To confirm the reactor preparations, the Faradaic efficiency of the H₂ evolution reaction (HER) was measured in several representative reactors without any microbial inoculation. The obtained Faradaic efficiencies of HER were approximately 40%–60%, which were lower than those reported previously (96%: Liu et al., 2016). It has to be noted that this difference in Faradic efficiencies could be due to the distinct experimental setup used in those studies. The Faradic efficiency was measured in a two-chamber electrochemical cell using 100 mM KPi buffer as the electrolyte in the prior study (Liu et al., 2016), while this study used the single-chamber reactor with the high-ionic-strength medium as the electrolyte (the same setting as the following enrichment experiments). Thus, the observed lower Faradaic efficiencies could likely be attributed to hydrogen recycling, where H₂ produced at the cathode was abiotically oxidized at the anode. Additionally, the presence of possible side reactions, such as interactions between metal ions in the medium and the electrodes, may have contributed to the reduced efficiencies. Yet, the results confirmed that the reactor performance was sufficient for the enrichment purpose.

To enrich halotolerant HOB, environmental samples from brackish/marine environments were inoculated into HBI reactors using a high-ionic-strength medium containing 180 mM phosphate buffer as the culture medium/electrolyte (Table 1). Then, the reactors were operated under the same condition (except the initial inocula). The pH of the medium was around 7.1 at the beginning of operations and no fluctuation was observed. To estimate HOB growth in the reactors, H₂ and CO₂ concentrations in the reactor headspace, the medium turbidity, and theoretical H₂ production (estimated from the current flowing through the reactor circuit) were monitored (Figure 1). When H₂ and CO₂ consumption were observed markedly, a few microliters of reactor medium were inoculated into a new reactor to start the next enrichment cycle. For each inoculum source, three enrichment cycles were performed. In general, initial lag periods with various durations were observed in the reactors, particularly in reactors H2_1 and H3_1. After lag periods, CO₂ and H₂ consumption were initiated simultaneously, and an increase in medium turbidity was observed in all reactors. Furthermore, the durations until the H₂ concentration in the headspace became undetectable were shortened in later enrichment cycles (approximately 400 and 200 h for the first and third cycles, respectively), indicating successful enrichment of halotolerant HOB in the reactors under a high-ionic-strength condition.

The phylogenetic diversity of bacterial communities enriched in the reactors was assessed using 16S rRNA gene amplicon sequencing. Alpha diversity was calculated using the amplicon sequence variant (ASV) data (Figure 2A). The number of ASVs observed in the third cycle was markedly lower than that observed in the first cycle, indicating that specific bacteria were selected through the enrichment cycles. Principal coordinate analysis based on Bary–Curtis distances (Supplementary Figure S1) showed a clear distinction between reactors H1, H3, and reactors H2 with H4, indicating that the origin of the initial inoculum significantly affects the resulting microbial enrichments.

Taxonomic annotation of the amplicon sequences revealed high relative abundances of seven genera in the reactors (Figure 2B). The genus Achromobacter was the dominant genus in the reactors of series H1, accounting for 72.32%, 76.54%, and 76.12% of the sequences from reactors H1_1, H1_2, and H1_3, respectively. In the reactors of the H2, H3, and H4 series, the genus Mycolicibacterium was highly abundant. In particular, the relative abundance of Mycolicibacterium increased after the enrichment cycles in series H2 (from 19.78% in H2_1 to 26.09% in H2_3) and H3 (from 56.35% in H3_1 to 67.38% in H3_3). Mycolicibacterium also showed the second highest abundance in reactor H4_3 (15.37%) behind Hydrogenophaga, which itself showed an increase in abundance in series H4 (from 13.89% in H4_1 to 46.72% in H4_3) and a high abundance in series H2 (63.25% in H2_1). In our previous study, HOB species of the genera Mycolicibacterium and Hydrogenophaga were enriched in HBI systems using a medium with 108 mM phosphate buffer (Feng et al., 2023). Additionally, three putative nitrate-reducing bacteria, including Ancylobacter (increased to 27.15% in H2_3), Nitratireductor (increased to 4.23% in H3_3), and Nitrobacter (increased to 5.02% in H4_3), showed an increase in abundance with increasing enrichment cycles.

