topological similarity between gene phylogenies, where crossing lines (“tangles”) indicate alternative evolutionary histories (42). Comparisons of PtxD, PtdC, and PtdF exhibited zero tangling, supporting a linked evolutionary history (SI Appendix, Fig. S4). Although the phylogenetic trees of individual DPO genes showed alternative branching patterns, this was expected, as genes with functional differences are subject to unique selective pressures.

Synteny provides an alternative metric to gauge the unison of ptx-ptd gene evolution because 1) linked genes tend to maintain organization throughout evolutionary history, and 2) closely related taxa show high genomic stability (43, 44). We found that the individual ptx and ptd genes were always codirectional in the order ptxED and ptdFC(G)HI, respectively (SI Appendix, Fig. S5). However, the directionality between the ptx and ptd gene clusters was variable and syntenic variation formed four distinct groups (groups I to IV; SI Appendix, Fig. S5) that correlated with host taxonomy. Groups I and IV do not contain ptdG, suggesting it is nonessential (SI Appendix, Fig. S5). While other genes were frequently missing from the ptx-ptd cluster, synteny analysis suggested this is due to fragmented contigs (Fig. 7D and SI Appendix, Fig. S5). For example, Synergistaceae SL3 was identified as a DPOM in our enrichments, but our pHMM search failed to identify its PtxD (Fig. 7). Synteny suggested that the ptxD sequence was truncated downstream of ptxE, which was confirmed by BLAST alignment (SI Appendix, Fig. S5).

Searching metagenome databases with alternative DPO marker genes would likely reveal other DPO contigs that were artificially separated from their ptxD gene. This was the case when we mined the IMG/M database for PtdC and identified five additional contigs with divergent PtxD phylogeny (Fig. 7B and SI Appendix, Fig. S4). While these divergent protein sequences may indicate further DPO diversity, their contigs showed noncanonical ptx-ptd neighborhoods and are not yet represented by validated DPO cultures. For those ptx-ptd clusters that confidently represent DPOMs, the predominance of vertical transfer was collectively supported by genomic features, taxonomy, and synteny.

Discussion

We used cultivation-based investigations coupled to high-resolution metagenomics to clarify many of the confounding factors that have precluded understanding of DPO. Results from our studies have expanded the known diversity of DPOMs 10-fold (from 2 to 21 genomes). Notably, phosphite oxidation coupled to CO₂ reduction appears to be the primary metabolic niche occupied by DPOMs. Although DPO coupled to any known inorganic electron acceptor (oxygen, manganese, perchlorate, nitrate, iron, sulfate, etc.) is thermodynamically favorable, DPOM genomes encode sparse electron transport machinery and are largely devoid of the enzymes required to reduce these compounds. CO₂ was the only exogenous electron acceptor provided to our sequenced enrichments, and physiological experiments demonstrated a CO₂ dependency. Yet DPOMs also lacked canonical carbon reduction or fixation pathways. The reductive glycine pathway was present in many DPOMs and may support CO₂ fixation, but the method by which CO₂ is fixed by the remaining DPOMs is unknown, as is the end product of CO₂ reduction (e.g., ethanol or lactate), begging future metabolomic analyses.

The highly specialized metabolic repertoire of DPOMs is analogous to that of syntrophs, corroborating the observation that DPOMs frequently belong to known syntrophic taxa (45). Thermodynamically, phosphite is too energetically favorable an electron donor to require a syntrophic partner, but such a codependency would explain their resistance to isolation (18). D. phosphitoxidans FiPS-3 remains the only cultured isolate to date, yet we failed to cultivate any close relatives of FiPS-3 in our enrichments, despite otherwise representing much of the DPO diversity present in global metagenomes. Furthermore, we found that FiPS-3 is phenotypically and genotypically anomalous when compared with other DPOMs. FiPS-3 exhibits greater metabolic versatility than typical DPOMs, reducing sulfate, thiosulfate, and nitrate as electron acceptors in addition to CO₂ (14, 17, 18). FiPS-3 is also one of only two examples by which ptx-ptd genes were likely acquired via HGT, suggesting that DPO is not its primary energy metabolism. Future efforts to cultivate DPOMs may consequently be informed by our DPO MAGs, whose metabolic features suggest a dependence on limited substrates and a potential requirement for microbial partnerships.

