The study was done in Henan, China, at the Yellow River Floodplain Ecosystems Research Station, located at 34°59'65” N and 113°25'05” E. Two plant communities (Phragmites australis (Cav.) Trin. ex Steud. and Erigeron canadensis L.) were selected from the floodplain. These two plant communities are the dominant species in the Yellow River floodplain ecosystem system. Three plots were set up in each plant community (Figure 1). In August and September 2020, water and sediment samples were collected every 2 weeks, four times in total. Sediment drilling was performed twice at 0–10 cm for the surface and 10–20 cm for the subsurface, and three points from each plot were selected and combined. In total, 48 sediments were taken (2 plant communities × 3 plots × 2 depths × 4 times). At each sampling plot, a 1.5 L water sample was collected, and the sum of 24 water samples was taken (2 plant communities × 3 plots × 4 times).

The samples were delivered to the lab in sanitized polythene bags and on ice. Some water portions were passed through a 0.45 μm filter for physical and chemical property determination. Another portion of the water proceeded through a 0.22 μm filter (Merck Millcore Ltd.) and was employed in order to identify the bacterial structural community. Two groups of sediment were distinguished. To ascertain the composition of the bacterial community, one half was kept at a controlled temperature of −80 °C. For the physicochemical inspection, the residual sample was air dried at an ambient temperature.

The total nitrogen and carbon contents of the sediment were examined using the Vario Max CNS elemental analyzer (Elementar Analysensysteme GmbH, Hanau, Germany). The pH was determined using a pH meter (Sartorius PT-21, Shanghai, China) and a soil-to-ddH₂O ratio of 1:2.5. Using a flow injection autoanalyzer, nitrate (NO3--N) and ammonium (NH4+-N) levels in sediment and water samples were determined (Westco Module 200, Smart Chem, France). Using an auto analyzer, the total phosphorus in sediment and water samples were also analyzed (AA3, Norderstedt, Germany). To evaluate the sediment moisture, 15 g of recent sediment was dehydrated in a 105 °C oven for 48 h. A TOC analyzer analyzed the total soluble organic carbon (TOC) in water samples (vario TOC cube, Elementar, Germany).

Implementing the guidelines provided by the manufacturer, an Omega Soil DNA Kit was used to obtain the total DNA of the water and sediment (M5635-02, Omega Bio-Tek, Norcross, GA, USA). In order to amplify the bacteria's V3-V4 hypervariable region sequences, the primers 338F (ACTCCTACGGGAGGCAGCAG) and 806R were used (GGACTACHVGGGTWTCTAAT). The MiSeq system performed pair-end sequencing with the Illumina MiSeq Reagent Kit v3 in Shanghai Personal Biotechnology Co., Ltd. (Shanghai, China). Illumina MiSeq-PE300 was used for all sequencing operations (Illumina, San Diego, CA, USA). Bioinformatics of the microbiota was performed using QIIME 2 (Bolyen et al., 2019). After the unprocessed measurements with paired ends had been demultiplexed using demux plugin, the primers were cut using the cut adapt plugin (Martin, 2011). DADA2 procedure was used to quality-filter and denoise the sequences during settings (Callahan et al., 2016). To control sequence quality and eliminate any sequencing problems, we used the default parameters for the learnErrors, derepFastq, dada, and mergePairs methods. Afterwards, the amplicon sequence variations (ASVs) of the singletons were discarded. Then, using the Silva v132 database (http://www.arb-silva.de), the taxonomic identities of bacteria were determined (Quast et al., 2013).

The BioProject numbers were included with raw sequencing information uploaded on the National Center for Biotechnology Information's Sequence Read Archive (SRA) database: PRJNA 898838 for the bacteria sequences.

We calculated observed species using indexes like Shannon, Chao1, the diversity of phylogeny (Faith's PD), Simpson, evenness (Pielou), and Good's coverage to evaluate Alpha-diversity variations among water and sediment. Differential patterns in the community composition were visualized using non-metric multidimensional scaling (NMDS). The comparison of similarities (ANOSIM) method was used to investigate how different bacterial communities existed in various groups. There are proven Spearman relationships between environmental variables and the variety and structure of bacteria in water and sediment. The effect size of linear discriminant analysis (LEfSe) was utilized to differentiate the abundant features of bacterial communities among groups.

The underlying mechanism(s) affecting the community assembly was identified by employing the average nearest taxon distance (MNTD) and the maximum pairwise phylogenetic distance (MPD) calculated (Stegen et al., 2012). The “comdistnt” function from the R package Picante was used to determine the abundance-weighted beta closest taxon index (NTI), which measures the variation in the mean of the random null expectation from the observed mean nearest taxon distance (MNTD) (999 for the number of randomizations) (Stegen et al., 2012). After that, the comparative impact of the assembly process and underlying processes (Variable selection, homogenous selection, dispersal limitation, dispersal homogenizing, and undominated) were assessed using the NTI and modified Raup-Crick (RC_Bray) metrics (Chase et al., 2011; Stegen et al., 2015).

