During the optimal season for nitrogen cycling processes, sediment sampling was conducted in July 2022 at three water reservoirs, namely Sanchaxi (SCX), Zhangnenzi (ZNZ), and Wenyang (WY), located in the Wuyanling National Nature Reserve in the Zhejiang Province, China. These reservoirs are crucial for hydroelectricity production, with SCX being the largest among them. A total of five sampling sites (SCX1–SCX5) were designated for SCX, while three sampling sites were designed for both ZNZ (ZNZ1–ZNZ3) and WY (WY1–WY3). The sampling points within each reservoir were sequentially numbered from upstream to downstream. Detailed information regarding the sampling sites can be found in Supplementary Figure S1. At each sampling site, surface sediments were collected from the 0–30 cm depth using a Petersen stainless steel grab sampler. To ensure sample representativeness, a composite surface sediment sample was created by combining three sediment samples collected in close proximity to the sampling site (Zhang et al., 2015; Gao et al., 2021). Additionally, corresponding water samples were collected using a vertical water sampler. The sediment samples were sealed in sample bags, and the water samples were stored in polyethylene bottles. All samples were then placed in ice boxes for transportation to the laboratory. Subsequently, the water samples underwent immediate analysis of their physicochemical characteristics, while the sediment samples were stored separately in sampling bags (at 4°C) and polypropylene tubes (at −80°C) for further analysis of physicochemical characteristics and microbial communities, respectively.

The sediment pH was measured using a PHSJ-4F pH meter (Leici, China) with a sediment suspension ratio of 1:2.5 (sediment to deionized water). Prior to measurement, the deionized water was purged with nitrogen gas for 15 min. The moisture content (MC) of sediments was assessed by oven-drying fresh sediment at 105°C until a constant weight was achieved (Wang et al., 2021). To extract exchangeable dissolved inorganic nitrogen, including ammonium (NH₄⁺), nitrite (NO₂⁻), and nitrate (NO₃⁻), from fresh sediments, a 2 M KCl solution pre-purged with high-purity nitrogen gas was utilized at a ratio of 1:5 (sediment to KCl solution, w/v). The resulting solution was filtered through a 0.45 μm membrane filter and then measured using a TU-1810 UV/VIS spectrophotometer, following the standard method (Li et al., 2019). The total organic matter (TOM) content was estimated by measuring the weight loss after combustion at 550°C using the LOI550 (loss on ignition at 550°C) method (Wang et al., 2021). Additionally, water quality parameters including pH, NH₄⁺, NO₂⁻, NO₃⁻, total phosphorus, total organic carbon, and total nitrogen were determined using standard methods (Rice et al., 2012). All the physicochemical characteristics were measured in triplicate. The summaries of the physicochemical characteristics of both the sediment and water samples are presented in Supplementary Tables S1, S2, respectively.

The total metagenomic DNA of the sediment samples was extracted using FastDNA Spin Kit for Soil (MP Biomedicals, United States). The concentration of the extracted DNA was determined using the TBS-380, while the purity was assessed using the NanoDrop2000. Equal masses of DNA from each of the three replicate samples were consolidated to reduce potential variability. The DNA extract was then fragmented to an average size of approximately 400 bp using the Covaris M220 to facilitate the construction of paired-end libraries. For library construction, the NEXTflex rapid DNA-seq kit (Bioo Scientific, United States) was utilized, with adapters that included the full set of sequencing primer hybridization sites ligated to the blunt ends of the DNA fragments. Paired-end sequencing was conducted on the Illumina Novaseq 6,000 (Illumina Inc., San Diego, CA, United States) at Majorbio Bio-Pharm Technology Co., Ltd. (Shanghai, China). The raw sequence data have been deposited in the NCBI Sequence Read Archive under the project accession number PRJNA1061319.