Our DPOMs spanned six classes of three bacterial phyla (Desulfobacterota, Firmicutes, and Synergistota). Such sparse representation across diverse taxa is typically indicative of broad–host-range HGT, but phylogenetic analyses of the ptx-ptd gene cluster showed that DPO metabolic gene evolution mirrored the host taxonomy. This indicates that vertical transfer is the predominant mechanism of inheritance. Small variations in synteny further support the correlation between gene order and taxonomy while also suggesting that ptx-ptd genes have coevolved as a metabolic unit specialized for DPO metabolism.

Given the diversity of DPOM lineages that likely inherited the ptx-ptd gene cluster vertically, it is tempting to speculate the biological timescale for when DPO metabolism originated. The last common node for all known DPOMs suggests that DPOMs arose before the divergence of monoderm and diderm bacteria (46). Mapping the divergence of these clades to geological timescales suggests that DPOMs evolved ∼3.2 Gya (47), contemporaneous with anoxygenic photosynthesis and ∼0.8 Gya after the evolution of methanogenesis (47). This is consistent with the suggestion that phosphite composed 40 to 67% of dissolved phosphorus species in Archaean oceans (>3.5 Gya) (48, 49). The half-life of oceanic phosphite under a reducing atmosphere is expected to be 0.1 to 10 billion y, which would have allowed phosphite persistence on early Earth, possibly supporting a robust chemolithoautotrophic DPO population.

One would expect such an ancient metabolism to be detected more broadly across all bacteria. However, oxygenation of Earth’s atmosphere since the great oxidation event (∼2.5 Gya) has likely depleted ancient natural phosphite reserves, as oxidizing radicals abiotically oxidize phosphite on geological timescales (3, 50). Phosphite would consequently be too rare for DPO in most contemporary environments, and lack of positive selection would promote widespread gene loss (51). Yet pockets of phosphite (0.1 to 1.3 μM) exist in diverse contemporary environments, and phosphite-oxidizing metabolisms still occur in various habitats on extant Earth (11, 23, 52, 53). Environmental metadata from global metagenomes identified DPOMs in multiple anoxic environments that represent relics of ancient Earth (i.e., oil reservoirs, deep subsurface aquifers) and serve as potential examples of contemporary phosphite accumulation (i.e., wastewater sludge, freshwater wetlands). A number of environments evidently continue to support phosphorus redox cycling. By coupling DPO to primary production via an uncharacterized CO₂ reduction pathway, DPOMs likely play a unique ecological role in any environment they inhabit.

Methods

Growth Conditions and Sampling.

Enrichment inocula were obtained from six wastewater treatment facilities in the San Francisco Bay area of California (Dataset S1). Serum bottles (150-mL volume) (Bellco) containing basal media (45 mL) were each inoculated with sludge (5 mL) and incubated at 37 °C. Anoxic medium was prepared by boiling under N₂/CO₂ (80:20, volume/volume [vol/vol]) to remove dissolved O₂, and dispensed under N₂/CO₂ (80:20, vol/vol) into anaerobic pressure tubes or serum bottles. These were capped with thick butyl rubber stoppers and sterilized by autoclaving (15 min at 121 °C). The basal medium was composed of (per 1 L of deionized water) 5 g NaHCO₃, 12 g Hepes buffer, 1 g NH₄Cl, 0.5 g KCl, 1.5 g MgCl₂, 0.15 g CaCl₂ (2H₂O), 0.5 g l-cysteine HCl, and 10 mL each of vitamins and trace minerals (54). Saline medium additionally contained 20 g/L NaCl. Salt solutions of Na₂HPO₃, Na₂SO₄, and NaNO₃ (10 mM) were added from sterile anoxic stocks as needed. Rumen fluid (Bar Diamond), prepared by degassing (30 min with N₂) and autoclaving (121 °C for 30 min), was added to the basal media as required. Heat-killed controls were autoclaved at 121 °C for 1 h. Samples for DNA extraction were pelleted by 30-min centrifugation at 10,000 relative centrifugal force and stored at −80 °C. Samples for ion determination were filtered and stored at 4 °C prior to ion chromatography using the method described previously (18). Cell growth was measured as optical density at 600 nm (OD₆₀₀) using a Genesys 20 Visible spectrophotometer (Thermo Scientific).

