Introduction

Reservoir microorganisms comprise a considerable number of bacterioplankton that play important roles in the aquatic ecosystem. Investigations of the community composition and function of bacterioplankton have far-reaching significance for the ecological management and maintenance of reservoir environments (Hanson et al. 2012; Guidi et al. 2016). Since the vast majority of microorganisms cannot be cultured (> 99%), modern culture-free molecular biology techniques have become an important means of studying microbial diversity (van Dijk et al. 2014). High-throughput sequencing can provide a more comprehensive picture of the composition of biomes than PCR-DGGE and library cloning methods because of its increased throughput for specific DNA fragments. Thus, this technique has been widely used to study the composition, distribution characteristics, and influencing factors of bacterioplankton communities (Faust et al. 2015; Stec et al. 2017). Studies have shown that bacterioplankton communities in water are affected by spatiotemporal factors (Liu et al. 2015; Li et al. 2017), pH (Xiong et al. 2012), temperature (Anders et al. 2010) and complex abiotic and biological processes, such as nutrient or phosphorus cycling (Anders et al. 2010; Logue et al. 2012) and eutrophication parameters (Zhao et al. 2016; Dai et al. 2017). However, these studies have focused mainly on the bacterioplankton community composition and diversity (α- and β-diversity), which cannot reveal the interactions among bacterioplankton (Wang et al. 2016; Zhao et al. 2016). Because microorganisms co-occur and interact through complex network structures in an ecosystem, understanding these interactions is a core topic of microbial ecology (Coyte et al. 2015). Network analysis is a systematic analysis method that analyzes the internal interactions in an ecosystem based on mathematical model analysis and has shown promising effects in studying interspecies interactions in biological communities (Faust and Raes 2012). Recently, network analysis has also been used to explore the network structures of microorganisms. In particular, the application of analytical techniques, such as high-throughput sequencing and gene microarrays, has provided a solid data foundation for the study of microbial networks. Network analysis has been applied to complex habitats, such as dental surface biofilms, microorganisms in humans, and soils (Bahram et al. 2014; Banerjee et al. 2016; Creamer et al. 2016; Shi et al. 2016; Chen et al. 2017, 2019). Network analysis has demonstrated obvious advantages in studying the interactions of microorganisms, the identification of key populations, and the response mechanisms of microbial community structures to environmental disturbance. At present, network analysis is rarely used to study microbial community compositions and functions and their roles in indicating water quality (Karimi et al. 2017; Xu et al. 2018). Recently, Zhao et al. (2016), Dai et al. (2017), and Wang et al. (2016) used molecular network analysis to investigate the interspecies interactions of bacterioplankton in different eutrophic lakes and the ocean acidification caused by different CO2 concentrations. These studies demonstrated that the use of network analysis could improve the understanding of the ecological functions of bacterioplankton communities in lakes and oceans.

The Middle Route of the South-to-North Water Diversion Project (MR-SNWDP) is a large-scale water diversion project that was implemented to alleviate the serious shortage of water resources in northern China. The Danjiangkou Reservoir is the core water source of the MR-SNWDP; thus, long-term monitoring of the water quality of this reservoir is critically important. Bacterioplankton play important roles in the aquatic ecosystem, but little research has been conducted on the bacterioplankton community in the Danjiangkou Reservoir area (Gao et al. 2018). In the present study, five typical ecological sites were selected in the Danjiangkou Reservoir area. High-throughput sequencing was used to analyze the bacterioplankton community diversity in surface waters and then investigate the community composition and distribution characteristics (Chen et al. 2018). Molecular network analysis was used to analyze the interactions of bacterioplankton. The aim of this study was to explore the ecological composition of the bacterioplankton community and its influencing factors in the Danjiangkou Reservoir and provide a reference for the protection of the water environment in this reservoir.

Materials and methods

Study site and sampling

According to the terrain and tributaries of the Danjiangkou Reservoir and the administrative regions in the corresponding section of Henan Province, five typical ecological sites were selected: Kuxin (K, the center of Danjiangkou Reservoir), Qushou (Q, the canal head of the MR-SNWDP), Songgang (S, located in a reservoir bay that was previously influenced by a shipping dock and tourists), Taizishan (T, the dividing line between Henan and Hubei provinces) and Heijizui (H, the last estuary to the Danjiangkou Reservoir after the confluence of the Danjiang River and Guanhe River). A total of 35 surface-water samples (seven replicates for each sample) were collected from the reservoir on May 1, 2016, and nine essential environmental factors were investigated (Fig. 1 and Table 1).

