Freshwater Bacteria Release Methane as a By-Product of Phosphorus Acquisition

Surface water was collected from Lake Matano, Indonesia, for DNA extraction and sequencing from January to March 2009. Water (100 liters) was pumped through 0.22-μm Steripak filters, which were frozen on-site and kept frozen until processing. DNA was extracted and sequenced using a 454 Life Sciences GS-FLX instrument with Titanium series reagents, as described earlier for water collected from the Lake Matano chemocline (40).

The raw reads were compared to the Clusters of Orthologous Groups (COG) database (41) using a BLASTx search with an E value cutoff of 10⁻¹⁰ and a minimum bit score of 50. The COG category of the best BLASTx hit was assigned to the raw reads. Odds ratios were calculated for each COG category identified in the metagenomic data set by calculating the ratio (A/B)/(C/D), where A is the number of reads in the metagenome that are in a given COG category, B is the number of reads in the metagenome that are in all of the other COG categories, C is the number of proteins in the COG database that are in a given COG category, and D is the number of proteins in the COG database that are in all of the other COG categories. COG categories with odds ratios of >1 were considered enriched (42), meaning that they are more highly represented in the metagenomic data than in the protein database. For comparison, odds ratios were calculated for the same COG categories in the metagenomic data from the chemocline in Lake Matano (40), where phosphate is not limiting (37).

Metagenomic reads were assembled into contigs using Newbler GS De Novo Assembler (software version 2.5.3) with the default settings for genomic projects. Contigs potentially including phosphonate degradation genes were identified by blasting the C-P lyase amino acid sequences from Escherichia coli (GenBank accession number NC_000913.3, locus tags b4092 to b4106 [https://www.ncbi.nlm.nih.gov/gene?LinkName=pubmed_gene&from_uid=1335942]) against the contigs. Contigs with at least three C-P lyase genes were analyzed using the open reading frame (ORF) finder (NCBI), BLAST (Basic Local Alignment Search Tool; NCBI), and CD-search (Conserved Domains; NCBI) to predict and annotate the genes on the contigs.

The PhnJ amino acid sequences from Pseudomonas stutzeri (NCBI accession no. AAR91738) and E. coli (NCBI accession no. AAC77059.1) were used as BLAST queries against freshwater metagenomic data sets publicly available at the Joint Genome Institute's Integrated Microbial Genomes website (https://img.jgi.doe.gov/cgi-bin/mer/main.cgi) and at the European Nucleotide Archive (http://www.ebi.ac.uk/ena) (43). Data sets were scored as positive for phnJ if they had at least one hit with an E value of less than 1 × 10⁻¹⁰.

Seven heterotrophic bacteria isolated from the surface water of Lake Matano (Table 1) (38) were screened for their ability to acquire P from MPn. The base medium was the National Botanical Research Institute phosphate growth medium (NBRIP) without added P [per liter: MgCl₂·6H₂O, 5 g; MgSO₄·7H₂O, 0.25 g; KCl, 0.2 g; (NH₄)₂SO₄, 0.1 g; glucose, 10 g] (44). To test for MPn utilization, strains were grown in NBRIP base medium with 0.2 mM MPn as the only P source. Production of methane from MPn was characterized in the four strains (LM-1, LM-5, LM6-1, and LM-Y) capable of using MPn as their sole source of P. MPn, K₂HPO₄, or both were added to the P-free NBRIP medium as P sources. The pH of the medium was adjusted to 7.0 with 1 mM solutions of NaHCO₃ or acetic acid for the medium with added K₂HPO₄ or 1 mM solutions of NaHCO₃ or NaOH for medium with added MPn, as appropriate. Prior to measuring growth curves, cells were grown in NBRIP with K₂HPO₄ as the only P source. Serum bottles (60 ml) containing 30 ml medium and air in the headspace were inoculated with cells to an optical density at 660 nm (OD₆₆₀) of ∼0.05 and incubated horizontally at 30°C with shaking at ∼150 rpm. Growth was monitored by measuring the OD₆₆₀.

For the initial growth experiments, MPn, KH₂PO₄, or both were added to concentrations of 0.2 mM each. For experiments characterizing methane release with different concentrations of MPn, no phosphate was added to the medium, and MPn was added to concentrations of 5 μM, 10 μM, 50 μM, 100 μM, or 200 μM. To determine the effect of phosphate addition on MPn degradation, the initial MPn concentration was 50 μM and the starting K₂HPO₄ concentration was 0 μM, 4 μM, 10 μM, 50 μM, or 100 μM.

