Twenty-three original samples of coalbed water were collected from the Erlian Basin (Supplementary Figure S1) in May, August, and October 2018, corresponding to the local spring, summer, and autumn. Meteorological conditions such as precipitation and temperature vary greatly between seasons in this region. In May, August, and October 2018, the total monthly precipitation was 6.6, 30.3, and 0.7 mm, and the mean temperature was 18.2, 23.1, and 4.4°C, respectively (Supplementary Table S2). The formation water samples were gathered from a depth of 400 meters. Water samples dedicated for chemical composition analysis were kept on ice. The basic physical and chemical properties of coalbed water were determined by the unconventional experimental center of China National Offshore Oil Corporation Energy Development Co., LTD. The pH of the samples ranged from 7.53 to 8.79, which was slightly alkaline. Nitrite ions were detected in only two samples (JM10O and JM13O). Nitrate and sulfate concentrations were also detected at low levels. Sodium concentration was detected at a high level based on the Chinese classification of mine water (Supplementary Table S1). Water samples dedicated for microbial analysis were stored and transported in ambient temperature. Water samples were transported by road, and the average transportation time was 3 days.

For original coalbed water samples, 500 ml water sample was required per sample for high speed centrifugation (5 min at 14,000 g at 4°C). For the sample of the culture experiment, 2 ml of culture at the end of the experiment was required for high speed centrifugation (5 min at 14,000 g at 4°C). After the supernatant was discarded, the precipitate was resuspended with 20 ml phosphate buffer, and the suspension was transferred to a new, enzyme-free spiral tube for further centrifugation. The precipitate obtained after final centrifugation was stored at −80°C until DNA extraction. DNA was extracted using a Water DNA Isolation Kit (Foregene).

High-throughput sequencing was performed on the extracted DNA samples by Beijing Nuohe Zhiyuan Co. Ltd. DNA concentration and purity were determined by ultramicro-spectrophotometer, and the quality of all DNA samples met the sequencing requirements. The V3-V4 region of the bacterial 16S rRNA genes was amplified using the bacteria-specific primers 341F_CCTAYGGGRBGCASCAG/806R_GGACTACNNGGGTATCTAAT. The V4-V5 region of archaea 16S rRNA genes was amplified using the archaea-specific primers Arch519F_CAGCCGCCGCGGTAA/Arch915R_GTGCTCCCCCGCCAATTCCT. Illumina HiSeq 2,500 was used for pair-end sequencing after the amplification products were recovered by gel cutting.

QIIME software was used to conduct quality control, chimeric removal, re-sampling, OTU (Operational Taxonomic Unit) clustering, selection of representative sequences, species annotation, and singleton removal for sequencing data from the sequencing unloading machine. Because of limited primer specificity, a small number of archaea sequences were found in the amplified fragments of bacterial primers, and bacterial sequences were also amplified by archaea primers. After the mismatched OTUs were deleted by Excel, the structure of the microbial community was analyzed. For original coalbed water samples, a total of 673,216 high-quality sequences were obtained using high-throughput sequencing, among which 502,724 were bacterial sequences and 170,492 were archaeal sequences (Supplementary Table S3). A total of 1,009 archaeal OTUs were obtained based on a 97% sequence similarity and the average number of OTUs of the samples was 44. The number of OTUs in samples JM3A (104) and JM4-1A (102) was higher than that of other samples. A total of 20,082 bacterial OTUs were obtained, and the average number of OTUs per sample was 873. The Shannon index indicated high bacterial diversity (3.57 ± 0.51; mean ± SD). Coverage for bacteria and archaea ranged from 0.96 to 0.98, indicating that sequencing depth was sufficient.

