Mercury and Other Heavy Metals Influence Bacterial Community Structure in Contaminated Tennessee Streams

Abstract

High concentrations of uranium, inorganic mercury [Hg(II)], and methylmercury (MeHg) have been detected in streams located in the Department of Energy reservation in Oak Ridge, TN. To determine the potential effects of the surface water contamination on the microbial community composition, surface stream sediments were collected 7 times during the year, from 5 contaminated locations and 1 control stream. Fifty-nine samples were analyzed for bacterial community composition and geochemistry. Community characterization was based on GS 454 FLX pyrosequencing with 235 Mb of 16S rRNA gene sequence targeting the V4 region. Sorting and filtering of the raw reads resulted in 588,699 high-quality sequences with lengths of >200 bp. The bacterial community consisted of 23 phyla, including Proteobacteria (ranging from 22.9 to 58.5% per sample), Cyanobacteria (0.2 to 32.0%), Acidobacteria (1.6 to 30.6%), Verrucomicrobia (3.4 to 31.0%), and unclassified bacteria. Redundancy analysis indicated no significant differences in the bacterial community structure between midchannel and near-bank samples. Significant correlations were found between the bacterial community and seasonal as well as geochemical factors. Furthermore, several community members within the Proteobacteria group that includes sulfate-reducing bacteria and within the Verrucomicrobia group appeared to be associated positively with Hg and MeHg. This study is the first to indicate an influence of MeHg on the in situ microbial community and suggests possible roles of these bacteria in the Hg/MeHg cycle.

Contamination of surface waters with mercury (Hg) and uranium (U) poses environmental and human health concerns. Uranium contamination results from milling processes for nuclear weapons and reactor fuel. Soluble U(VI) leaches from the mill tailings and migrates to surface water bodies, where it can biomagnify in food chains (76, 79). Reduced uranium [U(IV)] is of lesser concern because it can be immobilized in subsurface sediments (1, 15, 27, 54, 55). Microorganisms capable of U(VI) reduction include fermenters such as Clostridium sp. (32), Fe(III) reducers (IRB) such as Shewanella putrefaciens (58; R. Ganesh, K. G. Robinson, and G. D. Reed, presented at the 50th Purdue Industrial Waste Conference, 1995) and Geobacter spp. (13, 20, 53), and sulfate-reducing bacteria (SRB) (36, 55).

While the primary uses of U are limited, Hg has been used extensively in the automotive, electronic, agricultural, dental, and health care industries. The widespread occurrence of Hg in the environment is now a global concern. In nature, Hg is a toxic, rare element (25) that exists as a metal and in both inorganic (e.g., HgS) and organic (e.g., methylmercury [MeHg]) compounds. Total mercury levels in the environment are affected by agricultural and industrial wastes discharged into waterways or released into the atmosphere from coal burning or trash incineration (4, 5, 22, 39). Coal burning for electricity generation in particular continues to release tons of Hg into the atmosphere annually (34, 82). Once deposited into lakes and streams, about 98% of the Hg becomes immobilized in sediments.

Although inorganic mercury is neurotoxic, it does not bioaccumulate, and Hg poisoning is reversible with chelation treatment (16, 72). Methylmercury is several times more carcinogenic and neurotoxic than inorganic mercury (56, 69), and it does bioaccumulate (22, 75, 76, 79). Furthermore, MeHg poisoning is irreversible (22, 75), and it is particularly dangerous because it can cross the blood-brain barrier. The acute health effects include nervous system disease, brain and kidney damage, and damage to respiratory and gastrointestinal systems.

Bacterial mercury resistance has been described for many phyla, such as Firmicutes, Actinobacteria, and Proteobacteria (24, 60). A number of these organisms can reduce Hg(II) to Hg(0) and/or degrade MeHg, but Hg methylation is thus far restricted to SRB and IRB of the Deltaproteobacteria (19, 31, 44, 45). In streams, MeHg is produced under anoxic conditions, and the methylation potential depends on substrate availability and the presence/activity of Hg-methylating bacteria (11, 38). Although research into microbial mercury methylation has been ongoing for several decades, relatively few methylating bacteria have been identified (18, 31, 44-46). Bartha and coworkers were perhaps the closest to determining the genes involved in Hg methylation, using Desulfovibrio desulfuricans LS (10, 17, 18), but this strain was lost.

The Y-12 plant, located in the Department of Energy reservation in Oak Ridge, TN (Fig. 1), was constructed in 1942 to separate uranium-235 (²³⁵U) from the heavier ²³⁸U isotope by electromagnetic separation processes (12). In the years following World War II, the Y-12 plant took on a number of new activities and missions, including separation of naturally occurring stable isotopes of lithium by use of liquid Hg. Between 1950 and 1963, approximately 11 million kilograms of Hg were used at the Oak Ridge Y-12 National Security Complex for lithium isotope separation processes. About 3% of the Hg was lost to the air, soil, and rock under the facilities and to East Fork Poplar Creek, which originates in the plant site. Whereas a decrease in Hg concentration in East Fork Poplar Creek was observed over time, methylmercury concentrations in water and in fish have not declined in response to improvements in water quality and exhibit trends of increasing in some cases (12).

