Introduction

Microbial ecology seeks to find out what forces shape and maintain the composition and structure of microbial communities; elucidate the origin of community and guild diversity and their link to system function, stability and resilience; and identify the interactions that stabilize or constrain microbial communities; among others. Answering those questions requires access to microbial communities that can be observed, sampled and described at relevant spatiotemporal scales, and with sufficient replication, while the system’s environmental conditions and performance can be measured. Technical systems—where open microbial communities are subject to some degree of management—might provide useful alternatives to field systems—without suffering from the simplicity of laboratory-based model systems. Here, we introduce the microbial communities that populate rapid sand filters (RSFs), which are constantly fed with anoxic groundwater (GW) after vigorous aeration, as model technical systems, and present our initial analysis on their composition and ecological patterns.

Rapid gravity filtration is a conventional treatment process whereby GW is intentionally passed through a porous medium bed, often consisting of sand (then called rapid sand filtration), to produce high-quality drinking water. Rapid gravity filters receive continuous inputs from aquifers over multiple years to decades and this input may vary depending on the temporal and spatial dynamics of the aquifer (Griebler and Lueders, 2009). Conditions in source aquifers are similar in that they lack the photosynthetic activity, are at low temperature, carry no dissolved oxygen and typically have low contents of organic carbon, yet they can differ in their content of inorganic (e.g., NH4+, Mn2+, Fe2+, H2S) as well as some organic electron donors (mainly CH4) (Danielopol et al., 2000), all potential energy sources for microbial growth.

Studies on the diversity, abundance and distribution of microorganisms in GW-fed RSF have only recently been reported. Using 16S rRNA gene cloning, van der Wielen and van der Kooij (2009) and White et al. (2012) identified Nitrosomonas oligotropha, Nitrosomonas marina and Candidatus Nitrosopumilus maritimus as the dominant ammonia-oxidizing prokaryotes in three GW treatment plants. Pinto et al. (2012) and Albers et al. (2015), using 16S rRNA pyrosequencing, identified Proteobacteria, Acidobacteria, Actinobacteria, Nitrospira, Bacteriodetes and Plantomycetes as the major phyla in the filtration units. Using a 16S rRNA-targeted DGGE approach, de Vet et al. (2009) identified a shift from iron-oxidizing bacterial dominance in the subsurface-aerated GW to ammonia-oxidizing archaea in above-ground RSF. In a previous study, using 16S rRNA pyrosequencing, we identified Nitrosomonas europaea, Candidatus Nitrosoarchaeum and Nitrospira moscoviensis as dominant nitrifiers in a GW-based full-scale RSF (Gülay et al., 2014).

Reports on RSF microbial communities have typically involved limited sampling sites with insufficient replication within and across RSFs and across drinking water treatment plants (DWTPs) to allow inference of distribution patterns, identify the degree of filter-specific versus shared taxa across DWTPs or assess spatial variation of community structure within DWTPs.

We contend that further knowledge about the microbial community structure of RSFs and its core members is important, and RSF communities can provide suitable model systems to answer central microbial ecological questions. Core taxa have been identified in anaerobic digesters (Rivière et al., 2009), marine environments (Pommier et al., 2007) and activated sludge systems (Zhang et al., 2012; Saunders et al., 2015). Distribution patterns of shared taxa can reveal to what extent environmental conditions vis-à-vis dispersal drive microbial community structure across habitats (Shade et al., 2012) and help identify the microbial taxa that govern central system functions. In addition, the effects of dispersal limitation (Bell, 2010), environmental heterogeneity (Ramette and Tiedje, 2007), energy input (Bienhold et al., 2011) and system design and operation (Sundberg et al., 2013) on microbial community structure are central to many environments, and RSFs might be ideally suited for their investigation.

Here, we extensively characterize the microbial communities at five different DWTPs. Both the feed GWs and the serial filtration units (termed prefilters (PFs) and after-filters (AFs)) were analyzed via 16S rRNA amplicon sequencing. Replicate samples were taken within individual filters and between filters. We first identified and quantified taxa shared among DWTPs. We then evaluated to what extent energy availability could explain bacterial richness and evenness patterns within DWTPs. We also explored distance–decay relationship between aquifer microbial communities and the extent of dispersal within a DWTP. Finally, we examined the extent of local (within filter) spatial heterogeneity in microbial community structure at all RSFs as initiated before (Gülay and Smets, 2015).

