Comprehensive characterization of aerobic groundwater biotreatment media

Abstract

Aerobic biotreatment systems can treat multiple reduced inorganic contaminants in groundwater, including ammonia (NH₃), arsenic (As), iron (Fe), and manganese (Mn). While individual systems treating multiple contaminants simultaneously have been characterized and several systems treating one contaminant have been compared, a comparison of systems treating co-occurring contaminants is lacking. This study assessed the treatment performance and microbial communities within 7 pilot- and full-scale groundwater biotreatment systems in the United States that treated waters with pH 5.6–7.8, 0.1–2.0 mg/L dissolved oxygen, 75–376 mg CaCO₃/L alkalinity, < 0.03–3.79 mg NH₃-N/L, < 4–31 μg As/L, < 0.01–9.37 mg Fe/L, 2–1220 μg Mn/L, and 0.1–5.6 mg/L total organic carbon (TOC). Different reactor configurations and media types were represented, allowing for a broad assessment of linkages between water quality and microbial communities via microscopy, biofilm quantification, and molecular methods. Influent NH₃, TOC, and pH contributed to differences in the microbial communities. Mn oxidase gene copy numbers were slightly negatively correlated with the influent Mn concentration, but no significant relationships between gene copy number and influent concentration were observed for the other contaminants. Extracellular enzyme activities, community composition, and carbon transformation pathways suggested heterotrophic bacteria may be important in nitrifying biofilters. Aerobic groundwater biofilters are complex, and improved understanding could lead to engineering enhancements.

1. Introduction

Aerobic biological treatment (biotreatment) of groundwater utilizes naturally occurring microorganisms to oxidize and remove dissolved contaminants, typically reduced inorganic compounds such as ammonia (NH₃), arsenic (As), iron (Fe), and manganese (Mn), in the presence of oxygen. Biotreatment can achieve >90% contaminant removal to meet the effluent water quality targets, and co-occurring contaminants can be removed simultaneously if the necessary conditions are provided (Tekerlekopoulou et al., 2013). Advantages of biotreatment over physical-chemical treatment processes include decreased energy and chemical feed requirements, and reduced waste residuals streams. For Fe and Mn removal, biotreatment offers longer filter run times, higher loading rates, and less residual wastewater because biogenic Fe and Mn oxides tend to be more compact (Arturi et al., 2017; Bruins et al., 2015; McClellan, 2015).

Three process configurations are common for aerobic groundwater biofilter design: (1) open aeration followed by granular media filtration, (2) compressed air injection into pressurized granular media filters, and (3) bubble aeration up through a packed bed (i.e., aerated contactor) followed by granular media filtration. Aerated contactors are advantageous when treating water containing elevated NH₃ because nitrification has a high oxygen demand of 4.6 mg O₂/mg NH₃-N (excluding cell synthesis), and this design maintains saturated dissolved oxygen (DO) throughout the bed. Several studies have investigated the microbial communities in the first two designs, but research efforts to date on aerated contactors have focused on water quality, particularly the impacts of DO, pH, and phosphate (PO₄) on removal of NH₃, Fe, Mn, and As (Gülay et al., 2016; Lytle et al., 2013, 2020; Poghosyan et al., 2020).

In the U.S., aerobic biotreatment systems are less common for groundwater than for surface water where the treatment goal tends to be driven by organic carbon reduction rather than inorganic contaminant removal. Aerobic heterotrophic bacteria tend to be the most prevalent bacteria in surface water biofilters, and assessments of microbial parameters at multiple full-scale facilities have been reported (Keithley and Kirisits, 2018; Ma et al., 2020; Pharand et al., 2014). In contrast, there has not been a microbial community assessment of multiple aerobic groundwater biotreatment systems in the U.S. to date, and studies in other countries investigated aeration-filtration systems treating groundwaters with lower NH₃ concentrations than is found in some U.S. groundwaters (Gülay et al., 2016; Hu et al., 2020).

The removal of co-occurring contaminants in aerobic groundwater biofilters remains an active and important area of research. Contaminant removal is driven by specialized bacteria (e.g., NH₃- and Fe-oxidizing bacteria) that are widely phylogenetically distributed, although chemical oxidation and particle removal are also important. Questions of interest include how the contaminants affect biofilm structure and accumulated precipitates, if contaminant oxidation gene abundance is associated with the contaminant concentration, and how multiple contaminants affect the microbial community structure and functional potential. Answers to these and other questions will lead to improved biofilter design and operation.

Given the research needs and the value of biotreatment, the objective of this work was to compare the elemental coating, biofilm characteristics, and microbial community on filter media of systems treating one versus multiple biologically oxidizable contaminants at 7 systems in 4 states (Connecticut, Iowa, Massachusetts, and Ohio). Water and filter media samples were collected from pilot- and full-scale filters and aerated contactors that treated raw waters containing < 0.03–3.79 mg NH₃-N/L, < 4–31 μg As/L, < 0.01–9.37 mg Fe/L, and 2–1220 μg Mn/L. Contaminant removal effectiveness, biofilm composition, quantification of functional genes, composition of microbial communities and their functional potential were assessed. Comparisons across diverse systems provided a more detailed understanding of complex processes within aerobic groundwater biofilters.

2. Materials and methods

2.1. Study sites

One set of water and media samples was collected from 5 pilot- and 2 full-scale biotreatment systems, all of which had been in operation for at least 4 months and were at steady-state (Table 1). Three (1, 4, and 5) were in New England (Connecticut and Massachusetts), and four (2, 3, 6, and 7) in the Midwest (Iowa and Ohio). The treatment process at Sites 1, 4, 5, and 7 consisted of aeration followed by biofiltration, whereas Sites 2, 3, and 6 employed aerated contactors. The full-scale treatment plant at Site 4 consists of two filtration stages in series, with the first intended for removing Fe and the second for Mn. For this study, a pilot-scale filter operated in parallel to the second-stage Mn filter was sampled. Only the first stage of the treatment process was piloted at Sites 1, 2, 3, and 6b, all of which would include a downstream filter in the complete system. Sites 2, 3, and 6 included a PO₄ feed, whereas Sites 1, 4, 5, and 7 did not.

