Environ. Sci. Technol. 2010, 44, 5516–5522
Monitoring Microbial Community Structure and Dynamics during in situ U(VI) Bioremediation with a Field-Portable Microarray Analysis System D A R R E L L P . C H A N D L E R , * ,† ALEXANDER KUKHTIN,† REBECCA MOKHIBER,† CHRISTOPHER KNICKERBOCKER,† DORA OGLES,‡ GEORGE RUDY,† JULIA GOLOVA,† PHIL LONG,§ AND AARON PEACOCK| Akonni Biosystems, Inc., 400 Sagner Avenue, Suite 300, Frederick, Maryland 21701, Microbial Insights, Inc., 2340 Stock Creek Blvd., Rockford, Tennessee 37853, Pacific Northwest National Laboratory, Mail Stop K9-33, Richland, Washington 99354, and Haley & Aldrich, 103 Newhaven Road, Oak Ridge, Tennessee 37830
Received March 1, 2010. Revised manuscript received May 26, 2010. Accepted June 10, 2010.
The objective of this study was to develop and validate a simple, field-portable, microarray system for monitoring microbial community structure and dynamics in groundwater and subsurface environments, using samples representing site status before acetate injection, during Fe-reduction, in the transition from Fe- to SO42--reduction, and into the SO42--reduction phase. Limits of detection for the array are approximately 102-103 cell equivalents of DNA per reaction. Sample-toanswer results for the field deployment were obtained in 4 h. Retrospective analysis of 50 samples showed the expected progression of microbial signatures from Fe- to SO42- -reducers with changes in acetate amendment and in situ field conditions. The microarray response for Geobacter was highly correlated with qPCR for the same target gene (R2 ) 0.84). Microarray results were in concordance with quantitative PCR data, aqueous chemistry, site lithology, and the expected microbial community response, indicating that the field-portable microarray is an accurate indicator of microbial presence and response to in situ remediation of a uranium-contaminated site.
Introduction A universal theme that has emerged from two decades of fundamental microbiology research at contaminated Department of Energy (DOE) sites is the importance of metaland sulfate-reducing bacteria in subsurface bioremediation (e.g., 1-12). A typical strategy for radionuclide bioremediation involves injecting electron donors to stimulate metal reduction by microbial communities native to contaminated * Corresponding author tel: 734-428-0713; fax: 301-698-0202; eFax: 301-542-0120; e-mail:
[email protected]. † Akonni Biosystems, Inc. ‡ Microbial Insights, Inc. § Pacific Northwest National Laboratory. | Haley & Aldrich. 5516
9
ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 44, NO. 14, 2010
aquifers (1, 13), a strategy that has been tested at the DOE Integrated Field Research Center located in Rifle, CO (1, 10). If the fundamental science from DOE field sites is going to be translated into practical cleanup solutions for site reclamation or stewardship, then it becomes important for site engineers to understand which microbial groups are reducing (or reoxidizing) heavy metal contaminants in situ; which electron donors are most effective in stimulating their respective activities; and whether or not underlying changes in microbial community structure through time and space are diagnostic indicators of subsurface biogeochemical processes. Unfortunately, there are still very few methods for assessing in situ microbial community structure, activity, or remediation potential within a time frame or at a price point that impacts on-site treatment, remediation, or long-term stewardship decisions. Culture, phospholipids fatty acid analysis, and various forms of polymerase chain amplification dominate the remediation service industry, each of which provide useful information but suffer from limited coverage of the microbial community and/or ability to provide a more direct indication of remediation potential, activity, or efficacy. At the other end of the technology spectrum are microarrays, which provide unparalleled opportunities for multiplexed detection of nucleic acids and microorganisms. Several groups have now developed microarrays for the analysis of functional genes (14, 15) and mRNA (16) in the environment, 16S rRNA genes (17), and the direct detection of 16S rRNA (18-20). There is also an Affymetrix array containing probes targeting the entire Ribosomal Database Project (21), a system that has been applied to the analysis of microbial communities undergoing U reduction and reoxidation (22). Unfortunately, most of these microarray systems are too expensive, complex, labor intensive, and/or time-consuming (from sample acquisition and sample preparation through to data reporting) to find practical use in the industry, let alone in the field (see, e.g., 23). Rather than using microarrays as a fundamental science and discovery tool, an alternative approach is to take the accumulated biological knowledge of subsurface remediation processes, and use low-density arrays as indicators or diagnostics of microbial community structure and activity. In so doing, there is a corollary opportunity to reduce the financial and logistical burden of molecular monitoring, move array technology closer to the field, and provide information to site engineers in real time. The technical challenge posed by deploying microarrays in the field has been described elsewhere (23). The objectives of this study were therefore to take the fundamental knowledge from prior environmental science efforts, develop and validate a simple-to-use, fieldportable, microarray-based system for monitoring microbial community structure and dynamics, and monitor microbial community succession in response to in situ U bioremediation in a field experiment.
