Arsenic(V) Reduction in Relation to Iron(III) Transformation and Molecular Characterization of the Structural and Functional Microbial Community in Sediments of a Basin-Fill Aquifer in Northern Utah

Abstract

Basin-fill aquifers of the Southwestern United States are associated with elevated concentrations of arsenic (As) in groundwater. Many private domestic wells in the Cache Valley Basin, UT, have As concentrations in excess of the U.S. EPA drinking water limit. Thirteen sediment cores were collected from the center of the valley at the depth of the shallow groundwater and were sectioned into layers based on redoxmorphic features. Three of the layers, two from redox transition zones and one from a depletion zone, were used to establish microcosms. Microcosms were treated with groundwater (GW) or groundwater plus glucose (GW+G) to investigate the extent of As reduction in relation to iron (Fe) transformation and characterize the microbial community structure and function by sequencing 16S rRNA and arsenate dissimilatory reductase (arrA) genes. Under the carbon-limited conditions of the GW treatment, As reduction was independent of Fe reduction, despite the abundance of sequences related to Geobacter and Shewanella, genera that include a variety of dissimilatory iron-reducing bacteria. The addition of glucose, an electron donor and carbon source, caused substantial shifts toward domination of the bacterial community by Clostridium-related organisms, and As reduction was correlated with Fe reduction for the sediments from the redox transition zone. The arrA gene sequencing from microcosms at day 54 of incubation showed the presence of 14 unique phylotypes, none of which were related to any previously described arrA gene sequence, suggesting a unique community of dissimilatory arsenate-respiring bacteria in the Cache Valley Basin.

INTRODUCTION

Arsenic (As) is one of the most frequently detected contaminants in private domestic wells used for household drinking water (1) and public water supplies (2). In the United States, As concentrations in excess of the drinking water maximum contaminant level (MCL; 10 μg/liter) in public supply wells are distributed across the country, but 3/4 of these wells are in the western United States (2). In basin-fill aquifers in California, Nevada, New Mexico, Arizona (1), and Utah (3), 10% of the domestic wells tested had As in excess of the MCL. Modeling efforts by the USGS predicted that 43% of the area in the basin-fill aquifers of the Southwest might have groundwater that equals or exceeds the MCL for As (4). While public water suppliers are required to treat water to meet drinking water standards, there are no such requirements for private wells.

Common aspects of the basin-fill aquifers in the western United States are geothermal activity and volcanic rock. The source of As in the Cache Valley Basin, UT, is presumed to be the hydrothermal sulfide and arsenide deposits in the surrounding mountains. Due to low rainfall and high evapotranspiration, groundwater recharge does not result from precipitation in the basin but from precipitation in the surrounding mountains. The Cache Valley Basin is located 128 km northeast of Salt Lake City and is on the eastern edge of the Basin and Range Province. A survey of domestic wells in this basin showed that 27 of the 161 wells tested (16.8%) had As in excess of the MCL (5). Despite the elevated concentrations of As in this region, studies have been limited to groundwater surveys (3, 5), without investigating the mechanisms of As reduction or the role of dissimilatory arsenate-respiring bacteria (DARB) in producing elevated groundwater As concentrations. To the best of our knowledge, this is the first study in this region to examine As^V reduction in relation to both iron transformation and microbial community characteristics.

Under anoxic conditions at field sites and in laboratory studies, a dependency of As release on the microbial reductive dissolution of Fe^III oxide minerals has been observed (6, 7). This mechanism is proposed as the process driving As solubilization in groundwater in Southeast Asia (8–11). The released As^V may be resorbed onto minerals or precipitate as a new solid phase, or DARB may reduce the As^V in solution to As^III. As^V is also directly reduced by DARB without the need for dissolution of the Fe minerals (a decoupled process) (10, 12). Most of these studies have looked only at As^III and Fe^II in the aqueous phase, which is influenced by the amount of As^III and Fe^II precipitated or resorbed by the solid phase (7). In the current study, we have analyzed reduced As^III and Fe^II both in the aqueous phase and in association with HCl-soluble minerals.

Indigenous bacteria can directly reduce As^V to As^III for either detoxification (13) or energy production (14–16). DARB reduce As^V via the arsenate respiratory reductase enzyme (encoded by arrAB) that is present in the periplasmic space (15, 17) and can grow using As^V as an electron acceptor (18, 19). The arrA gene encodes the alpha-subunit of the dissimilatory arsenate reductase and is a key functional gene for the study of arsenate-respiring microorganisms.