Table 1.

Site descriptions.

Site	State	Scale	Sample Date	Process Configuration	Loading Rate (gpm/ft2)	EBCT1 (min)	Media Type and Size	Media Sample ID2
1	MA	Pilot	Nov-20	Aeration Pressure filter	3.8	7.4	Sand, 1.4 mm	1-F
2	IA	Pilot	Jan-21	PO4 feed (0.4 mg/L) Aerated contactor	1.8	16.3	GAC, 1.7–4.8 mm	2-C
3a3	IA	Full	Feb-21	PO4 feed (0.8 mg/L) Aerated contactors	1.0	27.8	Gravel, 4.6 mm	3a-C1,24
				Filters	1.8	10.6	Anthracite, 1.0 mm/Sand, 0.46 mm	–
3b3		Pilot	Feb-21	PO4 feed (0.8 mg/L) Aerated contactor	2.0	16	Ceramic, 2.0 mm	3b-C
4	CT	Full	Feb-21	Aeration Pressure filters	2.3	n/a5	Sand, 0.95 mm
				Aeration pH adjustment
		Pilot		Filter	6.0	5.0	Sand, 0.95 mm	4-F
5	MA	Full	Feb-21	Aeration pH adjustment Pressure filters	4.8	6.2	Sand, 0.95 mm	5-F
6a	OH	Pilot	Apr-21	PO4 feed (0.4 mg/L) Aerated contactor	3.0	6.3	Gravel, 12 mm	6a-C
				Filter	1.1	17.4	Anthracite, 1.0 mm/Sand, 0.46 mm	6a-F
6b		Pilot	Apr-21	PO4 feed (0.4 mg/L) Aerated contactor	1.1	13.8	Ceramic, 2.0 mm	6b-F
7	OH	Full	May-21	Aeration Retention basin Gravity filters	2	23.9	Anthracite/sand	7-F

¹EBCT: empty bed contact time.

²Media sample ID includes the Site number followed by the reactor type from which the sample was collected. C: aerated contactor, F: filter.

³Full-scale (a) and pilot-scale (b) contactors located at the same water treatment plant (Site 3) were sampled.

⁴Media samples from two parallel contactors were collected.

⁵n/a: Data not available.

2.2. Sample collection

Water samples were collected from raw water, filter influent and effluent for each stage in the biotreatment system, and finished water. Water samples designated for metals analyses were split into an unfiltered total metals sample and a 0.45-μm polyvinylidene fluoride syringefiltered sample representing dissolved metals. Historical data from each site were reviewed to check that these grab samples were within the range of expected values. Approximately 50 mL of filter media were collected in sterile tubes from the top of the biofilter. Thus, sampled media came from the influent end of the filters, which operated downflow, and the effluent end of the aerated contactors, which operated upflow. Precipitates that had accumulated in the packed bed were included in the sample.

Samples typically were collected by water system personnel using containers provided by EPA and then shipped overnight on ice to the EPA in Cincinnati, OH for analysis. At Site 7, EPA personnel collected and transported the samples.

2.3. Water quality analyses

pH and DO were measured onsite. Total inorganic NH₃ (350.1, USEPA, 1983), ${\text{NO}}_2^{-}$ and ${\text{NO}}_3^{-}$ (353.2, USEPA, 1983), and orthophosphate (365.1, USEPA, 1993) were measured via automated colorimetric methods. Alkalinity was measured via potentiometric titration (2320, APHA, 2005). Total organic carbon (TOC) was measured via combustion (5310 C, Standard Methods 2005). Metals were measured by inductively coupled plasma atomic emission spectroscopy (200.7, USEPA, 1994; Thermo Jarrel Ash, Franklin, MA). Hardness was calculated as the sum of the magnesium and calcium concentrations expressed in units of mg CaCO₃/L.

Contaminant removals across each biofilter were calculated (Eq. (1)). Removals of NH₃, Fe, Mn, and As were used to estimate the theoretical biomass production rate based on the yield for microbial oxidation using oxygen as the terminal electron acceptor and NH₃ as the nitrogen source for cell synthesis (Rittmann and McCarty, 2001). Calculated yields for each reaction are provided in SI.

$$\mathit{\text{Removal}}\left(g/m^3\mathit{\text{\hspace{0.17em}reactor}}/\mathit{\text{day}}\right)=\frac{\left[\mathit{\text{Influent\hspace{0.17em}Conc}}.\left(\frac{g}{L}\right)-\mathit{\text{Effluent\hspace{0.17em}Conc}}.\left(\frac{g}{L}\right)\right]*\mathit{\text{Flow\hspace{0.17em}Rate}}\left(\frac{L}{\mathit{\text{day}}}\right)}{\mathit{\text{Reactar\hspace{0.17em}Volume}}\left(m^3\right)}$$ (1)

2.4. Filter media analyses

Adenosine triphosphate (ATP) was measured via luminescence (LuminUltra, Fredericton, Canada). Extracellular polymeric substances (EPS) were extracted in duplicate using the protocol developed by Keithley and Kirisits (2018) except the shaking intensity was 100 rpm. Extracted EPS proteins were measured colorimetrically with bovine serum albumin (BSA) as the standard (Pierce BCA, Thermo Fisher Scientific, Waltham, MA). Extracted EPS polysaccharides were measured colorimetrically with glucose as the standard (Dubois et al., 1956). For the protein and polysaccharide colorimetric assays, absorbance was measured using a spectrophotometer (DR1900, Hach, Loveland, CO).