Materials and Methods Site Description and Sample Collection. The Rifle IFC site layout and geochemistry are described in detail elsewhere (1, 10), with the location of monitoring wells from the 2008 “Big Rusty” gallery illustrated in Suppl. Figure 1 (Supporting Information). Injection wells were installed to a depth of 6.1 m (20 ft) and screened from 1.5 to 6.1 m (∼5 to 20 ft) to encompass the entire saturated interval of the aquifer. Acetate injections during 2008 included 14 days of 50 mM acetate (Fe-reduction phase), a 7-day water flush (to slow the onset of sulfate reduction), 10 days of 50 mM acetate (transition 10.1021/es1006498
2010 American Chemical Society
Published on Web 06/18/2010
from Fe- to SO42--reduction), and 59 days of 150 mM acetate (SO42--reduction phase). One- to two-liter water samples were filtered through a Sterivex 0.2-µm filter at each phase and depth interval. Phases of the field experiment were confirmed by geochemical analysis of Fe(II), SO42-, and sulfide levels at all sampling depths (not shown). Microarray Design and Manufacture. The microarray used here (TruArray BER; Akonni Biosystems, Frederick, MD) is based on 16S rRNA-targeted arrays described in detail elsewhere (19, 24, 25), with the following modifications. Capture probe sequences were expanded to include fermentors, dechlorinators, dissimilatory metal-, sulfate-, and nitrate-reducer 16S rRNA sequences deposited in Genbank, utilizing full-length sequences for which an isolate is available in a public culture collection, and for which the link between phylogeny and function has been established via culturebased methods. The array can be used for both direct rRNA detection (25) and DNA detection (this study). In total, 150 thermodynamically balanced microarray probes serve as indicators for 37 genera, as described in Suppl. Table 1 (Supporting Information). Microarrays were manufactured by a single-step copolymerization technique and custom polymer at Akonni Biosystems. Probes were printed in quadruplicate at 0.125 mM concentration. Three complete arrays were printed per substrate and individually fitted with a Grace Biolabs frame seal gasket. Fabricated arrays were stored dry and in the dark under room temperature/ambient conditions until use. Positive Control Isolates. Nucleic acids were purified from Geobacter metallireducens, G. sulfurreducens, Clostridium acetobutylicum, C. glycolicum, Desulfovibrio vulgaris, and Thaurea aromatica according to standard techniques, and quantified by fluorometric analysis on a NanoDrop ND-3300 (Thermo Scientific, Wilmington, DE). DNA was prepared in 10-fold serial dilutions from 1 fg to 100 pg µL-1 in ultrapure water for protocol development and analytical performance evaluations. Nucleic Acid Extraction from Environmental Samples. For the 2008 deployment of the microarray system to the Rifle IFC field trailer, we utilized an Akonni SPT TruTip DNA/ RNA Extraction Kit (Frederick, MD) and MoBio PowerSoil DNA Isolation Kit (Carlsbad, CA) as per the manufacturer’s instructions. For subsequent microarray validation and to correlate microarray signals with quantitative PCR data, all samples were extracted with the MoBio kit. The TruTip has a binding matrix embedded in a 2-mL aerosol barrier pipet tip, such that the only equipment required for nucleic acid purification is a pipettor. Filter samples (representing 1-2 L of groundwater) were excised from their respective cartridges, cut into thin strips with sterile scissors and added in their entirety to a MoBio PowerBead tube containing 60 µL of MoBio lysis solution. After vortexing, the supernatant was divided evenly between the MoBio and TruTip procedures. The elution volume for both methods was 100 µL, representing 10-20 mL equivalent of groundwater per microliter purified extract. Purified nucleic acids were either used immediately (in the field deployment) or stored at -20 °C until use (validation study). Nucleic Acid Amplification and Labeling. For the infield experiment, each of two replicate extracts per filter sample was amplified in an individual 50 µL reaction; for the validation study, three separate amplifications were performed from each individual extract. Optimized amplification conditions were 1X Phusion Hot Start High Fidelity DNA Polymerase PCR buffer (New England Biolabs, Ipswich, MA), 0.25 mM of each dNTP, 10 µL of TruArray BER asymmetric primer mix, 1 µL of Phusion Hot Start High Fidelity DNA polymerase (New England Biolabs), and 3 µL of MoBiopurified nucleic acid (30-60 mL equivalents groundwater). The asymmetric primer mix contains one labeled universal