Moreover, co-occurrence networks of water, surface, and subsurface sediments were constructed. The core microbial amplicon sequence variants (ASVs) shared by more than 60% of samples and with a mean relative abundance of more than 0.1% were chosen to minimize the effects of site-specific ASVs upon the network layout. SparCC was then employed to create the co-occurrence networks. Nodes using a direct correlation value larger than 0.6 and a p-value lower than 0.01 were chosen to form the network. In order to see the network, Cytoscape 3.8.2 was used (Shannon et al., 2003). The degree of centrality, proximity of centrality, number of vertices and edges, and betweenness of centrality and modularity are a few examples of known metrics that were used to define the topology of the networks. Natural connectedness after “attacking” edges or nodes can be used to gauge how robust a network is: higher resilience to threats to natural connectivity are signs of a more strong or stable network (Wu et al., 2021).

The contributions of the designated source sites to the bacterial community composition of the designated sink were predicted with the Bayesian classifier software program SourceTracker26 using R package SourceTracker v1.0.1. The designated sink and source sites' precise configurations are shown in Supplementary Table 1. The ASV tables and files produced by quality control were utilized. It has been shown that Source Tracker analysis with default settings, which include a rarefaction Alpha (0.001), beta (0.01), and depth of 1,000. Provide relatively good sensitivity, specificity, reliability, and precision. Five separate Source-Tracker runs were used to calculate the mean contribution proportion and variance for each source to identify and prevent any potential false positives. The relative standard deviation (RSD), which is used to calculate the belief in the mean, was used to anticipate and utilize the ratio of averaged expected percentage. Source proportions were computed: Statistical evaluations were performed using the, stated R environment (v3.5.1; http://www.r-project.org/), unless otherwise noted.

The bacterial communities' alpha diversity within water and sediment is shown in Figure 2A. In water, the Chao1 index was between 333 and 959, the PD ranged from 25 to 92, and the Shannon index ranged between 3.7 and 7.6. The Shannon index varied between 8.7 and 9.8, the PD varied from 97 to 169, and the Chao1 index varied from 1,068 to 1,777 in surface sediment. The Shannon index was between 8.8 and 9.5, the PD varied from 86 to 173, and the Chao1 index ranged from 1,061 to 1,737 in subsurface sediment.

Bacterial community compositions in the water and sediment samples varied significantly, according to NMDS ordinations and comparison of similarity (ANOSIM) (P < 0.001; Figure 2B; Supplementary Table 2). There was a significant structural difference between the surface and subsurface sediment bacterial communities (Figure 2B). In sediment, the phyla Proteobacteria, Acidobacteria, and Chloroflexi have larger relative abundances (34.6–38.8%, 18.9–24.3%, and 13.2–13.9%, respectively) than in water (27, 0, and 0.3%, respectively). However, compared to sediment (2.3–4.6% and 2.4–4.8%, respectively), the relative amounts of Actinobacteria and Firmicutes were higher in water (30.9 and 36.2%Figure 2C). The Source Tracker results revealed that only 0.3% bacterial ASVs inside water came from the surface sediment and 0.08% from the subsurface sediment. Water is the source of 0.7% of bacterial ASVs in the surface sediment and 0.3% of the bacterial ASVs in the subsurface sediment (Figure 2D).

The examination of the null model found that deterministic processes took control of the bacterial communities in the water, whereas bacterial populations in the sediment were largely dominated by stochastic processes (Figure 3, Supplementary Table 3). Homogenous selection played a major role in controlling how bacterial colonies gathered in the water. The bacterial community in surface sediment assembly mainly comprises homogenous selection, dispersal homogenizing, and undominated processes (weak selection, weak dispersal, diversification, and/or drift). In subsurface sediment assembly, the bacterial community is primarily characterized by homogenous selection, dispersal limiting, dispersal homogenizing, and undominated processes.

Analysis of the co-occurrence network revealed that the number of nodes in the water was less than that in sediment, but the number of edges/nodes was greater than that in sediment (Figure 4 and Supplementary Table 4). The average number of neighbors, clustering coefficients, network density, network centralization, and the negative correlation ratio in water were also higher than those in sediment. We also performed a resilience test to gauge a network's resistance to node or edge attacks through changes in connectivity that occur naturally. We found that the natural connectivity of the water network reduced more significantly and fluctuated in severity more than sediment by deleting the identical number of nodes or edges, indicating weakened resistance. Natural connectivity based on robustness tests also revealed that surface sediment was more resistant to node or edge attack than subsurface sediment.

