All chemicals were of analytical or biochemical grade. All provided potentials refer to the standard hydrogen electrode (SHE) by conversion from Ag/AgCl saturated KCl reference electrodes (+0.197 V vs. SHE). Experiments were performed at room temperature (23 ± 2°C).

Four reactors were constructed using customary polypropylene (PP) tubes of 1 m length and 4.6 cm inner diameter (Marley Deutschland GmbH, Germany) (Figure 1A). Three reactors (reactors 1–3) served as experimental replicates. One reactor was used as control, hosting no sampling units to test if the incorporation of these sampling units influences reactor performance, for instance, by the intrusion of oxygen in the anaerobic zone of the electrified biotrickling filters. Titanium rods (grade 2, 5 mm diameter, Goodfellow Cambridge Ltd., UK) connected to titanium meshes (25 × 5 cm, 0.28 mm wire, Alfa Aesar, U.S.A.) were used as CC for the working electrode (i.e., cathode). A titanium rod and an Ag/AgCl electrode (saturated KCl, +0.197 V vs. SHE, SE 11, Xylem Analytics Germany Sales GmbH & Co. KG Sensortechnik Meinsberg, Germany) were integrated as anodic CC and reference electrodes, respectively. The in-house developed granule sampling unit consisted of 2 cylinders made of polyvinyl chloride (PVC), the outer cylinder of the sampling unit was made of acrylic glass and fixed with glue to the PP tube. The sampling window was 12 × 8 mm, and the sampling pocket of the inner cylinder had a depth of 5 mm (Quejigo et al., 2021). At the bottom of the reactor, a PVC cap with a tube connector as an effluent port and a titanium mesh (5 × 5 cm) were integrated to prevent clogging of this port by granules. All components were glued to the PP tube using vulcanizing glue (Tangit PVC-U Plus, Henkel AG & Co. KgaA, Germany).

Graphite granules (model 00514, Enviro-cell, Germany) were sieved to select granule sizes of 1.5–5.0 mm. Granules were consecutively washed with tap water, 1 M NaOH (for 24 h), tap water, 1 M HCl (for 24 h), and distilled water. The lower and the upper half of the reactors were filled with granules (703 ± 29 g in 50 cm resulting in a liquid net volume of 285 ± 5 ml) and glass beads (diameter 2 mm, 1,420 ± 42 g in 50 cm, resulting in a liquid net volume of 331 ± 24 ml), respectively. The simple and cost-effective open reactor design resulted in an aerobic zone in the upper half of the reactor enabling ammonia oxidation. Consequently, oxygen was consumed along the reactor so that the lower half was filled with conductive granules, which became anaerobic allowing bioelectrochemical denitrification therein. The control reactor was filled with 789 g granules (liquid net volume of 327 ml) and 1,383 g glass beads (liquid net volume of 365 ml) in the lower and upper half, respectively.

The medium was stored in a 20 L polyethylene canister and pumped with a four-channel pump (REGLO Analog, ISMATEC, Germany) using pumping tubes (inner diameter = 1.52 mm, Tygon LFL, SC0424, ISMATEC, Germany) and tubes (inner diameter = 1.6 mm; TYGON® E-3603, Saint-Gobain Performance Plastics, France). The tubes were sterilized by pumping 2.5% H₂SO₄ for 12 h, followed by cleaning with distilled water before use. The canisters were washed with 2.5% H₂SO₄ and distilled water before use.

In total, 2 L effluent from previously described biotrickling filters (Pous et al., 2021) were centrifuged (10 min, 10,000 g). Cell pellets were resuspended in 1.5 L synthetic aquaculture effluent, containing 50 mg L⁻¹ N-NH4+ and 50 mg L⁻¹ N-NO3-, which served as the medium for all experiments (0.162 g Na₂HPO₄, 1,072 g KH₂PO₄, 0.1 g MgSO₄ × 7 H₂O, 0.015 g CaCl₂, 0.25 g NaCl, 1.05 g NaHCO₃, 0.19 g NH₄Cl, 0.30 g NaNO₃, 12.5 ml vitamin solution, and 12.5 ml trace element solution dissolved in 975 ml distilled water) (Pous et al., 2021). In total, 350 ml of this inoculum solution were added to every reactor. The cathode potential (E_Cat) was controlled with a potentiostat (VSP, BioLogic Sciences Instruments, France) and initially adjusted to −100 mV vs. SHE. After 2 days without pumping, reactor liquid was pumped in a closed-loop mode with a hydraulic retention time (HRT) of 24 h. The inflow and outflow were at the top and bottom of reactors, respectively. On day 14, 350 ml of medium with 100 mg L⁻¹ of N-NH4+ and 100 mg L⁻¹ of N-NO3- was added, and the HRT was adjusted to 3 h for 2 more weeks for increasing mixing and facilitating biomass growth.