The potential activities of five extracellular hydrolases (β-d-glucosidase [BG], esterase [ACE], β-N-acetyl-glucosaminidase [NAG], leucine aminopeptidase [LAP], and alkaline phosphatase [AP]) were quantified. ACE and BG are involved in C acquisition through the breakdown of esters and glucans, respectively. NAG degrades chitin and peptidoglycan and is involved in both C and N acquisition. LAP is involved in N acquisition through the degradation of amino acids. AP is an esterase involved in P acquisition by breaking down phosphoesters. To measure extracellular enzymatic activity (EEA), first the filter media was gently rinsed with sterile 10 mM Tris (pH 8), and then biomass was extracted from the media into fresh Tris buffer using a bath sonicator. Activities were determined using fluorogenic substrates and appropriate standards, either 4-methylumbelliferone or 7-amino-4-methylcoumarin. Samples, standards, and negative controls were measured in duplicate in a black 96-well plate. Fluorescence was measured every 15 min for 1 h on a Synergy HT microplate reader using a 360 nm/460 nm filter (Biotek, Winooski, VT). Enzyme activity was calculated per German et al. (2011).

ATP, EPS, and EEA concentrations were normalized to the dry mass and volume of total solids used in the extraction. Total solids was quantified according to 2540 B (APHA et al., 2005). Bulk density was measured according to the method in Gülay et al. (2014).

Filter media were microscopically examined and elements were mapped under low-vacuum conditions using a Japan Electron Optics Laboratory (JEOL) 6490 LV scanning electron microscope (SEM) (Tokyo, Japan) equipped with an Oxford Instruments (Abingdon, United Kingdom) 50 mm² X-Max Silicon Drift Detector, running Aztec v.3.3 SP1 software. To investigate biofilm morphology under high-vacuum conditions, media samples were fixed using 2.5% glutaraldehyde in cacodylate buffer, washed and post-fixed with 1% osmium tetroxide, and washed again with distilled water. Dehydration was done with an ethanol dilution series of 25, 50, 75, 95 × 2 and 100% ethanol in water. Samples were dried in a desiccator for 48 h and sputter coated with gold using a Denton Desk IV (Denton Vacuum, Moorestown, NJ) vacuum evaporator for 90 s prior to imaging.

2.5. DNA extraction

For qPCR, DNA was extracted from the filter media samples (~0.25 g of each) according to the manufacturer’s protocol (PowerSoil DNA isolation kit; Mo-Bio, Carlsbad, CA, USA). For metagenomes, DNA was extracted using the PureFood GMO and Authentication cartridge-based purification kit with the Maxwell^® RSC 48 Instrument (Promega, Madison, WI) following the manufacturer’s instructions. DNA concentration was measured using a NanoDrop ND-1000 UV spectrophotometer (NanoDrop Technologies, Wilmington, DE). DNA extracts were stored at −20 °C.

2.6. Quantitative PCR

Thirteen qPCR assays (Table S2) to characterize nitrifying guilds and identify heavy metal oxidizers in the filter media samples were performed using a QuantStudio^™ 6 Flex system (Applied Biosystems). Reaction mixtures were prepared in duplicate with gBlock standards in MicroAmp Optical 96-well reaction plates with MicroAmp Optical Caps (Applied Biosystems, Foster City, CA). A standard curve for each qPCR assay was generated using serially diluted gBlock standards (IDT, Coralville, IA, USA). Undiluted DNA and 10- and 50-fold dilutions of all samples were used as needed to mitigate PCR inhibition. Detailed methods are provided in SI.

2.7. Sequencing and processing of metagenomic libraries

Paired-end genomic libraries were prepared using the Nextera XT Index Kit and sequenced on the HiSeq 2500 platform (Illumina Inc., San Diego, CA). Prior to assembly, the 150-nucleotide (nt) pair-end reads were subjected to quality filtering using the software package BBMap (http://sourceforge.net/projects/bbmap). The libraries contained an average of 23,594,235 ± 5,121,082 reads per sample. Libraries were de novo assembled using MEGAHIT v1.2.9 with default parameters but discarding contigs below 1500 nucleotides (Li et al., 2016). Contigs were annotated with MetaProkka (Seemann 2014; Telatin 2020).

Taxonomy was assigned using Kraken2 using the custom Genome Taxonomy Database release 202 (Parks et al., 2022; Wood et al., 2019). Taxonomy was summarized using Bracken (Lu et al., 2017). Metabolic reconstruction and the relative abundance of genes involved in key biogeochemical pathways were determined by DiTing and FeGenie (Garber et al., 2020; Xue et al., 2021). Normalized relative abundance was calculated for N, As, Mn, and C by dividing the relative abundance of a pathway in an individual sample by the sum of this pathway’s relative abundance in all samples; for Fe, calculated gene counts for each iron gene category were divided by the number of predicted ORFs in each metagenome.

To generate metagenome-assembled genomes (MAGs), binning was performed with MaxBin2 and MetaBat2 (Kang et al., 2019; Wu et al., 2016). Subsequently, the bins were optimized and dereplicated using DAS Tool (Sieber et al., 2018). Bins were consolidated using MetaWRAP (Uritskiy et al., 2018). Clusters were manually refined, and contaminants removed using MAGpurify (Nayfach et al., 2019). Bins were reassembled with MetaWRAP and continuously assessed for completeness and contamination with CheckM (Parks et al., 2015). Bins were de-replicated to generate a non-redundant set of MAGs using dRep (Olm et al., 2017). Taxonomy was confirmed using GTDB-Tk (Chaumeil et al., 2020), and relative abundance was assessed with CoverM (https://github.com/wwood/CoverM). Metabolic reconstruction of each MAG was performed with METABOLIC–C (Zhou et al., 2022). A phylogenetic tree of de-replicated MAGs was created with PhyloPhlAn using RAxML (Asnicar et al., 2020; Stamatakis, 2014). Detailed methods are provided in SI.