To ensure the reliability of the metagenomic data, adaptors and low-quality reads were trimmed. The remaining high-quality reads were then assembled using MEGAHIT, with contigs ≥300 bp chosen as the ultimate assembly. Gene annotation and prediction were performed on the contigs. Gene abundance quantification was conducted through the reads per kilobase per million mapped reads (RPKM) method. For each contig, open reading frames (ORFs) were predicted using MetaGene (Noguchi et al., 2006), and those with a length of ≥100 bp were extracted for translation into amino acid sequences using the NCBI translation table. Subsequently, a non-redundant gene catalog was built using CD-HIT, applying a 95% sequence identity and 90% coverage threshold. The high-quality reads were then aligned to this non-redundant gene catalog using SOAPaligner for determining gene abundance with 95% identity. For taxonomic annotations, representative sequences from the non-redundant gene catalog were aligned to the NR database using Diamond (Buchfink et al., 2015). The alpha diversity of the microbial community was assessed by calculating the number of reads at the genus level. The Conserved Domain Database (CDD) was utilized to annotate the structural domains of non-redundant genes (Jia et al., 2022; Shi et al., 2024b). The functions of non-redundant genes were annotated against the Kyoto Encyclopedia of Genes and Genomes (KEGG) database using Diamond (Buchfink et al., 2015).

The presence of rpf-like genes in the water reservoir sediments was investigated by analyzing the detected Rpf-like domains (Supplementary Table S3), including LT_TF-like, chitinase_GH19, lyz_endolysin_autolysin, chitosanase_GH46, pesticin_lyz-like, N-acetylmuramidase_GH108, lambda_lys-like, LT-like, Slt35-like, LT_IagB-like, Slt70-like, MLTF-like, LT_VirB1-like, LT_MltC_MltE, LT_Slt70-like, MltD-like, and Lyz-like.¹ Moreover, an investigation was conducted on the distribution of nitrogen-cycling genes (Supplementary Table S4) encompassing nitrogen fixation (nifH, nifD, nifK), nitrification (amoA/pmoA, amoB/pmoB, amoC/pmoC, hao, nxrB), denitrification (narG, nirS, nirK, nosZ, norB), DNRA (nrfA, nirB, nirD, nrfH), ANRA (nasA, nirA, narB), anammox (hzsA_1/hzsA_2, hdh/hzo, hzsB, hzsC), and ammonification (ureC, gdhA) in the water reservoir sediments (Kuypers et al., 2018; Zhang et al., 2022; Wang et al., 2022a). Additionally, the genes involved in VBNC formation (Supplementary Table S5) were also examined in the water reservoir sediments to explore their potential co-occurrence with rpf-like genes and nitrogen-cycling functional genes.

The physicochemical properties data obtained from the experiment in triplicate were presented as mean values and standard deviations. Principal coordinates analysis (PCoA) with Bray–Curtis dissimilarity was utilized to assess the dissimilarity of sediment microbial communities from three water reservoirs (Zhou et al., 2023). The significance of these differences was evaluated using the analysis of similarities (ANOSIM) test. To compare the microbial community variations among the three water reservoir sediments, the Kruskal–Wallis H test was performed, followed by the Tukey–Kramer test at a 95% confidence level. Redundancy analysis (RDA) was employed to examine the association between environmental variables and functional genes, as well as their associated microbial communities. The environmental variables considered in the RDA included pH, NH₄⁺, NO₂⁻, NO₃⁻, and TOM, which were selected to avoid collinearity and assessed using variance inflation factors. The significance of the RDA analysis was determined using Permutest (Shi et al., 2022). The relationships between functional genes and environmental factors, as well as dominant taxa and environmental factors, were evaluated using Spearman’s correlation coefficient (Wang et al., 2022a). Mantel test was employed to determine the correlation between different microbial communities and environmental factors. Additionally, to mitigate collinearity effects among variables, only functional genes or genera present in more than half (>50%) of all 11 sediment samples were included for network construction (Liu et al., 2023). Co-occurrence networks were generated using Spearman rank correlations, setting a threshold of |r| ≥ 0.7 and p < 0.01 to investigate the co-occurrence relationships among rpf-like genes, and genes related to nitrogen cycle and VBNC formation, as well as among the gene-associated genera. Spearman correlation coefficients were computed using the R package “psych,” and the resulting network diagrams were visualized using the Gephi 0.9.2 platform (Liu et al., 2023).