On day 30, the continuous operation was initiated with 12 h HRT and the synthetic aquaculture effluent. As bioelectrochemical denitrification requires anoxic conditions, 20 L of fresh medium were sparged with N₂/CO₂ (80%:20%) for 2 h. On day 54, the liquid level was adjusted to 75% of reactor height, and HRT was set to 24 h approximating operational parameters applied in the previous study (Pous et al., 2021). Due to the limitations of the potentiostat (refer to the Section Reactor Operation and Removal Performance), the initial E_Cat of −100 mV could not be applied during continuous mode and had to be gradually adapted. After 100 days, this adaptation phase was finished, and the operational phase started with an applied E_Cat of +250 mV. The pH of the influent during the operational phase was 7.0 ± 0.2.

Granules were extracted from day 129 onward for performing electrochemical and molecular biological analyses. Electrochemical data from electrified biotrickling filters were collected every 10 min using EC-Lab V11.31 (BioLogic Sciences Instruments, France).

Ammonium (N-NH4+) was photometrically measured with an alkaline dichloroisocyanurate-salicylate-nitroprusside-based assay at a wavelength of 660 nm (DIN ISO 105661999-4). Nitrite (N-NO2-) was photometrically measured with sulfanilamide and N-(1-naphthyl)ethylenediamine at acidic conditions at a wavelength of 540 nm (DIN EN 26777). Nitrate (N-NO3-) was determined by means of total oxidized nitrogen (TON) (EN ISO 13395). After reducing nitrate to nitrite with hydrazine under alkaline conditions, the absorbance at 540 nm was measured and related to TON. Subsequently, the nitrate concentration was calculated by subtracting nitrite concentration from TON. Ammonia, nitrate, and nitrite were determined with Gallery Plus Discrete (Thermo Fisher Scientific Inc., U.S.A.) using calibration curves.

Ammonium removal and total nitrogen (N-TN) removal were calculated using the concentration differences between influent and effluent of ammonium and all determined nitrogen species (i.e., N-NH4+, N-NO3-, and N-NO2-), respectively. The ammonium removal rate (N-NH4RR+) and total nitrogen removal rate (N-TN_RR) were calculated considering the total net volume of reactors and flow rate.

The dissolved oxygen concentration in the effluent was measured using a WTW FDO® 925 sensor (Xylem Analytics Germany Sales GmbH & Co. KG. Germany).

The e-clamp was an advancement from the originally reported version (Quejigo et al., 2018). The outer e-clamp parts were made of polyether ether ketone (PEEK) (Figure 1B). The grab hook consisted of four spring steel wires (V4A 1.4571, diameter 0.6 mm, Febrotec GmbH-Federn, Germany), which were pressed together in a stainless steel sleeve that also represented the electrical connector at the top of the e-clamp. The clamp mechanism was achieved by integrating two springs (stainless steel 1.430, wire diameter 0.45 mm, spring diameter 4.05 mm, and spring length 17.7 mm, Federntechnik Knörzer GmbH, Germany). The electrically conductive parts inside the PEEK sleeve and the grab hook were insulated with acrylic varnish. Only 1 cm of each grab hook wire was electrically conductive.

Granules were extracted for performing cyclic voltammetry using the sampling units, fixed with the grab hook of the e-clamp, and directly transferred to four-neck round-bottom flasks with an integrated three-electrode setup (Supplementary Figure S1) (Quejigo et al., 2021). The e-clamp was pierced through a silicone stopper and represented the working electrode, 250 ml medium was purged for 25 min with N₂/CO₂ (80%:20%) before the e-clamp was inserted. The headspace was constantly purged, the medium (synthetic aquaculture effluent containing 50 mg L⁻¹ N-NH4+ and 50 mg L⁻¹ N-NO3-) was stirred with a magnetic stirrer, and a temperature of 35°C was maintained during CV. 20 ml of the same medium but without NH4+ and NO3- was used in the counter electrode chamber. CV was conducted from −600 to +600 mV with a scan rate of 1 mVs⁻¹ using a potentiostat (MPG-2, BioLogic Sciences Instruments, France). Between CVs, the pH of the media in the main chamber was adjusted using 2 M HCl, 4 M KOH, and a pH meter (LAQUAtwin pH-22, HORIBA Advanced Techno, Japan). After experiments, granules were stored at −20°C. To determine the granule weight and thus gravimetric current density, all granules were dried at 80°C overnight after DNA extraction (Section Molecular Biological Analyses) and stored in a desiccator until weight stabilization. Electrochemical data were collected with EC-Lab V11.31 and analyzed with OriginPro 2019 (Version 9.6.0.172, 64-bit).