Materials and methods

Study sites

Filter materials and GW samples were obtained from five DWTPs in Denmark (Supplementary Figure S1): first line of Langerød (DWTP1, 55°41′45′′N, 11°40′44′′E); second line of Langerød (DWTP2); Sjælsø-1 (DWTP3, 55°52′25′′N, 12°28′36′′E); Sjælsø-2 (DWTP4, 55°52′20′′N, 12°28′32′′E); Islevbro (DWTP5, 55°42′14′′N, 12°27′21′′E). Study sites are detailed in Supplementary Information.

Sampling procedures and physicochemical characteristics of GW and discharge water

Typical values of the dominant chemical constituents in the GW and performance at the different DWTPs are in Table 1. Water quality data (between 2009 and 2011) of raw and discharge water were obtained from the JUPITER database (Hansen and Pjetursson, 2011) and from direct contact with the water utilities, respectively. GW samples were collected from the main influent lines under anoxic conditions before the aeration unit. Microbial mass was then harvested from 250 ml GW using a sterile 0.2 μm pore size (Millipore, Billerica, MA, USA) membrane filter and stored at −20 °C. Filter samples were collected by inserting a 60-cm-long core sampler at a single PF and two separate AF units at each DWTP 6 days AF backwash (Supplementary Figure S2). Sampling procedures are detailed in the Supplementary Information.

Table 1 Influent conditions and performance of the studied drinking water treatment plants
Full size table

DNA extractions, PCR amplification and pyrosequencing

DNA was extracted, amplified and subject to 16S rRNA amplicon 454 pyrosequencing, essentially following the protocol by Sundberg et al. (2013), as described in the Supplementary Information.

Real-time qPCR

Quantification of total bacteria (Eubacteria), the sum of Nitrosomonas plus Nitrospira, and Nitrospira was performed by 16S rRNA gene targeted quantitative PCR (qPCR) on replicates of DNA extracts and the results are described elsewhere (Tatari et al., in preparation). The qPCR primers, PCR conditions and procedures used are described in the Supplementary Information. Results, as reported by Tatari et al. (in preparation), are used here to calculate relative and absolute abundance values.

Bioinformatic and statistical analysis

Raw sequence data were quality checked (denoised) with Ampliconnoise (Quince et al., 2011) and all analyses were performed using the QIIME 1.5.0 software (Caporaso et al., 2010). Subsequent bioinformatics and statistical analyses are described in the Supplementary Information, and β-diversity significance testing was performed as described earlier (Gülay and Smets, 2015).

Nucleic acid sequences

Raw SFF files were deposited into the Sequence Read Archive at GenBank under the study accession number SRP045492.

Results

Overall measures of diversity

A total of 8 57 835 (74.7% of total) amplicon sequences (average length 421 nucleotides) passed all quality checks, including denoising and chimera removal, and were distributed across 33 019 operational taxonomic units (OTUs), defined at 97% sequence identity, hereafter called OTU0.03. We rarefied to 4000 sequences per sample for further diversity and taxonomic comparisons.

While rarefaction analysis suggests some undersampling of the metacommunity at all DWTPs based on individual or merged samples (Supplementary Figures S3 and S4), the large degree of replication provided many subsamples of the metacommunity. Comparison of microbial diversity in RSFs with other environments showed a high degree of richness in RSFs, notwithstanding the much lower input of nutrients (Supplementary Table S2).

The Shannon index of the GW community was twofold higher compared with that of the PF and AF communities; consistently, the Gini coefficient was lowest in the GW communities. Hence, the GW community had the most even distribution of OTU0.03s. The community richness as well as evenness was consistently higher in the AF compared with the PF (Table 2 and Supplementary Figure S5).

Table 2 Diversity estimates of the microbial communities at different locations and units at investigated waterworks
Full size table

Microbial community composition

Of all identified OTU0.03s, 99% (32 627) could be taxonomically assigned using NCBI BLASTN analyses (Altschul et al., 1997) at a minimum identity of 90%. Thousand seventy-five taxa were identified at the genus level from 21 941 OTU0.03s after sample size normalization. The communities were dominated by bacteria comprising 89.9%, 99.5% and 99.7% of all sequences in GW, PF and AF, respectively (Supplementary Table S3). At the phylum level, a large fraction of the sequences in PFs and AFs (89.56±19% and 87.62±15%, respectively) were Nitrospirae, Proteobacteria and Acidobacteria (Figure 1). Proteobacterial OTU0.03s in the PFs and AFs were distributed across the gamma (19.8±15%), alpha (13.9±5%), beta (7.4±5%) and delta (4.7±7%) subphyla.