The physiochemical properties (Supplementary Table S1) of the sediment samples exhibited notable spatial heterogeneity at a statistically significant level (p < 0.05). SCX sediments were marginally alkaline, while WY sediments were slightly acidic. ZNZ sediments had varying pH levels (6.80 to 8.51), with ZNZ1 upstream being alkaline, potentially due to proximity to a village and impact of domestic wastewater. WY sediments had the highest MC values (58.26–71.39%), followed by SCX (54.45–63.94%) and ZNZ (33.54–53.93%). The trend in ammonia concentrations was WY > SCX > ZNZ, with a decrease from upstream to downstream. Nitrite concentrations followed a pattern of ZNZ > SCX > WY. There were no significant differences in NO₃⁻ and TOM concentrations among the reservoir sediments. High-quality metagenomic sequencing data was obtained for sediment samples from the three reservoirs (Supplementary Table S6), with alpha diversity indices of microbial communities detailed in Supplementary Table S7. The Shannon diversity of sediment samples showed a significant positive correlation with NH₄⁺ (r = 0.76, p < 0.05). Previous studies have indicated that an appropriate range of ammonium nitrogen concentration can supply necessary nutrients for microbial growth, facilitating microbial reproduction in the absence of other nitrogen sources (Kuypers et al., 2018).

PCoA revealed significant differences in the microbial community compositions within sediment samples from the three reservoirs, as evidenced by ANOSIM test (r = 0.38, p = 0.014) (Supplementary Figure S2). Microbial communities at phylum and genus levels, with notable variations across the reservoirs, are illustrated in Figure 1. Proteobacteria (45.01–50.69%) was the predominant phylum in all sediment samples except ZNZ1, where Bacteroidetes (40.63%) dominated (Figure 1A). The prevalence of Proteobacteria in reservoir ecosystems is extensively documented (Li et al., 2019; Wu et al., 2021). Furthermore, significant differences (p < 0.05) in the relative abundance of 38 phyla were observed across the reservoirs (Figure 1B). In addition, pairwise comparisons indicated notable variations in Euryarchaeota between SCX and ZNZ (p < 0.01). Candidatus Omnitrophica exhibited significant differences in relative abundance between WY and SCX (p < 0.001), as well as between WY and ZNZ (p < 0.001). Lentisphaerae also demonstrated significant differences between WY and ZNZ (p < 0.05) (Figure 1B). Notably, Euryarchaeota, Candidatus Omnitrophica, and Lentisphaerae are known to play roles in geomicrobiological nitrogen cycling (Speth et al., 2016).

At the genus level, Bradyrhizobium (1.70–9.44%) and Anaeromyxobacter (0.52–7.03%) were the predominant genera in the three reservoirs, except for ZNZ1 (Figure 1C). These genera are known for their involvement in nitrogen fixation and the DNRA process in the nitrogen cycle (Nelson et al., 2016; Sun et al., 2023). In ZNZ1, the most predominant genus was the methane-producing bacterium Alkaliflexus (6.12%), favored by the alkaline conditions of ZNZ1, providing an optimal environment for the growth of Alkaliflexus (Wang et al., 2022b). Furthermore, Dechloromonas (2.95%), known as an organohalide-respiring bacterium, was the second most abundant genus in ZNZ1, associated with the degradation of aromatic compounds (Yuan et al., 2023). Additionally, based on multiple comparisons, the relative abundance of 261 genera, including Dechloromonas, exhibited statistically significant differences among the three reservoirs (p < 0.05) (Figure 1D). Pairwise comparisons revealed significant differences in Methanothrix between SCX and ZNZ (p < 0.05), and in Methanoregula between ZNZ and SCX (p < 0.01), as well as between ZNZ and WY (p < 0.01). Both Methanothrix and Methanoregula, members of the class Methanomicrobia, are known to play a crucial role in methane production (Mand and Metcalf, 2019; Wu et al., 2021). Significant variations in the relative abundance of the genus Opitutus, involved in the DNRA process (Nelson et al., 2016), were observed between all pairs of reservoirs (Figure 1D). It is noteworthy that a substantial proportion of the genera in the three reservoirs were unclassified (Figure 1C), indicating limited accessibility to most of them. Restoring the activity and culturability of VBNC microorganisms in the reservoirs is essential to bridge the significant gaps between functional strains obtained and those existing in the environments (Lennon and Jones, 2011; Zhou et al., 2023).