Sequence data have been submitted to the NCBI Sequence Read Archive (SRA) under the BioProject PRJNA883721 with BioSample numbers SAMN30987228–38.

3. Results

3.1. Water quality and treatment effectiveness

Table 2 summarizes the biofilter influent water quality and treatment performance. Additional water quality parameters throughout the treatment train are provided in Table S4. Typical treatment goals are NH₃ < 0.1 mg N/L, As < maximum contaminant level (MCL) of 10 μg/L, Fe < secondary MCL (SMCL) of 0.3 mg/L, and Mn < 20 μg/L. Most biofilters treated water containing multiple contaminants greater than these targets. Complete treatment systems at Sites 3a, 5, 6a, and 7 demonstrated excellent performance with treated effluent Fe ≤ 0.13 mg/L, Mn < 4 μg/L, As < 8 μg/L, and NH₃ < 0.15 mg N/L (Table S4). Only partial treatment trains were piloted at Sites 1, 2, 3b, and 6b, so the observed treatment performance might be less than would be achieved with the complete process.

Table 2.

Biofilter treatment performance and biomass production rates.

Biofilter	Influent Concentration					Removal (g/m3 reactor/day) [%]					Theoretical Biomass Production (g/m3 reactor/d)
	NH3 (mg N/L)	Fe (mg/L)	Mn (μg/L)	As (μg/L)	TOC (mg/L)	NH3	Fe	Mn	As	TOC
1-F	0.38	7.94	1136	23	4.9	4 (16%)	484 (92%)	−7.1 (−9%)	1.4 (91%)	126 (39%)	1.9
2-C	2.42	2.51	196	< 4	0.1	40 (59%)	16 (22%)	2.9 (53%)	0 (0%)	0 (0%)	13.2
3a-C1	3.22	0.09	< 1	< 4	1.1	26 (46%)	1.2 (76%)	0 (−33%)	0 (0%)	−1.8 (−10%)	8.6
3a-C2	3.22	0.09	< 1	< 4	1.1	26 (46%)	1.2 (76%)	0 (−33%)	0 (0%)	−1.8 (−10%)	8.6
3b-C	3.10	0.39	57	< 4	1.1	95 (>99%)	2.8 (23%)	1.6 (92%)	0 (0%)	−6.2 (−19%)	31.0
4-F	0.02	0.01	296	< 4	0.6	0 (0%)	1.1 (84%)	27.2 (>99%)	0 (0%)	3.7 (7%)	0.5
5-F	0.02	0.01	171	< 4	0.7	0 (0%)	0.4 (92%)	12.5 (>99%)	0 (0%)	3.5 (6%)	0.2
6a-C	3.79	3.18	16	20	5.6	322 (76%)	217 (59%)	0.8 (41%)	0.8 (37%)	15.3 (2%)	322.1
6a-F	0.89	1.31	10	12	5.5	111 (98%)	55 (99%)	0.3 (84%)	0.4 (84%)	4.3 (2%)	14.9
6b-C	3.79	3.18	16	20	5.6	176 (73%)	146 (72%)	0.2 (20%)	0.6 (46%)	9.8 (3%)	57.8
7-F	1.21	2.57	14	30	0.5	33 (88%)	75 (95%)	0.3 (74%)	0.7 (75%)	−9.4 (—66%)	10.8

Treatment goals: As < 10 μg/L, Fe < 0.3 mg/L, Mn < 20 μg/L, NH₃ < 0.1 mg N/L.

Detection limits: As = 4 μg/L, Fe = 0.001 mg/L, Mn = 1 μg/L, NH₃ = 0.03 mg N/L.

Filter 1-F treated the highest concentration of Fe and Mn and achieved the greatest Fe removal (484 g/m³ reactor/d) but negligible Mn removal, potentially because the DO and pH conditions were below the preferred region (DO > 7 mg/L, pH > 7.4) (Breda et al., 2019; Mouchet 1992). Filters 4-F and 5-F achieved the greatest Mn removal (27.2 and 12.5 g/m³ reactor/d, respectively) and produced effluent concentrations < 1 μg/L (Tables 2, S4). Notably, they were operated at a lower DO (5.4 mg/L) as compared to the other biofilters (6.9–10.6 mg/L), and the influent contained 0.8 mg ${\text{NO}}_3^{-}-\text{N}/\text{L}$ but no NH₃ (Table S4). Filter 7-F removed Fe and As and achieved nitrification. The aerated contactors treated higher concentrations of NH₃ than the filters (2.42–3.79 mg N/L vs. < 0.03–1.22 mg N/L) and removed 45–99% of the NH₃. Nitrification has a much higher oxygen demand including cell synthesis (3.9 mg O₂/mg NH₃) than the other contaminants (< 0.27 mg O₂/mg), and the contactors can maintain the needed DO levels (Table S1).

Differences in treatment performance were observed in gravel- and ceramic-packed parallel contactors at Sites 3 and 6. Site 3 piloted a contactor with ceramic media (3b-C) because the full-scale, gravelpacked contactors 3a-C1,2 were suspected to be clogged and not achieving the desired treatment performance. 3b-C achieved greater NH₃ removal (95 vs. 26 g N/m³ reactor/d) and better nitrification (99% vs. 46%) than 3a-C1,2, possibly due to better aeration or smaller media size. 6a-C achieved greater NH₃ removal than 6b-C, but nitrification was incomplete and the effluent contained 1.76 mg N/L nitrite $\left({\text{NO}}_2^{-}\right)$ (Table S4). 6a-C and 6b-C received the same influent water, so ${\text{NO}}_2^{-}$ accumulation likely resulted from differences in design and operation (e. g., shorter EBCT and larger media size resulting in a greater surface loading rate), but the exact cause was unclear. The nitrification performance suggests that ceramic media appears to be suitable for aerated contactors, but operational parameters such as loading rate, head loss, and backwashing need to be considered.

