DNA- and RNA-SIP Reveal Nitrospira spp. as Key Drivers of Nitrification in Groundwater-Fed Biofilters

With this study we provide the first in situ evidence of ecologically relevant ammonia oxidation by comammox Nitrospira in a complex microbiome and document an unexpectedly high H13CO3− uptake and growth of proteobacterial and acidobacterial taxa under ammonia selectivity. This finding raises the question of whether comammox Nitrospira is an equally important ammonia oxidizer in other environments.

Our previous studies on rapid gravity sand filters (RGSFs), used in potable water preparation from groundwater, revealed nitrifying microbial communities in which Nitrospira is far more abundant than Nitrosomonas (7), with several Nitrospira genomes containing genes for ammonia oxidation (8), and with an abundance of comammox (complete ammonia-oxidizing) amoA over ammonia-oxidizing bacterial (AOB) amoA genes (9). Together with the concurrent discovery of comammox Nitrospira strains by others (10)(11)(12), this suggested that comammox Nitrospira may drive ammonia oxidation in the examined groundwater-fed RGSFs. In addition, like Nitrospira, several acidobacterial, and gamma-and alphaproteobacterial taxa were at consistently higher abundance than Nitrosomonas, raising questions about their potential role in nitrification, as NH 3 is the primary growth substrate entering the filters (7,(13)(14)(15). Identifying the active ammonia-and nitrite-oxidizing organisms is essential not only for engineering purposes, but also for understanding the niches and biodiversity of nitrifiers. There has been a rapidly increasing documentation of global comammox Nitrospira occurrence across a myriad of habitats ranging from the subsurface, to soils, and sediments and from groundwaters to source and residual water treatment plants, but apparently excluding open oceanic waters (9,(16)(17)(18)(19), with occasional abundances that exceed those of canonical AOB (9,20,21). Nonetheless, it is yet to be shown whether comammox Nitrospira truly drives ammonia oxidation in open oligotrophic freshwater and soil environments, their presumed preferred habitat based on genomic and physiological evidence (22,23).
Here, we sought to identify the active ammonia and nitrite oxidizers in a groundwater-fed RGSF using RNA and DNA stable isotope probing (SIP) coupled to 16S rRNA amplicon sequencing. Lab-scale columns packed with filter material from a full-scale RGSF were fed with effluent water amended with NH 4 ϩ or NO 2 Ϫ and with 13 C-labeled or unlabeled HCO 3 Ϫ for 15 days in the presence or absence of inhibitors of autotrophic ammonia and nitrite oxidation (24). Our findings indicate that Nitrospira drives both ammonia and nitrite oxidation. In addition, several other taxa take up substantial HCO 3 Ϫ and their DNA and RNA increase in relative abundance when ammonia is the only provided energy source. This study provides the first in situ evidence of ammonia oxidation by comammox Nitrospira in an ecologically relevant complex microbiome.

RESULTS
Four different experimental treatments were designed to identify the microbes involved in ammonia and nitrite oxidation (Table 1) (columns 5 and 6), and (iv) 71 M NH 4 ϩ and 1 mM NaClO 3 (columns 7 and 8). All columns were operated with an influent containing 100% 13 C-labeled or 100% unlabeled bicarbonate for 15 days.
In columns 1 (H 13 CO 3 Ϫ ) and 2 (HCO 3 Ϫ ), the full-scale conditions were mimicked, with the aim to elucidate the complete in situ food web related to nitrification. In columns 3 (HCO 3 Ϫ ) and 4 (H 13 CO 3 Ϫ ), ATU was used to suppress bacterial ammonia oxidation while feeding at the same NH 4 ϩ loading as in columns 1 and 2 (25). Complete inhibition of bacterial ammonia oxidation with ATU has been observed at ATU concentrations of 8 to 86 M (26), while archaeal ammonia oxidation is less sensitive to ATU (27). The mechanism of ATU inhibition in AOB is proposed to be chelation of the Cu 2ϩ from the active site in the AMO enzyme (24). To identify taxa associated with nitrite oxidation, NO 2 Ϫ was fed to columns 5 (H 13 CO 3 Ϫ ) and 6 (HCO 3 Ϫ Ϫ -amended columns, NH 4 ϩ removal was severely inhibited (6% to 11%); removal of formed NO 2 Ϫ continued (from 70% to 54%), although accumulation of NO 3 Ϫ could not be detected. Nitrogen mass balances, based on influent and effluent NH 4 ϩ , NO 2 Ϫ , and NO 3 Ϫ concentrations closed for most experimental runs, minimizing the possibility of additional nitrogen cycling; N loss was only observed in the ATU-supplemented columns (columns 3 and 4) with ongoing treatment (Table 1; Fig. S3B and C).
