Introduction

The role of microorganisms in organic carbon cycling is of considerable interest, as these activities can impact carbon turnover rates and sequestration (Karl et al., 2012) and influence the fate of contaminants, such as petroleum (Hazen et al., 2010), uranium (Anderson et al., 2003) or arsenic (Islam et al., 2004). Sediments host a substantial proportion of the environmental bacterial and archaeal biomass, with an estimated 6–40% of the prokaryotic biomass inhabiting the terrestrial subsurface, and an estimated total carbon content rivaling that of the global plant biomass (Whitman et al., 1998). Most endogenous organic matter in terrestrial sediment is considered to be relatively refractory, consisting largely of polymers, such as lignin and cellulose, and plant-derived (or even petroleum-based) hydrocarbons (Hartog et al., 2004; Rowland et al., 2006). However, some fractions of complex organic matter can be degraded by fermentative or respiratory microorganisms (Benner et al., 1984; Leschine, 1995; Widdel and Rabus, 2001).

In anaerobic environments, such as the terrestrial subsurface, acetate is an important product of central carbohydrate degradation pathways, following the oxidation of pyruvate. It is also a common product of respiration or fermentation of other organic acids and alcohols by anaerobic microorganisms, including sulfate-reducing bacteria (SRB) that incompletely oxidize organic matter (Gibson, 1990). Characteristic pathways identified through which acetate oxidation to CO2 proceeds are the TCA cycle and the acetyl-CoA pathway (Thauer et al., 1989). Both pathways are reversible and can be used by autotrophic bacteria to catalyze the reduction of CO2 to back to acetyl-CoA (Fuchs, 1986). Acetate can be used in methanogenesis, and numerous bacteria can couple acetate oxidation to the reduction of inorganic compounds, such as Fe(III), U(VI), nitrate, S0 and sulfate (Thauer et al., 1989; Lovley et al., 2004). As a biologically relevant and non-fermentable organic compound, acetate is an attractive substrate to use in studying biogeochemical cycling linked to anaerobic respiration and bioremediation (Anderson et al., 2003).

While relatively few organisms proliferate when stimulating sediment microbial communities with acetate (Holmes et al., 2002), the enriched organisms constitute a community that is nonetheless intricate (Handley et al., 2012). Community processes can be expected to include syntrophic interactions, biomass recycling, and the precipitation of diverse and in some cases, cryptic biogeochemical cycles. Unraveling the individual roles of community members, and their metabolic pathways is non-trivial.

Here, we used genomics-informed proteomic information (proteogenomics, Ram et al., 2005) to resolve organism-specific activity and detect metabolic pathways utilized in an acetate-amended, sediment-hosted subsurface microbial community, a priori. The study provides a snapshot of community-wide functioning and interactions in an aquifer setting during acetate-induced uranium bioremediation under predominantly sulfate-reducing conditions, and builds on previous studies considering the role of bacteria, in particular Geobacter, in Fe(III) and U(VI) reduction within the aquifer (for example, Anderson et al., 2003; Holmes et al., 2009; Wilkins et al., 2009; Williams et al., 2011). Our proteogenomic data demonstrate the activity of bacteria linked to metal reduction, in addition to microbial utilization of the carbon, nitrogen and sulfur, and the indirect stimulation of autotrophic (or mixotrophic) bacteria in response to CO2 and sulfide generated through acetate-dependent sulfate reduction. Biogeochemical reaction modeling was also used to evaluate denitrification pathways.

Materials and methods

Experiment setup and sampling

Un-amended and acetate-amended sediments were sampled from an alluvial freshwater aquifer underlying the Department of Energy’s Integrated Field Research Challenge (IFRC) site at Rifle, CO, USA. In order to stimulate aquifer sediment with acetate in situ, and access sediment from the subsurface post-stimulation, we incubated sediment in a flow-through column in an existing groundwater well (P104; see well gallery in Williams et al., 2011). Un-amended sediment was first excavated from the aquifer using a backhoe, and sieved (to remove rocks) to a final particle size of <2 mm. This material was packed into a clear custom built PVC cylindrical column (5.1 cm wide × 10.2 cm long) and incubated ∼5 m below ground surface within the well (Supplementary Figure S1). The column equilibrated with subsurface conditions for 15 days before amendment, which have previously been shown to be anoxic with <16 μM of dissolved oxygen (Williams et al., 2011). Acetate (electron donor and carbon source) and bromide (conservative tracer) were injected into wells ∼0.5 m upgradient of well P104, obtaining final concentrations of ∼15 and ∼1.3 mM, respectively (method in Williams et al., 2011). Amended groundwater was pumped up through the column for 24 days, as described previously (Handley et al. 2012), during which this region of the aquifer was subject to its third consecutive summer of acetate amendment. Sediment from the entire column was homogenized and then flash frozen upon collection. Comparative analyses with a replicate column and other aquifer samples are described elsewhere (Handley et al., 2012).

Geochemistry analyses

Groundwater samples were filtered using 0.25 μm PTFE filters for geochemical analyses. Acetate, bromide, sulfate, chloride, uranium, Fe(II) and sulfide were measured, as described previously (Williams et al. 2011).

DNA extraction

Genomic DNA was extracted from 5 g of un-amended sediment and 300 g of acetate-amended sediment (5–10 g per tube) using PowerMax Soil DNA Isolation Kits (MoBio Laboratories, Inc., Carlsbad, CA, USA) with the following modification to the manufacturer’s instructions. Sediment was vortexed at maximum speed for an additional 2 min in SDS, and then incubated for 30 min at 60 °C instead of bead beating. Acetate-amended sediment extraction replicates were pooled in order to obtain ∼8 μg of DNA for downstream analysis. All eluted DNA was concentrated, as described by Handley et al. (2012).