Advances in molecular microbiology techniques, such as next-generation sequencing of the 16S rRNA gene, have allowed identification of microbial community structure at greater depth, along with detailed descriptions of shifts in community structure in response to different treatments. However, to obtain information about the potential As reduction function associated with these microorganisms requires the sequencing of the functional gene arrA. Previously, the arrA gene has been used to assess the diversity of DARB in a range of ecosystems, including water and sediments from Mono Lake and Searles Lake, CA (20–22), arsenic-contaminated aquifers in Cambodia (23, 24) and West Bengal (25), estuarine sediments from Chesapeake Bay (26), and streambed sediments from the Inner Coastal Plain, NJ (27, 28). However, information on the reduction of As^V in relation to Fe reduction and the microbial community (both structural and functional) responsible for these reactions within the basin-fill environment of the Southwestern United States has not been reported. Hence, the objectives of the current microcosm-based study were to (i) investigate the reductive dissolution of As^V in relation to Fe^III reduction, (ii) describe the bacterial communities associated with As and Fe reduction in both biostimulated and unstimulated systems, and (iii) determine the changes in the functional community composition of the DARB based on arrA gene sequences.

MATERIALS AND METHODS

Site description and sample collection.

Aquifer solids were collected from the Cache Valley Basin, UT, which is approximately 80 km long and 24 km wide, located on the Idaho/Utah border. The Cache Valley Basin is the eastern extension of the Great Basin, and its detailed geology has been previously described (29). Briefly, the basin consists of ancient Lake Bonneville sediments overlying volcanic deposits of the Salt Lake Formation. This formation is exposed in the surrounding mountains, the presumed source of As. Samples were collected in the center of the valley, located up gradient from the Logan City Municipal Landfill (41°44.02′N, 111°52.29′W). This study site was selected since there is an existing system of monitoring wells and historical As groundwater data associated with the landfill, as well as for ease of access, since the property is owned by the city of Logan or the state of Utah. A Geoprobe direct-push soil corer was used to collect aquifer solids from 13 locations within the study area. Aquifer solids were collected in 2.54-cm-diameter by 152-cm-long plastic sleeves within the saturated zone at depths of 2.1 to 4.9 m below ground surface.

The collection tubes were capped, delivered on ice to the Utah Water Research Laboratory within an hour of collection, and kept in a vinyl anaerobic chamber (100% N₂) (Coy Laboratory Products, Grass Lake, MI) in a constant-temperature room (15 ± 2°C) for 2 months while the initial processing and the establishment of the microcosms were carried out. For initial processing, sediment cores were sliced into two to four sections (43 sections in total from 13 cores) based on color and texture. These core sections were processed immediately to determine As and Fe oxidation states by extracting the solids (1 g) with 0.5 M HCl (30 ml; trace metal grade). The extracts were analyzed for Fe^II using ferrozine (30) with a Spectronic 601 spectrophotometer (Milton Roy, Rochester, NY) at 562 nm with a 1-cm cell and for As^III by hydride generation-atomic absorption spectroscopy (HGAA) (PerkinElmer Analyst 800). Inductively coupled plasma mass spectroscopy (ICP-MS) (Agilent 7500c) was used to determine the concentrations of total arsenic [As_tot = As^V + As^III] and Fe [Fe_tot = Fe^II + Fe^III]. Iron and As mineralogy were characterized using a modified six-step sequential chemical extraction procedure (31) (see Materials and Methods in the supplemental material) that operationally defines As associated with exchange surfaces, carbonate minerals, organic phases, Mn oxides, amorphous and crystalline Fe oxides, and sulfides and other insoluble minerals. Extracts were analyzed for Fe by flame atomic absorption spectroscopy (FAAS) and for As by ICP-MS. The solids were also characterized for pH, particle size distribution, organic carbon (OC), and carbonate content (32, 33).

Cluster analysis and microcosm establishment.

The 43 sections from 13 sediment cores were grouped into three clusters representative of the diversity of site conditions regarding As geochemistry by using cluster analysis (JMP 5.01 Statistical Software; SAS Institute, Cary, NC). The discriminating variables among the three clusters were HCl-extractable As^III, As associated with carbonates and amorphous iron oxide, Fe associated with carbonates, and organic carbon and sand content (Table 1). Since the bioavailability of As was of interest, three aquifer solids, one from each cluster, containing As associated with carbonates and amorphous iron oxides within the 95% confidence interval of the mean of each of the three clusters, were selected for the microcosm study.

TABLE 1.

Concentrations of the discriminative variables used for the cluster analysis of 43 sections from 13 sediment core samples of the Cache Valley basin-fill^a

Cluster (no. of sediment sections in cluster)	Fe associated with carbonates (mg/kg)	HCl-extractable AsIII (μg/kg)	As associated with carbonates (μg/kg)	OC (%)	Sand (%)	As associated with amorphous Fe oxides (μg/kg)
1 (28)	99.3 ± 17.1	225 ± 33.0	908 ± 150	0.18 ± 0.02	17.6 ± 5.1	2,789 ± 561
2 (11)	91.2 ± 31.3	78.2 ± 16.3	468 ± 82	0.14 ± 0.06	32.6 ± 5.6	825 ± 215
3 (4)	1,124 ± 698	38.2 ± 9.0	1,106 ± 335	0.42 ± 0.07	5.83 ± 5.0	1,070 ± 302

^aValues are 95% confidence intervals.