The theoretical biomass production rate was highest in aerated contactors and strongly correlated to the NH₃ removal rate (ρ = 0.91, p < 0.001) because NH₃ has a much higher yield than the other contaminants (0.33 vs. < 0.02 mg cells/mg donor, Table S1). Filters 4-F and 5-F, which only treated Mn, were expected to have low biomass production, and the actual rate might be lower due to abiotic removal mechanisms (Bruins et al., 2015).

3.2. Bulk media analyses

Established biofilms were observed on all media samples (Figs. 1, S1). Biofilters 2-C, 3a-C, 6b-C and 7-F, which treated NH₃ > 1 mg N/L, appeared to have the most surface coverage by visual inspection (Fig. S1), and the biofilms contained cocci embedded in polymeric sheets (Fig. 1B, C, H, I). Biofilm on the ceramic media in 3b-C was concentrated in the inner pores (Fig. 1D). Biofilm coverage in 4-F and 5-F, which treated high Mn waters, was more sporadic and mostly present as spherical clusters (Figure S1E, F). Twisted stalks and filamentous sheaths > 50 μm in length were visible in 1-F, 2-C, and 7-F (Figs. S1A, 1B, 1I).

Fig. 1.

SEM images from biofilters (A) 1-F sand, (B) 2-C GAC, (C) 3a-C2 gravel, (D) 3b-C ceramic, (E) 4-F sand, (F) 5-F sand, (G) 6a-F anthracite, (H) 6b-C ceramic, (I) 7-F anthracite.

Solid precipitates were observed on the media surface in every biofilter except 3a-C1,2, which did not treat Fe or Mn. Three morphologies were noted: small nodules present alongside cocci in 1-F, 2-C, 6a-F, and 6b-C; small nodules and thin rectangular sheets in 3b-C and 7-F; and flaky coral structures in 4-F and 5-F. The small nodules might be Fe oxides, and the coral structures might be the Mn mineral birnessite, as has been seen in other groundwater biofilters treating Mn (Bruins et al., 2015; Dangeti et al., 2020).

Elements detected on the media surface via EDS were generally consistent with the influent water quality (Table S5). Phosphorus was detected in biofilters where PO₄ was fed, and often concentrated P-rich masses overlapped with areas of high Fe concentration (Fig. S3). In 7-F, which did not supplement PO₄, P overlapped with both C and Fe in different localized areas, so some P might have been biofilm-associated, although inorganic and organic C could not be distinguished. PO₄ supplementation dose and location remain areas for optimization because it may improve nitrification in groundwater biofilters, but binding with Fe decreases its biological availability (de Vet et al. 2012; Lytle et al., 2013).

ATP, EEA, and EPS were analyzed to assess microbial activity and biofilm amount (Table 3). Media samples were collected from the influent and effluent ends of filters and contactors, respectively. In surface water and groundwater biofilters, biomass concentrations often are greater towards the influent end where substrate concentrations are greater (Gude et al., 2018; Pharand et al., 2014). However, aerated contactors have different flow and mixing compared to filters. Samples from the middle and influent end of 2-C and 3b-C also were collected and showed that these parameters did not vary substantially over the depth of the beds (data not shown). Thus, the difference in sampling location might have a limited impact on comparisons among the biofilters.

Table 3.

Media biofilm parameters.

Biofilter Sample ID	ATP (ng/cm3)	ACE (nmol/h/cm3)	BG (nmol/h/cm3)	NAG (nmol/h/cm3)	LAP (nmol/h/cm3)	AP (nmol/h/cm3)	EPS Poly. (mg glucose/cm3)	EPS Proteins (mg BSA/cm3)	PN: PS
1-F	97	250	n.d.	n.d.	126	12	0.03	1.60	47.8
2-C	18	24	0.13	0.17	n.d.	3	0.32	0.72	2.2
3a-C1	470	240	0.46	0.28	13	39	0.24	0.50	2.1
3a-C2	97	280	0.82	0.49	36	25	0.38	0.55	1.5
3b-C	120	360	0.55	0.84	49	60	0.06	0.21	3.8
4-F	34	49	n.d.	0.02	12	11	0.66	15.72	23.8
5-F	35	n.d.	0.01	0.12	6	3	0.39	15.77	40.8
6a-C	44	–1	–1	–1	–1	–1	–1	–1	–1
6a-F	820	1360	10.40	7.06	212	159	0.59	0.90	1.5
6b-C	620	1750	15.14	11.72	310	156	1.02	0.75	0.7
7-F	73	2240	4.09	2.48	144	196	1.65	1.44	0.9

ACE: acetate esterase, BG: β-glucosidase, NAG: β-N-acetyl-glucosaminidase, LAP: leucine aminopeptidase, AP: alkaline phosphatase, EPS: extracellular polymeric substances, PN:PS: EPS protein to EPS polysaccharide ratio, n.d.: non-detect.

¹Not measured due to insufficient sample volume.

ATP concentrations ranged from 18 to 820 ng/cm³ and the highest concentrations were measured in biofilters treating NH₃. This trend is consistent with these biofilters having greater theoretical biomass production rates, but the correlation was not significant (ρ = 0.55, p = 0.10). The biomass production rate did not include organic carbon transformations, which EEA assays and sequencing results suggested were important.