16S rRNA gene clone libraries

DNA was amplified using the general bacterial 16S rRNA gene primers 27f and 1492r (Lane, 1991), and a temperature gradient to minimize PCR bias, which comprised 11 PCR reactions at 11 different annealing temperatures. The PCR protocol was: 1 min at 94 °C; 25 cycles of 1 min at 94 °C, 30 s at 48–58 °C (11 temperature gradient) and 1 min at 72 °C; and 7 min at 72 °C. Amplicons were pooled, and precipitated as above. Clone libraries were constructed using the TOPO TA Cloning Kit (Invitrogen, Carlsbad, CA, USA) with electrocompetent cells. 16S rRNA genes from transformed colonies were PCR amplified using the protocol: 10 min at 95 °C; 25 cycles of 30 s at 95 °C, 30 s at 53 °C and 1.5 min at 72 °C; and 7 min at 72 °C. Inserts were screened for correct size (∼1400 bp) by gel electrophoresis, and sequenced using capillary electrophoresis (Applied Biosystems 3730xl DNA Analyzer, Foster City, CA, USA), with M13f (−21) and M13r (−24) primers (Invitrogen). Sequences were trimmed to remove Phred quality scores ⩽20, and forward and reverse strands were merged into near full-length sequences using Phrap (http://www.phrap.org/phredphrapconsed.html). USEARCH (Edgar, 2010) was used to check for chimeras—after which 98 sequences were retained for each sample (un-amended and acetate-amended)—and to cluster sequences into OTUs 97% similar. A representative from each OTU was BLASTed (Altschul et al., 1990) to a non-redundant version of SILVA SSURef102 (http://www.arb-silva.de/).

Metagenomic 16S rRNA gene sequence reconstruction

16S rRNA gene fragments from Illumina metagenomic sequencing (described below) were reconstructed into near full-length genes using EMIRGE (Miller et al., 2011) with 120 iterations and the non-redundant SILVA SSURef102 as the starting database. In the final iteration, 12 032 paired reads (0.05%) were reconstructed into 16S rRNA gene sequences. Sequences were clustered into OTUs⩾97% similar. To exclude less reliable rare sequences, OTUs with raw relative abundances ⩾0.5% were used. Representative sequences were BLASTed to the SILVA database. OTU abundances, calculated on the basis of a probabilistic accounting of read depth, were normalized by sequence lengths.

Phylogenetic analyses

For 16S rRNA phylogenetic analysis genes were aligned to reference sequences from GenBank with Clustal W (Thompson et al., 1994), and Maximum Likelihood trees were created (see Handley et al., 2012). Bootstrap consensus trees were inferred from 1000 replicates. Branches corresponding to partitions reproduced in <50% bootstrap replicates were collapsed. Trees were annotated with data sets in iTOL (Letunic and Bork, 2006).

Metagenome sequencing and assembly

Genomic DNA from the acetate-amended sediment was sequenced on one flow-cell of an Illumina Genome Analyzer IIx (Illumina, Inc., San Diego, CA, USA). A paired-end (PE) shotgun library with 1-kb insert size was prepared using Illumina’s Genomic DNA Sample Prep kit, according to the manufacturer’s instructions. Sequencing produced ∼7 Gb of sequence (versus ∼400 Mb with 454), and 29 million paired-end reads ∼125-bp long. See Supplementary Information for further details, and a description of 454 sequencing and assembly. Illumina reads were trimmed to remove low-quality bases from the 3′ ends, after which 87% of paired/single reads >60-bp long were retained. Reads were initially assembled eight times using Velvet 1.1 (Zerbino and Birney, 2008) with different parameters optimized on the basis of expected genome coverage, after which a reference-guided Velvet-Columbus re-assembly was undertaken using paired-end scaffolds from the most abundant organism (r9c1, see binning section below and Supplementary Information for details).

Genome annotation

Genes were predicted, and translated into protein sequences using Prodigal (Hyatt et al., 2010), and annotated using the pipeline, as described by Yelton et al. (2011).

Genomic binning of metagenomic assemblies

Emergent self-organizing maps of tetranucleotide frequencies and read coverage data were used to define bins (Dick et al., 2009). Sequence fragments 5-kb long were used to create the map and separate sequences into bins, after which 2-kb fragments were projected onto the map. Emergent self-organizing maps (ESOM) were created using Databionic ESOM Tools (Ultsch and Mörchen, 2005). PE-scaffolds >2-kb long were also evaluated on the basis of read coverage (that is, relative abundance), gene GC-content and BLASTP best matches to the Uniref90 database (Suzek et al., 2007) to help resolve unclear ESOM bin boundaries and classify low-abundance organisms. EMIRGE-reconstructed 16S rRNA sequences were assigned to genomic bins on the basis of the nearest BLAST-determined database match, read coverage and sequence abundance.

Genome completeness

Genome completeness was determined on the basis of 35 single-copy orthologous groups (OGs) (Raes et al., 2007). OGs were obtained from eggNOG v. 3.0, a database of OGs in 1133 taxonomically diverse organisms, including 943 Bacteria (Powell et al., 2012). Sequences were considered orthologous on the basis of reciprocal BLASTP analysis, if they had a minimum bit score of 60, an alignment length of ⩾70%, and ⩾30% shared identity.

Whole-genome comparisons

Relationships of assembled genomes to closely related genomes were measured using the amino-acid percent identity averaged across putative orthologs, determined using reciprocal BLASTP with the same criteria as for genome completeness estimates.