FIG. 1.

Locations of the sampling sites used in this study. The background site (HCK20.6) is located ∼23 km (linear distance) upstream from site EFK23.4. The industrial facility (Y-12 plant) is indicated schematically. (Adapted from reference 64 with permission of the publisher.)

While several studies have been conducted at U- or Hg-contaminated sites, few have examined in detail the microbial community composition, and none have examined the influence that these metals may have on the structure of the microbial community. Identifying the relevant methylating and demethylating populations and determining the influences of Hg/MeHg on community structure remain daunting challenges. The aim of the present work was to characterize the diversity and structure of bacterial populations in heavy metal-contaminated and control streams, with particular attention to the influences of U, Hg, and MeHg. The coupling of detailed geochemistry analyses with 454 pyrosequencing resulted in the identification of individual bacterial groups that may play a role in the mercury methylation process. This report provides the first comprehensive evidence that MeHg generation by the in situ active microbial community appears to directly influence the resultant composition of that microbial community along the mercury/methylmercury gradient.

MATERIALS AND METHODS

Site locations and sample collection.

Stream sediment samples were obtained from six locations situated in or near the Department of Energy reservation in Oak Ridge, TN (Fig. 1). Locations included five contaminated streams: three locations in East Fork Poplar Creek (EFK6.3, EFK13.8, and EFK23.4; the number indicates the stream distance, in kilometers, from the mouth of the stream), one location in Bear Creek (BCK12.3), and one location in White Oak Creek (WCK3.9). Hinds Creek (HCK20.6) was an uncontaminated stream with similar general chemistry and hydrology. The locations were selected for their proximity to sites that have been monitored for >25 years for Hg bioaccumulation and span a range of Hg and other contaminant concentrations. The long data record on bioaccumulation and water and sediment chemistry deepens the potential impact of this study when all data are considered.

Sampling occurred in October and November 2007 and in February, March, May, July, and September 2008. Samples were collected from all six locations, except in October 2007 and February and March 2008, when only EFK23.4 and HCK20.6 were sampled. Two samples were collected per location and per occasion: one from the stream middle and one adjacent to the stream bank (herein referred to as midchannel and near-bank sampling sites). Samples were collected by skimming the upper 2 to 3 cm of sediment with a wide-mouthed container and immediately transferring the material to a sterile wide-mouthed plastic 2-liter bottle. Separate water samples were collected for analysis of dissolved metals and anions, dissolved inorganic carbon (DIC), soluble reactive phosphorus (SRP), dissolved Hg and MeHg, and total sediment Hg. Sediment and water samples were placed on ice until return to the analytical laboratory (<5 h), where they were immediately centrifuged (3,700 × g, 4°C, 30 min). Gravel and pebbles were removed, and fine sediments were frozen (−80°C) until DNA extraction. A total of 60 sediment samples were collected. Water samples for anion or metal analyses were filtered (0.45 μm) or filtered and acidified to a pH of <2 with HNO₃, respectively, and were refrigerated until analysis.

Geochemical and physical parameters.

The anions Cl⁻, NO₃⁻, and SO₄²⁻ were analyzed by ion chromatography (Dionex DX 120 system, IonPac AS12A column, and NaHCO₃/Na₂CO₃ eluent). Twenty-five metals were quantified via inductively coupled plasma mass spectrometry (Perkin-Elmer Elan 600 instrument), including Li, Be, Na, Mg, Al, K, Ca, Cr, Fe, Mn, Ni, Co, Cu, Zn, Ga, As, Se, Sr, Ag, Cd, Cs, Ba, Pb, Bi, and U. DIC was quantified by combustion catalytic oxidation at 680°C (Shimadzu TOC-V CSH instrument). SRP was quantified by the molybdate blue method (3). Total Hg in stream water and sediments was determined using a modification of EPA method 245.7 (30). Mercury was converted to inorganic Hg(II) by oxidation with aqueous BrCl₂, reduced to Hg⁰ by using SnCl₂, and purged with Hg-free air to an RA 915+ Zeeman effect atomic absorption spectrometer. MeHg was analyzed by EPA draft method 1630 (29).

Parameters measured in the field included water pH, temperature, conductivity (Myron L Ultrameter II 6P, calibrated daily), turbidity (Hach 2100P turbidimeter), and dissolved oxygen (DO) (HQ20 Hach portable LDO device). Hourly temperature and precipitation data were obtained from a nearby meteorological station (Y-12 meteorological tower W [west]; lat 35.984664, long 84.265502; altitude, 326 m above mean sea level). The average data for 5 consecutive days before sampling events were used.

Environmental DNA extraction and pyrosequencing of bacterial 16S rRNA genes.