To calculate the total methane production, we assumed it to be in equilibrium between the gas and aqueous phases. In serum bottles with a headspace-to-liquid ratio of 30 ml/30 ml, the percentage of methane in the gas phase (f_g) is 96.87% at 30°C based on equation 1 (45), using a value of 0.0013 mol liter⁻¹ atm⁻¹ for Henry's law constant, K_H (35): where V_g and V_w are the volumes of the gas and aqueous phases, respectively, and R is the gas constant. The amount of methane produced was calculated by dividing the quantity of methane in the headspace by f_g.

Percent repression of methane production by phosphate was calculated as follows: where n = 4, 10, 50, or 100 μM.

Methane in the culture headspace was measured by gas chromatography coupled with a flame ionization detector (GC-FID), Agilent 7890A (G3440A). Gas was sampled from the culture headspace (25 μl for the LM-Y initial growth experiment; 100 μl for all other experiments) using a gas-tight syringe and injected into the GC-FID at an inlet temperature of 250°C. The carrier gas was helium, at a rate of 25 ml min⁻¹. The oven temperature was 90°C for 4 min, and then the temperature was increased by 120°C min⁻¹ for 1.25 min, until a final temperature of 240°C was reached. The temperature was maintained at 240°C for 1.5 min to remove any sample residue in the column. The detector temperature was 200°C, with hydrogen and airflows of 35 and 350 ml min⁻¹, respectively. MSDChem was used to integrate the peaks. Methane standards with 99 ppm methane in nitrogen gas (item no. 01-04-212; Scotty) and >99.9% methane (item no. GMT10015TC; Matheson) were used to prepare a standard curve of peak area and concentration (49.5 ppm, 99 ppm, 999 ppm, and 1,996.01 ppm). The methane concentration obtained based on the standard curves was then converted from parts per million to micromolar using the molar gas volume 24.87 liters mol⁻¹ at 30°C for ideal gas at 1 atm of pressure.

Initially, the primers PhnJoc1 and PhnJoc2 (Table 2) were used to amplify the phnJ genes from isolates (46, 47). Each PCR volume (25 μl) included 1× Mg-free buffer (item number D4545; Sigma-Aldrich), 1.5 mM MgCl₂, 0.25 μM each primer, 0.2 mM each deoxynucleoside triphosphate (dNTP), 0.75 U SigmaTaq (Sigma-Aldrich item number D1806), and 0.5 μl (7 to 30 ng) template DNA. PCR conditions were 94°C for 3 min, followed by 35 cycles of 94°C for 30 s, 52.5°C for 30 s, and 72°C for 1 min, with a final extension at 72°C for 5 min.

PCR products (∼400 bp) were extracted from agarose gels and sequenced by Sanger sequencing at the University of Delaware Sequencing and Genotyping Center. Primers for RT-PCR were designed based on these sequences (phnJ-F1/phnJ-R1 for strains LM-1, LM-5, and LM6-1 and phnJ-F2/phnJ-R2 for strain LM-Y; see Table 2 for sequences).

Liquid subsamples (1.5 or 3.0 ml) were removed from all cultures during the mid-exponential and stationary phases of growth on each P source (MPn, K₂HPO₄, or both) and stored at −20°C in RNALater until processing. For RNA extraction, cells were stabilized in RNA Protect Reagent (catalog number 76506; Qiagen) and then digested with lysozyme (15 mg ml⁻¹) and 20 μl proteinase K (800 U ml⁻¹, catalog no. P8107S; New England BioLabs) at room temperature for 15 min. RNA was then purified from the bacterial lysate using the Qiagen RNEasy minikit (catalog no. 74104; Qiagen) according to the manufacturer's instructions. Genomic DNA contamination was removed by Turbo DNase treatment (catalog no. AM1907; Ambion). RNA was used as the template for reverse transcription using the RETROscript kit (catalog no. AM1710; Ambion). RNA (∼1 μg) and random decamers (2 μl, 50 μM) were denatured at 75°C for 3 min and then immediately placed on ice. RT buffer (2 μl, 10×), dNTP mix (4 μl, 2.5 mM each), RNase inhibitor (1 μl, 10 U μl⁻¹), and Moloney murine leukemia virus reverse transcriptase (MMLV-RT; 1 μl, 100 U μl⁻¹) were added to the mixture to a volume of 20 μl. This solution was incubated at 44°C for 1 h and then at 92°C for 10 min to inactivate the reverse transcriptase (MMLV-RT). The single-stranded cDNA product was used as the template for quantitative PCR (qPCR).