Microcosms were constructed under anoxic conditions using the three selected sediments, NP1, NP3, and NP8. The two treatments were groundwater (GW), collected from a well at the site with less than 1 μg/liter As, and groundwater plus glucose (GW+G) at 2,000 mg carbon/liter (28 mM) (34). The selection of glucose as a C source was based on the results of a previous study in which we used a number of different C sources and the highest As reduction was detected in response to glucose addition (35). The groundwater quality parameters, distance between the sites, and depth of sampling for the sediments used to establish microcosms can be found in Materials and Methods in the supplemental material.

In each microcosm (n = 90), 10 g (dry wt) of sediment was mixed with 40 ml of groundwater or groundwater with glucose in 50-ml sterile disposable centrifuge tubes (9400RCF, high speed; Fisher Scientific). All solutions were filter sterilized (0.2 μm) before use and left in the anaerobic chamber (100% N₂) for a week to eliminate oxygen. Microcosms were incubated horizontally in the anaerobic chamber at 15 ± 2°C and sacrificed in triplicate for analysis at 0, 6, 12, 25, and 54 days of incubation. Sterile controls were established by autoclaving (twice) 10 g of sediment in individual centrifuge tubes. At the time of sampling, triplicate tubes were harvested, centrifuged (7,000 × g for 20 min), and returned to the anaerobic chamber. Supernatant was filtered through 0.2-μm nylon syringe filters and analyzed for Fe^II, As^III, As_tot, and Fe_tot. The aquifer solids were extracted with 0.5 M HCl and analyzed for As^III and Fe^II as described above. Ion chromatography (IC) (ICS-3000; Dionex, Sunnyvale, CA) was used to measure sulfate. The differences in the As^III and Fe^II concentrations for incubation times, sediment types, and/or treatments were explored by three-factor analysis of variance followed by Tukey's honestly significant difference (HSD) test (P ≤ 0.05) using the software package R (http://www.R-project.org). The remaining solids from the microcosms were stored at −70°C until analyzed for microbial characterization.

DNA extraction and PCR amplification.

Total genomic DNA was extracted in triplicate from each sediment sample of all treatments (1.0 g) at two time points (day zero and at the end of the study, i.e., day 54 of incubation) using the PowerSoil DNA isolation kit (Mo Bio, Carlsbad, CA). All extraction steps were followed according to the manufacturer's instructions, and DNA was stored at −20°C until analyzed.

16S rRNA gene barcoded pyrosequencing.

The extracted DNA samples were amplified with a set of primers targeting the V3-V4 hypervariable regions of the bacterial 16S rRNA gene. The primer pair F515 (5′-GTGCCAGCMGCCGCGG-3′) and R907 (5′-CCGTCAATTCMTTTRAGTTT-3′) used in this study has been previously tested for accuracy and covers more than 98% of the 16S rRNA gene sequences in the Ribosomal Database (36). Primers were designed with 8-base barcodes and 454 pyrosequencing adapters (Eurofins MWG Operon). Amplification reactions were performed in 50-μl reaction mixture volumes containing 1× buffer, 0.2 μM each primer, 1.8 mM MgCl₂, 200 μM deoxynucleoside triphosphates, 20 ng of template, and 1 μl FastStart high-fidelity PCR system enzyme (Roche Applied Sciences). The PCR conditions were 3 min at 95°C, followed by 30 cycles of denaturation at 94°C for 45 s, primer annealing at 54°C for 45 s, extension at 72°C for 1 min, and final extension for 7 min. Reaction mixtures from reactions performed in triplicate were purified with Agencourt AMPure beads (Beckman Coulter, Brea, CA). Purified, adapter, and unique-barcode-attached PCR products from each of the 27 DNA extracts were pooled such that the concentration of each product was the same in the mixture. Pyrosequencing was performed on the mixture with the 454 GS FLX sequencer (454 Life Sciences) at the Utah State University Center for Integrated Biosystems.

16S rRNA gene sequence analysis.

16S rRNA gene raw sequences were screened, and low-quality sequences were removed based on a minimum length of 360 bp, maximum homopolymer of >8 bases, sequences with unidentified bases (N), and sequences with more than one inexact match with the unique barcode identifier. Chimeric sequences were identified using DECIPHER (37) and removed using Mothur (38).

A total of 136,700 good-quality sequences were obtained from 24 of 27 samples (all samples except NP3 sediment samples at day zero), and the read number per sample varied between 5,000 and 8,675. Considering that species richness and other diversity indices can be influenced by the number of sequences, we randomly subsampled 5,000 sequences per sample using Python codes based on the sample with the lowest sequence number (total of 120,000 sequences). Complete data analysis was carried out both on subsampled and complete data. All sequences were analyzed using Mothur (38) and the Ribosomal Database Project (RDP; http://rdp.cme.msu.edu).