The total community genomic DNA (cgDNA) was extracted from 1 g (wet weight) of sediment by use of a PowerSoil DNA isolation kit (Mo Bio Labs, Inc., Carlsbad, CA). Pyrosequencing of cgDNAs isolated from 59 samples (cgDNA from one sample collected at HCK20.3 in November 2007 was lost during processing) was conducted using the method described at the RDP's (Ribosomal Database Project) Pyrosequencing Pipeline (http://pyro.cme.msu.edu/index.jsp). Briefly, the hypervariable V4 region (∼290 bp) of the 16S rRNA gene was amplified using primers identical to those in RDP's Pyrosequencing Pipeline, containing sequences (adaptors) required for GS 454 FLX pyrosequencing, with the forward primer containing a short tag sequence so that 40 samples could be analyzed in one sequencing run. PCR mixtures (50 μl) consisted of forward and reverse primers (1.5 μl; 10 μM [each]), 1 μl template DNA (10 to 80 ng μl⁻¹), and 0.6 μl (2.5 U μl⁻¹) high-fidelity AccuPrime Pfx DNA polymerase (Invitrogen, Carlsbad, CA). Samples were denatured (95°C, 2 min) and then amplified (30 cycles of 95°C for 15 s, 55°C for 30 s, and 68°C for 45 s), followed by a final extension step (68°C, 3 min). The PCR amplicons were purified using Agencourt AMPure solid-phase paramagnetic bead technology (Agencourt Bioscience Corporation, Beverly, MA). The purity, concentration, and size of the PCR amplicons were estimated using DNA 1000 chips and an Agilent model 2100 bioanalyzer (Agilent Technologies, Inc., Waldbronn, Germany). Sequencing reactions were performed on a Life Sciences GS 454 FLX genome sequencer (Roche Diagnostics, Indianapolis, IN). Raw 454 FLX data (∼235,670 Mb) were initially processed through RDP's Pyrosequencing Pipeline (21). The raw reads were sorted by tag sequence into original samples; the key tag and 16S rRNA gene primers were trimmed off, and low-quality sequences (N > 0; quality score of <20) were removed. The orientation of 16S rRNA gene sequences was checked and reverse complemented if needed. A total of 5,580 to 16,706 high-quality sequences of 200 to 220 bp were obtained per sample, and all sequences were deposited with the NCBI.

Phylogenetic analyses.

Sequences were aligned using the fast Infernal aligner, a SCFG-based, secondary structure-aware aligner (61), and clustered by the complete-linkage clustering (or farthest neighbor) method available at RDP's Pyrosequencing Pipeline (21). Before being submitted to the aligner, all sequences that contained dots or dashes, non-IUPAC characters, or sequences shorter than 150 bases were removed. The Shannon index and the Chao1 estimator of community diversity were calculated for each sample, at a 3% distance (http://pyro.cme.msu.edu/chao1/form.spr). Bacterial 16S rRNA gene sequences were assigned phylogenetically using a naïve Bayesian rRNA classifier, version 2.0, with a bootstrap cutoff of 80% (http://rdp.cme.msu.edu/classifier/classifier.jsp) (80). A graphical representation of taxonomic data was obtained via a heat map created using Genesis 1.0 software (74).

Statistical analyses.

Principal component analysis (PCA) was performed on the geochemical data to identify sample groups (CANOCO 4.5; Microcomputer Power). Constrained ordination techniques were used to identify patterns of 16S rRNA gene sequence variations among sites and correlations between 16S rRNA gene sequence distribution and geochemical measurements. Sequence abundances for each phylum were converted into weight percent values by dividing them by the total abundance for each sample; weight percent values were natural log transformed (ln + 1). Detrended correspondence analysis (DCA), an indirect gradient analysis based on segment length, was performed to determine the modality of the sequence data and environmental predictor variables. The analyses resulted in short (<2.0) segment lengths indicating linear data sets, so redundancy analysis (RDA) and partial redundancy analysis (pRDA) (CANOCO 4.5; Microcomputer Power) were performed. These constrained ordination analyses identified patterns of variation and correlated those patterns to environmental descriptors for 36 samples for which complete geochemical measurements, including Hg measurements, were available. Sequence data were used as the response variables, and the predictor variables were the measured environmental and geochemical parameters. Forward selection of the predictor variables followed by Monte Carlo permutation tests was used to prevent artificial inflation of variation due to autocorrelation in the constrained ordination model (50). Stream water temperature was included as a covariable in the pRDA, as the initial RDA (not shown) indicated that temperature accounted for 8 to 12% of the variation in microbial community structure. Since the sampling dates that have corresponding geochemical data occurred only from May through September (regionally warmer months), we eliminated the variability attributed to seasonality (i.e., stream water temperature) to focus on the geochemical parameters.

RESULTS

Site characteristics and geochemistry.

The streambed at 9 of the 12 sampling sites (EFK6.3 midchannel site, both EFK13.8 sites, both EFK23.4 sites, both WCK3.9 sites, and both HCK20.6 sites) was characterized by gravel and cobble deposits with fine-grained particulates entrained between or layered underneath the larger stones. At the other 3 sites (EFK6.3 bank site and both BCK12.3 sites), the streambed was comprised of poorly sorted sand (<1 mm) and smaller particles. For all sites, only fine sediments were retained for processing and analysis.