The abundance of the phnJ transcript was quantified in the cDNA products using qPCR. The qPCRs were performed on an Eppendorf real-time Mastercyclerep and processed in duplicates with 0.5 μl cDNA, 10 μl SYBR green FastMix for iQ (catalog no. 95071; Quanta), 0.25 μl each primer (25 mM), and 9 μl nuclease-free water. The reaction protocol was as follows: 95°C for 2 min 30 s, followed by 40 cycles of 95°C for 15 s, 62°C for 15 s, and 68°C for 30 s.

The metagenomic data obtained in this study have been deposited in the NCBI database with accession number SAMN03292416.

Heterotrophic bacteria isolated from Lake Matano were recently shown to utilize a diverse array of P sources, including 2-AEP (Table 1) and to make large changes in the RNA content and lipid composition of the cells when P starved (38). To assess the probability of methane production from MPn degradation in Lake Matano surface water, reads mapping to genes encoding the C-P lyase pathway (Fig. 1A) were quantified. All of the genes necessary for cleavage of MPn to methane and phosphate are present in the metagenomic data set from the surface water of Lake Matano (Fig. 1C). This pathway includes subunits of the phosphonate transporter (phnDEC), activation of the phosphonate by ATP (phnIGHL), release of diphosphate (phnM), and cleavage of the C-P bond, which results in release of methane (phnJ) (48).

The odds ratios compare the frequency of each protein in the metagenomic data set to its frequency in the COG database (42). An odds ratio greater than 1 indicates that a predicted protein is present in the metagenomic data set at a higher frequency than expected based on its frequency in protein databases (42). Below the chemocline in Lake Matano, the phosphate concentration increases to ∼10 μM, making the microbes at this depth less P starved than those in the surface water (37, 49). Because the odds ratios for nearly all of the genes required for MPn transport and degradation are less than 1 in the metagenomic data from the P-replete chemocline (Fig. 1B) but greater than 1 in the P-limited surface water (Fig. 1C), we hypothesized that MPn could serve as a P source for some organisms in Lake Matano, which would release methane as a by-product.

The Newbler GS De Novo Assembler was used to assemble the metagenomic reads from the Lake Matano surface water, and 530,076 reads (66.39% of the total number of reads) were assigned to 18,727 contigs. Amino acid sequences of C-P lyase proteins from E. coli (50) were used to query the contigs, and Contig00268 and contig00117 were chosen for further annotation. Contig00268 has 5,357 nucleotides and carries a partial phnI sequence, phnJKLM, a predicted inositol monophosphatase (IMPase), and a partial gene encoding the ATP-binding motif of an ABC transporter. Contig00117 has 7,056 nucleotides and carries the phosphonate transporter-encoding phnDCE, along with an alkaline phosphatase gene and a partial penicillin-binding protein-encoding gene (Fig. 2).

Based on blastn analysis, Contig00268 has more than 83% coverage and 65% identity to gene clusters from three actinobacterial strains: Rhodococcus opacus strain PD630, R. opacus strain R7, and Mycobacterium simiae strain MO323. The three actinobacterial strains lack the accessory phnNP and the regulatory gene phnO that are part of the C-P lyase gene cluster in E. coli (50). The three actinobacterial strains all carry an inositol monophosphatase (IMPase)-encoding gene downstream of the C-P lyase gene clusters, and M. simiae encodes a penicillin-binding protein and an inositol 1-phosphate synthase upstream of the C-P lysase, while the R. opacus strains encode it downstream of the C-P lyase. In the two R. opacus strains, a predicted ABC transporter lies immediately downstream of the C-P lyase cluster (not shown). The best hit of blastn analysis for Contig00117 was also most similar to an Actinobacterium, Isoptericola dokdonensis DS-3, with 32% coverage and 71% identity. The predicted PhnE amino acid sequences in Contig00117, R. opacus, and M. simiae appear to have two PhnE domains (TIGR01097), in contrast to the E. coli PhnE, which has only one (Fig. 2). Several glyphosate- and phosphite-utilizing microbes have phn gene clusters with two copies of phnE (51, 52). In the gene cluster on Contig00117, these appear to be fused: no stop codon between the two phnE genes could be found, so we predict that they encode a single polypeptide.