Figure 1
figure 1

Relative abundance and order-level taxonomic classification of 16S rRNA amplicons across all DWTPs against the Silva108 prokaryotic reference database. For each DWTP, different PFs (bright boxes) and AFs (dark boxes) are plotted along the x axis. The first and second box, in each color, refers to the top sample and composite depth sample of the filter, respectively. GW microbial communities (GW) are shown in first five columns.

Full size image

In PFs and AFs, absolute bacterial abundances, expressed in gene copies per gram of filter material, based on broad and clade-specific 16S rRNA-targeted qPCR ranged from 2 to 12 and 1 to 4 × 109 total bacteria, 6 to 23 and 0.4 to 15 × 107 AOB (Nitrosomonas plus Nitrosospira), 8 to 370 and 3 to 19 × 104 Nitrobacter and 2 to 24 and 1 to 9 × 108 Nitrospira, respectively, across the examined DWTPs (Tatari et al., in preparation). Comparing qPCR results on individual samples confirmed the high relative abundance of Nitrospira in all samples (Figure 2a). The qPCR estimated abundances are somewhat lower compared with those estimated by amplicon sequencing, but overrepresentation of abundant taxa when using universal primers in PCR-based surveys is well documented (Gonzalez et al., 2012). Nevertheless, even qPCR estimates reveal an extremely high Nitrospira (on average 17±11.6%) and much lower AOB abundance (2.26±2.79%) consistent with amplicon sequencing estimates (Figure 2b).

Figure 2
figure 2

Relative abundance (a) and inferred ratios (b) of Nitrosomonas plus Nitrosospira and Nitrospira clades measured by 16S rRNA gene targeted qPCR and inferred from 16S rRNA amplicon libraries.

Full size image

The microbial communities in the PFs were dominated (30.3±24%) by sequences of an uncultured genus of the Methylococcales (γ-proteobacterial methane oxidizers) and this taxon decreased in abundance in the AF communities (5.7±3%). The most abundant bacterial lineage in the AFs (45.2±23%) and the second most abundant in the PFs (27±23%) was the genus Nitrospira (phylum Nitrospirae), typically known as nitrite oxidizer (Pester et al., 2014).

Core taxa among water works

We speculated that similarities in influent, design and operational parameters of the DWTPs would lead to dominant DWTP-shared RSF microbial communities. Hence, shared taxa in the microbial communities of the GW, the PFs and the AFs at all DWTPs were determined. Shared taxa were classified at the phylum, class, order and genus level, and taxa (at the genus level) present at all DWTPs at more or >1% mean relative abundance were defined as dominant or rare core taxa (Gobet et al., 2010), respectively (Figure 3).

Figure 3
figure 3

Richness, abundance and identity of taxa shared across all DWTPs. (a) Richness and sequence coverage of shared taxa in GW, PF and AF units. (b) Shared genera and their absolute abundance. Sequence abundances were log transformed and colored from red-to-black to present higher-to-lower abundance. Heatmap boxes without any sequences were assigned 10−6 (darkest green). (c) Identification of shared genera as dominant (>1% relative sequence abundance) or rare (<1% relative sequence abundance).

Full size image

At the genus level, the GW comprised 73 core taxa (9.5% of all core genera) in 14 phyla accounting for 68.9±13% of the total sequence abundance (Figure 3a). In the PF microbial communities, 30 core taxa (4% of all core genera) were identified, comprising 87±5% of the total sequence abundance (Figure 3a). Most of the PF core taxa (72% of core genera) were also in the GW core (Figure 3b), although the habitats of anoxic GWs and highly oxic PFs are different. Twenty-nine taxa (3% of all core genera) constituted the core of AF microbial communities, representing 75±18% of the total sequence abundance (Figure 3a). Six of the rare core taxa in the PF were dominant core taxa in the AF, including Nitrosomonadales, OM190, Chloracidobacterium, Xanthomonadales, Anaerolineales and Da023. Both PF and AF community cores were dominated by Caulobacterales, Nitrospirales, Methylococcales, Rhizobiales and Myxococcales.

Although few genera constituted the core in both PF and AF communities, the number of OTU0.03s per core genus varied significantly. The number of OTU0.03s per taxon exponentially related to its abundance (i.e., sequence number) in both PF (r2=0.75; P=4.21e−08) and AF communities (r2=0.55; P=1.03e−04 with Da023, r2=0.82; P=3.39e−08 without Da023), except for genus Da023. The sequence abundance and intragenus OTU0.03 richness of Nitrospira, Methylosoma, Hypomicrobium, Rhizobiales (uncultured) and Caulobacterales (uncultured) was, thus, highest of all core taxa. Da023 displays an anomalously high OTU0.03 richness for its total sequence abundance of 8674 across all AF communities (on average abundance 131±69) (Supplementary Figure S6).