Detection of 13 C-labeled taxa from DNA-and RNA-SIP. DNA and RNA, extracted from column samples taken at the end of the experiments, were subjected to equilibrium density centrifugation, gradient fractionation, and 16S rRNA gene amplification. A total of 147 and 65 gradient fractions from DNA-SIP and RNA-SIP were sequenced using the Illumina Miseq and 454 pyrosequencing platforms, respectively (see Fig. S1a and b in the supplemental material). Operational taxonomic units (OTUs) were defined at 99% similarity, to minimize the effect of microdiversity, as similarities of 98.7% and lower represent the taxonomic levels of species, genus, and higher (30).
We first examined the incorporation of 13 C in OTUs in all treatments by comparing replicate columns with H 12 CO 3 Ϫ versus H 13 CO 3 Ϫ amendment. In DNA-SIP, where all SIP fractions were sequenced, we calculated the average shift in buoyant density of each OTU based on its relative sequence abundance and buoyant density in all fractions (see equation 2 in Text S1 in the supplemental material). As only selected fractions were sequenced in RNA-SIP, the mean buoyant density of each OTU in treatments with H 12 CO 3 Ϫ versus H 13 CO 3 Ϫ amendment was calculated using the standard deviation of the RNA distribution across the buoyant density gradient (Fig. S1C) Fig. S2 in the supplemental material) in the DNA-and RNA-SIP experiments, respectively. Only those OTUs that belonged to genera that contained OTUs that were 13 C labeled by both RNA and DNA-SIP were retained (filter 2; Fig. S2). A bootstrap resampling of labeled OTUs within each genus was then used to estimate taxon-specific 90% confidence intervals (Cls) for the buoyant density shift of a labeled genus (filter 3; Text S1 and Fig. S2).
Hence, after the 3rd filter step, 676 (57% DNA and 43% RNA of the total 13 C-labeled OTUs, NH 4 ϩ -fed treatment), 735 (67% DNA and 33% RNA, NH 4 ϩ -ATU-fed treatment), 529 (65% DNA and 35% RNA, NO 2 Ϫ treatment), and 43 (83% DNA and 16% RNA, NH 4 ϩ -ClO 3 Ϫ -fed treatment) OTUs were retained as significantly labeled. The fractional 13 C uptake of labeled OTUs was calculated by dividing the DNA and RNA buoyant density shift for each OTU by the total observed buoyant density shift for DNA and RNA, respectively. The abundances of labeled OTUs in the total community were estimated based on total (i.e., nonfractionated) DNA and rRNA extracts collected on day 15 (see Fig. S4A to D in the supplemental material).
Hypothesis: Nitrospira is an active ammonia oxidizer. H 13 CO 3 Ϫ was incorporated by Nitrospira in treatments fed with NH 4 ϩ , NH 4 ϩ -ATU, and NO 2 Ϫ (after filter 3, Fig. S4A to D). The observed labeling in individual treatments does not indicate whether labeled Nitrospira OTUs are capable of ammonia oxidation because both ammonia and nitrite oxidation occur in NH 4 ϩ treatment. We therefore performed a binary comparison between labeled Nitrospira OTUs detected in the NH 4 ϩ -versus NO 2 Ϫ -fed treatments (Fig. 1a). We assume that the labeled Nitrospira OTUs in NH 4 ϩ -fed treatment would include both comammox and nitrite-oxidizing Nitrospira, while the NO 2 Ϫ -fed treatment would exclude comammox Nitrospira based on observation that comammox Nitrospira growth is not supported by oxidation of environmental nitrite in the absence of ammonia (11).