After the third enrichment cycle, HOB were isolated from the culture media through single colony isolation, resulting in the successful isolation of 13 strains as follows: strains H1_3_1 and H1_3_2 from reactor H1_3; strains H2_3_1 and H2_3_2 from reactor H2_3; strains H3_3_1, H3_3_2, H3_3_3, and H3_3_4 from reactor H3_3; and strains H4_3_1, H4_3_2, H4_3_3, H4_3_4, and H4_3_5 from reactor H4_3 (Supplementary Figure S2). All 13 isolates were capable of chemolithoautotrophic growth in the high-ionic-strength liquid medium with an H₂/CO₂/O₂ (80:10:10) gas mixture. Additionally, chemolithoautotrophic growth in the low-ionic-strength medium (containing 36 mM phosphate buffer) and aerobic organotrophic growth (in Luria–Bertani medium) were also observed (data not shown). Therefore, all isolates were identified as facultative aerobic HOB that were capable of thriving under a high-ionic-strength condition.

The 16S rRNA sequences of the two isolates from reactor H1_3 were nearly identical to each other and had high similarity with species of genus Achromobacter. The other 11 isolates from reactors H2_3, H3_3, and H4_3 were also highly similar to each other and closely related to genus Mycolicibacterium. This result is consistent with the phylogenetic compositions of microbial communities enriched in the reactors, wherein Achromobacter and Mycolicibacterium showed high relative abundance. Two representative isolates, strains H1_3_1 and H4_3_1, were selected for further analyses.

To better understand the mechanisms underlying the chemolithoautotrophy, hydrogen oxidization, and halotolerance of the isolates, the whole genome sequences of representative isolates were determined. The strains H1_3_1 and H4_3_1 each contain a single circular chromosome of approximately 7.1 and 7.6 Mbp, which showed overall similarities of 70.7% and 92.4% (in the DNA–DNA hybridization values) with the type strains of A. xylosoxidans LMG 1863 (GCF_000508285.1) and M. mageritense JCM 12375 (GCF_010727475.1), respectively, indicating that the isolates belong to these species.

Functional annotation of the genomes revealed that both strains encode genes for a complete Calvin–Benson–Bassham cycle as well as genes for a complete tricarboxylic acid cycle and oxidative phosphorylation, indicating their capability of carbon fixation and aerobic respiration. The genome of A. xylosoxidans strain H1_3_1 encodes two O₂-tolerant hydrogenases, a Group 1d [NiFe]-hydrogenase and a Group 2b [NiFe]-hydrogenase, which are a membrane-bound respiratory H₂-uptake hydrogenase and H₂-sensing regulatory hydrogenase, respectively. The genome of M. mageritense strain H4_3_1 encodes one O₂-tolerant hydrogenase, a Group 2a [NiFe]-hydrogenase, which can use O₂ as the terminal electron acceptor. Both genomes also contain homologs of hyp genes, encoding proteins for the maturation of [NiFe]-hydrogenases (Greening et al., 2016; Fan et al., 2022).

Notably, a complete set of genes for ectoine/hydroxyectoine biosynthesis (ectA, ectB, ectC, and ectD) are found in both genomes, suggesting that these HOB strains cope with the high-ionic-strength environment in the reactor by producing the compatible solutes.

The chemolithoautotrophic growth of M. mageritense had been demonstrated in our previous study. On the other hand, so far chemolithoautotrophic metabolism has never been reported in A. xylosoxidans, while two members of genus Achromobacter, A. ruhlandii LMG 1866 and A. veterisilvae LMG 30378, have been reported to be HOB (Packer and Vishniac, 1955; Dumolin et al., 2020). Thus, the strain H1_3_1 was characterized and compared with the type strain of A. xylosoxidans.

A. xylosoxidans strain H1_3_1 were motile and small bacilli with a length of 1.0–2.0 μm and a width of 0.5–0.6 μm. Colonies were convex, nontransparent, cream color, with smooth margins and a diameter under 1 mm. Cells were Gram negative and spore formation was not detected. Growth was observed at all temperatures tested (28°C, 37°C, and 45°C), while growth at 45°C was relatively weak. Catalase and oxidase activities were detected. No fermentation/oxidization of glucose was observed and no acid/gas was produced from glucose. Growth on the MacConkey media tested positive. Nitrate reduction activity was detected. Glucose, potassium gluconate, capric acid, adipic acid, malic acid, trisodium citrate, and phenylacetic acid are assimilated. No indole production from L-tryptophane, fermentation of glucose, arginine dehydrolase, urease or β-galactosidase activities, esculin or gelatin hydrolysis, or assimilation of arabinose, mannose, mannitol, N-acetyl-glucosamine, and maltose were observed. Activities of esterase, leucine arylamidase, and acid phosphatase were detected. Weak activities of alkaline phosphatase and Naphthol-AS-BI-phosphohydrolase were also detected. No activity of esterase lipase, lipase, valine arylamidase, cystine arylamidase, trypsin, α-chymotrypsin, α-galactosidase, β-galactosidase, β-glucuronidase, α-glucosidase, β-glucosidase, N-acetyl-β-glucosaminidase, α-mannosidase or α-fucosidase was detected. Overall, the physiological characteristics of the strain H1_3_1 were highly similar to those of A. xylosoxidans LMG 1863, the type strain which is not HOB (Supplementary Table S2). Moreover, the cellular fatty acids composition of stain H1_3_1 showed overall similarity with that of the type strain (Vandamme et al., 2013) (Supplementary Table S1). Furthermore, the respiratory quinone composition was 99% ubiquinone 8 (Q8), 0.5% Q7, and 0.5 Q9, which are also similar to the type strain (Yabuuchi et al., 1998).