Physiological typing of the core taxa

Phylogenetic trees were constructed containing sequences from the core taxa and their closest cultured relatives. Assuming shared metabolic pathways among phylogenetically related organisms, this suggests a physiology of the core taxa (Langille et al., 2013). A high number of OTU0.03s within the core taxa of the PFs (eight taxa) and AFs (seven taxa) were closely related (⩾95% sequence similarity) to sequences of types with the ability to oxidize ammonium, nitrite, iron, manganese and methane as primary growth substrate (Figure 4). Furthermore, these OTU0.03s represent up to 79±2.3% and 66±0.6% of all sequences in the PF and AF core, respectively (Table 3).

Figure 4
figure 4

Phylogenetic trees of a subset of OTU0.03s of the shared taxa in PF and AF together with reference isolates with identified physiologies (AOB, ammonium oxidizers, NOB, nitrite oxidizers, MOB, methane oxidizers, IOB, iron oxidizers, MnOB, manganese oxidizers, SOB, sulfide oxidizers). The number of OTU0.03s in each cluster is within parentheses. The name for each clade was selected from a reference isolates in that clade. The scale bar represents 0.10 substitutions per nucleotide position. GenBank accession numbers of reference sequences are indicated.

Full size image
Table 3 Abundance of taxa (%) with known physiology in the core taxa at each waterworks
Full size table

In PFs and AFs, 31 OTU0.03s had Nitrosomonas oligotropha, a high-affinity chemolithotrophic ammonia oxidizer (Koops et al., 2006) as closest relative; in addition, Nitrosoccus mobilis, another oligotrophic ammonium oxidizer, was a close phylotype to many sequences in PF (25 OTU0.03s), and Nitrosomonas sp. BF16c52, a Nitrosomonas mobilis strain isolated a nitrifying freshwater system (Burrell et al., 2001) clustered with 73 OTU0.03s of AF core taxa (Figure 4).

Within the nitrite-oxidizing clades, a majority of sequences phylogenetically affiliated with uncultured representatives of the Nitrospira genus (169 OTU0.03s in the PF core, and 243 in the AF core). In addition, less OTU0.03s grouped with Cand. Nitrospira defluvii (25) and Nitrospira moscoviensis (80) in the PF and Nitrospira marina (36), Nitrospira calida (105) and Cand. Nitrospira defluvii (1) in the AF.

In PFs, 43 OTU0.03s of the core genera had Crenothrix polyspora, a methane-oxidizing bacterium originally isolated from an RSF (Wolfe, 1960), as a closest relative. The presence of such phylotypes indicates methane residuals in the influent water, although aeration units preceding the PFs have methane removal as central objective. C. polyspora-related phylotypes were not detected in the AF core; here, abundant taxa were identified with another methane oxidizer Methylovulum miyakonense (43) and uncultured representatives of the Methylovurum genus as closest relatives. OTU0.03s with iron- and manganese-oxidizing strains as closest phylotypes were mostly detected in the PF core. Thirty-two OTU0.03s had Pedomicrobium manganicum, a dimorphic prosthecate and oligotrophic manganese-oxidizing bacterium (Poindexter, 2006), as a closest relative, whereas 15 OTU0.03s clustered with sequences assigned to uncultured Gallionellaceae putatively strains with iron-oxidizing capability.

Relationships of diversity with energy availability

We investigated whether richness and evenness patterns in the RSFs related to energy availability within DWTPs. Energy sources for chemolithotrophic growth are gradually consumed between the serial filtration units (Lopato et al., 2013; Gülay et al., 2014), and within individual units removal is spatially stratified (Lopato et al., 2011, 2013; Tatari et al., 2013; Lee et al., 2014). Hence, a gradient in energy supply is present along the four sampling points in the DWTPs: from top to bottom of the PF and from top to bottom of the AF. In DWTP2, 3 and 4 decreases in total bacterial abundance were estimated from PF to AF, whereas in DWTP5, where typical AF units were absent, total bacterial density was clearly stratified with filter depth (Tatari et al., in preparation). Abundance of the Nitrosomonas plus Nitrosospira clade, followed the pattern of total bacterial abundance: decreasing from PF to AF units at DWTPs 2, 3, 4, whereas total abundance of the Nitrospira clade also decreased at DWTPs 3 and 4, but significantly increased from PF to AF at DWTPs 1 and 2 (Tatari et al., in preparation).

