Two laboratory-scale bioreactors (13 × 46 × 30.5 cm) were constructed from acrylic sheets, adhered using Weld-On® 3 Solvent Cement (IPS Corporation, Compton, CA) and sealed with 100% silicone sealant (Dow Corning, Midland, MI) (Figure 1). Water was pumped into the reactors through an inlet tube that extended into the woodchip bed. An adjustable overflow weir allowed control of water level. Within each woodchip bed, two aluminum mesh (0.635 cm) sampling cages were installed. The sampling cages could be lifted out of the reactors and opened at the bottom for woodchip sampling. They were installed so that their bottoms were at two different depths: with one sampling about halfway up the woodchip bed (referred to as top port throughout the manuscript) and the second one sampling near the bottom (bottom port). The reactors were filled to their full depth with woodchips collected from an active field-scale bioreactor and from a nearby above-ground pile (1:1 mixture). The initial woodchip bed porosity was 0.60 in the disturbed reactor and 0.62 in the constant reactor. During operation, additional woodchips from the same mixture were added to the top of the sampling ports as needed to maintain a consistent woodchip height across the reactor surfaces. No woodchips were added outside of the sampling ports.

The synthetic tile drain reactor feed was based on typical nitrate concentrations in central Illinois tile drainage (Kalita et al., 2006) and chemical analysis of actual tile drainage water (Blowes et al., 1994; Stone and Krishnappan, 1997). Sterile concentrated feed was mixed with dechlorinated municipal tap water for influent concentrations of 15 mg NO3−-N/L, 0.25 mL/L mineral solution (Tanner, 1997), 0.0625 mL/L trace metal solution (Tanner, 1997), and 2.5 mg/L yeast extract. Following an initial 2 week recirculating flow period and gradual step-wise increases in retention time, the theoretical retention time was maintained at 12 h for both reactors for the duration of operation. This HRT was chosen to achieve a moderate percent removal, allowing detection of both increases and decreases in performance.

The water level in reactor C was maintained at 19 cm, which corresponded to a 5.76 L woodchip bed void volume. The water level in reactor D had three phases. In Phase I (April 3–August 23, 2011) the water level was 20.5 cm, which corresponds to a 5.76 L woodchip bed void volume. In Phase II (August 23, 2011–April 12, 2012), the water level in reactor D was decreased to 11 cm, with a corresponding decrease in flow rate to maintain constant HRT. In Phase III (April 12–November 8, 2012), the disturbed reactor was operated with fluctuating water levels, switching between the high and low water levels every 3 weeks for 6 months.

Laboratory temperature and flow rates were measured every other day during reactor operation. The room temperature was typically between 19 and 22°C, with an average temperature of 21. On three occasions (October 2011, December 2011, and March 2012) the room temperature was elevated for 3–4 days, with temperatures between 23 and 27°C. To monitor nitrate removal, 5 ml samples of the influent (which fed both reactors) and of the effluent from each reactor were collected every other day, filtered through 0.22 μm nylon syringe filters, and stored in 10 ml polyvinyl vials (Dionex, Sunnyvale, CA) at 4°C until analysis. Woodchips (15 g wet weight) were sampled from the bottom of each sampling port in each lab reactor weekly for microbial community analysis and processed immediately. Woodchip samples (20 g wet weight) were also removed for activity measurements.

Tracer studies were performed on both reactors in November 2011, when the disturbed reactor was in Phase II, extended low water period. A step input of 100 mg/L bromide was used as the tracer. Effluent samples were collected every hour for the first 5 h, every half hour for the next 13 h, and every hour for the remaining 18 h.

Denitrification potential was measured through DEAs with the acetylene block method (Tiedje et al., 1989; Bernot et al., 2003). These assays were performed in duplicate for each sampling port. Ten grams of woodchips were incubated in a Wheaton bottle with 75 ml of solution containing 15 mg/L NO3− and 0.1 g/L chloramphenicol. Glucose was not typically added, but on three dates (June 14, June 28, and July 5, 2012), the assays were conducted with and without 1 mM glucose. The bottles were capped with chlorobutyl septa (Wheaton Science Products, Millville, NJ) and open top caps. The assay headspace was flushed with helium gas, and acetylene gas added to achieve a headspace concentration of about 20%. Samples were collected at times 0, 2, 4, and 6 h. Gas samples for nitrous oxide analysis were stored in 10 ml Vacutainers (BD, Franklin Lakes, NJ) until analysis. Liquid samples for nitrate quantification were filtered with 0.22 μl syringe filters and stored in 10 ml polyvinyl vials (Dionex, Sunnyvale, CA) at 4°C until analysis. After each sampling point, the extracted volume was replaced with 10:1 helium:acetylene gas mix. After the assay, the headspace volume was measured, and the woodchip samples were dried for 48 h at 105°C before measuring their dry weight.