Interestingly, in the genome of A. xylosoxidans strain H1_3_1, above mentioned genes for hydrogenases and their maturation proteins (hox and hyp genes) and enzymes of the Calvin–Benson–Bassham cycle (cbb genes) are located just proximal to each other (Supplementary Figure S3). Moreover, the hox-hyp-cbb cluster is flanked by two identical insertion sequences (IS256) with an ORF encoding putative integrase next to one of them, suggesting that the hox-hyp-cbb cluster is encoded on a transposable element. Thus, these results suggest that the strain H1_3_1 is a noble HOB strain of A. xylosoxidans, which likely acquired the hox-hyp-cbb cluster via lateral gene transfer.

To examine the production of the compatible solutes in the HBI system, A. xylosoxidans strain H1_3_1 and M. mageritense stain H4_3_1 were, respectively, cultivated in the reactors containing high-ionic-strength medium or low-ionic-strength medium for comparison. EIS confirmed that internal resistance of the reactor with high-ionic-strength medium was smaller than that of the reactor with low-ionic-strength medium (Supplementary Table S3). No bacterial growth was observed in abiotic controls or open-circuit biotic controls (data not shown). The two HOB strains were able to grow under both conditions, as shown in Figures 3A–D. Following cultivations in the reactors, the microbial cells were harvested, and ectoine/hydroxyectoine were extracted. Notably, accumulations of hydroxyectoine were detected in strains H1_3_1 and H4_3_1 cultivated in the reactor under the high-ionic-strength condition (Figure 3E), while accumulations of ectoine were below the detection limit (data not shown). The yields of hydroxyectoine were 1.32 μmol/L_{liquid medium} for A. xylosoxidans strain H1_3_1 and 0.06 μmol/L_{liquid medium} for M. mageritense strain H4_3_1. On the other hand, no ectoine/hydroxyectoine was detected in the cells cultivated in the reactors with the low-ionic-strength medium, suggesting that hydroxyectoine synthesis is induced under a high-ionic-strength condition and supports the notion that the compatible solute plays a crucial role in the halotolerance of HOB isolates.

Four samples were collected from saline environments and used as the initial microbial sources (Table 1). The samples were stored in 50 mL sterile tubes at 4°C until inoculation.

Synthesis of CoPi anode and Co-P cathode was operated by electrochemically depositing the catalysts on a 2 × 4 cm SUH316 stainless-steel mesh using previously described methods (Paseka and Velicka, 1997; Kanan and Nocera, 2008; Liu et al., 2016; Feng et al., 2023). The deposition process was performed using an electrochemical measurement system (HZ-7000, Hokuto Denko, Japan) referred to Ag/AgCl electrode. Following deposition, the electrodes were rinsed with deionized water.

Borosilicate glass cells (220 mL) equipped with the CoPi anode and Co-P cathode fabricated in Section 5.2 was used as the HBI system reactors. The high-ionic-strength medium, a modification of the minimal medium (Feng et al., 2023) that contains 180 mM phosphate buffer (33.7 g/L Na₂HPO₄·7H₂O and 7.5 g/L KH₂PO₄), was used as the culture medium. For experiments with a low-ionic-strength medium, the buffer strength of phosphate was decreased to 36 mM with 6.74 g/L Na₂HPO₄·7H₂O and 1.5 g/L KH₂PO₄. The reactors and medium were sterilized separately by autoclaving. Before inoculation, 100 mL of the sterilized medium was added to the reactor in an aseptic manner.