Proteomics

Proteins were extracted from ∼10 g of un-amended and acetate-amended sediment via a heat-assisted SDS-based method, followed by TCA protein precipitation/acetone washes, trypsin proteolysis, desalting and solvent exchange (Chourey et al., 2010). Digested peptides were loaded in triplicate onto 5 cm strong cation-exchange columns, and connected to a reverse-phase (C18) front column (Phenomenex, Torrance, CA, USA) with an integrated nanospray tip (New Objective, Inc., Woburn, MA, USA). Proteomes were analyzed via two-dimensional 24-hour separations with CH3COONH4 salt pulses followed by reverse-phase gradients (Dionex U3000 HPLC, Sunnyvale, CA, USA) with online electrospray tandem mass spectrometry (2D-LC-MS/MS, LTQ Velos Orbitrap, Thermo Scientific, San Jose, CA, USA). Data-dependent MS/MS spectra were acquired, with full scans (400–1700 m/z) at 30-K resolution (top-ten method; dynamic exclusion: one; see VerBerkmoes et al., 2009). MS/MS spectra were queried with SEQUEST (Eng et al., 1994), against a predicted protein database constructed from the Velvet-Columbus metagenome assembly, along with common contaminants. Identified peptide sequences were reassembled into proteins and filtered with DTASelect (Tabb et al., 2002) using the following parameters: Xcorr values >1.8 (+1), 2.5 (+2) and 3.5 (+3) and DeltCN values >0.08, and a requirement for ⩾2 peptides per locus. The DTASelect and metagenomic database files are available at https://compbio.ornl.gov/ersp_rifle/column_sediment_2009. Summed triplicate technical replicates and normalized spectral abundance factors (NSAFs, Florens et al., 2006) were used for subsequent analyses.

Modeling

The Supplementary Information describes a combined thermodynamic–kinetic modeling approach for microbially mediated rate laws on the basis of work of Jin and Bethke (2005) and Dale et al (2008).

Simulations of microbially mediated biogeochemical reactions (see below) were on the basis of physical and acetate-delivery parameters for well P104 and site geochemical data (Figure 1). The system was treated as one dimensional, with average flow and acetate delivery rates estimated from the arrival time of the bromide tracer (3.37 cm per day). Nitrate is present in micromolar concentrations within the aquifer near well P104 (Williams et al., 2011), and varies 2 orders of magnitude across the site (unpublished data). For modeling, we used upper and lower values of 72 μM and 5 μM nitrate, with the latter value being more representative of the aquifer local to P104.

Figure 1
figure 1

Plot showing concentrations of (a) acetate and bromide, (b) sulfate, (c) Fe(II) and sulfide, and (d) U(VI) in the column effluent (C, open symbols) and well bore (W, closed symbols) throughout the experiment. The column represents an extension of the amended aquifer, and no clear difference between C and W values is evident.

Full size image

The rate of denitrification is affected by the maximum rate per unit cell or mass of biomass, νmax, and the biomass concentration, Bmass. Lacking cell-count information for the sulfur-oxidizing bacteria, we assumed a rate constant that implicitly accounts for the biomass and results in substantial denitrification over the 0.5-m flow path. We assumed the autotrophic (sulfide-dependent) denitrification reaction rate to be five times slower than heterotrophic denitrification, following Koenig and Liu (2001 and references therein) and Cardoso et al. (2006) who showed S0-dependent autotrophic denitrification is 10 times slower than heterotrophic denitrification, and the rate of denitrification with sulfide is twice as fast as with S0. Comparisons were also made using an equal rate for autotrophic and heterotrophic pathways.

A half-saturation constant (that is, substrate concentration, where the specific growth rate equals half its maximum rate) of 2.1 μM NO3− (0.03 mg l−1 NO3−-N, determined for the reaction S0→sulfate; Batchelor and Lawrence, 1978) was used for the autotrophic denitrification pathway. This is close to a value, 3 μM, determined by Claus and Kutzner (1985), and to the values reported for aerobic sulfide oxidation (Sorokin et al., 2003). A range of half-saturation constants, 11.4 –142 μM NO3− (0.16–2 mg l−1 NO3−-N), were used for the heterotrophic pathway-based values from the literature determined with either sludge or methanol as an organic carbon source (Engberg and Schroeder, 1975; Æsøy and Ødegaard, 1994; Henze et al. 2000; Klas et al., 2006; Sarioglu et al., 2009).

The catabolic reactions considered in the modeling are:

with log equilibrium constants of 3.8305, 18.2813, 13.5785 and 19.1559, respectively.

Accession numbers

Reconstructed and cloned 16S rRNA gene sequences have been deposited in GenBank under the accession numbers JX125436–JX125454 and JX120370–JX120502, respectively. Metagenomic data are accessible through DDBJ/EMBL/GenBank (SRX149542 and AMQJ00000000 (version one: AMQJ01000000), BioProject PRJNA167727), and at http://banfieldlab.berkeley.edu/2009RifleSedimentMetagenome/.

Results and Discussion

Community composition and biogeochemical response to amendment

Previous studies have documented a period of Fe(III) reduction, followed by sulfate reduction in the Rifle aquifer during acetate amendment experiments, and concomitant reduction of U(VI) (Anderson et al., 2003; Williams et al., 2011). Periods marked by Fe(III) and sulfate reduction were accompanied by enrichments of Desulfuromonadales, and then Peptococcaceae followed by Desulfobacterales (Anderson et al., 2003; Handley et al., 2012). A legacy effect in the stimulated portions of the aquifer was observed to cause earlier onset of sulfate reduction upon subsequent re-stimulation, reducing the time from >1 month to <1 week over three subsequent experiments (Callister et al., 2010; Williams et al., 2011; Druhan et al., 2012). In this study (a third-time stimulation experiment), the onset of sulfate reduction occurred within 2 days of acetate being detected in the incubation well (Figure 1a). Sediment was recovered after at least 18 days of sulfate reduction, during which time acetate remained in excess, and U(VI)(aq) and sulfate concentrations were halved (Figure 1b and d). Sulfide accumulation corresponded to the depletion of aqueous sulfate and Fe(II) (Figure 1c). The color of sediment throughout the column transformed from brown to black owing to the precipitation of FeS.

