Microorganisms used in this study were retrieved from glycerol stock (-80 ºC) from a previous study on microbial isolation and characterization from environmental samples in West Java and Jakarta, Indonesia (25). Isolation was carried out on nitrate-rich liquid media under anaerobic conditions prior to colony selection on R2A agar plates (25). The isolates were checked for their possession of three central denitrification genes (nirS, nirK and nosZ) using primers available in the literature (26) and identified by 16S rRNA sequencing with primer pair 27F/1492R. Isolates were identified by BLASTn (bacteria and archaea) NCBI database (27) (https://www.ncbi.nlm.nih.gov/ ), aiming at a lower threshold identity of 98% (25). Soil bacteria were sampled from Depok, West Java, Indonesia, without any prior characterization.

Macronutrients were formulated following Bellini et al. (28), giving the composition of the media at: 1.641 g CH₃COONa, 0.02 g yeast extract, 0.85 g NaNO₃, 1.0 g NaCl, 0.5 g NH₄Cl, 0.0795 g CaCl₂·2H₂O, 2.7 g K₂HPO₄, 1.3 g KH₂PO₄, 0.235 g MgCl₂·6H₂O, 40 µL ethanol, 10 mL trace element solution, 10 mL vitamin solution (MEM vitamin solution; Sigma-Aldrich, Merck, St. Louis, MO, USA), 2 mL 0.2% resazurin indicator solution (Sigma-Aldrich, Merck), and distilled water to 1 L. Trace element solution was modified from Bellini et al. (28) from the initial formulation of Touzel and Albagnac (29). The solution consisted of: 1.24 g Titriplex III, 1.35 g FeCl₃·6H ₂O, 0.1 g MnCl₂·4H₂O, 0.1 g CaCl₂·2H₂O, 0.1 g ZnCl₂, 0.01 g H₃BO₃, 0.024 g Na₂MoO₄·2H₂O, 1.0 g NaCl and distilled water to 1 L. The MECs were filled with the medium prior to autoclaving. Heat-sensitive materials were sterilized by injection into each setup with 0.2-µm filters (hydrophilic Nylon₆₆ sterile syringe filter; Axiva Sichem Biotech, Delhi, India) once the reactors cooled down. Medium replenishment at 10% liquid volume of the setup was carried out once in the culture at the first current drop to 0.01 mA.

Single-chamber MEC reactors were built following Call and Logan (30). In this study, we increased the volumetric working capacity of the reactors to 100 mL (Fig. 1). Isomolded graphite plate (Graphitestore.com, Northbrook, IL, USA) was used as anode, connected to grade 2 Ti-wire (Ti-shop.com, London, UK). Stainless steel mesh, connected to stainless steel wire, served as a cathode. The electrodes were prepared following an existing protocol (30). External voltage was supplied at 0.7 V (P-3005 A; SUNSHINE Ltd., Guangzhou, PR China). Electrical measurements were continuously monitored over a 10 Ω resistor with a digital multimeter (APPA 109N; APPA, Taipei, Taiwan).

Anaerobiosis was achieved by vacuum-flush cycles of the reactor using ultra-high purity N₂ gas, as demonstrated in the literature (28, 30). Gases were filtered through 0.2-µm filters (hydrophobic politetrafluoroethylene sterile syringe filter; Axiva Sichem Biotech) during these cycles to maintain the sterility of the MECs. The setup was designed to be gas-tight and suitable for anaerobic culturing. To ensure proper sealing, we used a specific stopper designed for anaerobic culturing paired with a screw cap (GL-45 bromobutyl rubber stopper and GL-45 screw cap with aperture; DWK Life Sciences GmbH, Mainz, Germany). l-Cysteine HCl (Sigma-Aldrich, Merck) was added as a reducing agent in the medium to scavenge the leftover oxygen.

Headspace gas was monitored on a gas chromatography-thermal conductivity detector (GC-TCD) unit (GC-8A; Shimadzu, Kyoto, Japan) on an activated carbon column with argon gas as the mobile phase. Injector and column temperatures were set to 130 and 100 ºC, respectively. Headspace gas was injected at a volume of 1 mL using a gas-tight syringe (Hamilton, Merck, Reno, NV, USA). Syringe volume was calibrated with pure H₂ gas under normal atmospheric pressure (~101,325 kPa). Air, pure CH₄ and pure H₂ injections were made as peak identification controls.

Suspended bacterial growth was monitored periodically by measuring the absorbance at 600 nm on a spectrophotometer (UV-M90; BEL Engineering, Monza, Italy). At the end of the operation, we analysed the surface of the anodes by field emission scanning electron microscopy (FESEM; CMPFA Universitas Indonesia, Depok, Indonesia). Cells were fixed on the anodes following a modified chemical fixation protocol of Hrubanova et al. (31). Biofilms on the anodes were fixed using a 2.5% glutaraldehyde solution followed by repeated steps of ethanol dehydration, ending with air drying overnight at room temperature in a desiccator. Data presented in this study are average values of duplicate, independent MECs.