First, these sequences were aligned in Mothur using RDP reference alignment. Pairwise distances between sequences were calculated by using the furthest-neighbor algorithm, and operational taxonomic units (OTUs) were delineated at 97% sequence similarity. Differences in the microbial communities across the three sediment samples, as well as both GW- and GW+G-treated samples after 54 days of incubation, were evaluated through total species richness, Margalef's richness index, Shannon diversity, and the Chao1 estimator as described previously (39). To assess the overall changes in the microbial community structure, analysis of similarity (ANOSIM) of the square root-transformed community data (Bray-Curtis similarity) at a 97% OTU level was performed, and the differences were visualized with nonmetric multidimensional scaling (NMDS) plots generated in R (R Development core team; http://www.R-project.org) running the vegan community ecology package. The OTU-based DNA clustering data were also used to generate rarefaction curves in Mothur.

Phylogenetic community similarity based on the branch length data matrix was calculated using the unweighted UniFrac algorithm as described previously (39). Briefly, a phylogenetic tree of 120,000 16S rRNA gene sequences was constructed using FastTree 2 (40) and a principal coordinate analysis (PCoA) was performed using the Fast UniFrac matrix (http://bmf2.colorado.edu/fastunifrac/) (41).

Finally, the differences in the relative distributions of sequences at the phylum and genus taxonomic levels, characterized through RDP Naive Bayesian Classifier 2.5 (42) at a set confidence threshold of 80%, were used to generate bar graphs and pie charts.

Functional gene (arrA) cloning and DNA sequencing.

The arsenate reductase (arrA) gene was amplified from the genomic DNA using a nested PCR approach as described previously (26). Details of the primers and PCR conditions used are presented in Materials and Methods in the supplemental material. Amplified products (equal concentrations of DNA) from the three replicates of each treatment were pooled, purified with the QIAquick PCR purification kit (Qiagen, Valencia, CA), ligated into the TOPO TA pCR4.0 vector, and transformed into Escherichia coli One Shot TOP10 competent cells (TOPO TA cloning kit; Life Technologies, CA) as recommended by the manufacturer. A total of six independent clone libraries were generated and a minimum of 45 arrA gene clones were sequenced per treatment (total of 288 sequences). Sequencing was performed at the DNA Analysis Facility on Science Hill at Yale University.

A total of 444 arrA gene sequences, including 288 clone sequences from the current study and 156 sequences (≥630 bp in length) available in GenBank, were aligned and trimmed to uniform length (630 bp) using Sequencher 4.2.2 (Gene Codes Corp., Ann Arbor, MI), CLUSTAL_X, and MUSCLE (43, 44). The aligned arrA gene sequences were used to create a distance matrix, followed by clustering into OTUs at 97, 95, and 90% DNA similarity using Mothur (38). The rarefaction curves were also generated using the same program.

All the arrA sequences were also analyzed with four different phylogenetic methods, i.e., neighbor joining (NJ), maximum parsimony (MP), Bayesian, and maximum likelihood (ML). MEGA version 4 (45) was used for the NJ and MP methods, while Bayesian and ML analyses were completed using MrBayes 3.0 (46) and PHYML 3.0 (47), respectively. All of the detailed settings were the same as previously described (48, 49). The phylogenetic tree was collapsed at the 90% DNA similarity.

Nucleotide sequence accession numbers.

Partial arrA gene sequences were deposited in GenBank under accession numbers JN618463 to JN618505, JN704711 to JN704786, and KF009945 to KF010114. The 16S rRNA gene sequences were deposited to the NCBI Sequence Read Archive (SRA) under accession number SRP039499.

RESULTS

Characterization of the aquifer solids.

Physical and chemical properties of the three selected sediments are presented in Table 2. NP3 was rich in As. Although NP1 and NP8 had similar amounts of total As, the distribution of As (Fig. 1) among the solid phases, as defined by the sequential extractions, differed significantly. The ratio of Fe^II to Fe^III extracted with 0.5 M HCl defined the oxidation state of the sediments; NP3 was an oxidized sediment (97% of extracted Fe was Fe^III). For NP1, 38% of the extracted Fe was Fe^III. There was no extractable Fe^III in NP8. Irrespective of the Fe chemistry, the oxidation state of As extracted with the HCl was over 90% As^V for all three solids. Two of the sediments (NP1 and NP3) were from the redox transition zone, as evidenced by their red/brown coloration, characteristic of iron oxides, while the third sediment (NP8) was from the redox depletion zone, as indicated by its green coloration, reflective of the lack of HCl-extractable Fe^III.

TABLE 2.

Physiochemical characteristics of the three aquifer sediments selected for the microcosm study, which represent the three clusters generated from 43 sections obtained from the 13 sediment core samples of the Cache Valley site

Aquifer solid (cluster)	Depth (m)	pH	CaCO3 (%)	OC (%)	Sand (%)	Clay (%)	Total Fe (%)	FeII/FeIII (mg/kg)a	Total As (mg/kg)	AsIII/AsV (μg/kg)a
NP1 (2)	3.1–3.6	7.15	28.8	0.09	33.2	21.7	1.33	129/79	3.24	57.2/811
NP3 (1)	2.7–3.4	7.36	39.3	0.12	41.8	19.0	1.20	14.2/400	7.70	29.4/255
NP8 (3)	4.7–4.9	7.69	55.1	0.31	15.6	30.0	1.43	2,423/0	3.75	38.8/499

^a0.5 M HCl extraction.