The ranges of chemical and physical parameters for the six sampling sites are summarized in Table 1. Metal results are reported only for constituents that exceeded the minimum detectable concentration. All five Hg-contaminated sites, namely, EFK6.3, EFK13.8, EFK23.4, WCK3.9, and BCK12.3, exhibited higher dissolved and sediment-bound Hg concentrations as well as higher MeHg concentrations than those in the Hinds Creek (HCK20.6) control site. East Fork Poplar Creek received large amounts of Hg discharge over ∼13 years, beginning in the early 1950s, and thus is more highly Hg contaminated than White Oak Creek or Bear Creek. While the primary releases of Hg have decreased substantially, diffuse secondary sources continue to add Hg to the stream. White Oak Creek has a similar history, but smaller amounts of Hg were discharged. Additionally, both East Fork Poplar Creek and White Oak Creek receive process water treated with a Zn-containing corrosion inhibitor (Table 1). Corrosion inhibitors are commonly added to cooling water to protect equipment and ensure good heat exchange characteristics of the cooling system. The specific compositions of these additives are proprietary, but they have changed over time from acid chromate and now include zinc pyrophosphates. Greener alternatives to the zinc-based reagents are now being developed and introduced (http://www.prochemtech.com/Literature/Technical/Basic_Cooling_Water_Management_II.pdf). The presence of the corrosion inhibitor did not affect the metals analysis; in other words, there were no positive or negative interferences caused by its presence. A trend of decreasing inorganic mercury (138.28 to 40.1 ng liter⁻¹) but increasing MeHg (0.83 to 2.54 ng liter⁻¹) with increasing distance downstream from the Y-12 plant (sites EFK23.4 and EFK6.3, respectively) was observed (Table 1). Additionally, a uranium and nitrate groundwater contaminant plume is partially intercepted by Bear Creek. The effect of this discharge is observed in higher metal (Na, Mg, Ca, Ba, Sr, and U) and anion (Cl⁻ and NO₃⁻) concentrations, with the conductivity being more than two times greater than that at the other sites.

TABLE 1.

Chemical and physical parameters of the stream waters from each site for the May 2008, July 2008, and September 2008 sampling events^a

Parameter	Range of values for site
	HCK20.6b	EFK6.3	EFK13.8	EFK23.4	BCK12.3	WCK3.9
Element concn (mg/liter)
Na	3.6-9.0	14.9-17.7	8.6-9.2	9.9-10.3	40.5-50.2	11.1-15.7
Mg	14.1-19.0	10.2-12.3	10.9-12.3	11.9-12.9	19.5-25.5	13.1-16.6
Ca	39.4-44.3	41.3-4.60	36.3-43.6	36.5-42.5	128.4-154.4	42.2-48.6
Ba	64.1-99.9	35.6-70.4	45.5-65.4	40.9-56.8	167.9-230.0	47.9-64.2
Sr	0.1-0.1	0.1-0.2	0.1-0.1	0.1-0.2	0.5-0.5	0.1-0.2
Mn	2.0-150.9	24.3-3,770.0	15.9-184.0	24.7-81.3	1.3-15.9	6.3-43.7
Fe	0.1-0.8	0.1-0.8	0.1-1.7	0.1-0.3	0.2-0.3	0.3-0.8
Zn	1.4-8.1	5.6-14.2	3.0-18.5	11.2-33.0	1.5-6.5	12.7-38.5
Al	72.9-1,080.7	68.5-1,320.5	118.4-3,612.4	60.9-260.0	30.3-57.3	198.8-1,076.8
U	0.2-6.9	3.7-10.0	6.0-7.3	0.8-7.4	175.0-268.4	0.2-6.9
Dissolved Hg concn (ng/liter)	0.6-2.3	13.3-40.1	14.1-82.0	54.6-138.3	3.2-5.6	0.0-11.3
Sediment Hg concn (ng/mg)	0.0-0.1	10.6-20.2	11.9-17.7	30.4-47.1	1.4-1.8	2.5-15.1
Dissolved MeHg concn (ng/liter)c	0.0-0.1	0.5-2.5	0.2-0.4	0.4-0.8	0.0-0.1	0.0-0.9
Ion concn (mg/liter)
Cl−	4.0-4.2	10.0-14.0	4.0-6.0	6.0-8.0	4.1-54.1	8.0-16.3
NO3−	0-4.2	8.0-16.6	0.0-6.4	2.0-8.3	266.0-370.7	0.0-10.2
SO42−	16.0-18.1	34.0-46.8	32.0-36.6	34.5-38.0	38.0-50.0	34.2-98.8
DO concn (mg/liter)	7.0-9.0	8.0-8.3	8.4-9.1	9.0-10.1	8.3-9.4	8.3-9.2
DIC (mg/liter)	38.9-60.4	34.3-40.8	34.1-42.2	31.7-40.3	59.0-76.0	34.4-43.1
Conductivity (μS/cm)	348.0-518.0	438.0-626.0	334.0-556.0	340.0-540.0	1,170.0-1,835.0	395.0-592.0
pH	7.8-7.9	7.6-8.0	7.9-8.0	8.1-8.3	7.8-7.9	8.1-8.9
Temp (°C)	14.3-20.7	16.8-21.5	17.1-21.4	18.7-19.9	14.7-20.6	18.7-22.1
Turbidity (NTRUd)	9.9-25.4	4.3-14.0	6.6-23.7	3.6-19.4	1.5-12.6	3.7-5.0

^aSedimentary Hg is the exception, as sediments were analyzed in that case.