Soil microbiome was characterized at the end of the experiment through metagenomics approach on the culture medium using third-party services for gDNA extraction, next generation sequencing (NGS) of the V3-V4 region of the bacterial 16S rRNA, and bioinformatics analysis (Novogene, Singapore). The data is displayed as relative abundance. The phylogenetic tree of the designed consortium was generated from aligned sequences using MEGA7 software under the Clustal algorithm with a neighbour-joining method at a 1000 bootstrap value (32).

Hydrogen data fit to model and calculations of p-value were carried out using GraphPad Prism v. 8.4.3 for macOS (33). Gompertz model used to describe H₂ growth was obtained from the literature (34), formulated as follows:

where γ(H₂) stands for H₂ concentration (mg/L), γ(H₂)_max for maximum H₂ concentration (mg/L), r_max for maximum H₂ production rate (mg/(L·h)), λ for lag phase duration (h), and t for time (h).

Biohydrogen profile obtained using the system is shown in Fig. 2. At the beginning of the operation, the setup was run without external voltage. This period was designated as a preparatory period for the inoculum, during which no H₂ was produced (0–22 h) due to the lack of energy available to overcome the thermodynamic barrier of H₂ generation in MECs. Once the system was run with external voltage, both setups produced H₂ exponentially_. This is clear if we examine the 0–30 h period. MECs inoculated with native soil bacteria exhibited a decline after peaking at 35% of headspace H₂ concentration at 51 h of operation. The decrease in the headspace concentration when using soil inoculum started prior to medium replenishment. It continued even after the medium was replenished, unlike with designed consortium, which responded to medium replenishment with a slight increase in H₂ concentration. On the other hand, headspace H ₂ concentration with the designed microbial consortium stabilized at ~43% at the end of the observation (270–520 h), slightly under its peak value of 47% at 167 h. The two microbial consortia started to differ significantly in H₂ concentration at 99 h of the operation (p<0.05) and continued to do so until the end despite exhibiting similar profiles in the period leading up to this point (0–60 h).

The decrease of H₂ concentration in the headspace is likely to be attributed to a transformation into methane (methanogenesis) that is common in this type of reactor configuration. Methanogenesis was often found as a cause of failure to obtain biohydrogen in MECs. In the original study that inspired our setup, total conversion of H₂ to CH₄ occurred, leading to undetected quantities of H₂ in the headspace at the end of the culture (30). Our work has managed to maintain H₂ at a higher level throughout, ~45% with the designed consortium and ~18% with the native soil consortium over the observed culture period. Other works tried to avoid methanogenesis by physical or chemical means like adding antibiotics/inhibitors (35, 36), intermittent oxygenation ( 37) or ultraviolet irradiation (38). Each of these methods has its own benefits and limitations. For example, the addition of antibiotics in the culture medium poses a risk of a potential spill of the resistance trait over to the environment if care is lacking.

Aside from H₂, several other components of headspace gas were also detected (Fig. 3). These components are H₂, N₂ in the air, CO₂ and N₂O. Our method has a limitation in separating air into its molecular components of N₂ and O₂. Hence N₂ is referred to in the results as ‘N₂ in the air’. Additionally, O₂ is practically absent from the system due to the vacuum-flush cycle and the addition of oxygen-scavenging species in the media.

N₂O is an intermediary metabolite in the denitrification process (39). The existence of this gas in the headspace suggests that nitrate-reducing activity was present in the system. The presence of N₂ in the system is expected since the gas is used in the beginning to purge O₂ out of the system. Hence, it cannot be used as a marker for a complete denitrification process despite being the end product of the pathway.

Methane was interestingly absent from detection in this study. A possible explanation could be the transformation of methane into other metabolites like CO₂ in anaerobic methane oxidation, which could be the mechanism behind the significant jump in CO₂ concentration at the end of the cycle with soil MEC. The presence of CO₂ in MECs is otherwise normal since the breakdown of organic matter/substrate in the anode often results in CO₂ release (4). Methanogenic bacteria and denitrifying bacteria can interact in multiple metabolic pathways in the environment. For example, denitrifying anaerobic methane oxidation (DAMO) can be found in nature (40), which presents an opportunity for the same process to occur in the setup, leading to the transformation of produced methane to carbon dioxide. DAMO is currently of interest as a competing pathway to reduce methane in MECs ( 24). In the future, further characterization of the metabolic processes in the system is needed to confirm the presence of this interaction.