During the analysis of the pH influence on the normalized gravimetric current density (refer to the Section Analyzing the Influence of pH on Bioelectrochemical Denitrification), the Grubbs test was used to remove outliers at a significance level of 0.01 using OriginPro 2019.

Genomic DNA was extracted from sampled granules with the NuceloSpin Tissue Kit (Macherey-Nagel GmbH & Co. KG, Germany) for terminal restriction fragment length polymorphism (TRFLP) analysis as previously described (Korth et al., 2020) and metagenomics sequencing. By doing so, the amounts of lysis and extraction buffer were adjusted to cover all granules within a reaction tube. The amount of ethanol for DNA precipitation was adjusted accordingly. To perform genome-centric analysis of the functional potential present in the microbial community, two granule samples were submitted to GENEWIZ Germany GmbH (reactor 1, 1st sampling unit, day 253 and reactor 3, 1st sampling unit, day 283) for metagenome sequencing. The metagenomics libraries were prepared using NEBNext® Ultra (New England BioLabs Inc., U.S.A.) according to the manufacturer's instructions. After these metagenomics libraries were sequenced with an Illumina NextSeq 500 (Illumina Inc., U.S.A.) according to the manufacturer's instructions, ~25 million paired-end reads (150 bp each) were obtained prior to quality control. To recover metagenome-assembled genomes (MAGs), MetaWRAP version 1.1.7 (Uritskiy et al., 2018) was used according to the developer's instructions. Then, we selected only MAGs with quality scores higher than 50 (quality score = completeness – 5 × contamination) (Parks et al., 2017). Taxonomy was assigned to the MAG using GTDB-Tk v0.3.3 (Chaumeil et al., 2020). To define the operational taxonomic units (OTUs), the MAGs were grouped per unique taxonomy and clustered using hierarchical agglomerative clustering with the average nucleotide identity (ANI) distances. Generally, 0.99 ANI distance was used as the threshold as a proxy of the species level.

To study the functional capacity for the nitrogen cycle (KEGG modules: M00531, M00530, M00529, M00528, and M00804) (Kanehisa et al., 2016), the recovered genomes were annotated using Prokka version 1.14.5 (Seemann, 2014). To expand the default Prokka database, a custom database was created by downloading all the proteins related to KEGG modules (download in June 2021). After annotation, all predicted protein names were manually standardized, and a table of presence or absence was created for each OTU.

Three electrified biotrickling filters with integrated sampling units and one control reactor without were operated for around 300 days. The first 100 days were the inoculation and adaption phase in which the flow rate, liquid level, and cathode potential (E_Cat) were adjusted for balancing process parameters. The application-oriented, simple, and cost-effective open reactor design resulted in an aerobic zone filled with glass beads in the upper reactor part (Figure 1A). In this study, aerobic ammonium oxidation occurred according to the previous publication (Pous et al., 2021). However, if oxygen diffuses to the lower electrified reactor zone filled with graphite granules, it is electrochemically reduced. This leads to high negative currents that are unrelated to the desired process of bioelectrochemical denitrification. Therefore, the initially applied E_Cat of −100 mV, known to facilitate bioelectrochemical nitrate reduction (Pous et al., 2014), was stepwise changed during the adaptation phase to +250 mV. After 100 days, the operational phase started, and the data obtained therefrom were used for all analysis. Despite the more positive E_Cat, removal rates of ammonium (N-NH4RR+) and total nitrogen (N-TN_RR) of 34.0 ± 12.1 gN m⁻³ day⁻¹ and 43.4 ± 28.4 gN m⁻³ day⁻¹, respectively, were achieved (Table 1). An activity stratification within electrified biotrickling filters was in place by comparing process parameters derived from sampling influent, middle liquid port, and effluent. Specifically, although the main share of ammonium was oxidized in the upper aerobic zone, nitrate was mainly reduced in the electrified anaerobic zone at the bottom. Furthermore, considerable amounts of nitrite (30–40 mgN-NO2- L⁻¹) were detected in the middle at the beginning of the operational phase (~days 100–125). In the following days, only minor amounts of nitrite were detected in the reactor middle (1.9 ± 3.7 mgN-NO2- L⁻¹ for days 130–300) and the effluent (2.4 ± 3.9 mgN-NO2- L⁻¹ for days 130–300) indicating an improved total nitrogen removal due to growth of denitrifying microorganisms (see Supplementary Figures S2–S5 for the evolution of process parameters of all reactors).