Fig. 1
figure 1

Locations of the five sampling stations in the Danjiangkou Reservoir and the water conveyance canal of the Middle Route of the South-to-North Water Diversion Project (MR-SNWDP) in China. The station codes represent the first letters of the sampling station names: K: Kuxin, Q: Qushou, S: Songgang, H: Heijizui, and T: Taizishan. The map was generated using ArcGIS 10.0 (ESRI, Redlands, CA, USA: https://www.esri.com/software/arcgis)

Full size image
Table 1 Main physicochemical characteristics, water quality standards and the trophic level index (TLI) of the water samples (means ± SE)
Full size table

Physicochemical variables were measured according to the environmental quality standard for surface water in China (GB3838-2002). Water temperature (T), pH and dissolved oxygen (DO) were measured in situ using a YSI 6920 (YSI Inc., Yellow Springs, Ohio, USA). Secchi depth (SD) was determined with a 30-cm-diameter Secchi disk. Water samples for chemical analysis were transported to the laboratory within 24 h, stored in a refrigerator at 4 °C, and analyzed within one week after sample collection. The permanganate index (CODMn) was calculated using the potassium permanganate index method, and the chemical oxygen demand (COD) was measured by the potassium dichromate method. The total phosphorus (TP) was determined with acidified molybdate to form reduced phosphorus-molybdenum blue and measured spectrophotometrically. Total nitrogen (TN) was assayed via alkaline persulfate digestion and UV spectrophotometry, whereas ammonia nitrogen (NH4–N) was measured using Nessler’s reagent spectrophotometric method. Chlorophyll a (Chl a) concentrations were estimated spectrophotometrically after extraction in 90% ethanol.

The trophic status of the Danjiangkou Reservoir area was assessed by measuring the parameters TN, TP, CODMn, Chl a, and SD according to the improved Carlson’s trophic level index (TLI) (Wang et al. 2002).

DNA extraction and sequencing

Large plankton were removed by filtering through a 20-μm membrane filter, and bacterioplankton were collected by filtering through a 0.22-μm-diameter filter. The genomic DNA of bacterioplankton was extracted from the stored filters using the E.Z.N.A.® Water DNA Kit (OMEGA, USA) according to the manufacturer's instructions. Electrophoresis and a NanoDrop ND 2000 (Thermo Scientific, USA) were used to examine the quantity of the extracted DNA. The V3–V4 region of the bacterial 16S rRNA gene was amplified using 338F and 806R with sample-identifying barcodes (Lee et al. 2012). For further network analyses, PCR and sequencing were performed seven times for each sample. High-throughput sequencing (Illumina MiSeq PE300 platform, Illumina, USA) was performed by Shanghai Majorbio Bio-Pharm Technology Co., Ltd. (Shanghai, China). The MiSeq sequencing data were processed using the QIIME Pipeline (Caporaso et al. 2010). The operational taxonomic units (OTUs) were identified (at the 97% similarity level) using Usearch. The OTU number of each sample was used to represent species richness. Rarefaction curves and Shannon–Wiener indices were generated, and the ACE, Shannon, and Chao1 estimators were calculated to compare the bacterial richness and diversity. Taxonomic classification at the phylum and genus levels was performed using the Ribosomal Database Project (RDP) algorithm to classify the representative sequences of each OTU. Permutational multivariate analysis of variance (Adonis) was processed in R software with the “vegan” package to investigate the significant differences in community structures between treatments. Linear discriminate analysis (LDA) effect size (LEfSe) (https://huttenhower.sph.harvard.edu/lefse/), which is a statistical tool designed to find biomarkers from metagenome data, was used to identify potential statistically significant taxa between the different treatments (Segata et al. 2011). The sequencing data that support the findings of this study are available in Figshare with the https://doi.org/10.6084/m9.figshare.9917981.