FIG 1.

Sequential extraction of arsenic from three sediments (NP1, NP3, and NP8).

The results of analysis of variance (ANOVA) for both As and Fe reduction suggested significant differences for time, treatment, and sediment type (P < 0.05; see Table ST1A and B in the supplemental material).

As^V reduction.

An increase in As^III (aqueous + 0.5 M HCl-extracted As^III) was observed by the first sampling event (day 6) in all three sediments incubated with GW only (Fig. 2A to C). Arsenic reduction was supported in these low-organic carbon sediments (Table 2). Biostimulation with glucose resulted in significant increases in the concentrations of As^III in the NP1 and NP3 microcosms (Fig. 2A and B) compared to the concentrations in those treated with GW alone. The As^III concentration doubled (370 μg/kg sediment) at day 6 of incubation for NP1, while for the NP3 microcosm, As^III was 5-fold higher (1,940 μg/kg sediment) at day 25 of incubation than for treatment with GW alone. For NP8, the initial increase (day 6) in As^III was the same for both treatments, followed by a significant decrease in As^III to concentrations equivalent to those under initial conditions (Fig. 2C). The addition of glucose to the sediment from this redox depletion zone did not enhance As reduction. For all three sediments, the dissolved sulfate concentrations decreased with time, indicating that both stimulated and nonstimulated microcosms were under sulfate-reducing conditions (see Fig. SF1A to C in the supplemental material).

FIG 2.

Effects of supplemental electron donor on the reduction of arsenic and iron. Sediments were supplemented with groundwater (GW) and groundwater plus glucose (GW+G). The sediments used in microcosms were collected from sites NP1 (A), NP3 (B), and NP8 (C). Presented are the averages (Tukey's honestly significant difference) for triplicate microcosms harvested at different time points.

Effect of Fe reductive dissolution on As reduction.

In the samples treated with groundwater only, the initial As reduction was independent of Fe reduction. For all three sediments, As^III increased significantly by the first sampling event (day 6), while a statistically significant increase in Fe^II was not observed until day 12 (NP1), day 25 (NP3), or day 54 (NP8) compared to Fe^II at day zero. With the addition of glucose to NP1 or NP3, As reduction was concurrent with Fe reduction over the initial sampling intervals (Fig. 2A and B). For NP3, an increase in As^III concentration was strongly positively correlated with the increased reduction of Fe (R² = 0.87, P < 0.05). In the sediments from the depletion zone (NP8), no discernible Fe^II above time zero concentrations was observed until day 54 with or without glucose addition (Fig. 2C). Both treatments led to a temporary increase in As^III by day 6, but the increase was independent from Fe reduction.

Total microbial community characterization.

16S rRNA gene-based characterization of the microbial community at the phylum level showed that at day zero, 58% of the sequences for NP1 and 68% of the sequences for NP8 belonged to unclassified groups, suggesting subsurface sediment communities of not-yet-described bacterial phyla (see Fig. SF2 in the supplemental material). Among the classifiable sequences, Proteobacteria were the dominant group, representing 13 and 15% of the total sequences for the NP1 and NP8 sediments, respectively. However, the initial microbial community for the sediment NP3 remained uncharacterized because no 16S rRNA gene could be sequenced from this site. There was a clear shift in the microbial community in the microcosms in response to the 54-day enrichment with either the GW or GW+G treatment. In all three sediments treated with GW, Proteobacteria was the dominant group (range 52 to 60%), followed by Firmicutes (range 25 to 30%). Biostimulation with glucose, however, resulted in a complete shift in the dominance of bacterial groups toward the Firmicutes, which comprised up to 95% of the total microbial community in these three sediment samples. Proteobacteria-related sequences were detected only as a minor group (2 to 3%) with the biostimulated treatment (see Fig. SF2 in the supplemental material).

Differences in bacterial community composition at the lower taxonomic level (genus) were also obvious across different sites, as well as between treatments (Fig. 3). The majority of identifiable bacterial sequences at the genus level under carbon-limited conditions (GW treated) for microcosms of the NP1 sediment were affiliated with Desulfocapsa (12%) and Geobacter (10%). For the NP3 sediment sample, Geobacter species (24%) were the dominant group. Within the depletion zone sediment, NP8, Desulfocapsa (17%) and Shewanella (13%) were among the dominant bacterial genera. The glucose amendment resulted in a significant shift toward an abundance of Clostridium (Clostridium sensu stricto and Clostridium XIVa)-related sequences, represented by 80, 54, and 24% of the total sequences for the NP1, NP3, and NP8 sediments, respectively (Fig. 3; see also Table ST2 in the supplemental material).