16S rRNA primer and one unlabeled universal 16S rRNA primer in unequal ratios, which leads to a nonexponential amplification reaction and accumulation of a single-stranded, labeled product that can be hybridized directly to the microarray without additional fragmentation, labeling, or purification. PCR amplification was conducted in a Piko thermal cycler (Finnzymes, Woburn, MA) at 98 °C for 30 s, 25 cycles of [98 °C for 5 s, 59 °C for 5 s, 72 °C for 5 s], 35 cycles of [98 °C for 5 s, 64 °C for 5 s, 72 °C for 5 s], and a final extension at 72 °C for 1 min. Quantitative PCR. Quantitative PCR for specific bacteria or groups was based on Microbial Insights’ series of CENSUS qPCR assays, utilizing 3 µL of MoBio-purified environmental nucleic acid and PCR tests for eubacteria, iron- and sulfatereducing bacteria, Geobacter, and dissimilatory sulfite reductase. Results are presented as cell equivalents per mL of groundwater. Microarray Hybridization and Wash. For each replicate amplification, 15 µL of (unpurified) amplified product was brought to 30 µL total volume with microarray hybridization buffer, 1 µL of BSA (100 mg mL-1), 0.8 nM Cy3-labeled TruArray BER internal positive control, and ultrapure water. The internal positive control is a synthetic 92-mer with no known homology to any sequence in any database, and serves as a positive control for target hybridization only. During assay development and verification, experiments were also performed with chaperone/helper probes (one for each genus represented by the array) based on prior work (19, 24, 25) that showed chaperone/helper probes improve (planar- and bead-) array hybridization specificity and sensitivity by disrupting secondary and tertiary structures in the 16S rRNA target. Target nucleic acids were heat denatured for 3 min at 93 °C and 28 µL was applied to the array. The array was sealed with parafilm and then statically incubated for 3 h at either ambient temperature (20-24 °C, field experiment) or overnight at 37 °C in an MJ Research in situ tower (validation study). After hybridization, parafilm covers were removed, arrays were placed in a histology slide holder and washed in a bulk container containing 500 mL of 6X SSPE, 0.05% SDS for 10 min with intermittent agitation. Thereafter, the arrays were sequentially transferred through two containers of ultrapure water. Excess liquid in the frame seal chamber was removed with a Kimwipe, and arrays were air-dried (∼5 min) before imaging. Imaging and Data Analysis. Dried arrays were imaged on an Akonni field-portable TruDx 1000 imager for up to 10 s. Exposure time was adjusted to avoid pixel saturation. Images were saved as 16-bit raw .tif files and exported to Spotfinder v3.1.1 software. A custom-written PERL script automatically applied an analysis grid to the image, with the operator verifying that the grid was properly aligned prior to data extraction. Data extraction used an Otsu threshold for defining foreground from background pixels, and local background was subtracted from each gel element pixel. Integrated, background-corrected signal intensities were averaged across all replicate probes (including the nonsense probes, N), and a signal-to-noise ratio (SNR) was calculated relative to the average, background-corrected nonsense response. SNR data for the n ) 2 (field trial) or n ) 3 (validation) replicate arrays were then averaged, resulting in an average microarray profile for each sample. For the purpose of this study, we used an average SNR > 3 for declaring probes detectable over all sources of noise. Microarray data and profiles were visualized as “heat map” displays using TreeView v1.6 (26). For quantitative comparison to qPCR data, SNR values for all probes within a functional group were first summed and then log10transformed. Other comparisons and data analysis methods are described in text. VOL. 44, NO. 14, 2010 / ENVIRONMENTAL SCIENCE & TECHNOLOGY
9
5517