Network construction

Phylogenetic molecular ecological networks (pMENs), which were obtained by random matrix theory (RMT), offer a robust statistical means of analyzing networks because they provide solutions for common issues encountered during the use of high-throughput metagenomic data, including noise reduction and automatic network definition (Zhou et al. 2010, 2011; Deng et al. 2012). Network analysis was performed using the Molecular Ecological Network Analyses Pipeline (MENAP) (https://129.15.40.240/MENA/), which is freely available for use at the University of Oklahoma’s Institute for Environmental Genomics web server. For MENAP analysis, the relative abundance (RA) matrix was standardized and submitted for network construction. A cut-off value (similarity threshold, st) for the similarity matrix was automatically generated using default settings. Calculations of “global network properties”, “individual nodes' centrality”, and “module separation and modularity” were performed. The “output for Cytoscape visualization” procedure was performed in “greedy modularity optimization mode”. The data network was then exported for visualization using Cytoscape software. The “randomize the network structure and then calculate network” procedure was performed using the Maslov–Sneppen procedure to calculate the random network properties while maintaining the same number of nodes and links as in the empirical networks.

Statistical analyses

Detrended correspondence analysis (DCA) of the abundances of bacterioplankton in the community at the OTU level was performed using Canoco 4.5. The eigenvalues for the four axes were lower than 3.0, reaching a maximum of 1.141. Thus, we used redundancy analysis (RDA) in the linear model for analyses between bacterioplankton species and environmental factors. The physicochemical variables and bacterioplankton community diversity of water were compared by analysis of variance and Tukey’s test at the 5% significance level (P < 0.05) in SPSS V. 19.0 for Windows.

Results

Evaluation of the water quality and trophic status of the Danjiangkou Reservoir

The water quality monitoring results from May 2016 were generally good. All indicators met the Grade I water quality standards for the Environmental Quality Standards for Surface Water (GB38382-2002), with the exception of TN, TP and CODMn (Tables 1, S1). The TN contents in Kuxin, Qushou and Songgang exceeded 0.90 mg/L, which is the standard limit for Grade III surface water, and the TN contents at Taizishan and Heijizui were significantly higher than the levels at the other monitoring sites (P < 0.05), exceeding 1.00 mg/L, which is the standard limit for Grade IV surface water. The CODMn contents in the five typical ecological sites exceeded 3.00 mg/L, which is the standard limit for Grade II surface water.

The determination of the trophic status is a key part of ecological monitoring and evaluation of the Danjiangkou Reservoir (Shen et al. 2015). In May 2016, the trophic status of the Danjiangkou Reservoir area was assessed by measuring the TN, TP, CODMn, Chl a, and SD 5 parameters according to the improved Carlson’s trophic level index (TLI) (Wang et al. 2002). The detailed evaluation method and classifications are shown in the supplementary material and Table S2. The TLI of the Danjiangkou Reservoir area ranged from 40.10 to 43.29, reflecting a mesotrophic status, and the overall water quality of the Danjiangkou Reservoir was good (Table 1).

Bacterioplankton community composition

The high-throughput sequencing results showed that the average number of reads was 28,732, and the average number of operational taxonomic units (OTUs) was 350 for all samples collected from the five sites (Table 2). The bacterioplankton dilution curve is shown in Figure S1. With increasing amounts of sequencing data, the species richness increased at an early stage. Generally, the number of species leveled off when the number of reads exceeded 20,000. The estimation of Good’s coverage showed that all samples from the five sites had coverage above 99.63% (Table 2). The bacterioplankton community in the Danjiangkou Reservoir was evaluated using community richness indices (Chao1 and ACE), community diversity indices (Shannon index and Simpson index), and Good’s coverage. The samples from the five sites showed rich community compositions and high community diversity. The bacterioplankton community diversity at the five sites descended in the order Qushou > Songgang > Heijizui > Taizishan > Kuxin.