FIG 3.

Shifts in the microbial communities at the genus level in response to groundwater (GW) and groundwater-plus-glucose (GW+G) treatments in the microcosm samples from the three sediments collected from the Cache Valley site, identified by 16S rRNA gene sequences. The bacterial genera represented by >1% of total sequences are presented here.

Overall, the community structures (both taxonomically and phylogenetically) for both the subsampled data (120,000 sequences) (Fig. 4; see also Fig. SF3 in the supplemental material) and complete data (136,700 sequences) (see Fig. SF4) were similar. All subsampled sequences (120,000) could be grouped into 5,340 unique bacterial OTUs at the 97% sequence similarity level across all samples. Differences in the microbial taxonomic community structure assessed by NMDS analysis of OTUs at 97% DNA similarity showed a distinct clustering of the samples by treatment and sediment type with the Bray-Curtis similarity index (Fig. 4). Subsequent ANOSIM analysis confirmed that the observed differences were statistically significant (global R = 0.97, P < 0.01). Pairwise comparison among different treatments/sediments suggested that there were significant differences in the microbial community structure (P < 0.001) across all sediments for both treatments, with the exception of glucose-treated NP1 and NP3 sediments. Similarly, the phylogenetic community structures showed significant differences across different sediment types and treatments (see Fig. SF3).

FIG 4.

Multidimensional scaling plot, based on the Bray-Curtis similarity index (3% DNA dissimilarity), of 120,000 16S rRNA gene sequences (5,000 per sample) for three sediments at day zero (t0) and day 54 of incubation. The two treatments were groundwater (GW) and groundwater plus glucose (GW+G). From NP3 sediment at day zero, no 16S rRNA gene could be sequenced.

For the total number of unique 16S rRNA gene sequences, the Chao1 estimator, the Margalef's species richness index, and the Shannon diversity index suggested higher species richness and diversity at day zero than after 54 days of enrichment (see Table ST3 in the supplemental material). For sites NP1 and NP8, the samples amended with glucose had significantly lower levels of diversity than the samples treated with GW, while for the NP3 sediment, the diversity level remained unchanged. Similarly, rarefaction curve analysis indicates that the number of OTUs at 97% similarity did reach an asymptote for most of the treatments, except for the samples from NP8 sediments at day zero, which suggests that 15,000 sequences per treatment were sufficient to capture most of the microbial diversity present in these samples (see Fig. SF5).

Functional microbial community.

PCR amplification yielded an arrA gene product of the expected size (650 bp) from all sediments collected after 54 days of incubation in both treatments. Even though the same DNA extraction procedure and PCR conditions were used for all samples, no PCR product could be detected in any treatment at day zero, suggesting low background copy numbers of the arrA gene in these samples.

A total of 288 arrA gene clone sequences (at least 45 clones per treatment for each sediment type) were confirmed as arsenate dissimilatory reductase genes by blasting them against the NCBI databases. These sequences, along with the other 156 arrA gene sequences available in the GenBank database, from cultured as well as uncultured organisms, were aligned and clustered into OTUs/phylotypes. Clustering of a total of 444 arrA gene sequences at different percent similarity levels, i.e., 99, 97, 95, 90, 85, and 80%, resulted in 327, 87, 73, 41, 32, and 21 OTUs, respectively. In general, the differences among treatments were consistent regardless of the percent similarity cutoff values used (data not shown). For ease of presentation, we have reported only the results of the DNA sequences at the 90% similarity level (Fig. 5), even though cutoff levels as low as 85% similarity have been used for the clustering of this gene (27). Overall, the 288 arrA gene sequences from this study could be grouped into 14 distinctive phylotypes (clusters 1 to 14) and four singletons (Fig. 5). These phylotypes, consistently supported through high bootstrap values with different phylogenetic methods, were distinctively different from the arrA gene sequences obtained from GenBank for both cultured and uncultured organisms. All arrA gene sequences of uncultured organisms from GenBank could be grouped into nine major clusters/phylotypes (clusters A to E) (Fig. 5; see also Table ST4 in the supplemental material). The details of these sequences and the sites they were retrieved from are presented in Table ST4. All pure cultures except one Geobacter strain (AB-769875) were grouped together, and none of our sequences were related to them at the 90% DNA similarity level (Fig. 5).

FIG 5.