Five EEAs were quantified to assess the metabolic activity and nutrient condition (Sinsabaugh and Follstad Shah 2012). Strong positive correlations existed among all measured EEAs (ρ > 0.7, p < 0.02). There appeared to be a positive correlation between the total EEA and the ATP concentration, but it was not significant (ρ = 0.62, p = 0.054).

ACE activity was 2.1–2.8 orders of magnitude greater than BG activity across all sites, except Site 5 where no ACE activity and minimal BG activity were detected. ACE and BG activity were not related to the influent TOC concentration. The organic carbon content of groundwater generally is considered to be refractory and only 1-F achieved appreciable TOC removal (1.9 mg/L, 126 g/m³ reactor/d), although organic carbon transformations might occur in biofilters that are not reflected by the change in TOC concentration (Tables 2, S4) (Keithley and Kirisits 2019; McKie et al., 2015). Autotrophic microorganisms were expected to dominate these biofilters, but the comparatively high ACE activity among the assayed extracellular enzymes suggests that heterotrophic microorganisms might be important in groundwater biofilters.

The ratio among C-, N-, and P- acquiring EEA (i.e., ACE+BG: NAG+LAP: AP) may indicate the nutrient condition, with a ratio < 1 signifying a nutrient limitation (Sinsabaugh and Follstad Shah, 2012). There was no correlation between AP activity and the practice of PO₄ supplementation. The biofilters did not appear to be N- or P-limited with ACE+BG: NAG+LAP and ACE+BG: AP ratios >1 in all biofilters except 5-F, which had minimal microbial activity and was not expected to have a nutrient limitation.

EPS polysaccharides and proteins ranged from 0.03 to 1.65 mg glucose/cm³ and 0.21–1.62 mg BSA/cm³, respectively, across most biofilters. Biofilters 4-F and 5-F had much greater EPS protein concentrations of 15.7 mg BSA/cm³. The EPS protein to polysaccharide ratio (PN:PS) was much greater in biofilters where Fe or Mn was the primary contaminant (1-F, 4-F, and 5-F) as compared to nitrifying biofilters treating NH₃ > 1 mg N/L (23.8–47.8 vs. 0.7–3.8). The PN:PS typically is > 1 and sometimes as high as 20 in surface water biofilters and wastewater flocs and granules (Keithley and Kirisits, 2018; Liu and Fang, 2002). The nitrifying biofilters were more similar to these other engineering biotreatment systems, whereas the Fe- and Mn- biofilters were markedly different. EPS proteins have been associated with Fe and Mn sorption and oxidation, so the higher concentration in 1-F, 4-F, and 5-F might contribute to treatment performance (Sun et al., 2021). EPS and ATP concentrations were not correlated. To our knowledge, this is the first report of EPS concentrations in aerobic groundwater biofilters.

3.3. Microbial community and functional potential

3.3.1. Microbial community

Gammaproteobacteria, Nitrospiria, and Alphaproteobacteria were the top classes across all biofilters, and together they constituted 62–95% of the community, similar to other aerobic groundwater biofilters (Fig. 2A) (Gülay et al., 2016; Hu et al., 2020; Poghosyan et al., 2020). Archaea were present at very low relative abundance (< 0.1%) or were not detected. Frequently detected genera included nitrifying bacteria Nitrospira F, Nitrospira D, and Nitrosomonas and heterotrophic bacteria Xanthomonadaceae SCMT0 and Sphingopyxis (Fig. 2B). A genome-centric approach recovered 66 dereplicated MAGs from the 11 samples, and they were present at relative abundances ranging from 0.7 to 20.7% (Table S6). MAGs represent a microbial genome by binning assembled metagenomic data as a strategy to detect and genomically characterize organisms found in the samples (Yang et al., 2021). 18 of the MAGs were of nitrifying bacteria belonging to the genera Nitrospira F, Nitrospira D, Nitrosomonas, and Candidatus Nitrotoga. Other commonly detected MAGs included the methanotroph Methyloglobulus, the methylotroph Methylotenera, and Sphingopyxis and Sphingorhabdus B, which can degrade several organic contaminants (Deutzmann et al., 2014; Doronina et al., 2014; Glaeser and Kämpfer, 2014). Community diversity was similar across all biofilters (Shannon 3.58–5.01) (Table S7).

Fig. 2.

Microbial community composition at the (A) class and (B) genus levels revealed by metagenomics. Represented genera include the top 4 most abundant in each biofilter.

3.3.2. Nitrogen metabolism

Comammox bacteria appeared to be relevant nitrifiers in biofilters treating the full range of influent NH₃ concentrations based on qPCR results and recovered MAGs. Nitrospira were more abundant than Gammaproteobacteria or Nitrobacter in 6 of the 10 biofilters, which had influent NH₃ concentrations ranging from < 0.01–3.34 mg N/L (Fig. 3A). The microbial communities assembled via metagenomics similarly found that Nitrospira A, D, and F together were 2–300 × more abundant than Nitrosomonas in these 6 biofilters (Fig. 2). Comammox-amoA was detected in 4 of the 6 biofilters where Nitrospira was dominant as well as 2 others, which spanned the range of influent NH₃ concentrations, and it was highest in the mid category. Additionally, the amoA gene in the Nitrospira MAGs recovered from 4-F (one at 4.2% of the community) and 7-F (two at 16.7% and 3.9% of the community) clustered with the comammox clade. Gammaproteobacteria-amoA was detected in biofilters treating mid to high NH₃ concentrations (Fig. 3A, B). AOA-amoA was rarely detected, and its high concentration in biofilters treating a high NH₃ concentration was unexpected (Tatari et al., 2017). Low NH₃ concentrations are reported to favor comammox Nitrospira, and they have been found to be important nitrifiers in biofilters treating 0.2–0.9 mg NH₃-N/L, whereas Nitrosomonas were much more abundant than Nitrospira in a groundwater biofilter treating 0.9 mg/L NH₃-N (Gülay et al., 2016; Hu et al., 2020; Palomo et al., 2022; Poghosyan et al., 2020; Tatari et al., 2017). The results presented here indicate that comammox also may be important nitrifiers in biofilters treating NH₃ > 1.5 mg N/L, especially considering that NH₃ concentration gradients exist within the biofilter and biofilm.