Heat maps of labeled Nitrospira OTUs (Fig. 1b) CO 3 Ϫ . A large number of labeled Nitrospira OTUs are unique to the NH 4 ϩ -fed column, and few Nitrospira OTUs are shared between the NH 4 ϩ -and NO 2 Ϫ -fed columns, suggesting that most comammox Nitrospira do not readily switch from ammonia oxidation to nitrite oxidation. A large number of OTUs were also uniquely labeled in the NO 2 Ϫ -amended columns (25 in RNA and 66 in DNA), which suggests that, in the NH 4 ϩ -fed treatment, the produced NO 2 Ϫ was not sufficient to achieve labeling of nitrite-oxidizing Nitrospira due to complete nitrification by commamox Nitrospira.
Based on their 13 C labeling in NH 4 ϩ -and NO 2 Ϫ -fed treatments, 78 (8 in RNA-SIP and 70 in DNA-SIP) and 96 (25 in RNA-SIP and 71 in DNA-SIP) Nitrospira OTUs were identified as complete ammonia and nitrite oxidizing, respectively (Fig. 1c). All labeled Nitrospira OTUs belonged to lineage II, which comprises both comammox and non-comammox types. No clear branching between comammox and nitrite-oxidizing phylotypes was observed from the tree topology.
High H 13 CO 3 ؊ incorporation and growth by other bacteria. In our previous 16S rRNA amplicon-based analysis of the same and related RGSF communities, members of the Rhizobiales (Alphaproteobacteria), and Acidobacteria were consistently more abundant than Nitrosomonas, where NH 4 ϩ is thought to be the largest source of energy available for microbial growth (7). We observed that some of these taxa incorporated 13 HCO 3 Ϫ in both RNA-SIP and DNA-SIP (after filter 6; Fig. 2, left). In addition, several OTUs displayed higher buoyant density shifts than Nitrosomonas in NH 4 ϩ -fed columns (Fig. S4A). Finally, these OTUs also increased in relative abundance in both DNA and RNA over the course of the experiment (Fig. 3c).  Only Nitrospira was associated with both ammonia and nitrite oxidation (after filter 6; Fig. 2). Among the 13 HCO 3 Ϫ -incorporating taxa in the presence of NH 4 ϩ , subgroup 10 Acidobacteria, Nitrospira (Nitrospira), Pedomicrobium (Alphaproteobacteria), Rhizobacter, and Acidovorax (Betaproteobacteria) displayed higher shifts in relative RNA and DNA abundance compared to Nitrosomonas (Fig. 3c). With the exception of Pseudomonas (see Fig. S5A in the supplemental material), 13 C-labeled genera in SIP columns (Fig. S4A to D) differed from the dominant genera in the feed water, itself the effluent from the full-scale biofilter. Hence, invasion from feed water communities did not cause the increased relative abundance of subgroup 10 Acidobacteria, Pedomicrobium, Rhizobacter, or Acidovorax.
A metagenome, obtained from the same parent biofilter within the same year (2013 [8]) was further examined for its content of genes that are not canonical AOB amoA,   canonical AOA amoA, or comammox Nitrospira amoA, nor methanotrophic pmoA (8) (see Fig. S6 in the supplemental material) but showed homology with genes encoding AMOA protein family fragments from putative heterotrophic nitrifiers (PF05145 and IPR017516 from the Pfam and InterPro databases, respectively). Twenty-nine unique amoA gene fragments matching the PF05145 model were aligned with reference putative heterotrophic amoA gene fragments (see Fig. S7 in the supplemental material). However, no amoB or amoC genes were found on any of the contigs carrying these atypical putative amoA gene fragments (Fig. S7); furthermore, no similar gene synteny was detected between the other genes of PF05145 amoA-containing contigs and our metagenome amoA-containing contigs (Fig. S7).
The genus Nitrospira was the only taxon associated with nitrite oxidation (Fig. 2,  right).
Here, we examined the assimilation of H 13 CO 3 Ϫ coupled to nitrification in a RGSF using both RNA-and DNA-SIP. OTUs incorporating 13 C isotope label in the different treatments were unambiguously identified as those displaying significant buoyant density shifts between the H 12 CO 3 Ϫ and the H 13 CO 3 Ϫ replicates and were detected at high phylogenetic resolution (Ͼ99% pairwise identity [30]). A total of 200 gradient fractions were processed with sample size equalization.