Similar removal rates were observed for the control reactor (Table 1), showing a minor influence of the sampling units on the overall reactor performance. Moreover, the observed N-NH4+RR was similar and N-TN_RR even higher as reported by Pous et al. using the similar HRT and liquid level (Pous et al., 2021) (Table 1). Yet, it is noted that in addition to the cathode potential, the medium composition of both studies was different. The medium used in this study contained ammonium and nitrate. Typically, aquaculture effluents contain mainly ammonium but almost no nitrate (Yin et al., 2018). We amended the medium in this study by nitrate to foster an electrified zone that performs bioelectrochemical denitrification and to ensure a subsequent electrochemical analysis. Anaerobic conditions within reactors were verified by measuring the dissolved oxygen (DO) content in the effluent. It amounted to 0.23 ± 0.07 mg L⁻¹ (4 measurements per reactor between days 198 and 256) and thus was comparable to the proof of concept (Pous et al., 2021).

During the operational phase starting at day 129, graphite granules were collected from the electrified biotrickling filters via the minimally invasive sampling units. Subsequently, cyclic voltammetry was performed with sampled granules using the e-clamp in the presence of ammonium and nitrate (Supplementary Figure S1). In total, 93 granule samples were taken from the reactors during the operational phase. In addition, when terminating the reactor operation, all reactors were sampled along the vertical axes of the graphite granule bed (i.e., 6 sample spots regularly distributed along 50 cm of graphite granule bed, Supplementary Figure S6A), yielding further 24 granule samples. However, graphite granules are porous and only exhibit limited physical contact, resulting in an inhomogeneous redox potential (Newman and Tobias, 1962). Thus, the electrochemical driving force for EET and the ecological niche for EAM is not identical at all granules. This leads to different biomass growth at granules, and thus only ~60% of the sampled granule exhibited an electrochemical signal that could be analyzed.

During the electrochemical analysis of sampled granules (refer to the Section Cyclic Voltammetry With Sampled Graphite Granules), pH was sequentially changed in the range of 6–10, while CV was performed for each pH (Figures 2C,D). For analyzing pH dependence, j at the beforehand determined formal potentials E_f1 and E_f2 were quantified at every pH (j_pHi) and normalized to the gravimetric current density at pH 7 (j_pH7). Intriguingly, both redox sites exhibit a nearly constant current consumption in the pH range 6–8 (Figures 2E,F), representing the typical pH optimum for microbial denitrification (Tiedje, 1988). Only at pH 9, the normalized j decreased below 80% compared with pH 7, which was continued to pH 10 achieving around 46% (E_f1) and 65% (E_f2) activity. Therefore, electrified biotrickling filter seems to be an appropriate treatment method for aquaponics wastewater as pH 6–7 is the optimal operating range of aquaponic plants. In addition, it also provides a safety margin in case of unwanted alkalinization (Food Agriculture Organization of the United Nations, 2014). Although certain pH robustness is known for denitrifying microbiomes reporting microbial denitrification at up to pH 11–12 (Albina et al., 2021), this is, to our knowledge, not reported for bioelectrochemical denitrification. This broad pH range further supports the biotechnological potential of bioelectrochemical denitrification and extends its fields of application to moderate alkaline environments. As CV results indicate that the pH optimum is in the pH range 6–8, electrified biotrickling filters are suited for application in aquaponics systems. Plants prefer slightly acidic conditions (pH 6.0–6.5) improving nutrient availability and most fishes exhibit a pH tolerance range of 6.0–8.5 (Food Agriculture Organization of the United Nations, 2014).

In most cases, E_f1 and E_f2 shifted to more negative values for more alkaline pH during CV experiments (Figure 2C) following a typical Nernst behavior and the redox-Bohr effect already described for anodic EAM and their cytochromes (Morgado et al., 2012). To assess the effect of formal potential shift on the pH influence on bioelectrochemical denitrification, normalization of j was also performed by quantifying j at the shifted formal potentials for all studied pH values. This analysis revealed an even broader pH optimum. For E_f1, j was comparable for pH 6–9 and decreased to 56% at pH 10, but in the case of E_f2, no apparent differences could be observed for the tested range of pH 6–10 (Supplementary Figure S8). However, as the formal potentials shifted to more negative values, it cannot be excluded that the hydrogen evolution reaction contributed to the current. Furthermore, it can only be speculated if the applied pH range already led to conformational changes of the redox-active proteins and other cellular components affecting cathodic EET.

Sampled granules were also analyzed with molecular biological methods for investigating the microbial community composition within the granule bed of the electrified biotrickling filters. Due to the reactors' autotrophic conditions and the resulting limited growth, some granule samples were pooled for obtaining sufficient DNA required for molecular analysis (1–3 granule samples were used for DNA extraction).