Metagenomic Assembly, Binning, and Annotation.

Sequenced communities were grown in triplicate cultures amended with 5% rumen fluid with or without 10 mM phosphite (SI Appendix, Fig. S1). DNA was extracted from the no-phosphite triplicates in stationary phase (−Ps) and the 10 mM phosphite triplicates in exponential phase (+Pe) and stationary phase (+Ps) (SI Appendix, Fig. S1). Community R1 failed to reach stationary phase and was only represented by samples −Ps and +Pe. Communities LM1, R3, SL1, and SL3 failed to reproduce activity and were instead sampled from two previously active enrichments (E1 and E2) (SI Appendix, Fig. S1). DNA was extracted using the DNeasy PowerLyzer Microbial Kit (Qiagen) and sequenced with an Illumina HiSeq 4000 (150-bp paired-end reads) at the University of California Berkeley Vincent J. Coates Genomics Sequencing Laboratory. Reads were trimmed and filtered using Sickle v1.33 (quality threshold value of 20) (55). Gene-level taxonomy was assigned using Centrifuge v1.0.1-beta-27-g30e3f06ec3 (56). Reads for each of the 11 communities were combined and coassembled using MEGAHIT v1.1.2 (57) using the metasensitive preset. Reads were mapped to assembled contigs using BWA-MEM v0.7.17 (58) with default parameters. Contigs over 1,000 bp from each combined assembly were binned into individual genomes using Anvi’o v5.4.0 (59). Communities with <30,000 contigs (LM3, M1, R1, SM1, SM3, SV1, SV3) were binned manually using patterns of hierarchical clustering, sequencing coverage, GC content, and gene-level taxonomic assignments. Communities with >30,000 contigs (LM1, R3, SL1, SL3) were binned automatically using CONCOCT and then manually refined with the Anvi’o graphical interface (60). Quality of MAGs was measured from lineage-specific, conserved, single-copy marker genes (SCG) using the CheckM v1.0.18 lineage workflow (61). The resulting 11 coassemblies consisted of 1,900 Mbp, 1.99 million contigs, and 574 draft genomes (Dataset S2). Only draft genomes of medium quality or greater (>50% completion; <10% redundant) (24) were subjected to further study, resulting in 239 MAGs that represent 60% (647 Mbp) of the binned contigs (Dataset S3). Open reading frames were predicted from selected genomes using Prodigal v2.6.3 (62) and assigned taxonomy using the Genome Taxonomy Database Toolkit (26), which placed MAGs into protein reference trees using concatenated SCG sets. Contigs of interest were functionally annotated with Prokka v1.14.6 (63).

The DPO MAGs were also annotated with DRAM (64), a genome annotation tool that provides metabolic profiles for each input genome. For contigs of interest, these annotations were compared with Prokka v1.14.6 annotations (63). More detailed DRAM analyses are provided in Shaffer, Borton et al. (64). The raw annotations containing an inventory of all database annotations for every gene from each input genome are reported in Dataset S6. From the raw annotations, DRAM then summarizes key metabolisms across the genomes, with Fig. 5 and SI Appendix, Fig. S2 showing the DRAM Product output. All code for DRAM is available on GitHub: https://github.com/shafferm/DRAM.

Identification of Metagenomic DPO Proteins.