Our results provide the first in situ physiological evidence of ecologically relevant NH 4 ϩ oxidation by comammox Nitrospira in any environment. Other reports of in situ activity are inferred from bulk observations (ammonium removal when comammox Nitrospira bacteria are more abundant than AOB or AOA [20,41]) or from comammoxspecific amoA transcript analysis (42). Our former observations that Nitrospira was more abundant than Nitrosomonas and the discovery that Nitrospira harbors the complete nitrification pathway in a full-scale RGSF microbiome (8,10,11) are thus directly linked to the ammonia-oxidizing activity of Nitrospira in this environment. Nitrospira was the only genus incorporating H 13 CO 3 Ϫ in both NH 4 ϩ -fed and NO 2 Ϫ -fed treatments, indicating that Nitrospira is the only genus oxidizing both environmental NH 4 ϩ and NO 2 Ϫ in this system. Ammonia-and nitrite-oxidizing phylotypes of Nitrospira were compared phylogenetically; the resulting 16S rRNA tree topology shows no clear evolutionary separation of comammox and canonical Nitrospira. This is in line with previous studies that show that comammox Nitrospira bacteria are not evolutionarily distant from known canonical Nitrospira (Ͼ99% 16S rRNA nucleotide identity) (10). Furthermore, our phylogenetic analysis shows that the labeled-both ammonia-oxidizing and nitrite-oxidizing-Nitrospira OTUs branch within Nitrospira sublineage II, as reported in previous studies (8,(10)(11)(12). We did not identify the amoA clade affiliation of the active comammox Nitrospira phylotypes in this study, although separate investigations on this and related RGSFs have indicated a dominance of amoA clade B over clade A comammox Nitrospira (9,23).
It remains unclear whether comammox Nitrospira can switch between modes of ammonia and nitrite oxidation. However, the large numbers of Nitrospira phylotypes that were exclusively labeled in the NH 4 ϩ -versus NO 2 Ϫ -fed columns, respectively, suggests that comammox Nitrospira may not prefer to oxidize external NO 2 Ϫ alone, in agreement with observations in "Candidatus Nitrospira inopinata" (10,11).
Although ClO 3 Ϫ is a well-known competitive inhibitor of nitrite oxidoreductase (43), strong inhibition of both ammonia and nitrite oxidation was observed in NH 4 ϩ -ClO 3 Ϫfed columns. We expect that the inhibition of nitrite oxidation in comammox Nitrospira would be caused by ClO 3 Ϫ reduction to the toxic ClO 2 Ϫ , which would negatively affect overall metabolism in these organisms, including ammonia oxidation. Hence, it appears that the inhibitory effect of ClO 3 Ϫ on ammonia oxidation provides preliminary support for the contribution of comammox Nitrospira to ammonia oxidation, as we observed before (25). In addition, ClO 2 Ϫ may contribute to inhibition of other ammonia oxidizers as observed before (44). PTIO, an NO-chelating compound, has also been documented as a potent inhibitor of ammonia oxidation in "Candidatus Nitrospira inopinata," although its selectivity is unclear (45). ATU significantly suppressed ammonia oxidation in NH 4 ϩ -ATU-fed treatments, although the taxa assimilating 13 HCO 3 Ϫ did not change significantly compared to the NH 4 ϩ -fed treatment, excluding Azospira. This similarity in labeled taxa in the ATU-fed treatment may indicate that the ammonia monooxygenase (AMO) of the major ammonia oxidizers in this environment may be less sensitive to ATU than AOB at the given concentrations, as previously observed for AOA (46).
No archaeal taxa were 13 C labeled in any of the columns, although archaeal ammonia oxidizers (AOAs) are present, albeit at much lower abundance than Nitrospira (100-to 1,000-fold), in the RGSF used in this study (8,13,47). Columns were fed 71 M NH 4 ϩ to mimic full-scale conditions (14), while bottom layers of the full-scale biofilter receive very low ammonium concentrations due to removal in the top layers (14,15). The absence of AOAs in the 13 C-labeled taxa may be due to their low initial abundances or the elevated NH 4 ϩ concentrations applied during the experiment, as AOAs may thrive better under conditions of reduced energy supply consistent with their elevated abundance at bottom layers of the examined RGSF (48-51), even though a recently isolated comammox strain of "Candidatus Nitrospira inopinata" displayed higher NH 4 ϩ affinity than many of the characterized AOAs (22).