The presence of Rpf-like domains and associated genes in sediment samples from the three reservoirs is depicted in Figure 2A and detailed in Supplementary Table S3. Specifically, SCX, ZNZ, and WY contained 577, 1,246, and 327 genes respectively, spanning across 14, 17, and 14 Rpf-like domains. The samples SCX1 and ZN1, located closest to the upstream (near the village) in SCX and ZN, exhibited the greatest occurrence of rpf-like genes, while sample WY3 in the WY reservoir displayed the greatest occurrence of rpf-like genes. Notably, six Rpf-like domains, including MltD-like, Slt70-like, LT-like, MLTF-like, Slt35-like, and LT_Slt70-like, were particularly prevalent across the three reservoirs, collectively representing 87.58% of the total rpf-like genes. Moreover, a systematic assessment was conducted to investigate the relationship between rpf-like genes and environmental variables. RDA analysis showed that 45.93% of the total variation in rpf-like genes could be attributed to five environmental factors, including pH, NH₄⁺, NO₂⁻, NO₃⁻ and TOM. Importantly, pH and NH₄⁺ were identified as exerting significant influences (p < 0.05) on the distribution of rpf-like genes in the sediment samples (Figure 2B). Specifically, the relative abundance of lyz_endolysin_autolysin exhibited a significant positive correlation (r = 0.63, p < 0.05) with NH₄⁺ concentration, whereas the relative abundance of LT-like showed a significant negative correlation (r = −0.76, p < 0.01) with NH₄⁺ (Figure 2C). Additionally, pesticin_lyz-like and LT_VirB1-like were significantly associated with TOM, while LT_Slt70-like displayed a significant positive correlation (r = 0.63, p < 0.05) with NO₂⁻ concentration. These findings suggested the significant influence of NH₄⁺ concentration on shaping the distribution of rpf-like genes in sediment samples from water reservoirs.

In order to examine the distribution patterns of microbial communities harboring rpf-like genes in the reservoir sediments, an investigation into the microbial composition linked to the presence of rpf-like genes was carried out. At the phylum level (Figure 3A), Proteobacteria (31.56–78.04%) was the primary phylum harboring rpf-like genes in the reservoir sediments. Bacteroidetes, the second most abundant phylum containing rpf-like genes, exhibited an increased abundance from the upstream to downstream sections in the SCX and WY reservoirs. Remarkably, the presence of Bacteroidetes with rpf-like genes was significantly higher in the ZNZ1 (31.36%) compared to other sediment samples (0.51–10.81%). A range of 0.68 to 3.91% of Actinobacteria were also found to contain rpf-like genes, with the highest proportion detected in SCX1. Extensive documentation exists regarding the presence of rpf-like genes in high G + C Gram-positive Actinobacteria (Mukamolova et al., 2006; Shi et al., 2024b). At the class level (Figure 3B), Alphaproteobacteria (6.14–46.37%), a subgroup of the Proteobacteria, was the predominant class hosting rpf-like genes in the reservoir sediments, followed by Betaproteobacteria (6.47–32.62%). Additionally, the abundance of Bacteroidia (25.68%) containing rpf-like genes was notably elevated in the ZNZ1 in comparison to the other sediment samples, consistent with the findings from the phylum analysis. These results revealed that rpf-like genes in the three reservoirs were predominantly found in Proteobacteria, specifically in the Alphaproteobacteria and Betaproteobacteria.