^bSite HCK20.6 is considered uncontaminated, while sites EFK6.3, EFK13.8, EFK23.4, BCK12.3, and WCK3.9 are contaminated.

^cOnly May and July 2008 samples were analyzed for MeHg.

^dNephelometric turbidity ratio units.

The geochemistry data were described with two PCA ordination plots, since only 24 samples were analyzed for MeHg concentration. Results for the 36 samples in which Hg was measured are shown in Fig. 2 A, and the variation in geochemical parameters for the 24-sample subset, including Hg and MeHg concentrations, is shown in Fig. 2B. Both PCA scatter plots indicated different geochemical compositions between the uncontaminated (HCK20.6) and contaminated sites. The first PCA axes explain 44.5 and 41.6% of the variation in geochemistry (Fig. 2A and B, respectively) and clearly show that BCK12.3 differs significantly from the other four contaminated sites by elevated concentrations of metals (Mg, Sr, Ba, and U) and anions (NO₃⁻ and Cl⁻), DIC, and conductivity. The second PCA axes explain 22.4 and 25.2% of the site variations attributed to the concentrations of the dissolved and sedimentary Hg (Fig. 2A) and the dissolved and sedimentary Hg along with MeHg (Fig. 2B), respectively. Both uncontaminated Hinds Creek and mildly contaminated White Oak Creek were clearly separated from the more highly contaminated streams (Fig. 2A and B).

FIG. 2.

Scatter plot for PCA of geochemical data from 36 samples over three sampling dates (A) and from a 24-sample subset of data that includes MeHg data (B). The percentage of total variance explained by each axis is noted in the axis label. Individual parameters (r > 0.6) significantly associated with the variation are represented along each axis. Open symbols represent midchannel samples, and closed symbols represent near-bank samples.

Description of microbial community.

Total cgDNA isolated from the 59 samples yielded DNA concentrations ranging from 0.38 to 14.5 mg g⁻¹ sediment. The theoretical size of the prokaryotic cell population based on the total DNA recovered ranged from 9 × 10⁷ to 3.62 × 10⁹ cells g⁻¹ of wet sediment. These calculations were based upon the predicted effective genome size of 4.7 Mb for the soil bacterial/archaeal population (2, 65) and on the weight of 4.05 fg of a genome of this size (28). We did not include the eukaryotes in the population size estimate; however, inclusion of the eukaryotic component would reduce the cell population by 25% (65). The total bacterial community contained 23 phyla and unclassified bacteria (Fig. 3 ) obtained from 235 Mb of sequencing data. Sorting and filtering of the raw reads resulted in 588,699 high-quality sequences with lengths of >200 bp. These sequences were organized into a local BLAST database and are available for BLASTn search (http://genome.ornl.gov/∼cdx/form.html). According to the RDP Classifier, bacterial sequences constituted 99.84% (587,719 sequences) of the data set, while 0.16% (927 sequences) and 0.008% (51 sequences) of the sequences were assigned to the unclassified root and to archaea, respectively. The group of unclassified bacteria, which included sequences with <80% confidence for relationship to any known bacterial phylum, constituted 23.55% (range, 13.16% to 39.21%) of the community (138,410 sequences).

FIG. 3.

Heat map representation of phylogenetic data. The 16S rRNA gene sequences were assigned to phylogenetic bacterial taxonomic groups based on a naïve Bayesian rRNA classifier with an 80% confidence threshold, using RDP Classifier. The sample names are abbreviated as follows. The first letter or letter and number correspond to the site, as follows: H, HCK20.6; E1, EFK6.3; E2, EFK13.8; E3, EFK24.3; B, BCK12.3; and W, WCK3.9. The following letters MC or NB indicate a midchannel or near-bank sample. The last letter(s) indicates the month of collection, as follows: O, October; N, November; F, February; MR, March; M, May; J, July; and S, September.

Eleven phyla present in all samples included Proteobacteria (ranging from 22.95 to 58.33% of the community in each sample), followed by Acidobacteria (1.67 to 30.64%), Verrucomicrobia (3.42 to 31.03%), Cyanobacteria (0.22 to 32.01%), Bacteroidetes (0.03 to 14.42%), Chloroflexi (0.38 to 6.32%), Planctomycetes (0.03 to 3.92%), Gemmatimonadetes (0.05 to 1.34%), WS3 (0.01 to 1.17%), OD1 (0.04 to 0.80%), and TM7 (0.01 to 0.29%). The Firmicutes (0.03 to 5.26%), Actinobacteria (0.01 to 0.80%), and Chlamydia (0.01 to 0.25%) were found in all samples except those collected from the most Hg-contaminated site (EFK23.4) in March, February, and September 2008. The remaining 9 phyla were detected at lower abundances (0 to 0.41%) and in fewer samples (8 to 56 samples) (Fig. 3). There was no apparent relationship between the number of groups detected per sample and the site or collection time. The Proteobacteria were detected in all samples and were the most abundant. They were apportioned as Betaproteobacteria (8.60 to 30.70%), Alphaproteobacteria (1.13 to 24.78%), Gammaproteobacteria (3.33 to 16.21%), Deltaproteobacteria (0.31 to 3.47%), and unclassified Proteobacteria (1.66 to 5.52%), with members of the Epsilonproteobacteria being the least common (0 to 0.28%), found in only 35 (59%) of 59 samples (Fig. 3).