At community level, OTU richness correlated strongly and positively with sampling location in the DWTP (r2adj.=0.69, P=0.0001; Figure 5a), whereas a weaker and negative correlation was observed for OTU evenness (Gini coefficients, r2adj.=0.11, P=0.003; Figure 5b). This pattern was consistent across DWTPs and at each DWTP: OTU richness in all five DWTPs increased from PF to AF (r2adj.=0.57±0.2, P<10−2; Supplementary Figure S7) and Gini coefficient significantly decreased (r2adj.=0.54±0.12, P<10−2; Supplementary Figure S8).

Figure 5
figure 5

Ecological patterns in RSFs. Changes in OTU richness (a) and evenness (b) along sampling location. (c) Sequence abundance versus OTU richness of phyla at four sampling locations across DWTPs. Each individual dot represents a phylum at a sampling location. (d) Phylogenetic dissimilarity between Nitrospira OTUs across aquifers, PFs and AFs of DWTPs, based on unweighted UniFrac diversity estimates. (e) β-diversity analysis applied to six biological replicate samples from a filter unit based on weighted UniFrac diversity estimates. Corrected β-diversity values of replicate samples (black circles) and estimated meta-community subsamples (red circles) are plotted.

Full size image

For 4 and 21 core genera (of 30 total) in both PFs and AFs, evenness (0.17±0.04) and richness (0.53±0.24) were related to sampling location along the DWTP, respectively (Supplementary Table S4 and Supplementary Figure S9). Richness decreased with location for genera Akiw543, Crenothrix, Thiobacillus, Pedomicrobium, OD1, Pir4 and uncultured representatives of Burkholderiales, Nitrosomonadales, Rhodospirillales; richness increased with location for genera Haliangium, Chloracidobacterium, Da023, Nitrospira, Om190 and uncultured representatives of Xanthomonadales, Anaerolineales and Acidimicrobiales (Supplementary Figure S9).

Spatial heterogeneity

The microbial community structures at different sampling locations within an individual RSF were all significantly different (β-value exceeded the meta-community-derived threshold) at all DWTPs (Figure 1e), and the magnitude of dissimilarities were similar across PFs and AFs (Supplementary Figure S11). In PFs, however, communities were on average less spatially heterogeneous than in AFs, both in the horizontal plane and with depth. Community dissimilarity between DWTPs was greater than within DWTPs, and within a DWTP dissimilarities decreased along the treatment train (from GW to AF) (Supplementary Figure S12).

Microbial dispersal within DWTPs and between aquifers

Logically, microbial communities from GWs would seed and influence the taxonomic composition of the PFs communities; similarly, the PFs would seed the AFs. To assess this occurrence, each dominant OTU0.03 in a unit was checked for its occurrence in the upflow units. Dominant phylotypes (OTU0.03s) in the downflow units should also occur in the upflow units, if seeding were significant. On average, 75.4±20% of the dominant OTU0.03s in PF communities were detected in the GWs, be it as rare members. For the dominant OTU0.03s in the AFs, 99±2% were found as PF OTU0.03s, of which 49% were dominant and 51% rare (Supplementary Table S5). Clearly, the taxonomic inventory of the community in upflow process units, including the rare members, affects the downflow units.

Second, because deep aquifers are isolated ecosystems (Fredrickson and Onstott, 2001), a decay in taxon similarity is expected with distance. We investigated the degree of phylogenetic dissimilarities between Nitrospira OTU0.03s across DWTPs and the associated feed GWs. The pairwise Nitrospira dissimilarities increased with distance between aquifers (r2adj.=0.88, P=0.002; Figure 5d—GW). This distance effect was significantly less when comparing Nitrospira OTUs in the PFs (r2adj.=−0.11, P<2.2e−16; Figure 5d) or AFs (r2adj.=−0.11, P<2.2e−16; Figure 5d) at different DWTPs.