Table 2 Estimation of bacterioplankton community diversity (means ± SE)
Full size table

The assemblage characteristics of the bacterioplankton communities (at the OTU level) in the Danjiangkou Reservoir were analyzed by nonmetric multidimensional scaling (NMDS) ordination. The NMDS ordination of bacterioplankton had a stress value of 0.19, which could be represented by a two-dimensional scatter plot of the NMDS (Fig. 2). NMDS analysis can describe the community differences between different samples. We found that the Heijizui samples were distributed in the lower left, the Songgang samples were in the lower right, the Kuxin samples were in the lower center, the Taizishan samples were in the center, and the Qushou samples were in the upper center of the plot. These results indicated that the bacterioplankton communities differed to a certain degree across the different sites (Fig. 2). Clustering analysis was conducted based on the similarity of the bacterioplankton community compositions using the unweighted pair group method with arithmetic mean (UPGMA) based on unweighted UniFrac distances, and the results were similar to those obtained from the NMDS analysis (Fig. 3a).

Fig. 2
figure 2

NMDS results for bacterioplankton community diversity. The digital numbers represent seven biological replicates for each sample

Full size image
Fig. 3
figure 3

UPGMA tree (left) and the relative read abundance (right) of different bacterial community structures at the phylum level in the different treatments. The digital numbers represent seven biological replicates for each sample

Full size image

The differences in the overall community compositions of the bacterioplankton were examined using Adonis and Anosim tests, which showed that the bacterioplankton community compositions differed significantly across different sampling sites. Pairwise comparisons between different sites also revealed significant differences in the bacterioplankton community compositions (P < 0.05; Table S3). The high-throughput sequencing results showed that the bacterioplankton communities were composed mainly of 27 phyla and 336 genera of bacteria, including Actinobacteria, Proteobacteria, Bacteroidetes, Firmicutes, Verrucomicrobia, Cyanobacteria and Acidobacteria (Fig. 3b). There were 20 phyla and 230 genera in the Kuxin sample, 26 phyla and 296 genera in the Qushou sample, 23 phyla and 284 genera in the Songgang sample, 22 phyla and 243 genera in the Heijizui sample, and 22 phyla and 264 genera in the Taizishan sample (Fig. S2). The sum of reads belonging to Proteobacteria, Actinobacteria, Bacteroidetes, and Firmicutes accounted for 87.38–4.57% of the total reads; thus, these bacteria were regarded as the dominant populations (Fig. 3b).

To determine the differences in bacterial species richness among the different groups, we used the online statistical linear discriminant analysis effect size (LEfSe) tool to search for metagenomic biomarkers. Differential bacterial genera in different samples were calculated using the LEfSe method; information on the differences among all bacteria at the phylum, class, order, family, and genus levels are illustrated in a pie chart (Fig. 4). At the phylum level, significant differences were found in Firmicutes in the Kuxin sample; Bacteroidetes, Gemmatimonadetes, and Gracilibacteria in the Qushou sample; Actinobacteria and Cyanobacteria in the Songgang sample; Acidobacteria, Chlorobi, Planctomycetes, and Verrucomicrobia in the Heijizui sample; and Proteobacteria and Tenericutes in the Taizishan sample. Information on differential bacterial species in other taxa is shown in Fig. 4 and the Supplementary Dataset 1. These differential bacterial species belonged mainly to Proteobacteria, Actinobacteria, Bacteroidetes, Firmicutes, and verrucomicrobia.

Fig. 4
figure 4

Identification of the most differentially abundant taxa between different stations by LEfSe

Full size image

Relationship between the Bacterioplankton Community and environmental variables

The RDA results shown in Fig. 5 indicate that the first and second ordination axes explained 28.52% and 10.92% of the total variance, respectively. The physicochemical properties that were highly correlated with the first ordination axis were TN, COD, CODMn, NH4–N, SD, and Chl a; the association was significant (P < 0.05) or extremely significant (P < 0.01) for TN (R = − 0.7522, P = 0.001), COD (R = 0.8254, P = 0.035), and CODMn (R = 0.7302, P = 0.035). The physicochemical properties that were highly correlated with the second ordination axis were pH, TN, and CODMn; the association was significant (P < 0.05) for pH (R = − 0.9851, P = 0.027). The analysis results indicated that TN, pH, COD, and CODMn were significantly associated with the bacterioplankton community structure and were important factors affecting the bacterioplankton distribution.