Clone and metagenome-derived (that is, EMIRGE-reconstructed) 16S rRNA gene data revealed a less even, lower diversity community after amendment (Figure 2a), consistent with earlier deeply-sampled 16S rRNA microarray analyses that measured ∼40% lower OTU richness in the amended versus un-amended sediment (Handley et al., 2012). Amendment enriched primarily for Delta- and Epsilon-proteobacteria, in particular a deltaproteobacterium closely related to the acetate-oxidizing SRB Desulfobacter postgatei (99% 16S rRNA gene sequence identity; Widdel and Pfennig, 1981). Other less-enriched bacteria are related to Sulfurovum and Sulfurimonas, Geobacter and Desulfuromonas and Bacteroidetes (Figure 2b). In general, three methods used for determining community composition and/or abundance (clone library, EMIRGE, read coverage) agreed (Supplementary Figures S2 and S3). The metagenome comprises a sample of highly abundant microorganisms; no archaeal 16S rRNA genes were detected.

Figure 2
figure 2

Community composition. (a) Relative abundances of Classes in the un-amended (UA) and acetate-amended (AA) communities determined using 16S rRNA gene clone libraries (UA and AA) or EMIRGE-reconstructed 16S rRNA genes from the metagenome (AA). (b) Bootstrap consensus tree of EMIRGE-reconstructed 16S rRNA gene sequences (AA). The tree is rooted to Methanococcus vannielii SB (CP000742.1). The scale bar (LHS) indicates the number of nucleotide substitutions per site. Bootstrap values are shown as percentages. A similar topology was obtained with the initial ML tree. Black bars indicate the OTU relative abundances in the metagenome. Proteobacteria classes are abbreviated.

Full size image

Although previously un-amended sediment was used in the column, the community composition in both the column sediment, and in a split-phase sediment–quartz column incubated for the same period in well P104, differed substantially from compositions observed in first-time stimulation experiments after a similar period of amendment (Anderson et al., 2003; Handley et al., 2012). Less difference was observed between the column communities and those from second-time experiments, including post-amendment sediment collected from P104 in the previous year (Handley et al., 2012). This tends to suggest the sediment (and quartz) were influenced by the local, twice previously stimulated, aquifer community.

Assembled genomes

After assembling Illumina reads (Table 1), we obtained genomes from seven different phylogenetic groups (Figure 3), and allocated 99% of sequence length >2-kb long to a genomic bin (Figure 4a). Phylogenetic analyses indicate that the binned genomes belong to: Deltaproteobacteria (r9c1, r9c7-r9c8), Epsilonproteobacteria (r9c2-r9c3), Bacteroidetes (r9c4-r9c5) and Firmicutes (r9c6) (Table 2). All the genomes, except for the low-coverage Firmicutes genome, are represented by EMIRGE-reconstructed 16S rRNA sequences. Clone library data indicate the most abundant Firmicutes in the amended sediment were Clostridiales bacteria related to Gracilibacteraceae or Peptococcaceae (data not shown). However, greater similarity to a representative Clostridiales genome as opposed to Peptococcaceae is indicated by the average amino-acid identities of orthologs (Table 2).

Table 1 Summary statistics for the Illumina Velvet and Velvet-Columbus assemblies, and the 454 Newbler assembly
Full size table
Figure 3
figure 3

Genomic bins from the Illumina assembled data distinguished by (a) average read coverage per PE-scaffold (‘other’ includes bins r9c8-r9c12 and unk; ‘unb’ comprises unassigned fragments 500–2000 bp long), (b, c) ESOM (repeating tiled view—the black line demarks one representation of each genomic bin), and (d) GC content per gene. In the ESOM, several major genomic bins occur as distinct clusters demarked by topographic highs (beige lines, c). Two minor fragmentary genomic bins (r9c8 and r9c12) were not substantial or distinct enough from clusters 1 and 7 to form separate clusters in the ESOM, but both group with the Deltaproteobacteria. The three small ESOM clusters, r9c9–r9c11, are putative plasmid sequences, characterized by genes associated with transmembrane regions, DNA polymerase, phage, coiled-coil proteins, and a large number of hypothetical genes. ‘Unk’ are phylogenetically unassigned fragments >2 kb long. (c) ESOM without data points. Dark green denotes areas of high similarity. Epsilonprot, Epsilonproteobacteria; Deltaprot., Deltaproteobacteria.

Full size image
Figure 4
figure 4

Plots showing (a) the length of sequence in Illumina- and 454-based genome bins; and (b) the estimated number of genomes per bin based on BLASTP searches for conserved OGs (bin lengths divided by estimated genome lengths). The sum of OGs detected in each bin was divided by the number of OGs queried (i.e. 35 via BLAST). ‘Other’ comprises bins r9c9-r9c12. ‘Unb’ comprises unassigned fragments 500–2000 bp long that contain predicted genes (‘unk’ are unassigned fragments >2 kb). Comparisons largely agree between estimates based on OGs and those determined based on the genome size of closely related organisms (R2=0.86; Supplementary Figure S4) even though genome sizes of related organisms can only be considered as an approximate guide.