Results and Discussion Analytical Sensitivity and Specificity. Some of the assay optimization data are shown in Suppl. Figure 2 (Supporting Information) for a 3 h static hybridization at 37 °C. Limits of detection (LoD) for the microarray method against a G. metallireducens genomic DNA target were 10 fg (approximately 2 cell equivalents) per reaction (Suppl. Figure 2A and B). Chaperone probes in the hybridization mix led to a general decrease in microarray LoD (from 10 fg to 1 pg), as shown in Suppl. Figure 2B for all Geobacter-specific probes. Limits of detection for other positive control isolates were generally 100 fg to 1 pg per reaction in the absence of chaperone/ helper probes (not shown). We therefore estimate the LoD to be between ∼100 and 1000 genomes per reaction for a 3 h hybridization protocol, a result that is comparable with realtime PCR data, known copy-number requirements for reproducible PCR, and intrinsic biomass of subsurface environments prior to in situ remediation. The microarray response (the totality of detectable signals over all probes) was fairly specific to the target organisms (Suppl. Figure 2C and 2D). At 100 pg DNA per reaction and in the absence of chaperone probes, strong cross-hybridization (SNR > 10) beyond the targeted genus was only observed for a few probes, as described in the Suppl. Figure 2C legend. No template reactions were negative over all probes or only intermittently positive at a SNR > 3, including all tests conducted in the field. As in prior studies with 16S rRNAtargeted arrays (19, 24), chaperone/helper probes improved the analytical specificity of the array response, but did not entirely eliminate the observed cross-hybridization (not shown). We therefore chose to exclude chaperone probes from the method and analysis of environmental samples, favoring the improvement in LoD over some loss in analytical specificity The limited extent of cross-hybridization is not necessarily a technology flaw or impediment to diagnostic use in an environmental setting, as discussed in detail elsewhere (25). For community profiling applications, we advocate that individual probes are simply indicators of the cognate gene or organism, that correlated samples are required for interpreting microarray data, and that it is the relative change in probe A versus probe A across the correlated sample set that carries biological or ecological information. In the event that a specific “name” of an organism or group of organisms is important to the user, then the microarray result should be followed by a targeted qPCR assay. Thus, the analytical data are consistent with prior 16S rRNA-targeted arrays, and display analytical sensitivity and specificity on par with more complex, lab-based microarray analysis systems (e.g., 15, 21, 22, 27). Field Deployment. The microarray system was deployed to the Rifle IFC field trailer for a one-week trial in July 2008. Approximately 100 microarray analyses were conducted by a single technician over a 4-day period. Some data for the MoBio and Akonni sample preparation comparison are shown in Figure 1 for background well U01 and downstream well D01 at t ) 0 and t ) 4 days. Average SNRs for these data ranged from 3.1 to 64.2. No template reactions were either negative for all probes or sporadically positive for random probes (i.e., no consistently positive probes for any no template reaction; not shown), indicating no cross-contamination between samples (or from other 16S rRNAtargeted PCR amplification activity occurring simultaneously in the field trailer). Relative to the MoBio sample preparation method, TruTip-purified nucleic acids showed an increase in the number of reactive probes and the detection of the expected microbial community shift toward Fe reduction at t ) 4 days. We naturally expect that different sample preparation methods will lead to different lysis, extraction, and purification efficiencies and, ultimately, molecular 5518
9
ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 44, NO. 14, 2010
FIGURE 1. In-field microarray analysis of microbial community response to acetate injection, showing an expected bloom in Geobacter at t ) 4 days. Only average signal-to-noise ratios (SNR) > 3 are plotted in the heat map. Band intensity is correlated with the absolute average SNR values for each sample, but is not a quantitative reflection of SNRs due to the number of bins used to generate the heat map. signatures (28-30). Considering both techniques utilized the same (MoBio) lysis method, differences between the two sample preparation methods are most likely related to extraction and purification efficiency. What is more important (and encouraging) about these results is that the vast majority of microarray signatures detected from the MoBio extract were also detected via the TruTip method (Sørensen similarity index )0.81 for TruTip-generated profiles and 0.71 for MoBiogenerated profiles at t ) 0), which indicates that a 5-min, in-field TruTip sample preparation procedure is efficient and effective for filtered groundwater samples. Logistically, sample acquisition took approximately ∼15 min per filter, and the entire assay (sample-to-answer) was translated into microbial community profiles within 4 h of sample receipt. We conclude from these experiments that the microarray and field portable system can provide ecologically relevant information to site engineers within a single shift, and in the field. Analytical method complexity, equipment infrastructure, and assay turnaround time are therefore no longer an impediment to field deployment of microarray technology and real-time monitoring of microbial community response to environmental conditions. Assay Validation and Microbial Community Dynamics During in situ Bioremediation. To validate the microarray,