The product of phnJ is responsible for cleavage of the C-P bond in phosphonate compounds and thus for release of methane from MPn (53), so the distribution of this gene in freshwater metagenomic data sets was assessed. Metagenomic data derived from freshwater samples and archived at the Integrated Microbial Genomes resource (https://img.jgi.doe.gov/) or the European Nucleotide Archive (http://www.ebi.ac.uk/ena) were queried for the presence of phnJ. Eighteen of the 23 data sets (78%) had homologs of phnJ (Table 3), suggesting that phosphonate bond cleavage may occur in a variety of freshwater environments, including lakes, lake sediments, bogs, rivers, and streams.

Methanogenesis has been observed in oxic surface waters of lakes (4–6, 8, 10). However, it has not been directly observed in Lake Matano surface waters, so we assessed the metagenomic data set from 10 m to determine whether surface water methanogenesis seemed likely. No reads mapping to archaeal rRNA sequences (5S, 16S, or 23S) are present in this data set. Ninety-five reads (of a total of 798,463 reads) mapped to archaeal genomes. Most of these (75 of 95 reads) mapped to the class Halobacteria, the members of which are not methanogenic. The remaining 20 reads were initially mapped to the methanogen genera Methanococcus and Methanomicrobium; however, when these reads were used as blastn queries against the NBCI nonredundant database, only 4 were identified as similar to sequences from methanogen genomes (Methanoculleus and Methanococcus). After the metagenomic reads were used as queries in a blast search against the COG database, the results were screened for reads mapping to methanogenesis pathways, but none were identified. Additionally, the amino acid sequence of McrA from Methanosarcina barkeri DSM804 (RefSeq: WP_011305916.1) was used as a query against the metagenomic data set, and no reads with homology to McrA were identified. We thus conclude that if water column methanogenesis occurs in the epilimnion of Lake Matano, it is either localized deeper than 10 m or carried out by a very small number of organisms not detected in our metagenomic survey.

Seven isolates were screened for growth on MPn. Four were capable of growth on MPn as the sole P source and were selected for further analysis (Table 1). All four strains grow at the same rates on MPn, K₂HPO₄, or both and reach stationary phase at approximately the same time (Fig. 3, dashed lines). These isolates do not appear to be capable of consuming the methane released as a carbon or energy source, since they do not grow when MPn is provided as the sole P, C, and energy source (data not shown).

Methane was produced only in the presence of MPn (Fig. 3, solid lines). In the presence of both MPn and 200 μM K₂HPO₄, strain LM-Y did not produce any methane (Fig. 3D). However, the other isolates produced some methane in the presence of K₂HPO₄ when MPn was also present (Fig. 3A to C). No methane was produced by any strain when MPn was not provided, indicating that the methane is a product of MPn metabolism.

To further investigate the relationship between MPn concentration and methane production, strain LM-Y was grown in NBRIP with different concentrations of MPn (Fig. 4). This strain has been shown to decrease RNA content and replace its phospholipids with amino- and glycolipids when P is limited, so the cells may increase their P content without necessarily altering the growth yield (38). As the MPn concentration increases, the cell yield and total methane production also increase, suggesting that availability of additional P promotes additional growth, as expected (Fig. 4). Both yield and methane production are similar when 100 or 200 μM MPn is added, suggesting that at concentrations of >100 μM MPn, sufficient P is available to maximize growth yield in NBRIP and the excess P_i in solution may even inhibit further degradation of MPn (54).

Addition of 200 μM phosphate to the cultures with initial MPn concentrations of 200 μM reduced the amount of methane produced by strains LM-1, LM-5, and LM6-1 by 40 to 60% but completely repressed methane production in the culture of LM-Y. To better understand the effect of phosphate on phosphonate consumption by this strain, LM-Y was grown with 50 μM MPn and 0 to 100 μM KH₂PO₄ (Fig. 5). Methane production was completely inhibited in cultures grown with 50 μM MPn and 50 or 100 μM KH₂PO₄ (Fig. 5A). In cultures grown with 50 μM MPn and 0 μM, 4 μM, or 10 μM P_i, the amount of methane produced decreased as KH₂PO₄ increased (Fig. 5A). The relationship between added KH₂PO₄ concentration and percent repression of methane production is linear (Fig. 5B), and we extrapolate that methane production would be completely repressed at P_i concentrations above 30 μM.