After strong filters were applied to remove heterotrophic OTUs, we retain the following taxa with substantial H 13 CO 3 Ϫ incorporation in the NH 4 ϩ -fed treatments: Acidobacteria subgroup10, Pedomicrobium, Rhizobacter, and Acidovorax; these taxa also show a greater shift in relative abundance in DNA and RNA during the experiment than Nitrosomonas and Nitrospira OTUs ( Fig. 3c; see Fig. S8 in the supplemental material). Most heterotrophic microbes can engage in CO 2 assimilation via carboxylation reactions (52,53). However, CO 2 assimilation via anaplerotic metabolism (54) typically results in only 3 to 8% of the cellular carbon assimilated by heterotrophs, which would be insufficient for label detection by DNA-SIP (52,55). Thus, heterotrophic carbon assimilation would not explain the greater extents of H 13 CO 3 Ϫ incorporation (higher density shifts; see Fig. 2 and see Table S1, parts A and B, in the supplemental material) relative to Nitrosomonas OTUs, known ammonia oxidizers. Furthermore, the cellular mass and activity supported by cross-feeding decay products from autotrophs (56) would be significantly less than the chemolithoautotrophic biomass and activity itself. Thus, the observations of 13 C-labeled genera with higher buoyant density shifts and higher shifts in DNA and RNA abundance shifts ( Fig. 2 and 3) compared to Nitrosomonas and Nitrospira are difficult to explain by cross-feeding alone.
Can a plausible explanation for the high H 13 CO 3 Ϫ assimilation of these taxa be nitrification? An earlier metagenome from the same parent material revealed the presence of putative amoA genes (Table S1, part C) that could not be classified as amoA from canonical AOB, canonical AOAs, or comammox Nitrospira (8) (Fig. S6), yet were phylogenetically related to PF05145, purported to contain AMOA-encoding genes in heterotrophic bacteria (57). In addition, the phylogeny of 10/30 of these aberrant amoA genes indicated their presence, among others, in Hyphomicrobiaceae and Comamonadaceae. Of the highly labeled taxa (in both RNA-SIP, and DNA-SIP) in the NH 4 ϩ -fed treatment, Pedomicrobium, and Acidovorax (but not Acidobacteria subgroup 10 or Rhizobacter) belong to the Hyphomicrobioaceae and Comamondaceae. While it is tempting to speculate that we have identified novel ammonia-oxidizing bacteria, we were unable to identify additional amo genes that would constitute a complete amo operon on any of the metagenomic contigs. In addition, recent doubt has been cast on the assignment of PF05145 as encoding a putative ammonia monooxygenase (58), and careful physiological or genomic evidence of heterotrophic nitrification remains elusive (59). On the other hand, some of the acidobacterial metagenome-assembled genomes (MAGs) that were retrieved from the studied RGSF metagenome contained CO 2 fixation pathways (i.e., CG15 encoded a near-complete reductive tricarboxylic acid [rTCA] pathway [8]).
The second step of nitrification, the oxidation of nitrite to nitrate, is known to be performed by nitrite-oxidizing chemolithoautotrophs such as Nitrotoga, Nitrospina, Nitrobacter, Nitrolancea, and Nitrospira (60-62), which use nitrite oxidoreductase (NXR) as the key enzyme. The known autotrophic nitrite oxidizer Nitrospira was identified as the only active nitrite oxidizer in the studied system.
In summary, comammox Nitrospira and Nitrosomonas are the chemolithoautotrophic drivers of ammonia oxidation in the groundwater-fed biofilter, and comammox Nitrospira make the greatest contribution. The fundamental niche of comammox Nitrospira, however, remains poorly defined. While kinetics and modeling suggest that these bacteria thrive in environments with low ammonium concentrations (22,63), as observed in this study, they are equally abundant in some environments with higher ammonium content, such as fertilized soils and wastewater treatment systems (64). AOAs did not contribute significantly to nitrification, and Nitrospira bacteria were the only nitrite oxidizers identified in this environment. Hence, we provide the first in situ evidence of ecologically relevant ammonia oxidation by comammox Nitrospira in a complex microbiome and document an unexpectedly high H 13 CO 3 Ϫ uptake and growth of proteobacterial and acidobacterial taxa under ammonium selectivity.

MATERIALS AND METHODS
Sampling sites and procedure. Filter material samples were collected from a rapid gravity sand filter (biofilter) at the Islevbro waterworks (Rødovre, Denmark) in May 2013. The influent and effluent water quality is reported elsewhere (13,14,65). Filter material was collected from three random horizontal locations of the biofilter using a hand-pushed core sampler. From the extracted filter material core, the top 10 cm was aseptically segregated on site and stored on ice for further use. A portion was frozen on-site in liquid nitrogen for RNA extraction.