Maximum-likelihood phylogenetic tree based on partial sequences (630 bp) of the arsenate respiratory reductase genes (arrA) of the 288 gene clones from the libraries generated from microcosms and of 146 and 10 arrA gene sequences from uncultured organisms and pure cultures, respectively, from GenBank. Sequences belonging to the same OTU (at 90% similarity), as assigned by Mothur, were clustered together, and branches within clusters were collapsed to show the overall relationships of the clusters to one another. The size of the triangle is proportionate to the number and variation of sequences within a cluster. Asterisks at nodes reflect bootstrap support values above 70% or posterior probability values from at least three of the four phylogenetic methods maximum likelihood, maximum parsimony, neighbor joining, and Bayesian analysis. Clusters labeled 1 to 14 are represented by the arrA gene sequences from the current study, and clusters A to E are from the GenBank sequences from previous studies, with the detection site and number of clones indicated. Numbers for each cluster (clusters 1 to 14) in the corresponding table represent the numbers of arrA gene clones (%) from groundwater/groundwater-plus-glucose treatments (GW/GW+G). The first three columns are for the three sites (NP1, NP3, and NP8), and the fourth column represents the total number of the clone sequences in that cluster. The arxA gene from Magnetospirillum was used as an out group, as previously used for the arrA gene (23, 26).

Biostimulation with glucose altered the dissimilatory arsenate-respiring microbial communities considerably compared to the GW-treated samples. About 54% of the arrA gene sequences (78 clones) from all three sites clustered together in phylotype 6 following glucose amendment (Fig. 5), which corresponds well with the dominance of Clostridium-related sequences identified based on the 16S rRNA gene sequencing.

The rarefaction curve analysis indicates that the number of OTUs did approach an asymptote for all samples at the 90% similarity level, which suggests that additional sequencing effort would have not impacted these findings at the major group level (see Fig. SF6 in the supplemental material). However, at the lower taxonomic level (99% similarity), the dissimilatory arsenate-respiring microbial community may have been undersampled (see Fig. SF6), and additional sampling could have captured more diversity.

DISCUSSION

Even though the organic carbon (OC) content of these aquifer solids was low, ranging from 0.09 to 0.31% of total sediment dry weight, and the added groundwater had 6.9 mg/liter OC, the carbon and energy resources were adequate for As and sulfate reduction in the samples treated with GW alone. Many areas of the world with As-contaminated groundwater, including West Bengal (25, 50), Cambodia (51), and Hungary and Romania (52), have low OC in both the groundwater and the aquifer solids. Several microcosm studies also reported the observation that low-OC solids can support microbial reduction of As (25, 50, 53). In the current study, the low bioavailability of mineral Fe^III in these OC-limited systems revealed that As reduction was independent of Fe reductive dissolution. This was more evident in the reduced sediment, NP8, which contained no chemically defined reducible Fe^III minerals (Table 2); As reduction was independent of Fe reduction in both stimulated and nonstimulated systems.

Microbial community characterization based on the 16S rRNA gene sequencing for these aquifer sediments showed that the addition of GW promoted bacterial community shifts characterized by an increase in sequences related to members of the Proteobacteria genera Desulfocapsa, Desulfovibrio, Geobacter, and Shewanella and the Aquificae-related Desulfurivibrio, which include organisms well characterized for As, Fe, and sulfate reduction (25, 54–59). Interestingly, despite the dominance of these well-characterized Fe reducers (especially Geobacter and Shewanella), the reduction of As was independent of any Fe reduction under C-limited conditions, suggesting that the reduction of As occurred primarily through direct mechanisms. Previously, the abundance of these groups has been reported in response to the addition of a labile C source (10, 23, 51). Parallel arrA gene sequencing from the microcosms of these carbon-limited sediments suggested the presence of diverse dissimilatory arsenate-respiring bacteria, and the dominant phylotype/cluster was unique for each sediment type (Fig. 5). This implies that the reduction of As^V under C-limited conditions was carried out by distinct microbial groups (Fig. 5).

In contrast to GW treatment, the addition of glucose in the NP1 and NP3 sediments promoted reduction of As^V concurrent with reductive dissolution of Fe (Fig. 2A and B), supporting the model of coupled reductive dissolution of iron minerals and As release to solution (10, 27, 35, 60–65). A strong correlation between the increases in the concentrations of Fe^II and As^III for the NP3 sediment amended with glucose suggests that the reduction of As^V was likely due to the direct activity of DARB on both adsorbed arsenic and that released into solution by the reduction of Fe^III. This site had the highest concentration of As^V, and most of it was associated with iron oxides and amorphous iron (Fig. 1). The microbial community responded to biostimulation with a significant increase in the Clostridium-related sequences, which was correlated with the higher rates of both As and Fe reduction, predominantly in the microcosms established from the sediment samples collected from the oxidized-sediment zone (NP1 and NP3). The dominance of the Clostridium-related sequences suggests a possible role of this group in both As and Fe reduction. Further studies using isolation and/or single-cell genomics of these organisms would broaden our knowledge about the exact role and mechanism of As reduction by these bacteria. Previous studies based on the single-strain inoculation of different Clostridium strains have shown that, under fermentative conditions, they can reduce As^V through dissimilation, by respiring both Fe^III and As^V as terminal electron acceptors (66), and via detoxification mechanisms (6, 67). In the current study, the reduction of As occurred primarily through the dissimilation mechanism, as evidenced by an increase in the arrA gene at day 54. Dissimilatory arsenate reductase gene sequencing from these biostimulated samples showed a shift in the functional community dominance toward phylotype/cluster 6 (Fig. 5), with 54% of the total arrA gene sequences from the three sediments grouped in that cluster. This occurred in concert with the dominance of Clostridium-related sequences identified using 16S rRNA gene sequencing. We cautiously speculate that the phylotype/cluster 6 organism or organisms belonged to the Clostridium group and that they served an active role in the reduction of As and Fe in the biostimulated samples. This requires confirmation through the isolation of these organisms in pure cultures, followed by sequencing of both 16S rRNA and arrA genes.