The genomic DNA was randomly sheared into short fragments. The obtained fragments were end repaired, A-tailed and further ligated with Illumina adapter. The fragments with adapters were PCR amplified, size selected, and purified. The library was checked with Qubit and real-time PCR for quantification and bioanalyzer for size distribution detection. Quantified libraries will be pooled and sequenced on Illumina platforms, according to effective library concentration and data amount required. The raw reads were dereplicated and quality-trimmed using Trimmomatic (Bolger et al., 2014). High-quality metagenomic sequences were de novo assembled using SPAdes (v.3.12.0) with the default k-mer size (Nurk et al., 2017). The assembled scaffolds were binned using MetaBAT (v.2.12.1) based on tetranucleotide frequency, GC content, and coverage (Kang et al., 2019). Partial and near-complete genomes were recovered after binning. The completeness, contamination, and strain heterogeneity of MAGs were evaluated using CheckM (v.1.0.11) based on lineage-specific marker sets (Parks et al., 2015). The taxonomic assignment of the different bins was conducted with the GTDB-Tk package (v.1.3.0; Chaumeil et al., 2019), and bins were translated by Prodigal using the “-p meta” parameters (Hyatt et al., 2010). For each predicted coding sequence (CDS), protein function was annotated using the KEGG server (BlastKOALA) and eggNOG-mapper (Kanehisa et al., 2016).

Water sample JM13M was selected for methanogenic simulation culture. After shaking well, the 180 ml water sample was added to the 300 ml sterilized and dried serum bottle. The pH was adjusted to neutral. Na₂S·9H₂O (1.0 mM), resazurin (0.0005 g L⁻¹), trace elements (2.0 ml/L), and vitamins (V284/VB1/VB12; 4.0 ml/L) were supplemented. Vitamin and trace element solutions were prepared as described previously (Lu and Lu, 2012). The headspace gas was flushed with N₂ for 30 min to remove oxygen. Fourteen methanogenic precursors were selected for methanogenic simulation culture:H₂/CO₂ (50 ml/25 ml), sodium formate (20 mM), dimethyl sulfide (20 mM), dimethylamine (20 mM), methanol (20 mM), sodium acetate (20 mM), sodium propionate (20 mM), sodium butyrate (20 mM), ethanol (20 mM), 1-propanol (20 mM), 2-propanol (20 mM), 1-butanol (20 mM), 2-butanol (20 mM), and Choline hydroxide (20 mM). No precursor was added to the control. Three parallel were set for each treatment and incubated at 35°C in the dark.

The CH₄ content in the headspace of all culture bottles was measured using gas chromatograph 7820A (Agilent Technologies, United States) with Porapak Q capillary columns (3 m) at 65°C. High-purity hydrogen (99.999%) was used as the mobile phase. The column flow rate was 27 ml/min. TCD (thermal detector) at 130°C, the inlet temperature was 105°C, the reference flow was 30 ml/min, and the tail blow flow was 2 ml/min. The gas detection time was 1.5 min, and the injection volume was 0.2 ml. The relative proportions of N₂, CH₄, and CO₂ in the sample gas were calculated using the corrected area normalization method for the peak diagram. All samples were tested at 5-day intervals.

The datasets generated for this study have been deposited under NCBI. For original coalbed water samples, 16S rRNA genes of archaea: PRJNA681683. 16S rRNA genes of bacteria: PRJNA681635. For the sample of the culture experiment, 16S rRNA genes of archaea and bacteria: PRJNA923302. Metagenomic raw data: PRJNA682244 (Supplementary Table S9).