DPO proteins (PtxD, PtdC, PtdF) were identified from publicly available metagenomes. The largest metagenomes (representing 90% of proteins from each ecosystem category) in the JGI IMG/M database were collected (n = 17,888) on 1 August 2018 (65) (Dataset S7). Sequence data from the IMG/M database were produced by the US Department of Energy Joint Genome Institute (https://www.jgi.doe.gov/) in collaboration with the user community. The FiPS-3 PtxD, PtdC, and PtdF were searched against all proteins using BLASTP with bit score thresholds of 270, 300, and 250, respectively. Positive hits were aligned using MUSCLE v3.8.1551 (66) and constructed into an approximately maximum-likelihood phylogenetic tree using FastTree v2.1.11 (67) with 1,000 bootstrap resamplings. DPO proteins were defined as those that 1) formed a phylogenetically distinct clade with proteins from experimentally validated DPOMs, 2) were found on a contig near at least one other putative DPO gene, and 3) were at least 90% of the length of their homolog protein in FiPS-3. Protein sequences from the identified ptx-ptd gene clusters were used to create profile hidden Markov models for each of the PtxDE-PtdCFGHI proteins using HMMER v3.2.1 (68, 69). These pHMMs are available as supporting files (SI Appendix, Files S1–S7). Bit score thresholds for stringent de novo identification of DPO proteins were determined by a reciprocal pHMM search on a subset of the IMG/M database (Dataset S9). To compare the evolutionary relationships between PtxD, PtdC, and PtdF, members of the DPO clade were dereplicated with CD-HIT v4.8.1 (70) by clustering proteins with 100% sequence similarity and selecting the largest contig to represent each gene cluster in a simplified phylogenetic tree. Tanglegrams comparing PtxD with PtdC and PtdF were generated with Dendroscope v3.7.2 (71). Gene synteny was visualized with SimpleSynteny (72), where genes were identified with BLAST and annotated according to our custom pHMMs.

Characterization of DPO Genomes.

Annotated protein sequences from all MAGs were searched for known DPO proteins (PtxDE-PtdCFGHI) with our custom pHMMs. MAGs were operationally considered capable of DPO if they included at least one gene from the ptx-ptd gene cluster. The ptx-ptd genes that were absent from MAGs were searched for in all remaining contigs of the respective community.

A pHMM for rpS8 was obtained from Wu et al. and applied to all DPO MAGs (69, 73). The rpS8 gene has been shown to effectively represent whole-genome average nucleotide identity values (27) and was present once in each DPO MAG. Each rpS8 was BLAST-searched against the National Center for Biotechnology Information (NCBI) GenBank database to identify the closest relative, closest isolated relative, and informative representatives for phylogenetic analysis. Identified close relatives corresponded to the multigene taxonomy assignments of the GTDB (Dataset S4). Sequences were aligned using MUSCLE v3.8.1551 (66), and an approximately maximum-likelihood phylogenetic tree was constructed with 1,000 bootstrap resamplings using FastTree v2.1.11 (67). Trees were visualized using FigTree v1.4.4 (tree.bio.ed.ac.uk/software/figtree/).

The 16S rRNA gene for each community was reconstructed from metagenomic reads using default parameters in EMIRGE with 50 iterations (25). Reconstructed genes were classified using SILVA (36) and mapped back to the 16S rRNA gene fragments of DPO MAGs. The novelty of each DPO MAG was determined by the rank of closest relatives in the GTDB, NCBI (rpS8), and SILVA (16S rRNA gene) databases (Dataset S4). A DPO MAG was considered novel at the specified rank (i.e., species, genus) based on the following thresholds: 1) GTDB, considered novel if there were no logged relatives for that rank; 2) NCBI (rpS8), considered a novel species if the closest relative was <98.3% identity; and 3) SILVA (16S rRNA gene), considered a novel species if the closest relative shared <96.7% identity and a novel genus if the closest relative shared <94% identity (27). The novelty of a DPO MAG was assigned based on the lowest resolved taxonomic rank between all searched databases.

Data Availability

All metagenomic reads, assemblies, and curated metagenome-assembled genomes reported in this paper (quality metrics >50% complete and <10% redundant) have been deposited in the NCBI BioProject (accession no. PRJNA655520).