Fig. 5
figure 5

RDA ordination biplot between bacterioplankton species and environmental factors

Full size image

To further elucidate the correlation between bacterioplankton and environmental factors, we performed Spearman’s correlation analysis on the most differentially abundant bacterial taxa identified in the LEfSe analysis and the physicochemical indicators of water quality. The results showed that among the physicochemical indicators of water quality, TN, pH, COD, and CODMn were the primary components that were significantly correlated with the assayed bacterial taxa (Supplementary Dataset 2). Among the 99 identified bacterial genera, 35 were significantly correlated with TN, of which Brevundimonas (R = 0.703, P = 0.003), Corynebacterium (R = 0.657, P = 0.008), and Pseudarcicella (R = 0.552, P = 0.033) of the phyla Proteobacteria, Actinobacteria and Bacteroidetes, respectively, and four other genera were positively correlated with TN, whereas the remaining genera were all negatively correlated with TN. The genera Tabrizicola, Sandarakinorhabdus, and Pseudospirillum of the phylum Proteobacteria and Candidatus_Planktoluna and Gemmatimonas of the phyla Actinobacteria and Gemmatimonadetes, respectively, showed highly significant negative correlations with TN. The effects of COD and CODMn on bacterioplankton communities exhibited essentially identical trends. COD was significantly correlated with 19 bacterial genera, while CODMn was significantly correlated with 28 bacterial genera. Highly significant positive correlations were observed between COD and the genera Tabrizicola, Sulfuritalea, Aeromonas, Paracocccus and Hirschia of the phylum Proteobacteria and between CODMn and the genus Lacibacter of the phylum Bacteroidetes. In contrast, highly significant negative correlations were observed between COD and the genera Canteidatus_Cryptoprodotis and BD1-7_clade of the phylum Proteobacteria and between CODMn and the genera Enterococcus and Oceanobacillus of the phylum Firmicutes. The pH was significantly correlated with 19 bacterial genera, with highly significant positive correlations observed for Silanimonas, Acidibacter, and Simplicispira of the phylum Proteobacteria; Gemmatimonas of the phylum Gemmatimonadetes; and Owenweeksia of the phylum Bacteroidetes. These results were similar to the RDA results at the OTU level because TN, pH, COD, and CODMn were important factors affecting the bacterioplankton community composition.

Bacterioplankton Community interactions

Five molecular networks were constructed using the RMT method with the RAs of bacterioplankton at the five sites in the Danjiangkou Reservoir area. The visualized network structure diagram shows the role of bacterioplankton at different sites in the entire network and their interactions with other bacteria (Fig. 6). In the visualized network structure diagram, each node represents an OTU, and a red or blue line between nodes represents a positive or negative correlation between OTUs (Figs. 6 and S3). All five networks had similarity thresholds > 0.90, which were higher than the thresholds of most networks constructed using this method; thus, these networks could be used for the subsequent analysis of interspecies interactions between microorganisms (Deng et al. 2012) (Table 3). Generally, a complex system has important features, such as scale-free, small-world, modular, and hierarchical networks (Deng et al. 2012). In the present study, the bacterioplankton networks of the five sites were composed of 167–188 nodes. The R2 of the power law ranged from 0.723 to 0.757, and the connectivity conformed to the power-law model, indicating the scale-free nature of the networks (Table 3). The geodesic distance ranged between 5.990 and 7.751, which was similar to that of a small-world network summarized by Brown et al. (2004). In addition, the average clustering coefficients of the experimental networks (0.255–0.387) were markedly higher than the coefficients of the corresponding random networks (0.008–0.033), indicating that the bacterioplankton networks of the five sites were small-world networks. The modularity was between 0.675 and 0.789, which was similar to the values of modular networks described by Deng et al. (2012). Thus, the networks constructed in the present study were also modular networks.

Fig. 6
figure 6

Overview of the bacterioplankton networks at the five sampling stations. Each node represents an OTU. The colored circles indicate the OTUs affiliated with particular phyla (color code on the right). The sizes of the nodes represent the relative abundances of the OTUs. A blue edge indicates a negative interaction between two individual nodes, whereas a red edge indicates a positive interaction. The digital number indicates modules with more than five OTUs for each sample

Full size image
Table 3 Topological properties of the co-occurrence networks of the bacterioplankton communities in the Danjiangkou Reservoir and their corresponding random networks
Full size table