Among the environmental parameters for water, the NO3--N demonstrated the most positive associations with Cyanobacteria, Chloroflexi, Planctomycetes, and the Chao1 index, and the most negative associations with Firmicutes (Supplementary Figure 1). The NH4+-N demonstrated the most positive associations with Acidobacteria, Verrucomicrobia, and the Shannon index. In surface sediment samples, however, the NO3--N was most significantly connected with positive Acidobacteria, Gemmatimonadetes, and the Simpson index, and most negatively correlated with Actinobacteria and Firmicutes. The NH4+-N had the most powerfully positive correlation with Proteobacteria, and the most negative correlations with Firmicutes and Rokubacteria. TN had the most significant negative correlations with Chloroflexi. In subsurface sediment samples, the NH4+-N was most significantly connected with positivity with Acidobacteria, and most strongly inversely associated with Chloroflexi. The Simpson index and the TP had the strongest positive correlations, whereas the pH had the strongest positive correlation with Nitrospirae.

In the sediments (Surface and subsurface), a significant difference was found between the Erigeron canadensis L and Phragmites australis (Cav.) Trin. Ex Steud (Figure 5). Phragmites australis (Cav.) Trin. Ex Steud's bacterial community was indexed by Chao1, Shannon, and Simpson, which were significantly lower than those of Erigeron canadensis L in both the surface and subsurface sediment. The results of the ANOSIM test and NMDS analysis revealed no significant correlation. While a considerable variation was discovered between both the surface and subsurface sediment of the two plants, there were no differences in the bacterial population in the water between the two plants (Figure 5 and Supplementary Table 5).

Significant variation in bacterial community composition was detected between different plant communities (p < 0.001) (Supplementary Figure 2). The relative abundance of Subgroup7, Subgroup6, Anaerolinea, AKYG1722, Gitt GS136, C0119, Gemmatimonas, Sphingomonas, Ramlibacter, and Ellin6067 in the surface sediment of Australian Phragmites (Cav.) was found. Compared to the other villages, Trin. Ex Steud was much higher (p < 0.01). The proportional prevalence of PAUC26f , Subgroup 22, Subgroup 9, A4b, S085, Desulfuromonas, MBNT15, Azoarcus, PLTA13, Pseudomonas, and Rokubacteriales in the subsurface sediment of Erigeron canadensis L was substantially higher than that in the Phragmites australis subsurface silt (Cav.) Trin. Ex Steud. The relative abundance of Latescibacteria, Haliangium, and MND1 in the subsurface sediment of Phragmites australis (Cav.) Trin. Ex Steud was higher than that of other communities.

The alpha diversity of the fourth water sample shows a significant difference, although there is no significant difference in the samples of surface and subsurface sediments (Supplementary Figure 3). The Chao1, Simpson, and Shannon indexes among the four samples showed the number of microbiological colonies was the highest in the second water sample and lowest in the fourth. Results of the NMDS analysis and the ANOSIM test showed that there were notable variations in the bacterial communities of water between the four times samples (Figure 6 and Supplementary Table 6), while there was no significant difference in the bacterial communities of the surface and subsurface sediments between the four times samples (Figure 6 and Supplementary Table 7).

Most of the microorganisms in the communities were Firmicutes, Actinobacteria, Bacteroidetes, Proteobacteria, and Cyanobacteria in water (Supplementary Figure 4). We discovered that the abundance of these dominant bacteria fluctuated in the four samples we collected. The results of Source Tracker revealed that 52% of the ASVs in the water sampled for the second time came from the water sampled for the first time, 44% of the ASVs in the water sampled for the third time came from the water sampled for the second time, and 91% of the ASVs in the water sampled for the fourth time came from the water sampled for the third time (Supplementary Figure 5). The majority of the microorganisms in the communities were Proteobacteria, Acidobacteria, Chloroflexi, Gemmatimonadetes, the bacterial genus Bacteroidetes, and Firmicutes in surface and subsurface sediments, and the relative abundance in the sediments sampled four times is relatively stable (Supplementary Figure 6).

LEfSe revealed that the abundance of Alsobacter in the first water sample compared to the other three periods was considerably higher. However, the abundance of Acidobacteria, Chloroflexi, Cyanobacteria, and Verrucomicrobia in the second water sample, compared to the other groups, was substantially higher (Supplementary Figure 7). In comparison to the other three water samples, the fourth had a much higher number of Actinobacteria. The Venn diagram showed that there are only 242 common ASVs in the fourth water sample, accounting for 0.3% of the total ASV, while there are 74,364 and 74,359 common ASVs in the surface and subsurface sediments of the fourth water sample, accounting for 96.8 and 96.8%, respectively (Figure 6).

The null model analysis revealed that the third and fourth water samples had greater deterministic bacteria community formation processes than the first and second water samples (Supplementary Table 8).