Discussion

Microbial communities in RSFs

Proteobacteria were the dominant phylum in the PFs and remained abundant in the AFs, consistent with other, more limited, surveys of GW-fed RSFs (Gülay et al., 2014; Albers et al., 2015), and suggest their central role in RSF processes. Largest decreases in abundance from PF to AF were noted for phylotypes of the Nitrosomonadales, an ammonium-oxidizing Beta-proteobacterial taxon, and Methylococcales, a Gamma-proteobacterial methane consuming taxon in line with the removal of their substrates from PF to AF (Macalady et al., 2013; Tatari et al., 2013; Lee et al., 2014). Even though dissolved oxygen concentrations in AFs are typically near saturation, Chloroflexi of the order Anaerolineales, assumed to be strict anaerobes (Yamada et al., 2006), were major phylotypes in AFs, suggesting anoxic microniches or another physiology. We report for the first time the presence of Gemmatimonadetes phylotypes in RSFs, but the ecology and physiology of this phylum is undescribed (DeBruyn et al., 2011).

Nitrospira was consistently the most abundant phylum in the AFs and the second most abundant phylum in the PF amplicon libraries. This high abundance was confirmed by targeted qPCR (Figure 2), although the high Nitrospirae abundance caused a potential overrepresentation in the amplicon libraries (Figure 2) as observed before (Gonzalez et al., 2012). Significantly high Nitrospira occurrence, be it at lower abundance, has been seen in full-scale RSFs fed with GW (detectable Nitrospira DGGE bands (de Vet et al., 2009) and 6% of 16S rRNA amplicons (Albers et al., 2015)), fed with GW/surface water mixtures (2% of 16S rRNA amplicons (Pinto et al., 2012) and 50.4% of 16S rRNA clone libraries (White et al., 2012)) or fed with surface waters (13% to 21% of 16S rRNA amplicons (LaPara et al., 2015) and 17% of 16S rRNA clone libraries (Feng et al., 2012)). Nitrospirae were also dominant in a model drinking water distribution system carrying water from a GW-fed RSF (5–27% in biofilm and 30–40% in bulk water clones; Martiny et al., 2005).

All detected Nitrospirae OTU0.03s belonged to the Nitrospira genus (275 and 814 for PFs and AFs, respectively). The subgenus level analysis showed a broad diversity (Figure 4), suggesting a number of ecophysiologically distinct Nitrospira species, as observed elsewhere (Daims et al., 2001; Pester et al., 2014). In PFs, 214 and 39 OTU0.03s belonged to Nitrospira lineage I and lineage II, respectively, whereas 39 Nitrospira OTU0.03s were of undefined lineage (Supplementary Figure S13—PF). In contrast, in the AFs, the majority of the Nitrospira OTU0.03s (328) were of undefined lineage, with few OTU0.03s belonging to lineage I(2), II(5), IV(2) and VI(36), respectively (Supplementary Figure S13—AF).

Of all possible co-occurrence relations among the Nitrospira OTU0.03s (45.958±15.228) in each DWTP community, few were detected as significant and positive (5590±3951, 13±8%) or significant and negative (123±95, 0.30±0.28%) (Supplementary Table S6). Hence, the vast majority of relations between Nitrospira OTU0.03s appear neutral instead of competitive. Taken together with the high intragenus diversity, this suggests a high degree of specialization within the Nitrospira clade, permitting the exploitation of slightly distinct niches.

The high abundance of these Nitrospira sequences in the community (Figure 2a) is puzzling given the low influent nitrite concentrations (<0.01 mg l−1) and the relatively low abundance of known ammonium-oxidizing prokaryotes (Nitrospira over AOB ratios: 45±90 and 210±431 based on qPCR and amplicon libraries, respectively; Figure 2b) in the filters. Unusually high Nitrospira/AOB ratios were recently reported by others in other GW-fed filtration systems (Nitrospira: 33.4±8%, Nitrosomonas: 3.7±2.5%—16S rRNA amplicons; Nitzsche et al., 2015; Nitrospira: 16.9±12%, Nitrosomonas: 3.2±2%— 16S rRNA clones; White et al., 2012), and even in surface water-fed filters (Nitrospira: 17.3%, Nitrosomonas: 0.82%—16S rRNA clones; Feng et al., 2012 and Nitrospira: 17±5%, Nitrosomonas: 0.07%–, 16S rRNA amplicons; LaPara et al., 2015). Differences in 16S gene copy number (Nitzsche et al., 2015), differences in specific growth rates of Nitrosomonas vs Nitrospira (Feng et al., 2012), presence of unidentified AOBs (LaPara et al., 2015), multiple energy metabolisms in Nitrospira (LaPara et al., 2015; Nitzsche et al., 2015), primer bias in PCR (Nitzsche et al., 2015) and the presence of dormant Nitrospira cells (Martiny et al., 2005) have all been invoked to explain these unexpected ratios. While all these explanation may contribute to high Nitrospira/Nitrosomonas ratios (Figure 2), the observed ratios in our study remain much higher than previously reported.