The topological role of the nodes can be represented by the connectivity of node i within a module (Zi) and the connectivity of node i among modules (Pi). Zi describes the degree of connection of a node to other nodes in the module, whereas Pi reflects the degree of connection of a node to different modules. Network nodes are generally classified into four groups as follows: peripheral nodes (Zi ≤ 2.5, Pi ≤ 0.62), connectors (Zi ≤ 2.5, Pi > 0.62), module hubs (Zi > 2.5, Pi ≤ 0.62), and network hubs (Zi > 2.5, Pi > 0.62) (Deng et al. 2012) (Fig. 7). Peripheral nodes have few connections and are always connected to other nodes within the same module. Module hubs are nodes that are highly connected within modules and are rarely connected to other modules. Connectors have a very low degree of connection within their module but are highly connected to several other modules. Network hubs have a dual role as modular hubs and connectors. By analyzing the topological role of network nodes at different sites, the vast majority of the nodes in the bacterioplankton networks were identified as peripheral nodes, whereas some nodes were identified as module hubs and connectors (Fig. 7). The Kuxin sample had three nodes identified as connectors, the Qushou and Songgang samples had one and two nodes identified as module hubs, respectively, and the Taizishan sample had three nodes identified as module hubs and two nodes as connectors. The Heijizui sample had five nodes identified as connectors and two nodes as module hubs. The above results suggest that Zi and Pi were higher in the bacterioplankton network structures of Taizishan and Heijizui than in those of the other samples.

Fig. 7
figure 7

Zi-Pi plot showing the distribution of OTUs based on their topological roles

Full size image

In the topological structure of the molecular network, the topological roles of different nodes can be used to identify key microorganisms. Herein, Zi = 2.5 and Pi = 0.62 were taken as the thresholds according to Deng et al. (2012). We defined all nodes with Zi ≥ 2.5 or Pi ≥ 0.62 as key bacteria that played an important role in connecting the bacteria within and among the large modules. The key bacteria were hgcI_clade (OTU233) and Bacillus (OTU716) in the Kuxin sample; uncultured Verrucomicrobiaceae (OTU540) in the Qushou sample; Psychrobacter (OTU525), alphaI_cluster, and norank (OTU733) in the Songgang sample; Sandarakinorhabdus (OTU573) and Sphingorhabdus (OTU622) in the Taizishan sample; and hgcI_clade (OTU233), Flavobacterium (OTU580), Ralstonia (OTU601), CL500-29_marine_group (OTU724), and uncultured Planctomycetaceae (OTU94) in the Heijizui sample. These bacteria were affiliated with Proteobacteria, Actinobacteria, Firmicutes, Cyanobacteria, Verrucomicrobia, and Tenericutes (Fig. 7 and Table S4). Among them, the key bacterial species of Sandarakinorhabdus, Bacillus, Flavobacterium, CL500-29_marine_group, Sphingorhabdus, and hgcI_clade were among the most differentially abundant taxa in the LEfSe analysis results.

Relationships between modules and environmental variables

A network is divided into numerous modules, each of which is often treated as a functional unit in biological systems (Deng et al. 2012). The molecular network analysis of bacterioplankton in the Danjiangkou Reservoir area showed that the bacterioplankton network structure of Heijizui and Taizishan had modularities of 0.789 and 0.756, respectively, which were higher than those of the other samples; the lowest modularity was found in the Songgang sample, at 0.675 (Table 3). The Heijizui sample had 27 modules in the molecular bacterioplankton network, which was higher than the numbers of nodules in the other samples (15–17). There were nine modules with more than five OTUs in the Heijizui, Kuxin, and Songgang samples and seven modules with more than five OTUs in the Qushou and Taizishan samples (Fig. 6). To analyze the effects of the physicochemical properties of water on the modules in the bacterioplankton network, Module-EigenGene analyses were performed on modules with more than five OTUs to evaluate the correlations between the physicochemical properties of water and the network modules. For the Qushou sample, SD (R = − 0.81, P = 0.03) and COD (R = − 0.86, P = 0.01) were significantly negatively correlated with modules Q4 and Q6 in the molecular network, respectively. In the Heijizui sample, TN (R = − 0.80, P = 0.03) was significantly negatively correlated with module H4, whereas NH4–N (R = 0.77, P = 0.04) was significantly positively correlated with module H4 in the molecular network. In the Kuxin sample, a highly significant negative correlation was found between DO (R = − 0.9, P = 0.005) and module K1 in the molecular network. In the Songgang sample, TN (R = − 0.85, P = 0.01) was significantly positively correlated with module S1 in the molecular network; in contrast, pH had a significant negative correlation with modules S1 (R = − 0.79, P = 0.04) and S2 (R = − 0.77, P = 0.04), and SD also had a significant negative correlation with modules S2 (R = − 0.81, P = 0.03) and S2 (R = − 0.80, P = 0.03).