In the current study, our assumption that a significant increase in As^III in both treatments was due to the microbially mediated reduction of As^V was based on two observations. First, the addition of glucose resulted in enhanced activity of both dissimilatory As^V- and Fe^III-reducing microbial communities, based on the observation that higher As^V reduction occurred in the glucose-amended samples than in those treated with groundwater alone. Second, for all treatments, amplification of the arrA gene in the microcosms after 54 days of enrichment but not at the time of initiation (day zero) indicated an increase in the population of DARB over time. It is noteworthy that we observed the presence of microbes in our autoclaved biological control samples, which suggests that our sterilization effort was not successful, and hence, we cannot quantify any abiotic reduction of As^V in these systems. Work by others using a similar setup has reported minimal to no reduction of As^V in their sterilized control samples (23, 28, 61). However, limited abiotic reduction of As^V under specific conditions has also been reported (63, 68, 69).

Overall, the broad-spectrum differences observed in the structural and functional microbial communities for both treatments in the three different sediments are likely to be due to the differences in physiochemical conditions and carbon amendments. Others have reported that soil physiochemical conditions and carbon addition can directly influence the soil microbial community composition (10, 70, 71). The observed decreases in the overall bacterial species richness and diversity in the microcosm samples in response to enrichment and biostimulation can be explained by the selective growth (48) and resource heterogeneity (34, 72) hypotheses.

Interestingly, the characterization of the functional microbial community based on the arrA gene sequencing at the 90% DNA similarity led to the clustering of 288 sequences into 14 unique phylotypes and four singletons (Fig. 5). None of these sequences were related to any of the previously sequenced arrA genes from diverse environments, including West Bengal (25), Chesapeake Bay (24), and Crosswick Creek, Coastal Plain, NJ (28). This suggests a unique biogeographic distribution of DARB at the Cache Valley site (Fig. 5; see also Table ST4 in the supplemental material).

Like As and Fe reduction, differences in sulfate reduction were also observed, and the microcosms treated with GW progressed to sulfate-reducing conditions faster than those that received the GW+G treatment. The higher rates of sulfate reduction development correlate with the dominance of the well-characterized sulfate-reducing microorganisms, including Desulfovibrio, Desulfurivibrio, Desulfocapsa, and Desulfosporosinus (57) under carbon-limited conditions. In contrast, the limited sulfate reduction observed in the biostimulated microcosms could be due to the intermediate products of fermentation that act as electron shuttles and chelators enhancing the bioavailability of Fe minerals (35). The addition of glucose to any of the aquifer solids increased the bioavailable iron (Fig. 1A to C). As this bioavailable Fe continued to be utilized by microbes, sulfate reduction was limited (see Fig. SF1 in the supplemental material). Previously, it has been suggested that in the presence of abundant ferric minerals, iron-reducing bacteria can outcompete sulfate-reducing microorganisms by reducing the availability of electron donors for sulfate reduction (73).

We saw that the concentrations of reduced arsenic either decreased or reached a steady state after an initial increase (Fig. 2). A decrease in As^III with time could be due to its sorption or precipitation onto non-HCl-soluble minerals. HCl extraction was used to define As^III associated with mineral surfaces or As^III in various minerals soluble in 0.5 M HCl. With incubation, As^III may precipitate with sulfide, forming minerals not soluble in this acid extract. Precipitation of arsenic as As-sulfide phases is an important sink for As^III in reduced environments (25, 63, 73, 74).

In conclusion, we observed both As and Fe reduction in microcosms with relatively low concentrations of natural organic matter. This is consistent with observations from aquifers with low organic matter at other, widespread locations. In these low-carbon sediments, As reduction was independent of Fe^III reduction despite the dominance of bacterial genera, including Desulfocapsa, Geobacter, and Shewanella, that have members well characterized for their role in Fe reduction. The coupled, reductive dissolution of Fe minerals and As release in glucose-stimulated, oxidized sediments (NP1 and NP3) concomitant with increased densities of Clostridium populations suggests that these bacteria, which are generally considered to have fermentative metabolism, may be responsible for this activity.

Within the relatively small area of the Cache Valley Basin (∼6.4 ha) from which samples were collected for this study, the guild of arsenate-respiring genes and, probably, the bacteria that carried them had unique phylotypes depending on the sediment geochemical characteristics. This site had at least 14 distinctive phylotypes that have not been reported in previous arrA gene-based studies.