Linear regressions were used to calculate the rates of nitrate removal and nitrous oxide production. The time zero samples were not included in the regression, because they were usually not linear with the other time points. Rates were normalized per dry gram of woodchips for individual assays. The denitrification potential values for multiple categories of disturbance and water level were averaged together. Significance of values was evaluated using paired t-tests for unequal variance in Microsoft Excel (Microsoft Corporation, Redmond, WA).

Nitrate was quantified using an ICS-2000 ion chromatography system with Chromeleon Chromatography Management Software (Dionex, Sunnyvale, CA). An IonPac AS18 4 × 250 mm hydroxide-selective anion-exchange column was used. A standard curve was created for each run using standards with 3.125, 6.25, 12.5, 25, 50, and 100 mg/L NO3−. Samples collected during the tracer studies were also analyzed for bromide with this configuration.

Nitrous oxide was quantified in gas samples collected during DEA using a Shimadzu GC-2014AFsc gas chromatography system with packed columns and GCSolution software (Shimadzu Corporation, Nakagyo-ku, Kyoto, Japan). Preconditioned, custom packed HayeSep N 80-100 columns were included in the set-up (Supelco, Bellefonte, PA). A standard curve was created for each run using standards with 0.11, 0.73, 1.09, 1.80, 3.67, and 7.33 mg/L N₂O.

Sample processing included shaking with glass beads, extraction with the FastDNA Spin Kit (MP Biomedicals, Solon, OH), and CTAB purification, as described previously (Porter et al., 2015). Automated ribosomal intergenic spacer analysis (ARISA) (Fisher and Triplett, 1999) was used to analyze the composition of the overall bacterial community. Quantitative PCR was used to measure the abundance of clade I and clade II nitrous oxide reductase (nosZ) genes (Henry et al., 2006; Jones et al., 2013). Due to the low abundance of clade II nosZ (<8 copies/ng DNA, data not shown), the bacterial denitrifying community structure was analyzed using terminal restriction fragment length polymorphism analysis (tRFLP) of the clade I nosZ (Rich et al., 2003). Microbial community analysis was carried out as described previously (Porter et al., 2015), except that 100 ng of DNA was used in each nosZ PCR and 30 cycles were carried out. The ABI GeneScan ROX 1000 size standard was used for capillary electrophoresis, and electropherograms were analyzed using GeneMarker 2.2.0 (SoftGenetics, LLC, State College, PA).

Bacterial communities were analyzed using PRIMER 6, version 6.1.10 (Primer-E, Ltd., Plymouth, UK). To visually analyze the patterns in community structure, non-metric multi-dimensional scaling (MDS) plots were generated. To generate similarity values among different groups of samples, the ANOSIM function (Clarke and Green, 1988) was used. The DIVERSE function was used to determine taxa richness.

Both reactors showed nitrate removal throughout the entire study period. The two reactors removed nitrate similarly when they both had the same high water level (Figure 2, Phase I). The constant reactor had an average of 31 ± 7% nitrate removal, and the disturbed reactor had an average of 31 ± 8% nitrate removal. However, the performance of the disturbed reactor unexpectedly improved in Phase II, when the water level was lower, even though the flow rate was adjusted to maintain a constant theoretical HRT (Figure 2). At the lower water level, the disturbed reactor had an average of 59 ± 11% nitrate removal.

During the period of regular disturbance (Phase III), the performance of the disturbed reactor continued to respond to differences in water level, showing similar nitrate removal to the constant reactor at high water level and increased nitrate removal at low water level (Figure 2, Phase III). There was also a consistent but transient spike in nitrate removal, up to 70–90%, observed in the disturbed reactor immediately following the decrease in water level. Within a few days at low water level, the performance decreased, but still remained better than performance at the high water level for either reactor.