To validate the H₂ profile obtained in this study for the designed consortium, which resembles a regular growth curve of batch culture, we chose a common growth model to fit the data. Natural soil consortium possesses a different H₂ profile, likely due to the consumption of H₂, which does not suit the chosen growth model well. In batch culture, bacterial growth rate followed these well-known stages: lag, acceleration, exponential, slowing down, stationary and death phases. Growth models may take a linear form like Monod or non-linear forms, such as Gompertz and logistic models (41). The Gompertz model was selected for this study, relying on simple information related to H₂ evolution in the system. Meanwhile, Monod was unsuitable since it requires additional information related to substrate consumption. Nonetheless, the H₂ production in MEC with designed consortium can be described well using Gompertz model as presented in Fig. 4. The modified Gompertz model was first formulated by Zwietering et al. (42) and adjusted to describe H₂ in newer studies (34).

To consider the period of preparation before running the MEC, this period was excluded from the model (0–22 h). The correlation coefficient (R²) of the model fit was 0.973. Using the model, several parameters were obtained: r_max (mg/(L·h)), γ_max (mg/L) and λ (h). To determine the rate of H₂ generation r_max can be used. γ_max corresponds to the maximum H₂ concentration, while λ is related to the lag phase after initialization of the system. For this study, model fit values for r_max, γ_max, and λ were 0.247 mg/(L·h), 8.605 mg/L and 20.98 h, respectively. A confidence band based on a confidence interval of 95% was used to graphically present the true location of the curve ( Fig. 4). The fit of the model, assessed from its correlation coefficient, corresponds well to more than 20 values presented by Wang and Wan (34) in the range of 0.90–1.0 despite coming from a different setup than the batch fermentation method that is traditionally used to produce biohydrogen. This result suggests that H₂ production in MEC can be modelled like a traditional batch fermentation process, given that the H₂ profile matches those of ordinary batch growth/processes.

Characterization of the soil community is available in Fig. 5. Soil bacteria were detected based on metagenomics approach. This approach was chosen since it can provide a more expansive overview of genetic materials present in environmental samples, including from microbes that may be difficult to isolate and preserve in laboratory settings. The sensitivity of the method provides insight into the complexity of soil microorganisms.

The MEC inoculated with soil contained mostly bacteria of the class Gammaproteobacteria (48.54%) that includes both Pseudomonas and Acinetobacter genera. Pseudomonas and Acinetobacter are classified as Gram-negative bacteria, much like the other genera dominating natural soil consortia. The medium used in this study contains high nitrate concentration, requiring bacteria to possess the necessary adaptive ability to survive. For designed consortium, the bacteria were preselected based on their capability to survive in a nitrate-rich environment as well as to metabolize nitrate using the denitrification pathway (25). The use of denitrifying bacteria as competitors to methanogenesis is based on the idea that metabolites released from denitrification may act as inhibitors to methanogens in soil samples (17).

Differences in H₂ content in the headspace can be attributed to the microbiome inside the system (Fig. 2). Soil microbiome was dominated by several genera, in descending order: Pseudomonas, Brucella, Achromobacter, Bordetella, Klebsiella, Lachnoclostridium 5, Stenotrophomonas, Clostridium sensu stricto 18, Lactobacillus and Acinetobacter. On the other hand, the designed consortium consists only of several species belonging to two genera, Pseudomonas and Acinetobacter, which are also present among the top ten genera in the soil microbiome. This is in agreement with the fact that the two genera were originally isolated from soil samples in Indonesia. Hence, we drastically reduced the complexity of the community by reducing a rich source of microbiome to only two genera. Future adjustments of the composition of the designed consortium could include well-documented electroactive bacteria like Geobacter to further facilitate electron transfer and H₂ production (43, 44).

Exoelectrogenic microorganisms may transfer electrons to their environment by physical or chemical means. Physical mechanisms include the presence of structural nanowires, a term for electrically conductive pili (45). Chemically, electron transfer may occur through mediators secreted by the bacteria, i.e. pyocyanin by Pseudomonas aeruginosa (46). In MECs, electroactive species present in the medium can spontaneously attach themselves to form biofilms on the surface of the anodes (47). In this study, the physical states of anodes post-operation were analysed by SEM imaging. The anodes were treated prior to imaging following a modified approach from existing literature (31) to prevent structural degradation of biofilms, as presented in Fig. 6. Chemical methods utilized to fix the biomass prior to imaging involve repeated washing, which may degrade the extracellular matrix of biofilms present on the surface (circled in yellow). However, it is a more straightforward method than cryotreatment, and suitable for surface imaging (31).

The presence of biofilms on the anodes suggests that in both consortia physical transfer of electrons is possible. Additionally, given that Pseudomonas comprise the majority of the designed consortium and a major fraction of the soil consortium, it is also possible that pyocyanin-secreting species are present, hence allowing mediator-based electron transfer. In future studies, it would be interesting to analyse and compare the composition of microbiomes found on the surface of the anodes with microbes suspended freely in the medium. A similar imaging approach was used in literature (23) where they managed to show nanowires used for direct interspecies electron transfer. This approach is interesting to use for consortium with distinct morphological differences. In our case, the bacteria were morphologically similar.