Possibly, novel physiological features support the presence of these Nitrospira phylotypes. This is in line with the co-occurrence analysis on all core taxa; significant co-occurrence patterns of Nitrospira with other taxa were exclusively negative, suggesting competitive interactions (Supplementary Figure S14). It has been documented that Nitrospira-like bacteria in activated sludge can assimilate pyruvate (Daims et al., 2001), and simple organic compounds can stimulate growth of Cand. Nitrospira defluvii and Nitrospira marina (Watson et al., 1986; Spieck et al., 2006). In addition, the genome of Cand. Nitrospira defluvii encodes pathways for the degradation and assimilation of acetate, pyruvate and formate (Lucker et al., 2010), and Nitrospira can efficiently use formate as carbon and energy source (Gruber-Dorninger et al., 2015). It is possible, then, that part of the Nitrospira abundance is supported by their growth on assimilable organic carbon substrates, present in the GW. In addition, the very recent identification of novel Nitrospira species carrying and expressing the amo operon (Cand. Nitrospira inopinata (Daims et al., 2015), Cand. Nitrospira nitrosa and Cand. Nitrospira nitrificans (van Kessel et al., 2015) suggest that Nitrospira may well be involved in ammonium oxidation in these RSFs. Confirmatory experiments are needed to support the actual ecophysiology of the Nitrospira clade in the RSFs.

The candidate division OD1 phylum, present at 1.8±1%, has not previously been reported as dominant in GW-fed filters, although it has been detected at low densities in raw surface waters (Eichler et al., 2006) and treated surface and GW mixtures (Pinto et al., 2012). Marine members of OD1 phylum have been implicated in sulfur cycling (Harris et al., 2004); a similar role is possible in RSFs. Little is known about the physiology of Actinobacteria phyla beyond the general notion that Actinobacteria can use a range of labile organic carbon compounds (Pernthaler, 2013).

Identification of core taxa and their likely physiological role

Filter-specific OTU0.03s accounted, on average, for 2.91±2.3% and 6.41±4.8% of the total sequence abundance within PFs and AFs, respectively; most (96±2.1%) were singletons or doubletons. The relative abundance of the shared genera at each DWTP was strikingly high with 87±5% and 75±18% of all sequences in PFs and AFs, respectively. The existence of core genera shared across the DWTPs indicate that environmental conditions in the RSF are sufficiently similar, and DWTP-specific conditions—or source water communities—have a secondary influence on the microbial community composition. Abundance of few shared taxa has been reported before in technical systems: 70 core genera (of 744 total) accounted for 63.7% of all sequences in 15 activated sludge communities, sampled from around the world (Zhang et al., 2012), 86 core genera accounting for 68% of all sequences in activated sludge communities of 13 Danish wastewater treatment plants (Saunders et al., 2015), 23% of all sequences in seven anaerobic digesters across Europe could be assigned to six core OTU0.03s (Rivière et al., 2009). Core taxa were abundant at individual DWTP locations—suggesting a functional role (Fuhrman, 2009)—and would indicate selection as a dominant force for community assembly.

By clustering OTU0.03s in the PF and AF core taxa with the closest cultured strains beyond the genus level (>95% pairwise similarity), putative physiologies could be assigned. Physiological types included ammonium, nitrite, methane, manganese and iron oxidizers, which are expected in the DWTPs (Figure 4). These phylotypes dominated the core taxa in both PFs (73.6±6%; Table 3) and AFs (61.4±21%; Table 3), supporting the notion that core genera are functionally important in RSFs. Nevertheless, for a fraction of OTUs in the core taxa (21±0.6% of all OTUs in the PF core and 34±0.4% of all OTU0.03s in the AF core), no close cultured isolates could be identified. Most of the unassigned OTU0.03s in the core belong to the taxa Cand. Chloracidobacter, Da023 (Acidobacteria), Myxococcales, Rhizobiales (α-proteobacteria), Xanthomonatdales (γ-proteobacteria) and Anaerolineales (Chloroflexi). These taxa are also dominant in other GW-fed RSFs (Albers et al., 2015; Nitzsche et al., 2015). Acidobacteria are abundant in a suite of environments (Barns et al., 1999), harbor broad metabolic capabilities and can cope with nutrient limitations (Ward et al., 2009; Hartmann et al., 2015); they may be central to carbon cycling in RSFs. Myxococcales are widely distributed in terrestrial and marine environments, are often surface attached (Ganesh et al., 2014) and have a predatory lifestyle or live off polymer hydrolysis (Shimkets et al., 2006; Huntley et al., 2011); they might have a central role in recycling microbial decay products in RSFs. The physiology of Xanthomonadales is largely undescribed, but some are, similarly, known to have high extracellular hydrolytic activity (Jacobson Meyers et al., 2014).