A total of 24 bacterial phyla, including the candidate divisions, were detected from coalbed water samples (Supplementary Figure S2). Proteobacteria was the most abundant phylum, and the average relative abundance of Proteobacteria in the 23 samples was 77.3% (range 25.6–93.0%). Other bacteria were mainly affiliated with the phyla Bacteroidetes (10.1%) and Firmicutes (6.7%; Supplementary Figure S2). Bacteria with a relative abundance of more than 5% (genus level) in each sample were selected, as shown in Figure 1A. The dominant Gammaproteobacteria genera were Methylomonas (15.1%), Methylobacter (2.8%), Thioalkalimicrobium (2.1%), Arenimonas (2.0%) Methylomicrobium (0.6%), Aeromonas (0.4%), Rheinheimera (0.3%), and Shewanella (0.3%). Epsilonproteobacteria included Arcobacter (5.2%), Sulfuricurvum (2.9%), Sulfurimonas (1.5%), and Sulfurospirillum (1.3%). Betaproteobacteria included Methylotenera (3.8%), Hydrogenophaga (1.8%), Acidovorax (1.1%), Malikia (1.1%), and Thiobacillus (0.6%). Alphaproteobacteria included Porphyrobacter (2.5%), Rhizobium (1.3%), Roseovarius (0.7%), Hyphomonas (0.6%), and Hoeflea (0.4%). Bacteroidetes included Flavobacterium (1.8%), Algoriphagus (1.3%), and Barnesiella (0.5%). In Firmicutes, only Acetobacterium (1.1%) showed slightly higher relative abundance.

OTUs with a relative abundance of >5% included primarily methanotrophs, methylotrophs, acetogenic bacteria, chemoautotrophic bacteria, and fermentative bacteria. The aerobic methanotrophs were Methylomonas and Methylobacter, and Methylomonas dominated the coalbed water with the highest relative abundance (up to 40%; Figure 1A). Methylotrophic bacteria included Methylotenera, whose substrate spectrum is narrow and could only metabolize methylamine, methanol, and other methyl compounds (Kalyuzhnaya et al., 2012; Mustakhimov et al., 2013). Acetobacterium can generate acetic acid using H₂/CO₂, short-chained alcohols, and various carbohydrates (sugars; Kremp et al., 2018; Westphal et al., 2018). Chemoautotrophic bacteria included Thiobacillus, Sulfurospirillum, Sulfurimonas, and Sulfuricurvum. Thiobacillus (Kellermann and Griebler, 2009), Sulfurospirillum (Stolz et al., 1999; Luijten et al., 2003; Hubert and Voordouw, 2007), and Sulfurimonas (Inagaki et al., 2003; Sievert et al., 2008) were the non-strict chemoautotrophic bacteria, which utilize reductive sulfide, hydrogen, and organic matter (glucose, volatile fatty acid, amino acid, etc.) as electron donors, nitrate or sulfate as electron acceptors, and carbon dioxide as a carbon source. Sulfuricurvum, however, is a strictly chemoautotrophic bacteria, and can only use hydrogen and reductive sulfides as an electron acceptor.

PCA results showed that the bacterial community structure of coalbed water samples in the Erlian Basin was clustered according to sampling months. There was no overlap or crossover between the bacterial communities in the coalbed water during the three sampling periods (Figure 1B). This indicated that the community structure of bacteria in coalbed water in this block differed among seasons. A large number of aerobic bacteria (Methylomonas, Methylobacter, Methylotenera, Sulfurospirillum, Acidovorax, and Thioalkalimicrobium) were detected in the coalbed water, which may be related to atmospheric precipitation. Atmospheric precipitation leads to surface runoff into coal seams and brings in oxygen, nutrients, and electron receptors. However, the periodicity of rainfall leads to an unsustainable supply of oxygen, while the metabolic activity of aerobic microorganisms causes periodic changes in oxygen concentration, resulting in co-existence of aerobic and anaerobic bacteria in coalbed water (Bates et al., 2011). The meteorological data from the National Meteorological Science Data Center¹ show that there were significant differences in rainfall and mean temperature among the sampling sites during different seasons (Supplementary Table S2). However, due to the lack of reliable measured data, we are still unable to attribute bacterial community changes.

At the phylum level, archaea were dominated by Euryarchaeota (relative abundance 74.6–92.3%). Only some of the archaea in samples JM2 and JM9 were distributed in Nanoarchaeota (relative abundance 8.1–8.6%) and Thaumarchaeota (relative abundance 5.2–6.9%). At the genus level, the dominant archaea were hydrogenotrophic methanogens Methanobacterium, and the average relative abundance of the 23 samples was 61.6% (range 6.6–98.2%; Figure 2A). In addition, the dominant archaea in sample JM4A were hydrogenotrophic methanogens Methanocalculus (relative abundance 62.9%), and the dominant archaea in JM11O, JM1_2M, and JM14M were hydrogenotrophic methanogens Methanocorpusculum, with relative abundances of 87.7, 77.0, and 87.9%, respectively. The methylotrophic Methanolobus (Doerfert et al., 2009; relative abundance 36.9%) in JM9O was also relatively high.