Column experiments and stable isotope labeling. Experiments were conducted using a continuous-flow lab-scale system consisting of glass columns (2.6 cm diameter, 6 cm long) filled with parent filter material (26.5 cm 3 ) as described previously (14). Effluent water from the investigated waterworks was used as the medium in all experiments to avoid interference of other autotrophic processes and approximate full-scale conditions.
The experimental design consisted of 4 treatments applied to columns fed with 13 C-labeled or unlabeled HCO 3 Ϫ (at 1 mM). The influent and effluent pHs of all treatments were 7.5 to 7.6. The experiments were organized in two phases of 4 columns each; filter material was sampled for DNA and RNA extraction just before the onset of each experimental phase. In the 4 treatments, the influent waters were spiked with (i) NH 4 ϩ (NH 4  ϩ -N loading rates (approximately 1.5 g N/m 3 /h) experienced by the full-scale parent biofilter (14). Test and control columns were operated for 15 days with continuous feeding to allow sufficient 13 C label incorporation. Further details are given in Text S1.
Analytical methods. Column effluents were sampled daily, filtered (0.2-m-pore cutoff), frozen and analyzed colorimetrically for NH 4 ϩ and NO 2 Ϫ as described in Tatari et al. (25). Colorimetric analysis of ammonium in samples containing ATU underestimated the NH 4 ϩ concentration (25), and thus NH 4 ϩ in these samples was quantified by flow injection analysis (66). NO 3 Ϫ was quantified by ion chromatography (Dionex, ICS 1500) with a device fitted with a guard column (Dionex, AG 22) and an analytical column (Dionex, Ion Pac AS22). NH 4 ϩ removal (%) was calculated by subtracting effluent from influent NH 4 ϩ concentration and normalizing for the influent NH 4 ϩ concentration. NO 2 Ϫ removal (%) was calculated as the difference between produced NO 2 Ϫ concentration and effluent NO 2 Ϫ concentration, after correcting for trace NO 2 Ϫ present in the water (ca. 0.3 M NO 2 Ϫ ) and normalization for the produced NO 2 Ϫ concentration. The NO 2 Ϫ produced by ammonia oxidation was estimated as the difference between influent and effluent NH 4 ϩ concentrations. NO 3 Ϫ accumulation (%) was calculated from the difference between the effluent and influent NO 3 Ϫ concentrations, normalized for the produced NO 3 Ϫ concentration. The NO 3 Ϫ produced was estimated as the difference between influent and effluent NH 4 ϩ (or NO 2 Ϫ in the case of columns 5 and 6) concentrations. Nucleic acid extraction and SIP. Filter material samples collected from the full-scale biofilter and the sacrificed columns were subject to DNA and RNA extraction. Genomic DNA was extracted from 0.5 g of drained filter material using the MP FastDNA Spin kit (MP Biomedicals, LLC, Solon, OH) according to manufacturer's instructions. The concentration and purity of extracted DNA were checked by spectro-photometry (NanoDrop Technologies, Wilmington, DE). RNA was extracted from frozen filter material samples (Ϫ80°C) with a MoBio PowerSoil total RNA isolation kit (no. 12866-25) according to the manufacturer's instructions. The RNA was further purified with a Qiagen AllPrep DNA/RNA minikit (Hilden, Germany) and quantified with a Ribogreen RNA-quantification kit (Invitrogen, Eugene, OR). Extracted rRNA (approximately 650 ng) was mixed well with cesium trifluoroacetate solution to achieve an initial density of 1.790 g/ml before ultracentrifugation at 38,400 rpm for 72 h at 20°C in a Beckman VTi 65.2 rotor (67). Centrifuged rRNA gradients were fractionated into 250-l fractions, the buoyant density of each fraction was measured by refractometry, and rRNA was precipitated from fractions as described previously by Whiteley et al. (68). The concentration of purified RNA was determined using a Ribogreen RNA quantification kit.