At the genus level (Figure 3C), most of the genera containing rpf-like genes in the reservoir sediments were categorized as unclassified. In the SCX reservoir, Bradyrhizobium (5.24–10.20%) was the predominant genus, except for SCX1 where Anaeromyxobacter was dominant. Significant differences were observed in the distribution of genera containing rpf-like genes from the upstream to downstream sections of the ZNZ reservoir. The predominant genera were Thermotoga, Nitrospira, and Anaeromyxobacter in ZNZ1, ZNZ2, and ZNZ3, respectively. In the WY reservoir, Nitrospira (4.20–5.17%) was the dominant genus, except for WY3 where Dechloromonas was prevalent. It is noteworthy that genera harboring rpf-like genes, such as Bradyrhizobium, Anaeromyxobacter, and Nitrospira, have been extensively reported for their involvement in nitrogen cycling (Wang et al., 2021; Sun et al., 2023). Although several studies have reported the presence of rpf-like genes in Streptomyces and Mycobacterium (Mukamolova et al., 2006; Kana et al., 2008), Streptomyces with rpf-like genes were only found in SCX1 (0.08%) and SCX2 (0.33%), and Mycobacterium with rpf-like genes were only detected in SCX5 (0.08%) and ZNZ3 (0.13%).

The association between genera harboring rpf-like genes and environmental factors was examined through heat map analysis (Figure 3D). The relative abundance of Aromatoleum exhibited a significant (p < 0.05) and negative correlation with pH, while Dechloromonas showed a significant (p < 0.05) and negative correlation with NO₃⁻ concentration. Similarly, the relative abundances of Hyphomicrobium and Mycobacterium displayed significant (p < 0.05) and negative correlations with NH₄⁺ concentration. The relative abundance of Permianibacter was significantly (p < 0.05) and negatively correlated with NO₂⁻ concentration. In contrast, the relative abundance of Smithella was significantly and positively correlated with MC (p < 0.05) and TOM (p < 0.01). The relative abundance of Plasticicumulans exhibited a significant (p < 0.05) and positive correlation with NO₃⁻, while the relative abundance of Serpentinimonas showed a significant (p < 0.05) and positive correlation with pH. The heat map analysis revealed that environmental factors had diverse effects on species harboring rpf-like genes, thus influencing the microbial community structure. Furthermore, by integrating local environmental variables with community structures, the correlation between environmental factors and community compositions containing rpf-like genes was identified. As illustrated in Figure 3E, the first two components explained 41.18% of the microbial diversity containing rpf-like genes. The RDA results supported the significant influence of pH (p < 0.05) on the distribution of microbial communities with rpf-like genes in the reservoir sediments. Additionally, pairwise Spearman’s correlation and partial Mantel tests were further employed to reveal the environmental drivers of communities harboring rpf-like genes in the three reservoirs (Figure 3F). The microbial communities containing rpf-like genes in SCX and WY reservoirs were significantly (r = 0.58, p < 0.05; r = 0.52, p < 0.05) correlated with pH, indicating that pH exerted a primary influence on the species compositions containing rpf-like genes.

Based on the KEGG annotation, nitrogen-cycling functional genes associated with nitrogen fixation, nitrification, denitrification, DNRA, ANRA, anammox, and ammonification were annotated in all sediment samples (Supplementary Figure S3 and Supplementary Table S4). Denitrification was found to be the predominant function, constituting 34.67% of the total nitrogen transformation functions (Supplementary Figure S4). Ammonification and DNRA genes accounted for 18.16 and 16.73% of the total nitrogen transformation functions, respectively, while nitrogen fixation genes exhibited the lowest proportion at 7.23%. Compared to denitrification, the DNRA process prefers highly reduced environments (Zhang et al., 2023c). The dominance of denitrification over DNRA in the reservoir sediments may be attributed to the less reduced environments. Notably, the key genes, hzsA, hzsB, and hdh, known to be involved in anammox (Kuypers et al., 2018; Wang et al., 2022a), were absent in all samples. This absence could be explained by the low tolerance of anammox bacteria to oxygen, which inhibits their activity even under low oxygen levels. Molecular oxygen can directly poison redox enzymes, particularly Fe-S proteins with low redox potentials (such as ferredoxins and hydrogenases), essential for electron transfer. Furthermore, during the oxygen reduction process, intracellular reactive oxygen species are generated due to electron leakage, leading to oxidative stress that further hampers the activity of anammox bacteria (Zhang et al., 2023b).