Fig. 3.

Categorical gene copy number concentrations versus reactor influent concentrations for (A) ammonia 16S assays, (B) ammonia amoA assays, (C) ammonia nxrB assays, (D) manganese assays, (E) arsenic, and (F) iron. Gray bands indicate the “Mid” category for qPCR assays with 10⁵–10⁷ gene copies/cm³. Lighter shaded points were below the detection limit.

Nitrospira and Cand. Nitrotoga were the most abundant nitrite-oxidizing bacteria. Nitrospira-type nxrB was detected in every sample at low to mid concentrations, but Nitrobacter-type nxrB was not detected in any sample (Fig. 3C). Neither assay was related to the influent NH₃ concentration. 1-F, 2-C, 6a-F, and 6b-C contained Nitrospira 16S and Nitrospira-type nxrB but not comammox-amoA, suggesting that the Nitrospira in these biofilters were canonical nitrite-oxidizing bacteria. The Nitrospira-type nxrB concentration was 3–20 × greater than the comammox-amoA concentration in most biofilters that contained comammox bacteria, except 7-F where the comammox-amoA concentration was 90 × greater than the Nitrospira-type nxrB concentration. Interestingly, Nitrobacter was not detected in any sample via metagenomics, but the nitrite-oxidizing bacteria Cand. Nitrotoga was detected in all samples and constituted 10–12% of the community in biofilters 1-F, 2-C, and 3b-C. Cand. Nitrotoga tend to be cold-adapted and possibly have a competitive advantage over Nitrospira in high DO environments, such as those in an aerated contactor (Spieck et al., 2021). To our knowledge, Cand. Nitrotoga have not previously been identified in groundwater biofilters, possibly due to lack of recognition by 16S rRNA-based analyses (Navada et al., 2020).

Nitrification was the dominant pathway in the nitrogen cycle in all samples, and the relative abundance appeared to be related to the influent water quality, with 3a-C1,2 having the greatest and 4-F having the least (Fig. 4B). Functional genes for denitrification and dissimilatory nitrate reduction to ammonia (DNRA) also were present, indicating the potential functional diversity within the biofilters.

Fig. 4.

Normalized relative abundance of key metabolic pathways. Classification of low, medium, or high relative abundance applies across all biofilters within a contaminant category. Normalized relative abundance calculated for N, As, Mn, and C by dividing the relative abundance of a pathway in an individual sample by the sum of this pathway’s relative abundance in all samples; for Fe, calculated gene counts for each iron gene category divided by the number of predicted ORFs in each metagenome. ND: not detected.

3.3.3. Arsenic oxidation

The aioA gene was detected in all biofilters that treated waters containing As above the MCL of 10 μg/L and sometimes was detected when the water did not contain detectable As (Fig. 3E). The prevalence of arsensite oxidation genes among the sites as identified by metagenomics was similar to that identified by qPCR (Fig. 4D). Genes involved in arsenate reduction, which often is a detoxification mechanism, were much more abundant than genes involved in arsenite oxidation, as expected (Dunivin et al., 2019). The ability to oxidize arsenic is widely phylogenetically distributed, and some known As-oxidizing bacteria, such as Hydrogenophaga and Nitrospira, were present in all biofilters (Cavalca et al., 2019; Szyttenholm et al., 2020; Yamamura and Amachi 2014). The biofilters harbored As-oxidizing bacteria when As was present in the source water, which suggests that they played a critical role in the observed As removal by oxidizing As(III) to As(V).

3.3.4. Iron oxidation

Evidence of biological Fe oxidation was apparent in all biofilters except 7-F, although it is difficult to distinguish Fe removal mechanisms. Fe-oxidizing bacteria Gallionella and Crenothrix were detected in 1-F, 2-C, 3a-C1,2, 4-F, and 5-F via metagenomics and/or qPCR, and 4-F and 5-F contained genes related to Fe oxidation (Fig. 4C). The presence of Gallionella in aerated contactors suggests that some Fe may be oxidized biologically even under conditions conducive to chemical oxidation. 1-F had the greatest influent Fe concentration (8.51 mg/L), was operated with low influent DO and pH to favor biological Fe oxidation, and contained the greatest Gallionella 16S concentration (1.57 × 10⁹ gene copies/cm³). Amongst all biofilters, the Gallionella 16S concentration was not related to the influent Fe concentration (Fig. 3F). Greater than 95% of the influent Fe to 7-F was present as particulates, likely having been oxidized in the upstream aerators, so it is unsurprising that Fe-oxidizing bacteria were not detected.

Other Fe metabolism pathways, including gene regulation, transport, storage, and siderophore transport, were more abundant than oxidation and did not show a pattern among the biofilters (Fig. 4C). Importantly, no genes involved in Fe reduction were detected in any biofilter, which suggests that microbial activity might not contribute to Fe oxide destabilization if a process upset caused low DO conditions.