The phnJ gene encodes the enzyme that catalyzes release of methane from the intermediate α-d-ribose-1-methylphosphonate-5-phosphate (53). We examined the expression of this gene in all four bacterial strains grown with 0.2 mM phosphate, 0.2 mM MPn, or both in the medium. LM-Y, the only strain in which methane production from MPn is abolished in the presence of phosphate (Fig. 3), appears to express phnJ only when MPn is present and phosphate is absent (Fig. 6). The other three strains, which produce methane when MPn is provided, whether or not it is the only P source (Fig. 6), express phnJ under both conditions. However, the phnJ transcript is less abundant in the presence of phosphate (Fig. 6 and Table 4).

Here, we show that freshwater heterotrophic bacteria are capable of producing methane from MPn and that the pathway for phosphonate degradation is widespread in freshwater lakes. We further demonstrate that in Lake Matano, Alphaproteobacteria, Gammaproteobacteria, and likely Actinobacteria can take up phosphonate compounds and cleave the C-P bond to acquire P and that this activity is repressed in the presence of P_i. Similarly, the phnJ gene is expressed in these strains only in the presence of MPn, and its expression level is modulated by addition of P_i. In sum, this work shows that methane production in freshwater systems may occur as a result of phosphate limitation, as microbes acquire P from phosphonates.

Phosphonates, including MPn, may comprise up to 10% of dissolved organic phosphorus (DOP) in some lakes (55–57), though many freshwaters have no detectable phosphonates (58, 59) or only very small quantities thereof (60, 61). However, homologs of phnJ are present in 18 of 23 metagenomic data sets from freshwater environments (water and sediment), including bogs, lakes, ponds, rivers, and streams (Table 3). The number of data sets in which phnJ is found and the diversity of environments indicate that the ability to degrade phosphonates is widespread in freshwater environments. Additionally, based on the abundance of phnD, which encodes one subunit of the phosphonate transporter, picocyanobacteria in the Great Lakes are predicted to be capable of phosphonate uptake (62). Given the broad distribution of phosphonate uptake and degradation pathways, phosphonates may be undetectable in freshwater systems not because they are not present but because they are rapidly broken down. In fact, phosphonates in marine systems have been shown to be highly reactive (63), and the same may be true in freshwater environments, which are typically P limited (64). This P limitation may result in microbial acquisition of P from a wide variety of sources (38), ultimately leading to release of organic by-products such as methane.

The steady-state concentration of methane at 40 m in Lake Matano is ∼3 μM, 3 orders of magnitude greater than the saturation concentration (7). Many freshwater lakes are supersaturated with regard to methane, often more than can be explained by methanogenesis in anoxic sediments or bottom waters (5, 6, 10). In some of these systems, surface water methane is produced by methanogenesis in oxic surface waters (5, 10). However, in severely P-limited systems, demethylation of P-containing compounds may also contribute to surface water methane production. Because Lake Matano has less than 50 nM total P_i in the surface water (37), planktonic organisms there must be capable of immediate uptake and utilization of P in any available form (38). Both physiological experiments with isolates and bioinformatics analysis of metagenomic data indicate that heterotrophic bacteria in Lake Matano surface water are capable of producing methane as a by-product of acquisition of P from phosphonate compounds. Given the prevalence of genes encoding phosphonate biosynthesis and degradation in freshwater metagenomic data sets, P metabolism in freshwater systems may have unexpectedly large effects not only on P cycling but also on release of greenhouse gases. The current models for methane cycling in fresh waters use lake size, shape, nutrient status, temperature, and primary production to predict rates of methane production (19, 21, 65) but do not include methane production by any pathway in oxic water columns, even though this phenomenon has been observed in many lakes (4–6, 8, 10) and is known to be important in the ocean (12, 13). Because lakes may release as much as ∼100 Tg methane globally each year, or ∼20% of total annual natural methane emissions, which is more than the emissions from the world's oceans (1, 2), understanding the environmental and biological factors contributing to methane emissions from freshwater systems is critical to making more-accurate predictions of both freshwater and global methane emissions (9, 18–21).