It should be pointed out that, while an RSF core is evident at the genus level, the core taxa still retain subgenus variation across DWTPs, potentially due to dispersal limitation between the source aquifers (Gibert et al., 2009), leading to different evolutionary trajectories for any clade. Indeed, the observed distance–decay relationship for Nitrospira OTU0.03s in the source aquifers is consistent with previous studies on aquifer communities (Franklin et al., 1999; Hug et al., 2015) and a weakened signal remains in the PF and AF Nitrospira clade.

Ecological patterns in RSFs

The consistent decrease in community evenness and richness from GW to filter might be due to the strong environmental changes (shift in DO level from below detection to saturation), resulting in loss of many, potentially oxygen-intolerant, taxa from GWs, and enrichment and selection of few taxa, adapted to the new environmental conditions. While the total OTU richness decreased, the OTU richness of few PF genera strikingly increased.

At the community level, OTU richness and evenness were consistently higher in the AF compared with the PF communities; this is surprising as available energy would decrease (because of microbial removal of energy sources) moving down-gradient in the DWTP (Lopato et al., 2011, 2013; Tatari et al., 2013; Lee et al., 2014) and available energy would promote richness by its positive effect on community size, per the species-energy theory (Wright 1983). Increasing community richness with decreasing energy input contradicts previous observations on natural and technical systems (Bienhold et al., 2011 and Supplementary Table S2).

Direct evidence of total community size decrease from PF to AF was difficult to obtain (small decreases were detected for DWTP2, 3 and 4 with qPCR), and suggests a significant degree of density redistribution by detachment/deposition, among other processes. The pattern of diversity increase is detected at the community, phylum, order and even at the genus level (Supplementary Table S7). OTU richness did, however, remain correlated with relative (Supplementary Table S8) abundance (Nitrospira and Nitrosomonas) within specific taxa, confirming that increasing taxon abundance increases taxon diversity (Figure 5c). These taxa were, primarily, the dominant members of the AF community.

Negative relationships between diversity and energy input have not been reported at the community level, but they have been observed at the level of specific taxa (Bienhold et al., 2011). Oligotrophy was suggested as the cause for taxa to exhibit negative relationships with energy input (Bienhold et al., 2011), and could be an explanation for the increase of abundance, and hence richness, with decreasing energy levels in RSFs. The abundance of Nitrospira as well as genera Xanthomonadales, Nkb5, Haliangum, A0839 (Proteobacteria), Acidomicrobiales, Cl500 (Actinobacteria), Anaerolineales (Chloroflexi), Cand. Chloroacidobacter, Da023 (Acidobacteria) and OM190 (Planctomycetes) significantly increased with sampling location. Potentially, Nitrospira OTUs become more dominant and diverse in AFs (Figure 3b), as they are favored under low substrate conditions (Schramm et al., 1999; Attard et al., 2010; Lucker et al., 2010) (Supplementary Table S9). On the other hand, the relative abundance of core genera with putatively assigned physiology, such Nitrosomonadales, Crenothrix, Thiobacillus, Methylosoma, Hyphomicrobium and Azospira, decreases with sampling location in a DWTP, consistent with consumption of their specific energy substrates (Lopato et al., 2011, 2013; Tatari et al., 2013; Gülay et al., 2014; Lee et al., 2014).

Here, we extend our previous observation (Gülay and Smets, 2015) that RSF microbial communities display spatial heterogeneity within a single PF or AF at all DWTPs. This heterogeneity is more pronounced in AFs than PFs. In all cases, rare taxa (which are more dominant in AFs than PF) explain the dissimilarities (Table 1 and Gülay and Smets, 2015), as observed in other environments (Youssef et al., 2010). The contribution of observed community dissimilarity to differences in filter activity (Lopato et al., 2011, 2013) is uncertain, especially given the abundance of the core taxa, shared across all samples. Across DWTPs, communities are as a result less dissimilar in the RSFs than in the GWs that feed them (Supplementary Figure S12), confirming similar selective forces driving community structure (Table 1), and an identifiable resultant RSF core microbiome.