PCA showed that the archaea communities of the 23 coalbed water samples were partially overlapping and were dispersed in different areas in the PCA diagram, with no obvious clustering (Figure 2B). This indicated that the community structure of archaea in coalbed water in this block was relatively stable. The archaea community structure is affected by many environmental factors. All methanogens are strict anaerobes, not documented to use O₂ as an electron acceptor during energy-conserving metabolism (Horne and Lessner, 2013). Nutrient availability is one of the key factors determining the abundance of methanogens (Zhou et al., 2015). Salinity (Waldron et al., 2007; Zhang et al., 2020) and temperature (Lu et al., 2015) are also primary environmental factors regulating methanogenic community assemblage. The different responses of archaea and bacteria to seasonal variation in coalbed water reflect their distinct ecological niches.

In order to verify the methane production potential of microorganisms in coalbed water and assess its methanogenic pathway, coalbed water sample JM13M was selected to study methanogenic potential using different precursor compounds. H₂/CO₂ and formate (Figure 3A), methyl compounds (methanol, dimethylamine, dimethyl sulfide, choline; Figure 3C), and short-chain fatty alcohols (propanol, isopropyl alcohol, butanol, and isobutanol; Figure 3B) may be used by microorganisms in JM13M to produce CH₄, while short-chain fatty acids cannot be used (Figure 3D). Both the lag phase and the specific growth rate of methane production were different under different substrate treatment (Supplementary Table S4). The dominant archaea were the hydrogenotrophic methanogens Methanobacterium (relative abundance 2.9–91.7%) and Methanocorpusculum (relative abundance 4.9–91.6%), and the methylotrophic Methanolobus (relative abundance 1.2–32.9%; Figure 3E). The methanogenic activity of coalbed water was confirmed by the simulation culture (Figure 3). The bacterial community mainly consisted of Desulfomicrobium and Desulfovibrio (Figure 3F), which might be related to the addition of sodium sulfide.

We also collected published and available coalbed archaea data from different basins, including Qinshui Basin (Guo et al., 2017; Yang et al., 2018), Hailar Basin, Jingmen-Dangyang Basin (Wei et al., 2014), Ordos Basin (Guo et al., 2012a), and Powder river Basin (Huang et al., 2017; Davis et al., 2018), and combined them with the Erlian Basin data for analysis (Figure 4). We found that most of these basins were dominated by hydrogenotrophic methanogens Methanobacterium (except the Powder River basin and JM-DY Basin, which were dominated by acetotrophic methanogens). Methanobacterium is widely distributed in anaerobic habitats such as marine and freshwater sediments, soils, animal gastrointestinal tracts, anaerobic sewage digesters, and geothermal habitats (Liu, 2010). Methanobacterium are typically autotrophic and use H₂ and CO₂, while some also use formate as a substrate for methanogenesis (Liu, 2010). Formate dehydrogenases that supply reductant from formate oxidation and that yield intracellular CO₂ to allow for methanogenesis to proceed under otherwise dissolved inorganic carbon (DIC) limited conditions (Fones et al., 2021). Recent studies have shown that Methanobacterium are also able to engage in direct interspecific electron transfer (Zheng et al., 2020), which facilitates the formation of a more efficient connection between Methanobacterium and syntrophic bacteria. These properties allow Methanobacterium to thrive in a variety of methane-producing environments, including coalbed water. Methanocorpusculum was also dominant in some coalbed water samples, and was the most prominent genus in a coal bed of the Illinois Basin (Strapoc et al., 2008) and in shale in northern Michigan (Waldron et al., 2007). Methanocorpusculum belong to the order Methanomicrobiales within the archaeal phylum Euryarchaeota. Similar to Methanobacterium, Methanocorpusculum use either H₂/CO₂ or formate as substrate for methanogenesis. Compared to other strains of Methanomicrobiales, Methanocorpusculum labreanum was the only strain showing evidence of genome downsizing (Browne et al., 2017), which leads to niche specialization (Lee and Marx, 2012). However, M. labreanum’s putative specialization remains unclear.