The heatmap illustrating the correlations between nitrogen-cycling genes and environmental factors is presented in Figure 4. Nitrogen fixation genes were predominantly located downstream of the WY reservoir, particularly in WY3, with a minor distribution observed in the downstream sections of SCX4 and SCX5 in the SCX reservoir. The abundance of nifH, involved in nitrogen fixation, demonstrated a significant positive correlation with the TOM. The relatively higher abundance of nifH in the downstream section may be attributed to reduced flow, resulting in TOM deposition (Wakelin et al., 2008). Nitrification genes were predominantly present in the WY reservoir, with the amoB gene, responsible for ammonia oxidation, exhibiting a significant positive correlation with NH₄⁺ concentration, a relationship previously observed by Hallam et al. (2006). Denitrification and DNRA genes were mainly distributed in the ZNZ2, with a high distribution also observed in the WY reservoir. The abundance of nirS and norB genes was significantly associated with pH, while the fluctuation in nirD corresponded to that of NO₂⁻ concentration. Furthermore, ANRA genes were primarily distributed downstream of the ZNZ reservoir, with significant negative correlations observed between the abundance of nasA and nirA genes and NH₄⁺ concentration. The narB gene exhibited significant correlations with NO₃⁻ and NO₂⁻ concentrations. Additionally, the abundance of ammonification genes in the three reservoirs displayed a decreasing trend from the upstream to downstream sections.

Following taxonomic assignment, a total of 617 genera within 61 phyla containing nitrogen-cycling genes were identified in all sediment samples. At the phylum level (Figure 5A), the majority of nitrogen-cycling functional genes were associated with Proteobacteria (47.40–63.66%), followed by Acidobacteria (1.64–15.67%), Actinobacteria (0.84–11.49%), Nitrospirae (0.62–8.58%), Bacteroidetes (0.40–16.72%), and Chloroflexi (1.40–5.84%). Notably, the abundance of Bacteroidetes in ZNZ1 was significantly higher compared to the other sediments. At the genus level (Figure 5B), apart from unclassified genera, 23 genera with relative abundance greater than 1% were detected, with Nitrospira (0.07–7.07%), Bradyrhizobium (0.34–7.32%), Anaeromyxobacter (0.03–13.93%), and Dechloromonas (0.30–12.53%) being the most dominant. To elucidate the primary chemical characteristics of sediment supporting the nitrogen-cycling microbial community, RDA was conducted to assess the sediment chemical properties exerting the most significant influence. RDA1 and RDA2 could explain 44.70% of the total variation (Figure 5C), with SCX and WY samples exhibiting greater clustering, while ZNZ samples appeared more dispersed. The composition of sediment nitrogen-cycling microbial communities in the SCX, ZNZ, and WY exhibited significant correlations with NH₄⁺ (p < 0.01) and pH (p < 0.05). Furthermore, partial Mantel tests were employed to further elucidate the environmental drivers of nitrogen-cycling microbial communities in the sediments (Figure 5D). The community structure of denitrifiers was significantly influenced by pH (r = 0.41, p < 0.05) and MC (r = 0.42, p < 0.05), while the nitrogen-fixing microbial community was primarily affected by TOM (r = 0.61, p < 0.01). Figure 5D also illustrated the correlations among different sediment parameters, with the exclusion of high collinearity (|r| > 0.8) (Ding et al., 2022). Specifically, TOM was found to be significantly correlated with MC (r = 0.75, p < 0.05). The results suggested that TOM and MC may jointly or independently impact the composition of nitrogen-fixing microbial communities (Zhang et al., 2023a).