VOL. 44, NO. 14, 2010 / ENVIRONMENTAL SCIENCE & TECHNOLOGY
9
5519
FIGURE 2. (A) Heat map display of average (n ) 3 arrays per sample) microarray response for all probes with a SNR > 3. Each set of four profiles is organized by distance down-gradient from the injection gallery (Suppl. Figure 1 in Supporting Information), by depth, and by phase of the acetate injection. (B) TaqMan qPCR data (n ) 1) for each groundwater sample. The dotted line across all graphs represents the approximate limit of detection for the microarray, as estimated from Suppl. Figure 2B. (C) Group-specific microarray signal intensity for those functional groups that were also measured by qPCR. Signal to noise ratios over all probes within the described functional group were first summed, and then log10 transformed to coincide with the qPCR data scale. Charts and individual points in panels B and C are aligned with the corresponding well, depth, phase, and microarray heat map signature in panel A.
FIGURE 3. Fold-change of genus-level microarray SNR relative to the correlated U02 background sample. Total SNR for each functional group was summed over each depth interval within the well, and then averaged across the three downstream wells to arrive at a global average SNR for each genus during each phase of the field experiment. Averaged SNR values were then divided by the corresponding genus-level SNR in up-gradient sample U02 to calculate a fold-change in array response. For those genera that were undetectable in the upstream or downstream gradients (SNR < 3), the SNR was set to 1 prior to the calculation to avoid a “divide by zero” error. Thereafter, calculated fold-changes were log2 transformed as an estimate of an underlying doubling effect in the detected genera. *The Ferribacterium probe also has perfect sequence identity with Dechloromonas aromaticum, and is therefore not in and of itself an absolute indicator of Ferribacterium presence. we applied the system to a retrospective analysis of 50 groundwater filter samples associated with the U02 to D10 transect (Suppl. Figure 1 in Supporting Information). The overarching purpose of the 2008 field experiment was (in part) to understand the microbiological processes contributing to U(VI) removal from groundwater during the in situ transition from iron- to sulfate-reduction and into the sulfatereduction phase. Changes in microbial community composition as a function of distance from the injection gallery, depth, and phase of in situ remediation are plotted as heat maps in Figure 2A. Matching no template controls (n ) 33) were all negative or sporadically positive for one or two random probes (as in the field deployment). The expected bloom in Geobacters is clearly evident in the Fe-reduction phase, with a corresponding reduction in Geobacter signatures (number of probes) and intensity as the system transitioned to and entered sulfate reduction. Wells and all depth intervals nearest the injection gallery showed the strongest Geobacter response, with well D10 showing a lag in community response during all phases of the field experiment, a response that is consistent with site lithology and groundwater flow through the gallery. The general pattern of Geobacter probe response is also reflected in the qPCR data (Figure 2B) and by plotting the log10(Σ average SNR) over all Geobacter-specific probes (Figure 2C). Importantly, only those samples with >102 Geobacter cell equivalents mL-1 were detectable by the microarray, consistent with the estimated lower limit of detection for the array-based method (Suppl. Figure 2B). Equally important is that the total Geobacter-specific microarray signal was highly correlated with the Geobacteraceae qPCR results (R2 ) 0.84, Suppl. Figure 3). 5520
9
ENVIRONMENTAL SCIENCE & TECHNOLOGY / VOL. 44, NO. 14, 2010
As measured by qPCR, the total eubacterial community fluctuated between ∼105 and 107 cell equivalents mL-1 whereas the combined iron- and sulfate-reducer assay varied over 5 orders of magnitude (Figure 2B). The dsr gene was consistently detected at