Full size image
Table 2 Summary data of genomic bins on the final Illumina Velvet–Columbus assembly
Full size table

We estimate that the seven major genomic bins comprise ∼6 near-complete and three at least half-complete genomes (Figure 4b), of which, r9c1 and r9c2 each contain a single near-complete genome, with the latter being in two PE-scaffolds (1.9 and 0.1-Mb long). R9c3 and r9c5 each contain at least two near-complete genomes. The r9c7 bin contains partial-genome fragments closely related to Geobacter and Desulfuromonas. Similarity between these bacteria in terms of phylogeny, GC-content and abundance levels makes it difficult to resolve r9c7 genomes into separate bins.

Ortholog identities of >85% have been shown to correlate with DNA–DNA hybridization values of >70%, indicative of species level assignment (Goris et al., 2007). On the basis of very high (96%) ortholog average amino-acid identity shared between r9c1 and D. postgatei, we suggest that they belong to same species. Excluding the small collection of r9c8 genes, which are very closely related to Desulfomicrobium baculatum, the other genomes have identities too low to sequenced genomes to suggest species assignment (Table 2).

Proteogenomics

A total of 1695 proteins and 112 606 peptides (93% unique peptides) from the acetate-amended samples were identified by liquid chromatography–tandem mass spectrometry using the metagenome as a protein reference database (Supplementary Dataset). Little digestible protein was present, and few (∼31) proteins were identified in the un-amended sediment after searching peptides against the same database (Supplementary Information), and are not discussed further.

Examination of organism-resolved proteins enabled detection of six bacterial groups (r9c1-r9c5 and r9c7) active during the amendment experiment (Figure 5; Supplementary Information/Data set). All groups expressed proteins associated with acetate utilization or carbon fixation via the TCA cycle. Among other key enzymes detected were those used in ribosomes, cell division, glycolysis/gluconeogenesis, acetate/acetyl-CoA formation, electron transport, nitrogen fixation, denitrification, sulfur respiration, phage resistance, chemotaxis and motility and response to oxidative stress. The majority of proteins detected (75% of proteins) and protein expression levels (95% of peptide spectral counts) are attributable to the dominant r9c1 genome.

Figure 5
figure 5

Proteins expressed in key functional groups (Supplementary Dataset) are shown as (a) number of proteins identified per group, (b) averaged or (c) summed unique NSAFs, and (d) relative proportions of summed unique NSAFs with r9c1 and unb removed. Little difference was evident between non-unique (not shown) and unique NSAF charts, except the overall contribution of TCA cycle peptides was 0.5% less in the latter. No proteins within these groups were matched to the r9c6 Firmicutes genome. Abbreviations and definitions: (3) oxidative/reductive TCA cycle proteins (4) conversion between pyruvate and phosphoenolpyruvate (PEP); (5) conversion between acetate and acetyl-CoA; (6) NADH/NADPH dehydrogenase; (7) glycolysis/gluconeogenesis; (11) Cas proteins associated with Clustered Regularly Interspaced Short Palindromic Repeats, CRISPR; (12) cytochrome-c3 and cytochrome-c3 hydrogenase; (13) nitrogenase (NifABDEHKU); (14) putative nitrogenase accessory proteins RnfCDEG; (15) ammonium transport AmtB; (16) sulfate reduction proteins ATP-sulfurylase (Sat2), adenylyl sulfate (APS) reductase (AprAB), dissimilary sulfite reductase (DsrABCEFH); (17) putative accessory proteins heterodisulfide reductase-like menaquinol-oxidizing enzyme (HmeCDE) and quinone-interacting membrane-bound oxidoreductase (QmoABC); (18) sulfur oxidation proteins SoxCY and SQR; (19) NO3- (NapAB), NO2- (NirS), NO (NorC), N2O (NosZ) reductases (further details in Supplementary Text and Dataset).

Full size image

Activity and genomic potential of enriched bacteria

Acetate-oxidizing SRB

Amending the system with acetate created a niche ecosystem for Desulfobacter (r9c1), which proteomic data indicate actively respired sulfate coupled to acetate oxidation (Figures 5a–c). Desulfobacter species (almost) dedicatedly couple anaerobic acetate oxidation to the reduction of sulfate, and were originally considered marine bacteria, before the isolation of a freshwater sediment strain (Widdel and Pfennig, 1981; Widdel, 1987). Within the Rifle aquifer, acetate-stimulated enrichments of Desulfobacteraceae appear to supersede enrichments of sulfate-reducing Peptococcaceae (Anderson et al., 2003; Handley et al., 2012), although the reason for this is currently undetermined.

In this study, key proteins essential for the activation and reduction of sulfate were only detected for r9c1 (Figures 5 and 6a), including proteins potentially involved in the transfer of electrons from the membrane to APS and sulfite (QmoABC and HmeCDE; Mander et al., 2002; Pires et al., 2003; Mussman et al., 2005). Genes were identified for sat1 and hmeB, but were not represented in the proteome. In addition, almost all proteins necessary for gluconeogenesis were detected (Supplementary Information/Data set). Proteomic data (Figure 5) further indicate that r9c1 was actively dividing; chemotactic; motile by means of flagella and twitching pili; uptaking ammonium; and fixing nitrogen, possibly assisted by Rnf in electron transport (Schmehl et al., 1993). Nitrogenase and ammonium transporter amtB have been shown to be co-expressed by acetate-stimulated blooms of Geobacteraceae in the Rifle aquifer during peak Fe(III) reduction (Mouser et al., 2009). Expression of amtB transporter (and sensor) increases as ammonium concentrations approach zero (Javelle et al., 2004; Mouser et al., 2009) and nitrogen fixation becomes feasible (Helber et al., 1988; Holmes et al., 2004). While the high relative abundance of r9c1 was probably supported by nitrogen fixation, this seems unlikely to have conferred a substantial competitive advantage to r9c1 over other resident SRB (Handley et al., 2012)—many of which are associated with genera containing diazotrophs (Zehr et al., 2003).