Additionally, the relative contributions of different taxa to nitrogen-cycling genes across various nitrogen transformation processes were evaluated (Figure 6). The majority of nitrogen fixation genes were associated with Anaeromyxobacter, Geobacter, and Methanoregula. Notably, Methanoregula was found to harbor the nifH gene, which encodes iron-containing electron transfer proteins (Kuypers et al., 2018), and exhibited significant enrichment in downstream sites, regulating the downstream nitrogen-fixing process. Regarding the nitrification process, amoA, amoB, amoC, hao, and nxrB were predominantly assigned to the genus Nitrospira, known for its ability to oxidize ammonia in ammonia-limited environments (Kits et al., 2017). The prevalence of Nitrospira in the SCX reservoir suggested its potential involvement in the complete ammonia oxidation process. Furthermore, the main denitrifiers, namely, Bradyrhizobium and Nitrospira, exhibited higher abundance downstream of the reservoirs compared to upstream. Bradyrhizobium, also identified as the predominant genus in the ANRA process, was mainly associated with the abundance of nasA, encoding the catalytic subunit of assimilatory nitrate reductase (Hu et al., 2022), and demonstrated increased prevalence at downstream sites. Hidalgo-García et al. (2019) demonstrated the generation of nitrous oxide through the interplay of the ANRA and denitrification processes, indicating the promotion of the initial step of the ANRA process at downstream sites, subsequently providing more substrate for denitrification. Meanwhile, most of the ammonification genes were also assigned to the genus Bradyrhizobium, as well as Anaeromyxobacter. In the DNRA process, nrfA and nrfH were predominantly attributed to the genus Anaeromyxobacter, and unclassified genera in Deltaproteobacteria and Anaeromyxobacteraceae, while nirB was mainly assigned to unclassified genera in Betaproteobacteria. Previous studies have highlighted the role of DNRA in generating NH₄⁺ to enhance primary productivity in oligotrophic marine and soil systems (Lam et al., 2009; Shan et al., 2016), a process that may also occur in water reservoir sediments.

The network analysis based on Spearman correlation revealed co-occurrence patterns among rpf-like genes, nitrogen-cycling functional genes, and VBNC-related genes with strong correlations (Figure 7A). The network consisting of 81 nodes connected by 638 edges, with 88.1% positive and 11.9% negative interactions. Notably, the rpf-like genes demonstrated 11 positive connections and 4 negative connections with nitrogen-cycling functional genes. Among these, the MLTF-like gene, which showed high abundance in all three reservoirs, displayed the highest number of connections with nitrogen-cycling functional genes and primarily positive correlations, indicating its crucial role in the positive correlation between rpf-like genes and nitrogen-cycling functional genes. Furthermore, the rpf-like genes demonstrated 33 positive links and 13 negative links with VBNC-related genes, with the LT_Slt70-like gene specifically linked to VBNC-related genes through 28 positive connections. Additionally, VBNC-related genes showed the highest self-co-occurrence rate, comprising around 74.5% of the occurrences.

To unravel the intricate microbial co-occurrence network, correlation network analysis was conducted on the gene-associated genera. As depicted in Figure 7B, the network consisted of 364 nodes connected by 1874 edges, with 82.0% representing positive correlations and 18.0% reflecting negative correlations among these genera. A modularity index exceeding 0.4 is widely acknowledged as indicative of a modular structure (Jiao et al., 2016), and the observed modularity index of 0.58 suggested a modular construction of functional microbial populations in the water reservoir sediments. Within the network, the key genera Pseudolabrys, Bradyrhizobium, Nitrospira, Anaeromyxobacter, and Rubrivivax, which all contained nitrogen-cycling functional genes, rpf-like genes, and VBNC-related genes, collectively constituted 5.4, 4.9, 3.3, 2.2, and 1.7% of the network, respectively. Previous studies have documented the participation of Pseudolabrys in denitrification processes (Kämpfer et al., 2006; Li et al., 2022). Notably, Pseudolabrys appeared to be a significant genus within the network, possessing all three types of genes, with a higher degree (degree: 105) and betweenness centrality. Dechloromonas was noted for harboring both nitrogen-cycling functional genes and rpf-like genes. Furthermore, 11 genera were found to contain nitrogen-cycling functional genes and VBNC-related genes (Supplementary Table S8), suggesting the possibility of microorganisms involved in the nitrogen cycle entering the VBNC state. The results indicated that these nitrogen-cycling functional microbial populations included microorganisms with rpf-like genes and VBNC-related genes. Overall, the interaction between rpf-like genes and genes involved in nitrogen cycle and VBNC formation, along with their associated genera, revealed a strong correlation of Rpf homologs with functional microbial communities in water reservoir sediments.