Bacterial community composition in relation to stream geochemistry.

Since geochemistry data that included Hg were obtained only for samples collected in May, July, and September 2008 (Table 1), statistical analysis between bacterial community structure and stream geochemistry was based on this 36-sample subset. The subset contained 359,307 sequences (61.03%) of the total of 588,699 sequences (from all 59 samples), with 358,689 (99.83%) of the sequences assigned to Bacteria at the ≥80% confidence threshold. The unclassified root and archaeal sequences numbered 587 (0.16%) and 31 (0.009%), respectively. Diversity indices calculated at a 0.03 distance level for each of 36 samples ranged from 6.0 to 7.2 (Shannon index) and from 1,567.2 to 4,166.9 (Chao1 index). The diversity indices for midchannel and near-bank samples from East Fork Poplar Creek varied slightly for each site and sampling event (Table 2).

TABLE 2.

Diversity indices for samples collected in May, July, and September^a

Index	Collection date (mo)	Diversity index
		HCK20.6		WCK3.9		BCK12.3		EFK6.3		EFK13.8		EFK23.4
		MC	NB	MC	NB	MC	NB	MC	NB	MC	NB	MC	NB
Shannon	May	6.29	5.98	6.41	6.25	6.39	6.63	6.37	6.28	6.44	6.40	6.21	6.16
	July	6.68	6.88	6.12	6.41	6.82	6.76	6.52	6.57	6.77	6.62	6.92	6.67
	September	6.71	6.82	6.73	6.58	6.88	6.78	6.73	6.87	7.24	6.89	6.33	6.38
Chao1	May	2,371	1,887	2,520	1,567	1,811	2,840	3,193	3,427	2,582	2,870	2,135	2,294
	July	3,041	3,816	2,181	3,178	3,110	3,185	3,151	3,261	2,995	2,359	3,057	2,976
	September	3,094	4,167	3,462	3,439	2,961	2,765	3,069	3,768	4,088	3,705	2,819	2,811

^aDiversity indices were calculated at a 3% distance. MC, midchannel sample; NB, near-bank sample.

Canonical axes 1 and 2 in the pRDA graph describe 17.3 and 2.5% of the variation in the bacterial community structure, respectively (Fig. 4). Stream water temperature explains an additional 8.8% of the variation (P = 0.002; F = 6.36). This triplot of samples, geochemical variables, and bacterial phylogenetic taxonomic groups indicates significantly different microbial communities in the sediments of Bear Creek from those at the other sites. The Bear Creek samples correlated with an increase in stream water U concentration. Uranium and other correlated geochemical parameters differentiated BCK12.3 from the other sites (Fig. 2A and B), and its geochemistry appeared to positively influence the presence of Chlamydiae and unclassified Bacteria. Many of the geochemical parameters measured in this study were autocorrelated, so isolation of the influences from individual geochemical parameters was not always possible. Forward selection of the independent variables followed by Monte Carlo permutations during the RDA selected the most significant (P < 0.05) environmental variable from the group of correlated variables and discarded the remainder. For example, U is important for describing the influence of particular geochemical parameters on a given subset of the microbial community (Fig. 4). However, since U was significantly correlated with NO₃⁻, Ca, Ba, Cl⁻, Mg, Sr, DIC, and conductivity values (r > 0.75), each of these parameters showed a similar statistical influence on the subset of the community.

FIG. 4.

Triplot of RDA for bacterial phyla from the 36 stream sediments at six sites located on or near the Oak Ridge Reservation, with forward selection of predictor variables followed by Monte Carlo permutations. Solid arrows represent predictor (geochemical) variables significantly associated (P < 0.05) with the variation in the bacterial community structure. Dashed arrows represent individual phyla (r > 0.6) significantly associated with the variation among samples. The length of each arrow is correlated with the degree of relationship between the response variables. The arrows point in the direction of the maximum change for the associated variable. Open symbols represent midchannel samples, and closed symbols represent near-bank samples.