The previous high-throughput sequencing results revealed high relative abundance of aerobic methane-oxidizing bacteria in coalbed water. This implied that the methane oxidation reaction was very active. However, the mechanisms by which aerobic methane-oxidizing bacteria act in coalbed water warranted further study. For this, three coalbed water samples (JM1-2M, JM5M, and JM13M) were selected for metagenomic sequencing (Supplementary Table S5). After sequencing and quality control of the sequences, taxonomic classification was performed using two different tools, Kraken2 and MetaPhlAn2. Overall, bacteria (relative abundance >1%) accounted for 60.0–76.9% of the genera identified by Kraken2 and 96.7–97.4% of the genera identified by MetaPhlAn2 (Supplementary Figure S3). However, the relative abundance of archaea in metagenomic data was too low (<0.02%), as shown in the figure.

Kraken2 and MetaPhlAn2 produced similar mock community composition profiles. According to Kraken2, Methylomonas (19.0–51.7%) occurred in a higher proportion in bacteria (Supplementary Figure S3B), and the same result was found using MetaPhlAn2 (51.8–87.1%; Supplementary Figure S3A). Such a high abundance of Methylomonas was also consistent with our high-throughput sequencing results. In addition, Methylotenera, Pseudomonas, and Thiomonas occupied a certain proportion in some samples. Among them, Pseudomonas was particularly noteworthy due to its unique value to research and application to PAH and coal biodegradation (Foght and Westlake, 1988; Tang et al., 2011; Poblete-Castro et al., 2012).

Metagenomic data were used to assemble methane-oxidizing bacteria genomes. Seven high-quality genome bins (MAGs) were obtained (Supplementary Table S6), and after dereplication with dRep with an ANI cutoff of 97%, five cluster MAGs were obtained for subsequent analysis (Supplementary Table S7). Phylogenetic trees constructed with the genomes of methane-oxidizing bacteria (Supplementary Table S8) showed three clusters affiliated with Methylomonas (Supplementary Figure S4).

These three cluster MAGs contained complete metabolic pathways for aerobic methane oxidation. In the first step of methane metabolism, methane is captured and converted to methanol by granular methane monooxygenase (pMMO), an enzyme complex that uses oxygen to oxidize the C-H bonds in methane. In the second step of methane oxidation, methanol is catalyzed by methanol dehydrogenase (MDH) to form formaldehyde. There are two known MDH enzymes, the canonical and long-studied MxaF type and a novel type, XoxF (Chistoserdova et al., 2009). These three cluster MAGs encode both types of methanol dehydrogenases. In the next step, formaldehyde is a key metabolite. Formaldehyde can be converted to formate via the tetrahydromethanopterin (H₄MPT) pathway or the methylene-tetrahydrofolate (methylene-H₄F) pathway. Another part of formaldehyde can be metabolized through the RuMP cycle and the serine cycle. All genes coding for these pathways were detected in our MAGs (Figure 5). Succinate dehydrogenase (complex II, sdhABCD) was found in the MAGs, together with genes encoding cytochrome c oxidase. The proton motive force generated by the respiratory chain can be used by ATPase (complex V). Nitrogenase (NifDHK), dissimilatory nitrite reductase (NiRk), and nitric oxide reductase (NorB) were also found in these MAGs. This means that these genomes may have the ability to fix nitrogen and convert nitrite into ammonium through denitrification (Figure 5). Denitrification may support methane oxidation in environments with low or below-detectable oxygen concentrations (Padilla et al., 2017; Smith et al., 2018; Smith and Wrighton, 2019).