Figure 6
figure 6

(a) Schematic of biogeochemical cycling and physiology inferred from genetic (white boxes) and proteomic (yellow boxes) data. (b) TCA cycle with enzymes necessary for the reductive cycle shown in green. The dashed line indicates the path with acetyl-CoA transferase (forming succinate and acetyl-CoA). (c) Summary biogeochemical redox reactions inferred from proteogenomic data. Colored circles correspond to the organisms in (a). rTCA, reductive TCA cycle; mTCA, modified TCA cycle; TMAIII, trimethylarsine gas; dehyd, dehydrogenase.

Full size image

We detected all but one r9c1 protein needed to oxidize acetate using a modified TCA cycle (Table 3). This cycle is used by Desulfobacter species instead of the acetyl-CoA pathway, which is used by other acetate-oxidizing SRB (Thauer et al., 1989). The modified cycle makes use of 2-oxoglutarate:ferredoxin oxidoreductase and a reversible citrate lyase, which generate reduced ferredoxin instead of NADH and an extra ATP, respectively (Brandis-Heep et al., 1983; Müller et al., 1987; Müller-Zinkhan and Thauer, 1988). Neither the expected membrane-bound malate:quinone oxidoreductase (EC:1.1.5.4; Brandis-Heep et al., 1983) nor the typical TCA cycle malate dehydrogenase (NADH-forming) was identified in the r9c1 genome by homology searches. While this could be an artifact of genome incompleteness, divergent homology or an unidentified analog, proteomics identified a pyruvate-forming malate dehydrogenase (NAD+/NADP+), which may potentially bypass this step, and regenerate oxaloacetate via pyruvate carboxylase or phosphoenolpyruvate (PEP) (cf. Hansen and Juni, 1979). Proteins for succinyl-CoA synthetase and for acetate activation by acetate kinase and phosphate acetyltransferase were identified by proteomics; however, identification of acetyl-CoA transferase suggests that r9c1 also used CoA transferase/hydrolase for both acetate activation and succinate formation (Figure 6b), as has been demonstrated for D. postgatei, Desulfuromonas acetoxidans and Geobacter species (Brandis-Heep et al., 1983; Gebhardt et al., 1985; Wilkins et al., 2009).

Table 3 TCA cycle and related components identified for genomic bins r9c1-5 and r9c7
Full size table

In contrast to r9c1, Desulfomicrobium (r9c8) was less enriched. Members of this sulfate-reducing genus are incomplete oxidizers and are only known to use acetate mixotrophically with H2 (Dias et al., 2008). Assuming that r9c8 was also reducing sulfate, or other inorganic compounds respired by Desulfomicrobium species (for example, thiosulfate and nitrate), then it would likely have been substrate limited, and dependent on the production of H2 or organic acids by other community members. Most Desulfomicrobium species are also able to ferment select organic acids (Dias et al., 2008).

Other acetate oxidizers

Few proteins were identified from Desulfuromonadales r9c7, owing partly to incomplete genomes (Table 2, Figure 4b). Proteins and genes identified may be attributed to either the Geobacter- or Desulfuromonas-like bacteria comprising the r9c7 bin. Some genes, but no proteins, were detected for nitrogen fixation, ATP synthase, cell division, gluconeogenesis and motility by pili or flagella, and only one protein was detected for chemotaxis. Several key TCA cycle proteins were detected (Figures 5a and d; Table 3), demonstrating that Desulfuromonadales were actively consuming acetate, while sulfate reduction prevailed as the major terminal electron-accepting process. These bacteria do not reduce sulfate, but are well-known for their ability to reduce metals (including iron and uranium) and other inorganic elements or compounds (Lovley et al., 2004).

Studies have demonstrated a correlation between the addition of simple organic substrates, such as glucose and organic acids like acetate, to terrestrial aquifers and the enrichment of Fe(III)-reducing Geobacter species (Snoeyenbos-West et al., 2000; Holmes et al., 2002, 2007). Repeat experiments amending the Rifle aquifer with acetate have consistently produced blooms of Geobacter associated with Fe(III) and U(VI) reduction (for example, Anderson et al., 2003; Holmes et al., 2007; Wilkins et al., 2009). These enriched Geobacter and co-enriched Desulfuromonas species persist as significant community members, although at lower relative abundances, after sulfate reduction becomes prevalent (Handley et al., 2012), and Geobacter are key candidates for the ongoing reductive immobilization of U(VI) observed during sulfate reduction in this study (Figure 1d). Despite the capacity for abiotic reduction of sulfide, microbial Fe(III) reduction may proceed concurrent with sulfate reduction (Sörensen, 1982; Tugel et al., 1986; Canfield, 1989). It is therefore plausible that either genera attributed to the Desulfuromonadales r9c7 may have also coupled acetate-oxidation with Fe(III) reduction. Alternatively, r9c7 bacteria may have gained energy from S0 reduction (Pfennig and Biebl, 1976; Lovley et al., 2004), which has been shown to accumulate during acetate amendment of the Rifle aquifer. Williams et al. (2011) identified 4–5 mmol kg−1 S0 (below 3-m depth) in the core collected to create well P104 (used in this study) after 110 days of acetate amendment (second year amendment). Comparably, little S0 was identified in less-stimulated sediment cores collected further from the acetate source.