Similarly, pH was positively associated with the Gemmatimonadetes and Cyanobacteria, but since decreasing pH significantly correlated with SO₄²⁻ concentrations, the latter had a comparable effect. Deltaproteobacteria and Chloroflexi were associated with decreasing pH values. Dissolved Hg (and the autocorrelated variable sediment Hg) displayed a significant (P = 0.002) correlation with the Verrucomicrobia. Seasonal variation at the Hg-, U-, and NO₃⁻-contaminated BCK12.3 stream was evident. May samples displayed higher abundances of Proteobacteria, Actinobacteria, Planctomycetes, and TM7, whereas Gemmatimonadetes were more abundant in July. Chlamydiae and Acidobacteria levels were highest in the September samples (Fig. 3). Conversely, bacterial communities in the uncontaminated and Hg-contaminated May samples were characterized by an increase of Verrucomicrobia; however, July samples showed increases in Deltaproteobacteria, Chloroflexi, and unclassified Bacteria (Fig. 3 and 4).

Influence of total Hg and MeHg on microbial populations.

One of the most novel findings in this study was that certain bacterial groups are positively correlated not only with Hg but also with MeHg concentrations. pRDA canonical axes 1 and 2 (Fig. 5) describe 29.4% of the variation in the bacterial community from the 24-sample subset that included MeHg data. Stream water temperature explains an additional 12.3% of the variation (P = 0.01; F = 3.27). We observed a similar cluster pattern for the set of 36 samples presented in Fig. 4. Again, Bear Creek was the most distinct from the other sites, primarily because of the strong positive (P = 0.01) correlations between U and NO₃⁻ concentrations and abundances of Chlamydiae and unclassified Bacteria. Samples from East Fork Poplar Creek (EFK13.8 and EFK23.4 [nearest to the contamination source]) and White Oak Creek (WCK3.9; collected in May) were significantly (P = 0.01) associated with dissolved Hg concentrations, displaying higher abundances of Deltaproteobacteria. The Deltaproteobacteria were present in all samples, although they made up a very small portion (0.75 to 3.47%) of the total bacterial community. The samples furthest downstream from the contamination source in East Fork Poplar Creek (EFK6.3) displayed the highest MeHg concentrations and the highest abundances of Verrucomicrobia and Epsilonproteobacteria (Fig. 5). Overall, the abundances of Epsilonproteobacteria were low, ranging from 0.00 to 0.28%, and these bacteria were detected in only 23 (64%) of 36 samples. The abundances of Verrucomicrobia varied among samples, from 4.70% to 31.02%, with the lowest abundances in the U- and NO₃⁻-contaminated BCK12.3 samples (5.21 to 9.15%), and were slightly higher at the reference HCK20.6 site (11.42 to 18.96%). However, the highly Hg-contaminated areas of East Fork Poplar Creek (3 sites) and the less-contaminated WCK3.9 site displayed greater abundances of Verrucomicrobia (4.70 to 31.02% and 11.57 to 25.60%, respectively), although they did vary between sampling times.

FIG. 5.

Triplot of RDA for bacterial phyla of 24 stream sediments from six sites located on or near the Oak Ridge Reservation, TN, with forward selection of predictor variables followed by Monte Carlo permutations. Solid arrows represent predictor (geochemical) variables significantly associated (P < 0.05) with the variation in the bacterial community structure. Dashed arrows represent individual phyla (r > 0.6) significantly associated with the variation among samples. The length of each arrow is correlated with the degree of relationship between the response variables. The arrows point in the direction of the maximum change for the associated variable. Open symbols represent midchannel samples, and closed symbols represent near-bank samples.

DISCUSSION

The goals of this study were to determine the microbial community structures at Hg- and U-contaminated areas and to determine associations between these communities and their geochemical profiles. Furthermore, we tested if there was a relationship between particular members of the microbial community and MeHg, particularly since no such relationship has been determined previously. In general, the results indicated that the composition of the streambed bacterial community was associated with season and stream water geochemistry. Previous studies have shown seasonal effects on MeHg production in sediments, with the highest rate occurring in the summer, and have linked the methylation process to increasing microbial activity, which is usually enhanced at higher temperatures (38, 73). Overall, higher abundances of Deltaproteobacteria, OD1, Chloroflexi, and unclassified bacteria were observed on the warmer sampling dates, while in cooler months an increase of Gammaproteobacteria and Cyanobacteria was evident. These results are not surprising in that a semiannual shift of ∼15°C could produce such differences in diversity. The other likely contributing factor is that during the colder months, there is less leaf cover in the canopy of this heavily forested area, and thus there is a larger amount of sunlight for the photosynthetic Cyanobacteria. Whether this also explains the increase in the Gammaproteobacteria is unknown at this time.

While Bear Creek also experienced temperature fluctuation, it was distinct from the other sites, not only geochemically but also in the composition of the microbial community. Given the high levels of U, Hg, and NO₃⁻ in Bear Creek, it is not surprising that these parameters had a significant and complex influence on the microbial community. Based upon the microorganisms known to reduce NO₃⁻ and/or U(VI), it could be assumed that the Deltaproteobacteria would have been more prevalent here than at the other sites (1, 15, 27, 32, 33). However, this group was consistently less abundant (Fig. 3). Instead, Alphaproteobacteria and unclassified members of the Proteobacteria correlated positively with these contaminants. The Alphaproteobacteria class includes bacteria typically found in soils but not usually associated with U or NO₃⁻ contamination, such as species of Agrobacteria, Caulobacter, or Bradyrhizobium. However, it also includes members of the Sphingomonadaceae family which are known for their biodegradative capabilities (37, 48), including Sphingomonas sp. BSAR-1, which bioprecipitates uranium in alkaline solutions (62). A total of 237 Sphingomonas-like sequences were identified in Bear Creek samples, and future experiments could be targeted to isolate and test the possible contribution of this organism to uranium immobilization.