Discussion

The water transfer system for the MR-SNWDP was started in 2014 in China. As a result, the normal water storage level of the Danjiangkou Reservoir, which is a water source area for the MR-SNWDP, was raised from 157 to 170 m; thus, the aquatic ecosystem of the Danjiangkou Reservoir underwent restructuring. Given the important role of bacterioplankton in the aquatic ecosystem, long-term monitoring of the bacterioplankton community composition in the Danjiangkou Reservoir should be conducted; however, related work is rare (Gao et al. 2018). In the present study, Illumina MiSeq sequencing was used to study the bacterioplankton community composition at five sites in the Danjiangkou Reservoir area in May 2016. We found that the bacterioplankton community consisted mainly of 27 phyla, including Actinobacteria, Proteobacteria, Bacteroidetes, Firmicutes, Verrucomicrobia, and Cyanobacteria, and 336 genera, including hgcI_clade, Fluviicola, CL500-29_marine_group, Limnohabitans, Comamonadaceae unclassified, LD12_freshwater_group_norank, Sediminibacterium, and Roseomonas. The bacterioplankton community composition in the Danjiangkou Reservoir was similar to the typical bacterial community composition in previously studied lake waters (Haukka et al. 2006; Zhao et al. 2016; Li et al. 2017; Xu et al. 2018).

Studies have shown that bacterioplankton communities in water are affected by complex abiotic and biological processes (Logue et al. 2012; Li et al. 2017; Peter et al. 2017). The long-term monitoring program based on our results and those of other researchers showed that the water quality of the Danjiangkou Reservoir was generally good (Li et al. 2009; Chen et al. 2015, 2018; Gao et al. 2018; Pan et al. 2018). However, agricultural nonpoint source pollution, industrial wastewater, and domestic sewage from villages and towns were present, resulting in TN values that notably exceeded the standard limit (Li et al. 2009). The water quality monitoring results of this experiment showed that the TN and CODMn contents were high, and the TN contents at the Taizishan and Heijizui sites were significantly higher than those at the other sites (Table 1). These two sites are where two important tributaries, the Han River and the Danjiang River, merge into the Danjiangkou Reservoir (Fig. 1). The excessive levels of nutrient salts in the tributary water bodies enter the reservoir water body at these two sites, causing TN levels that are higher than those at other sites. The high concentration of nitrogen nutrient salts may affect the composition and diversity of bacterial communities in the Danjiangkou Reservoir, and the analysis of the composition of bacterioplankton communities confirmed this concept. Cluster analysis based on the UPGMA method showed that the community compositions at the Taizishan and Heijizui sites were rather different from those of the other samples (Fig. 3a); analyses based on the diversity index also indicated that the diversity of bacterial communities in the Taizishan and Heijizui samples was lower than that in the Qushou and Songgang samples. Spearman’s correlation analysis between the different abundances of planktonic bacterial taxa at the different sites and the physicochemical indicators of water quality indicated that 35 bacterial genera were significantly correlated with TN, among which Corynebacterium, Pseudarcicella, Staphylococcus, and uncultured Verrucomicrobiaceae in the Heijizui samples showed a significantly positive correlation with TN, as did Brevundimonas and Candidatus_Rhodoluna in the Taizishan samples. These bacteria play a role in nitrogen fixation (Sebastian et al. 1987) and denitrification (Zhou et al. 2019). The bacteria observed at the other sites mostly showed a significantly negative correlation with TN. These results indicate that the TN input in the upper reaches of the Danjiangkou Reservoir can affect the community composition of bacterioplankton as well as that of microbes involved in the nitrogen cycle, which may play an important role in nitrogen cycling in water bodies. In this study, Spearman correlation analysis between TN, COD, and CODMn showed that TN was significantly negatively correlated with COD (R = − 0.528, P = 0.023) and CODMn (R = − 0.663, P = 0.007). COD and CODMn were also important factors affecting the bacterioplankton community compositions in the Danjiangkou Reservoir.