Due to the unexpected performance improvement at low water level, a tracer study was conducted during Phase II. The measured HRT was 7.5 h for the constant reactor, while the disturbed reactor showed a slightly longer HRT of 8.8 h at the low water level.

In the DEAs, nitrate removal was measured under standard conditions over 6 h to determine the denitrification potential. In most cases the nitrate removal rates observed in this assay were not significantly different between the reactors or ports (Table 1), despite differences in water level, temporal proximity to water level changes, and performance. The exception was the top port of the disturbed reactor during Phase II, which had an average nitrate removal rate of 0.004 ± 0.001 mg NO₃-N/h/dry g woodchips. This removal rate is significantly lower than those from the same port during Phase III and from any other port during any phase.

The denitrification potential was also measured using the more common metric of nitrous oxide production, which accumulates in DEAs due to the inhibition of nitrous oxide reductase by acetylene. Nitrous oxide production rates were highly variable, ranging from 0.00002 to 0.004 mg N₂O-N/h/dry g woodchips (for average values see Table S1). The high level of uncertainty in these measurements makes it difficult to compare the nitrous oxide production rates from different reactor conditions. However, nitrous oxide production rates were consistently about ten-fold lower than the nitrate removal rates.

To test whether organic carbon was limiting activity in the DEAs, samples from three dates were also assayed with and without glucose amendments. The average nitrate removal rates increased slightly when measured with amended glucose for both ports of both reactors (Figure S1), although only the comparison for the top port of the constant reactor was statistically significant (t-test, unequal variance, p = 0.03). This trend was also true for most of the individual DEAs.

The bacterial communities observed in the denitrifying bioreactors included 323 unique Operational Taxonomic Units (OTUs). Individual samples contained a median of 76 OTUs, and the median number of OTUs was similar for both reactors. These results are only slightly lower than those observed in field bioreactors, where for example 435 unique ARISA fragments were found in a temporal study (Porter et al., 2015) and the median number of OTUs was 101.

The fluctuating water level had an effect on bacterial community composition. Comparing the constant and disturbed reactor, the bacterial communities of the two reactors first became significantly more distinct from each other during Phase III. This difference is illustrated by MDS (Figure 3) and confirmed by ANOSIM. The ANOSIM R value comparing the communities of both reactors before Phase III was 0.211 (p < 0.001). This included samples from Phase I, where the water height was identical, and from Phase II, where the disturbed reactor had a lower water level. After operation with fluctuating water levels (Phase III), the ANOSIM R value for comparison of communities between reactors was 0.417 (p < 0.001). The higher ANOSIM R value indicates that the bacterial communities became increasingly different between the two reactors once the recurring disturbance period began. This divergence was not observed between the ports of the disturbed reactor, despite their differences in environmental conditions (Table 2). Specifically, within the disturbed reactor, the ANOSIM R value comparing the top port communities at low and high water levels was not significant (p = 0.325), nor was the value comparing the bottom port communities at low and high water levels significant (p = 0.833), meaning that the communities are not significantly different from each other. However, the top port of the disturbed reactor, which experienced the most fluctuations in environmental conditions, did have the most distinct bacterial community.

The richness of the denitrifying bacterial community in the bioreactor samples was high, with a total of 272 unique AluI terminal restriction fragments and 172 unique HhaI terminal restriction fragments observed. Each sample contained between one and 69 AluI fragments and one and 59 HhaI fragments, with a median of 11 fragments per sample for AluI digests and 12 fragments per sample for HhaI digests. The disturbed reactor had medians of 10.5 AluI and 12 HhaI fragments, and the constant reactor had medians of 11 AluI and 13 HhaI fragments. Using the same primers and restriction enzymes, the median number of terminal restriction fragments in field bioreactors was 9 or 11, for HhaI and AluI respectively (Porter et al., 2015).

In contrast to the overall bacterial communities, the denitrifying communities did not show a shift to distinct assemblages between the two reactors following the disturbance period (Figure S2 and Table S2). The ANOSIM R values comparing the reactors during these operational periods were insignificant, indicating indistinctness from each other. The ANOSIM R values comparing the denitrifying communities from the four different ports were also insignificant, indicating that the communities were indistinct from each other. Finally, there was no distinction between the communities of an individual port sampled during high and low water conditions.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.