Other sites contaminated with Hg but not with U or NO₃⁻ displayed overall higher abundances of Deltaproteobacteria, Epsilonproteobacteria, and Verrucomicrobia that were positively correlated with increased Hg and MeHg concentrations (Fig. 4 and 5). These data suggest that these bacteria tolerate Hg/MeHg or may play a role in Hg biotransformations. However, with the exception of the Deltaproteobacteria (9, 26, 44, 47, 51, 69), there is no evidence for other microorganisms having associations with Hg methylation. However, relatively few microbial species (<50) have actually been tested for the ability to methylate Hg, and this remains an area that should be addressed to more fully understand the diversity and abundance of bacteria that can methylate or reduce Hg or demethylate MeHg. The constant decrease of Hg(II) concentrations along the downstream stream course and the concomitant increase in MeHg concentrations over the ∼17-km distance suggest that Hg is both bioavailable and accessible to the resident microorganisms (6-8).

Even though Epsilonproteobacteria have not previously been associated with Hg, the presence of these bacteria was correlated with Hg. Genera in this class include Wolinella, Campylobacter, Arcobacter, and Helicobacter, which are found in diverse environments, such as the mammalian digestive system (http://www.nlm.nih.gov/cgi/mesh/2009/MB_cgi?mode=&term=Epsilonproteobacteria) and extreme environments such as deep-sea hydrothermal vents (59). Other members include sulfur-oxidizing bacteria, such as Lebetimonas acidiphila DSM 16356 and Sulfurimonas denitrificans DSM1251 (71, 77). The sequenced genome of S. denitrificans revealed genes similar to the transposase accessory protein genes found in the mercury resistance (mer) transposons of other bacteria (70), and our 454 pyrosequencing data revealed 197 sequences related to this organism.

Finally, the most abundant phylum in the Hg/MeHg-contaminated sites was the Verrucomicrobia. This little-known group is widely distributed in soils, sediments, and aquatic and gastrointestinal habitats (49, 67, 81) and makes up as much as 12% of soil bacterial communities (68). In comparison, the Verrucomicrobia constituted up to 31% (EFK6.3) of the communities explored in this study. The ecological roles of this phylum remain largely unexplored (35), but its occurrence has been correlated with increased concentrations of chemical elements and nutrients (37). Verrucomicrobia sequences have been recovered from numerous hydrocarbon-, mercury-, uranium-, and pesticide-contaminated environments, suggesting that they are resistant to a number of contaminants, possess metabolic plasticity, and may play a significant role in the decontamination process (14, 57, 66). Culturable representatives display diverse metabolic characteristics and range from aerobic heterotrophs involved in organic carbon transformations (41, 67) to organisms thriving in metal-rich oligotrophic aquatic environments (78). While determining similar physiologies from phylogenetic data by using shorter reads is difficult, overall, 1,116 pyrosequences in this study clustered with known Verrucomicrobia species, using a cutoff of 0.15, with 750 sequences clustered with Opitutaceae sp. TAV2 at a distance level of 0.11. Furthermore, a recently identified novel acidophilic methanotroph, Methylacidiphilum infernorum, clusters phylogenetically closely to Opitutaceae TAV2, which possesses an elaborate system of heavy metal resistance and Hg(II) reduction genes (42). In order to determine if this clustering is the case, future efforts will include isolation of Hg-resistant Verrucomicrobia strains and testing of their resistance to elevated Hg and their ability to transform Hg and MeHg. Given the methanotrophic properties of some phylum members (23, 42, 43, 66), they may also participate in MeHg demethylation in order to use the methyl group as a carbon source via C₁ metabolism. While the organomercurial lysase (merB) is widely known as the methylmercury demethylase, other genes may be responsible. For example, the recently completed genome sequencing of the Hg-methylating and MeHg-demethylating species Desulfovibrio desulfuricans ND132 by our group yielded no mer genes upon the initial annotation (unpublished data).

Mercury contamination and biotransformation to methylated mercury by indigenous bacteria are of great concern globally given the neurotoxic effects to both animals and humans. Efforts such as the present study are working to better understand the diverse bacteria capable of these biotransformations and the influence that cocontaminants and other geochemical parameters have on the generation of MeHg. This study provides the first evidence of a positive correlation between substantial and sustained levels of MeHg and specific members of the in situ microbial community. Given the body of literature showing that MeHg generation is exclusive to the Deltaproteobacteria, it is surprising that this group made up such a small percentage of the microbial community for which higher MeHg values were detected. Just as surprising was the high incidence of Verrucomicrobia and the Epsilonproteobacteria. These results provide a valuable baseline study and suggest that in situ mercury methylation may be more widespread than previously believed, while also suggesting that the Deltaproteobacteria may not be the major methylating organisms within the community.