Microorganisms co-occur and interact through complex network structures in the ecosystem, and therefore, understanding these interactions is a core topic of microbial ecology (Coyte et al. 2015). Network analysis has shown promising effects in studying the interactions in biological communities (Faust and Raes 2012). However, few studies have used this method to investigate the interactions of bacterioplankton, identify key populations, and assess their indicative roles in water quality (Karimi et al. 2017; Xu et al. 2018), Zhao et al. (2016) constructed bacterioplankton networks for six eutrophic lakes in Nanjing, China, and analyzed the interspecies interactions between Cyanobacteria and other bacteria; the results improved our understanding of the ecological functions of Cyanobacteria in eutrophic lakes. Xu et al. (2018) also analyzed the interspecies interactions between free-living (FL) and particle-attached (PA) bacterioplankton in Taihu Lake using network analysis. Herein, we constructed molecular networks of bacterioplankton in the Danjiangkou Reservoir area. The molecular network analysis also showed that the bacterioplankton network structures differed in the Danjiangkou Reservoir area. The bacterioplankton network structures of the Taizishan and Heijizui samples, which had higher TN contents than the samples from the other sites, had higher Zi and Pi values than the structures of the other samples (Fig. 7 and Table S4). In the topological structure of the molecular network, all nodes with Zi ≥ 2.5 or Pi ≥ 0.62 were defined as key bacteria. Thus, the number of key bacteria in the Taizishan and Heijizui samples was also higher than the numbers in the other samples. These key bacteria were found in Proteobacteria, Actinobacteria, Firmicutes, Cyanobacteria, Verrucomicrobia, and Tenericutes (Fig. 7 and Table S4). In the molecular network analysis, modules are often considered functional units in the biological system (Guimerà and Nunes Amaral 2005). The correlations between the physicochemical properties of water and the network modules can be further analyzed using Module-EigenGene analyses to reveal the effects of the physicochemical properties of water on the functional units of the bacterial community in the molecular network (Deng et al. 2012). By analyzing the modules in networks constructed for FL and PA bacterioplankton and the environmental variables of water, Xu et al. (2018) found that the environmental variables had varying effects on the FL and PA modules and therefore speculated that the FL and PA networks had different preferences for low and high concentrations of nutrients (e.g., nitrogen and phosphorus). In the present study, we selected modules with more than five OTUs in the molecular bacterioplankton networks in the Danjiangkou Reservoir area and analyzed the correlations between the physicochemical properties of water and the network modules using Module-EigenGene analyses. Our results showed that TN, NH4–N, pH and COD were significantly correlated with the network modules. These results indicate that TN, pH, and COD affect the bacterioplankton community composition and might influence the functional units of the bacterial communities in the molecular bacterioplankton networks of the Danjiangkou Reservoir. In this study, molecular network analysis was used to investigate the links between numerous taxa; however, these links often have difficulty providing any evidence of such interactions using biochemical or other standard microbiological tests referring to living microbes (Faust and Raes 2012). Thus, follow-up experimental validation is required to confirm the true bacterial interactions in the Danjiangkou Reservoir (Wang et al. 2017).

Conclusion

The water quality monitoring results in May 2016 showed that the water quality of the Danjiangkou Reservoir was generally good, but the TN values exceeded the standard limit. Illumina MiSeq sequencing found that the bacterioplankton community consisted mainly of 27 phyla and 336 genera, which was similar to the typical bacterial community compositions in previously studied lake waters. The molecular network analysis of bacterioplankton in the Danjiangkou Reservoir area revealed that the number of key bacteria was higher at the Taizishan and Heijizui sites, which were associated with high TN contents. The combination of high-throughput sequencing and network analysis showed that TN, pH, COD, and CODMn were important factors affecting the bacterioplankton community composition, and TN, COD and pH also might affect the functional units of the bacterial communities. This study explored whether TN might directly or indirectly affect the community structure and interspecies interactions of the bacterioplankton community in the Danjiangkou Reservoir and provide a reference for the protection of the water environment in this reservoir.