Bacteroidetes, on the other hand, are well-known for their ability to degrade carbohydrates and other complex organic compounds using respiratory or fermentative metabolisms (Holmes et al., 2007; Lee et al., 2010; Thomas et al., 2011). A number of genes identified here are associated with mannose metabolism (mannose-1-phosphate guanyltransferase, mannose-6-phosphate isomerase, GDP-mannose 4,6-dehydratase, phosphomannomutase), xylan degradation (candidate β-xylosidase, endo-1,4-β-xylanase) and cellulose degradation (family 3 candidate β-glycosidase), including identification of Bacteroidetes-like unbinned partial fragments (0.5–2 kb long) resembling candidate β-glycosidase and cellobiose phosphorylase. However, few expressed proteins indicative of these metabolisms were identified. Only two proteins (with peptide counts just above detection) were identified that are suggestive of cellobiose degradation and mannose metabolism (candidate β-glycosidase hydrolase, phosphomannomutase), and only two glycolysis proteins were detected.

Instead, detection of some key TCA cycle enzymes, including citrate synthase (Figures 5a and d; Table 3), suggests that the Bacteroidetes may have been using the cycle heterotrophically to oxidize acetate (cf. Xie et al., 2007; Zhang et al., 2009), although use of the reductive cycle cannot be excluded. Genomic data indicate that Bacteroidetes r9c5 has a complete oxidative TCA cycle (near-complete in r9c4), and both r9c4 and r9c5 have genes that may support a reductive cycle (that is, citrate lyase, 2-oxoglutarate synthase, pyruvate synthase, PEP synthetase and carboxylase). The electron acceptor for this reaction is not evident; however, on the basis of genomic evidence, one possibility is that Bacteroidetes coupled acetate oxidation to the reduction of nitrogen species, as genes were identified for nitric-oxide reductase (norABD) and nitrous-oxide reductase (nosLZ).

Genes were also identified for a formate-dependent, ammonium-forming nitrite(/polysulfide) reductase (nrfADH); selenate reductase ygfK (r9c4 only); sulfur metabolism, namely SQR (r9c4 only), a Chlorobium luteolum-like flavocytochrome-c sulfide dehydrogenase and a phsC-like cytochrome-b561 thiosulfate reductase; and arsenic detoxification (similar to r9c2-r9c3; Figure 6a). Although the experimental condition under which Bacteroidetes r9c4 and r9c5 were growing was anoxic, genes for cytochrome-c and cytochrome-bd oxidase and cytochrome-cbb3 oxidase (ccoPQNOS) indicate a capacity for these bacteria to also grow (micro)aerobically (Preisig et al., 1993; Trumpower and Gennis, 1994; Visser et al., 1997; Kulajta et al., 2006).

Sulfur oxidizers: acetate-induced syntrophy

Previously, researchers have demonstrated syntrophic growth between oxygen-dependent, sulfide-oxidizing Thiobacillus thioparus and sulfate-/S0-reducing, Desulfovibrio desulfuricans in co-culture, such that a positive-feedback cycle was established between the reductive and oxidative processes of these two bacteria (van den Ende et al., 1997). In another study, geochemical and functional gene-based evidence for a cryptic microbial sulfur cycle, suggests that this process may have important biogeochemical consequences for sulfate recycling in oceanic oxygen minimum zones (Canfield et al., 2010). Proteogenomic results here indicate that a similar process was operating in the acetate-amended Rifle aquifer sediment, and suggest that syntrophic growth of autotrophic or possibly mixotrophic Sulfurovum- and Sulfurimonas-like bacteria (r9c2 and r9c3) were supported by sulfide and CO2 generated by heterotrophic respiration.

Species within these two epsilonproteobacterial genera are autotrophic, and oxidize sulfide, sulfur and thiosulfate either with oxygen and/or nitrate serving as electron acceptors (Hoor, 1975; Inagaki et al., 2003, 2004). Sulfurimonas denitrificans oxidizes sulfide by reducing nitrate completely to N2 (Hoor, 1975). According to genomic studies of Sulfurovum sp. NBC37-1 and S. denitrificans DSM1251, sulfur oxidation in these organisms may proceed via the sulfur oxidation pathway (SoxCDYZXAB), forming sulfate, or SQR, forming S0 from sulfide (Nakagawa et al., 2007; Sievert et al., 2008).

Similarly, r9c2 and r9c3 possess respiratory genes for a complete denitrification pathway (napABDFGHL, nirSF, nirNJ for r9c3 only, norBC, nosZ), sulfur oxidation (SQR, soxCDYZ) excluding soxXAB (Supplementary Figure S5), and also the reductive TCA cycle for CO2 fixation (Table 3; Figure 6b). Proteomic data infer that, during amendment these Epsilonproteobacteria performed nitrate-dependent sulfide/sulfur oxidation coupled to CO2 fixation via the reductive TCA cycle (Figures 5a and d). However, simultaneous operation of an oxidative TCA cycle to support mixotrophic growth with acetate cannot be excluded (cf. Tang and Blankenship, 2010). Proteomics further show r9c2 and r9c3 were dividing, and r9c3 was chemotactic and motile. Genes for cytochrome-cbb3 oxidase (ccoPQONS), cytochrome-b561 (r9c3; Murakami et al., 1986) and cytochrome-bd (r9c2) suggest that these bacteria could also oxidize sulfur (micro)aerobically.