For the initial enrichment cycle, approximately 10 mL of an environmental sample was filtered through a 0.45 μm nitrocellulose filter (Nalgene Nunc International, United States) in an analytical filter unit under vacuum conditions. The residue trapped on the filter was rinsed three times with a sterilized medium and inoculated into the reactor. After sealing the reactor with a butyl rubber plug and an aluminum closure, the ambient gas in the headspace was replaced with an 80:20 mixture of N₂ and CO₂. The reactor was kept at 30°C, continuously agitated with a magnetic stirrer, and subjected to a constant 2.0 V voltage via a power supply (3645A, Array Electronics, China) during the incubation. For the second and third enrichment cycles, a few microliters of the liquid medium from the previous enrichment cycle were collected using a sterilized inoculating loop and inoculated into a freshly prepared reactor. Thus, the experimental conditions of reactor series H1, H2, H3, and H4 were identical except the initial inocula (i.e., four different environmental samples: Table 1).

The voltage of a 1.0 Ω resistance across the electrodes was monitored with a multimeter (34970A, Agilent Technologies, United States). The electrical current flowing through the circuit was calculated based on Ohm’s law. The headspace pressure was monitored using a pressure sensor (APC40, Keyence, Japan). Gas chromatography with a thermal conductivity detector (GC-2014, Shimadzu, Japan) and argon as the carrier gas was used to analyze the gas composition. To monitor microbial growth, the optical density (OD⁶⁰⁰_1cm) of the culture medium was measured using a spectrophotometer (Ultrospec 6,300 pro, GE Healthcare, United States). Electrochemical impedance spectroscopy (EIS) was employed to determine the internal resistance of the HBI reactor as described in Kim et al. (2021). The measurement was performed using an electrochemical measurement system (HZ-7000). The solution resistance (R_ohm) was determined using a two-electrodes setup, where Co-P alloy served as the working electrode and CoPi as the counter electrode. To measure the charge transfer resistance of each electrode (R_{act_anode} and R_{act_cathode}), a three-electrode setting was used with Ag/AgCl as a reference, then the values were determined from the semi-circular from the Nyquist plot. The total internal resistance R_int was then calculated as the summation of the charge transfer resistances and the solution resistance.

The theoretically produced hydrogen n_H2__theoretical (mol) was determined according to Equation 1:

I(t) is the current (A) at sampling time interval t (sec), T is the experiment duration (sec), and F is Faraday’s constant (96,485 C/mol).

Microbial cells were extracted from approximately 50 mL of the reactor medium using vacuum filtration through a 0.45 μm nitrocellulose filter in an analytical filter unit (Nalgene Nunc International). DNA extraction was operated following the protocol described in Feng et al. (2023). 16S rRNA genes were amplified by PCR using the extracted DNA as the template and primers U789F (5′-TAGATACCCBGGTAGTCC-3′) and U1068R (5′-CTGACGRCRRCCATGC-3′) (Baker et al., 2003) as described previously (Feng et al., 2023). The resulting PCR products were subjected to library preparation using Nextera XT DNA Library Prep Kits (Illumina, United States) according to the manufacturer’s instructions, and the resulting amplicons were sequenced on a MiSeq system (Illumina) at Biken Biomics (Japan). The resulting raw reads were processed on the QIIME2 platform (version 2021.4.0) using DADA2 to cluster them into amplicon sequence variants (ASVs) (Callahan et al., 2016; Bolyen et al., 2019). The alpha diversity (observed features) was calculated, and the taxonomy was assigned using the Blast+ algorithm against the Sliva (138 SSURef NR99) 16S rRNA database (Camacho et al., 2009; Robeson et al., 2021).

The medium from the reactors at the end of the third enrichment cycle was collected and serially diluted to 1:100,000 with the sterilized medium. Subsequently, 100 μL of each diluted sample was spread onto plates containing the high-ionic-strength medium solidified with 1% (w/v) gellan gum. The plates were cultured under an 80:10:10 H₂/CO₂/O₂ atmosphere at 30°C. After around one week of incubation, colonies were isolated by picking, spreading, and cultivating them on new plates. This procedure was repeated thrice to guarantee the purity of the isolates.

The nearly full-length 16S rRNA gene sequences were obtained following the procedures described in Feng et al. (2023), using the primer pairs 8F (5′-AGAGTTTGATYMTGGCTCAG-3′) and 1492R (5′-CGGYTACCTTGTTACGACTT-3′) (Grabowski et al., 2005). The amplified fragment sequences were determined using standard Sanger sequencing at Macrogen Japan (Japan) and compared against the EzBioCloud database (Yoon et al., 2017).