Studies have shown that sediments can act as reservoirs, absorbing and subsequently depositing microorganisms from surrounding areas, such as soil and plant detritus (Crump et al., 2012). In addition, sediments are formed over a long period through soil deposition and erosion. In this process, the selective transport of light particles and labile organic components occurs (Mueller-nedebock et al., 2016). Deposition increases moisture and nutrient content simultaneously and may change pH (Berhe and Kleber, 2013). These modifications made in sedimentary environments may greatly enhance bacterial diversity. According to the Spearman correlations analysis, many environmental conditions impact bacteria co-existing in sediment and water. The bacterial communities clearly distinguished between the habitations in the water and the communities in the sediment, indicating that the habitat was primarily responsible for the differences in community structure. The results of SourceTracker also show that the bacterial ASVs in water were rarely from sediment, and the bacterial ASVs in sediment were rarely from water, which suggested that their interactions were minimal.

In addition, we believe that the difference in community assembly among different habitats is also a factor that will cause variations in the composition of bacterial communities. While stochastic mechanisms dominated the bacterial communities in the sediment, the bacterial species in the water were dominated by deterministic mechanisms. The non-random, niche-based mechanisms that underlie the deterministic process include environmental filtering and numerous biological interactions (Zhou and Ning, 2017). Stochastic process refers to the random change of community structure caused by birth, death, immigration, and/or historical contingencies (Zhou and Ning, 2017). In general, the process of deterministic assembly corresponds to low nutrient conditions, and large environmental changes lead to deterministic assembly (Yu et al., 2019). The stochastic assembly is more likely to occur in small habitats with abundant species, high production, little disturbance, and/or little predation. Due to the elimination of certain functions, the deterministic process predominates in low-diversity communities while the stochastic process predominates in communities with a high level of diversity (Xun et al., 2019).

The current study proved that the network's complexity increased in water compared to sediment. We also discovered that water strengthened the network's members' negative correlation. The rise in the negative correlation ratio indicated increased competition between soil microbial communities. We also found that compared to sediment, the bacterial network in water exhibits unstable properties. Earlier research has revealed that improved node connection (Fan et al., 2018), centrality (Jordán, 2009), and complexity (Schoener, 1974) are linked to diminished network stability. According to the robustness assessments based on natural connection, compared with sediment, the robustness of the bacteria network in the water against node or edge attacks was reduced.

Previous studies have shown that the soil properties of the Yellow River floodplain changed significantly with soil depth (Yu et al., 2020; Chen et al., 2022). The main reason for the difference in depth may be that the surface soil is greatly affected by the input of root exudates and plant litter, while the nutrient source of the subsurface soil is mainly the decomposition of dead roots (Jobbagy and Jackson, 2000). Wetland is impacted by long-term flooding, and the root biomass is the largest in the surface soil (Zeng et al., 2016). The difference between the surface and subsurface sediment environments also profoundly affects the bacterial community structure. Moreover, we also found that, compared with the surface sediment, the robustness of the bacterial network in the subsurface sediment was also reduced. As a result, the bacterial population in the surface sediment is more resistant to environmental change than the community in the deeper soil layers.

In addition, we also noticed the different responses of bacterial communities to plant communities in water and sediment. The plant community in the floodplain ecosystem significantly affected the bacterial diversity and structure communities in sediment (surface and subsurface), but did not affect the structure of the bacterial community in water. Some studies throughout the Yellow River Delta found that microbial metabolic activities were closely related to plant covers, and the β-diversity patterns and function of the group of bacteria differed significantly between soils with different plant covers (Cao et al., 2019; Wei et al., 2020). In order to control growth conditions, by secreting bioactive compounds into the rhizosphere and changing the soil environment, plants can affect the soil microbiota (Hu et al., 2018).

The patterns of the water's bacterial community exhibit dynamic temporal alterations. The results of Source Tracker analysis on the source of bacteria ASVs in all four water samples revealed that the bacteria ASVs in water fluctuated with sampling time. Venn diagrams indicated that the proportion of unique ASVs of bacterial communities in water at different sampling times was high, and the proportions of shared ASVs were low, whereas the opposite was true in sediment. This shows that the water at different times has a unique bacterial community composition and the water environment continuously changes. Our results indicate that the water selects particular assemblages of microorganisms that do not happen randomly or occur in a predictable way over time, while the sediment environment is relatively stable and the bacterial communities gather randomly. The research area was in the center and bottom of the Yellow River floodplain, and its hydrodynamic conditions are complex. During the two adjacent sampling times, due to the influence of the river flow discharge power, the water flow at the sampling location exchanged violently, resulting in significant differences in the bacterial communities of the different water samplings. In comparison to water, the exchange rate of sediment was slow, and the bacterial community was relatively stable.