The missing soxXAB genes (not detected by homology searches) are normally required for thiosulfate (and presumably sulfide and sulfur) attachment to SoxYZ, activation to sulfane, and release after oxidation by SoxCD (Friedrich et al., 2005; Sauvé et al., 2007; Zander et al., 2011). Hence, the exact function of Sox in sulfur oxidation by r9c2 and r9c3 cannot be deduced from our data. Green sulfur bacteria lacking soxCD (but with functional soxXAB and soxYZ genes) are able to use dsr with or without apr genes in reverse to oxidize sulfide or sulfite, respectively (Sakurai et al., 2010; Gregersen et al., 2011). We identified no genes indicative of a reverse sulfate reduction pathway in r9c2 and r9c3. Nevertheless, in vitro enzyme assays by Rother et al. (2001) suggest sulfide or S0 oxidation may proceed, although at a slower rate (13–19 or 3–17 times less, respectively), without either SoxXA or SoxB. The same is true for SoxXAB without SoxCD or SoxYZ.

Synteny is shared between soxXAB and soxCDYZ in the model organism, Paracoccus pantotrophus, and several other Alphaproteobacteria (Friedrich et al., 2005). However, they form two non-syntenous gene clusters in bacteria closely related to r9c2 and r9c3, Sulfurovum sp. NBC37-1 and S. denitrificans DSM1251 (Nakagawa et al., 2007; Sievert et al., 2008; Supplementary Figure S5). In as much as conservation of gene order tends to suggest conservation of gene function, loss of synteny tends to suggest a loss of co-dependence between the gene clusters, possibly occurring with increased evolutionary distance (Yelton et al., 2011). This may explain the apparent loss of soxXAB genes in r9c2 and r9c3.

While it is possible that r9c2 and r9c3 completely re-oxidized sulfide to sulfate, an autotrophic denitrifying community from anaerobic sludge has been shown to only partially oxidize sulfide (probably to S0) when placed under nitrate-limiting conditions, but completely oxidize sulfide to sulfate with unlimited nitrate (Cardoso et al., 2006). Considering the micromolar concentrations of nitrate available in the aquifer (Williams et al., 2011), this could also reasonably explain the enrichment of bacteria in this study (r9c2 and r9c3) that are potentially only able to oxidize sulfide to S0. Any formation of S0 by these bacteria would likely be re-cycled by putative sulfur-reducing bacteria, r9c7.

Autotrophic versus heterotrophic denitrification

Simulations of biogeochemical processes occurring during acetate amendment captured the general features of the acetate and sulfate breakthrough curves at well P104, and estimate nitrate conversion to N2 (Supplementary Figures S6a and b). Although simulations for autotrophic (sulfide-dependent) and heterotrophic (acetate-dependent) denitrification use estimates for nitrate and specific growth rates, and exclude biomass, analyses indicate that the autotrophic denitrification pathway, involving sulfide oxidation to S0 or sulfate, is thermodynamically feasible. Despite its greater standard Gibbs energy of reaction, the autotrophic sulfide to sulfate pathway is less favored than the sulfide to S0 pathway (Supplementary Figure S6c) because of the high sulfate concentration in groundwater (Figure 1b).

Although heterotrophic denitrification is thermodynamically favored over autotrophic denitrification (ΔG of ca. −220 versus −178 kJ per electron per mole), the effect is insignificant if the thermodynamic factor (FT) formulation in Equation 3 (Supplementary Information; Supplementary Figure S6c) is assumed to be correct, as the system is effectively far from equilibrium with respect to both pathways, and FT is very close to 1. This implies that behavior is controlled by the Monod kinetic terms in Equation 2 (Supplementary Information). Principally owing to the low half-saturation constant used for autotrophic denitrification, simulations for the sulfide–sulfur reaction tend to indicate that the autotrophic pathway dominates over the heterotrophic pathway given a concentration of 5 μM (at all considered heterotrophic versus autotrophic rates), and vice versa for the higher 72-μM concentration when assuming a slower autotrophic rate (Supplementary Figure S6d–i). While a potential impact of sulfide toxicity on community composition and function cannot be excluded, modeling results may explain why, with acetate in excess, autotrophic sulfide-dependent denitrification by Epsilonproteobacteria r9c2 and r9c3 out-competed heterotrophic denitrification in this experiment. Another factor may be possible mixotrophic growth by Epsilonproteobacteria r9c2 or r9c3, which would likely increase cell growth and denitrification rates (Cardoso et al., 2006).

Conclusions

We reconstructed the genomes of members of a subsurface sediment community enriched during acetate amendment. Proteomics identified organism-specific function and syntrophic interactions among community members (Figures 6a and c). While the dominant process identified was acetate-fueled sulfate reduction, excess acetate was also respired by enriched Desulfuromonadales, probably supporting concurrent Fe(III)-, U(VI)- and/or S0-reduction. Gene-based evidence and TCA cycle proteins detected for Bacteroidetes suggest that they contributed to acetate degradation, and have the capacity for reducing nitrogen species. Co-enrichment of Epsilonproteobacteria and expression of proteins associated with sulfide oxidation and carbon fixation imply that products of heterotrophic acetate oxidation—sulfide and CO2—were used as a carbon and energy source by autotrophic or mixotrophic Epsilonproteobacteria. Sulfide-dependent denitrification may have been favored over the heterotrophic pathway owing to nitrate-limiting conditions within the Rifle aquifer, such that kinetic factors govern outcomes. In turn, reaction products of epsilonproteobacterial metabolism, such as N2, were probably fixed and sulfate or S0 respired by Desulfobacter or Desulfuromonadales bacteria. These results suggest acetate-amendment promoted complex organismal and metabolic processes and interactions involved in carbon, sulfur, metal and nitrogen cycling.