The HOB isolate was propagated in 20 mL of the high-ionic-strength medium under an H₂/CO₂/O₂ atmosphere (80:10:10) at 30°C for two days and gathered via centrifugation. Genomic DNA was extracted from the cell pellet using Genomic-tip 20/G (Qiagen) with the Genomic DNA Buffer Set (Qiagen) for Achromobacter species or the MasterPure Gram Positive DNA Purification Kit (Epicentre, USA) for Mycolicibacterium species. The DNA template library was constructed using the SMRTbell gDNA Sample Amplification Kit (PacBio, USA) and SMRTbell Express Template Prep Kit ver.2.0 (PacBio) and bound to Sequel II polymerase 2.2 (PacBio) using the Sequel II Binding Kit 2.2 (PacBio). Sequencing was performed using the Sequel IIe platform (PacBio) at Bioengineering Lab (Japan). After trim of the overhand adapter and removal of low-quality and short reads, the resulting high-quality reads were assembled using Flye (ver.2.9.1-b1780) (Kolmogorov et al., 2019). Annotation was performed using Prokka (ver. 1.14.5) to identify relevant features and generate protein sequences (Seemann, 2014), which were further analyzed using eggNOG-mapper (ver. 2.1.9) against the Kyoto Encyclopedia of Genes and Genomes (KEGG) database (Huerta-Cepas et al., 2019; Cantalapiedra et al., 2021). Digital DNA–DNA hybridization values were estimated using the Genome-to-Genome Distance Calculator (ver. 3.0) (Meier-Kolthoff et al., 2022). Hydrogenase genes were classified using HydDB (Greening et al., 2016; Sondergaard et al., 2016).

Achromobacter xylosoxidans strain H1_3_1 was grown on nutrient agar (Oxoid) at 30°C under aerobic conditions for 2 days and observed using a light microscope (Olympus BX50F4). Conventional phenotypic tests were performed as described in Barrow and Feltham (1993). The activity of constitutive enzymes and other physiological properties were determined using the API 20NE and API ZYM microtest systems (bioMérieux) according to the manufacturer’s instructions. For cell fatty acid analysis, strain H1_3_1 was cultivated on TSBA (Becton Dickinson) at 30°C under aerobic conditions for 24 h. The cells were freeze-dried, and fatty acid methyl esters were extracted and analyzed using the Sherlock Microbial Identification System (ver.6.0) according to the manufacturer’s procedures. Analysis of respiratory quinones was carried out by Deutsche Sammlung von Mikroorganismen und Zellkulturen GmbH (DSMZ, Germany). Respiratory quinones are extracted from freeze-dried cell material using hexane, purified by a silica-based solid phase extraction, and analyzed by HPLC recording absorption spectra (Vieira et al., 2021). For relative quantification, 270 nm for ubiquinones and 326 nm for menaquinones are used.

The isolate was precultured in 20 mL sterilized medium in a glass vial under an H₂/CO₂/O₂ atmosphere (80:10:10) at 30°C for 2 days. Subsequently, 10 mL of the preculture was transferred into a freshly prepared HBI system reactor and incubated as described above. The bacterial cells were harvested via centrifugation. To extract hydroxyectoine, 200 μL of bidistilled water was added for every 10 mg of biomass. The hydroxyectoine concentration was determined following a method described in Laryea et al. (1998). The supernatant (extract) was recovered via centrifugation at 10,000 g for 5 min, and 50 μL of the extract was mixed with 50 μL of 100 mM KH₂PO₄, which was further mixed with 900 μL of the labeling solution (2.5 mM 18-crown-6 and 50 mM 4-bromophenacyl bromide in acetonitrile). The labeling reaction was performed for one hour at 80°C. The mixture was vortexed and centrifuged at 1000 g. 200 μL of the supernatant containing the phenacyl esters of hydroxyectoine was directly injected into HPLC (LC-20 AD, Shimadzu) equipped with a Supelcosil LC-SCX column (Sigma-Aldrich, United States). The mobile phase was 22 mmol/L choline in 900 mL/L acetonitrile and 100 mL/L water at a 1.5 mL/min flow rate. The sample was eluted isocratically over twenty minutes and monitored at 254 nm and 28°C. To prepare the standard curve, hydroxyectoine (Sigma-Aldrich) was used at concentrations of 500, 100, 20, and 4 μM.