Density gradient ultracentrifugation of DNA isolated from columns and full scale was performed according to Neufeld et al. (69). Briefly, 1.6 g of DNA in CsCl with a final density of approximately 1.725 g/ml was subject to ultracentrifugation at 44,800 rpm for 44 h, 20°C in a Beckman ultracentrifuge with a Beckman VTi 65.2 rotor (Beckmann). Gradients were fractionated into 250-l fractions, density was determined by refractometry, and DNA was recovered by precipitation with PEG. DNA concentration was determined using a Picogreen high-sensitivity double-stranded DNA (dsDNA) quantification kit (Invitrogen).
PCR amplification and tag sequencing. RNA samples purified from density gradient fractions and from sacrificed column experiments and full-scale biofilters were reverse transcribed using reverse primer 1492R with the Sensiscript reverse transcription (RT) kit (Qiagen) according to the manufacturer's protocol. Ten nanograms of cDNA or DNA from direct DNA extracts (Table 1) was used to amplify the V3 and V4 regions of bacterial 16S rRNA genes using the Phusion (Pfu) DNA polymerase (Finnzymes, Finland) and 16S rRNA gene-targeted (rDNA) modified universal primers PRK341F and PRK806R (70). PCR was performed as described in reference 7. All fractions (a total of 145 [ Fig. S1A]) from DNA-SIP and selected fractions (a total of 62 [ Fig. S1B]) from RNA-SIP experiments were sequenced on an Illumina MiSeq and GS FLX pyrosequencing platform, respectively. Pyrosequencing was applied in a two-region 454 run on a 70-75 GS PicoTiterPlate using a Titanium kit (7); paired-end 16S rRNA amplicon sequencing was done on the Illumina MiSeq platform with MiSeq reagent kit v3 (2 ϫ 301 bp; Illumina). All sequencing was performed at the National High-Throughput DNA Sequencing Center (Copenhagen, Denmark).
Bioinformatic and statistical analysis. All bioinformatic and statistical analyses are described in detail in Text S1. Briefly, raw 454 sequence data from RNA-SIP samples were quality-checked (denoised) with Ampliconnoise (71) and chimeras were removed with UCHIME (72) using default settings. Raw Miseq Illumina sequence data from DNA-SIP samples were quality-controlled with MOTHUR (73), and chimeras were removed with UCHIME (72) using a reference data set. Sequence libraries were combined and trimmed to 418 bp. All further sequence analyses were performed in QIIME 1.9.1 (74).
A total of six filter steps were applied to identify ammonia-and nitrite-oxidizing phylotypes (Fig. S2). Detailed steps are described in Text S1. OTUs incorporating H 13 CO 3 Ϫ were determined by the following: (i) filter 1, comparing the mean buoyant density of each OTU in columns with and without 13 C amendment (31); (ii) filter 2, identifying OTUs affiliated with genera that are present in both DNA and RNA SIP; and (iii) filter 3, selecting OTUs with buoyant density shifts higher than genus-specific 90% CIs for buoyant density shifts. The remaining filter steps were applied to assess ammonia and nitrite oxidizing phylotypes. Cross-feeders and taxa performing heterotrophic CO 2 assimilation (i.e., carboxylation) were largely removed by (iv) filter 4, excluding OTUs with lower buoyant density shift than the maximum buoyant density shift value of labeled Nitrosomonas and Nitrospira OTUs, respectively, (v) filter 5, selecting the genera that contained OTUs in both RNA and DNA-SIP, and (vi) filter 6, comparing the labeled OTUs between treatments (NH 4 ϩ fed, NH 4 ϩ -ATU fed, NO 2 Ϫ fed, or NH 4 ϩ -ClO 3 Ϫ fed). To identify ammonia-oxidizing phylotypes, labeled OTUs in all treatments, excluding the one fed only with NH 4 ϩ , were removed from the labeled OTU library of NH 4 ϩ -fed treatment. To identify nitrite-oxidizing phylotypes, labeled OTUs in the treatment fed with ClO 3 Ϫ were removed from the labeled OTU library of the only-NO 2 Ϫ -fed treatment. As an additional step, detected genera were ranked according to the increase in their relative abundance in both total DNA and RNA from the beginning (day 0) to the end (day 15) of the experimental runs.
Data availability. R codes for all bioinformatics and statistics, including the detection of labeled OTUs in DNA-and RNA-SIP, can be found in https://github.com/ardagulay. All sequence data have been deposited at NCBI GenBank under Biosample accession numbers from SAMN12227610 to SAMN12227705.