3.3.5. Manganese oxidation

Genera containing known Mn-oxidizing bacteria, such as Hyphomicrobium, Sphingopyxis, and Hydrogenophaga, were detected at all sites (Fig. 2) (Tebo et al., 2005). However, Pseudomonas spp., which often are present in other Mn-oxidizing biofilters, were infrequently detected (Bruins et al., 2017; Hu et al., 2020). Mn oxidizing genes were detected at all sites except 5-F (Fig. 3D). The three assays were positively correlated to each other, and their relative concentrations were the same across all biofilters: mopA > moxA > mofA. Although gene copy numbers were expected to be higher when the influent Mn concentration was higher, negative correlations between influent Mn and moxA (ρ = −0.72, p = 0.020) and mopA (ρ = 0.84, p = 0.0024) were observed, similar to Hu et al. (2020). Mn oxidation pathways were present in all biofilters except 3a-C1,2, which did not contain influent Mn (Fig. 4E). The lack of Mn oxidizing genes in 5-F and the low abundance of Mn oxidation pathways in 4-F and 5-F were surprising because Mn was the only contaminant in these biofilters and they removed > 99% of the influent Mn without oxidant application. These results highlight the important role of abiotic Mn removal mechanisms on aged biofilter media (Bruins et al., 2015; Breda et al., 2019).

3.3.6. Carbon transformation

The microbial communities contained both autotrophic and heterotrophic bacteria. Among the 18 most abundant genera across all samples, heterotrophs comprised 8–56% of the community. They were more abundant in 1-F (46%), which achieved high TOC removal (126 g/m³ reactor/d), and biofilters treating high NH₃ (21–56%, median = 35%), which had limited TOC removal (Fig. 2B, Table 2). The most abundant carbon metabolic pathways were central metabolism, fermentation to succinate and ethanol, methane oxidation, and carbon fixation (Fig. 4A). The relative abundance of carbon fixation pathways tended to be highest in biofilters treating NH₃ > 1.5 mg N/L. Methane is a common contaminant in anoxic groundwaters such as those included here, and Site 6 reported that methane was present in the groundwater. qPCR confirmed that the particulate methane monooxygenase gene, pmoA, was detected in 1-F, 3a-C1, 3b-C, 4-F, 5-F, and 6a-F at concentrations ranging from 6.5 × 10³ - 4.8 × 10⁶ gene copies/cm³. Additionally, MAGs of Methyloglobulus and Methylotenera were recovered from biofilters 5-F and 7-F (Table S6).

4. Discussion

Although differences in media elemental coatings, biofilm parameters, and community were expected among the biofilters given the differences in location, source water quality, treatment process configuration, and media type, trends were expected between the target contaminants and the microbial community composition and function. Fig. 5 shows how the biofilters clustered based on microbial community composition and the contributions of NH₃, pH, and TOC concentrations. Eleven water quality parameters were evaluated, and these three were selected because they had a more substantial impact. Biofilters 4-F, 5-F, and 7-F clustered together and were separate from the contactors that treated high NH₃. Nitrospira F, which included comammox MAGs in 7-F, and Palsa-1315 (class Nitrospiria) were more abundant in these biofilters and contributed to their dissimilarity. 6a-C and 6a-F had greater TOC concentration and pH compared to the biofilters at Site 3 (Tables 2, S4). The relative abundance of Nitrosomonas and two heterotrophs (Ga0077550 [class Polyangia] and Hydrogenophaga) were greater in the two biofilters compared to those at Site 3, which had a greater relative abundance of Cand. Nitrotoga and three heterotrophs (SCMT01 [class Gammaproteobacteria], Sphingopyxis, and Mycobacterium). Interestingly, 1-F clustered with the contactors at Site 3, which treated high NH₃ and low TOC, not with the other biofilters that primarily treated metals or the contactors that treated high TOC.

Fig. 5.

Non-metric multidimensional scaling of all biofilter communities at the species level based on Jensen-Shannon distance. Arrow size and direction represent the contribution of NH₃, pH, and total organic carbon (TOC) concentrations to community dissimilarity.

Biofilters 4-F and 5-F, which primarily treated Mn, had lower theoretical biomass production, ATP, and EEA but much higher EPS proteins as compared to the other biofilters (Table 3). The media surface contained Mn and trace amounts of Fe, and the biofilm appeared to sparsely cover the surface, exist in clusters, and contain flakey precipitates. A positive association between Mn-oxidizing gene abundance and Mn influent concentration was not observed, and Mn-oxidizing bacteria were not more abundant in these filters. 2-C also removed Mn, but its microbial community and other biofilm parameters were different from those in 4-F and 5-F. It is possible that there was less microbial activity in these biofilters as compared to those treating NH₃ due to less yield and greater physical/chemical Mn removal, but the biological activity nevertheless was integral to treatment performance, as seen in other groundwater Mn biofilters (Bruins et al., 2015).

Biofilters treating NH₃ had greater theoretical biomass production, biofilm coverage, ACE activity, relative abundance of organic carbon metabolic pathways, and relative abundance of heterotrophic bacteria capable of degrading complex organic carbon compounds, but TOC removal was limited (Table 2). Furthermore, 1-F was the only biofilter to achieve substantial TOC removal and its microbial community was similar to those in 3a-C1,2, 3b-C, and 6b-C (Fig. 5). Together, these results suggest that organic carbon transformations occurred within biofilters that treated high NH₃. Nitrifiers produce soluble microbial products that support the growth of heterotrophic bacteria, and this relationship might exist in groundwater biofilters treating elevated NH₃ (Kindaichi et al., 2004; Rittmann et al., 1994). To date, the role of soluble microbial products in groundwater biofilters has received little attention.

5. Conclusions

Groundwater aerobic biofilters treating primarily inorganic contaminants are complex, with various biofilm and precipitate morphologies and EPS concentrations, different degrees of metabolic and nutrient acquisition activity, diverse microbial community compositions, and expansive genetic functional potential. The influent composition affected the biofilm and microbial community, but all biofilters also harbored metabolic potential unrelated to the target contaminants.

Data availability

data will be made publicly available at EPA ScienceHub upon publication.