Background

Rivers play an important part in coupling biogeochemical cycles between continents and oceans [1]. River water flow acts as a carrier of biotic and abiotic substances, whereas sediment serves as a sink or source in the cycling of nutrients. Previous studies on riverine biodiversity mostly focused on invertebrates or fish, and very limited reports were related to bacterial composition along river networks [2,3,4,5]. Bacteria hold key roles in microbial communities and contribute significantly to biogeochemical processes and the cycling of nutrients in river ecosystems [6,7,8]. Recent studies have shown that the biogeography patterns of bacterial communities in complex freshwater networks can be explained by their origins in upstream freshwater [9] and terrestrial sources [10]. Planktonic bacterial taxa arise as the sum of multiple upstream sources of bacteria that grow in rainfall, lakes, groundwater, and soil. The resulting planktonic bacterial community is vulnerable to changes in its composition and structure. Sedimentary bacterial taxa develop from long-term cumulative processes of sediment erosion and deposition under ambient conditions [11]. The spatiotemporal distribution of planktonic and sedimentary bacterial communities in rivers could be quite different. Moreover, bacterial diversity is significantly altered by varying fluvial landforms and severe human interference along a large river. A better understanding of bacterial responses to the changing environment of river ecosystems is useful in the context of riverine cycles of nutrients, e.g., carbon and nitrogen, which are highly relevant to emission or sequestration of greenhouse gases [12,13,14].

Until recently, riverine bacterial communities have proved highly diverse and variable. Spatial and temporal variability of bacterioplankton composition in rivers has been demonstrated [2,3,4,5, 4: Figure S3) and co-occurrence of season-associated OTUs within the same phylum in network analysis (Additional file 5: Figure S4, Fig. 1b, d). In each phylum, positive association (red lines) dominates negative association (blue lines), indicating the uniform seasonality response of season-associated OTUs within a phylum. This uniform response is quantified by the network density value (d) (“Season-associated taxa analysis” section). Phyla that respond consistently are identified by having a higher density value implying a denser distribution of the association among OTUs. For example, Verrucomicrobia and Spirochaetes are the two most tightly correlated clusters (d = 0.433 and 0.455, respectively), whereas Acidobacteria in sediment samples and Actinobacteria and Bacteroidetes in water samples are more dispersed. The co-occurrence network analysis reveals almost exclusively or overwhelmingly positive correlations, consistent with the general pattern of dominant season-associated OTUs in most phyla (Fig. 1a, c).

In summary, the results show that microbial communities in Yangtze River are particularly sensitive to the season and are more prevalent in the autumn.

Biogeography patterns of bacterial communities

Non-metric multidimensional scaling (NMDS) was applied using unweighted UniFrac distance to identify the community compositions of all samples (Fig. 2). The first axis revealed that the bacterial communities of water samples were different from the corresponding sediment samples regardless of sampling sites and season. Bacterial communities of water samples demonstrated clear seasonal groups. However, bacterial communities of sediment samples did not form two separated clusters by season. The consistency of the results was confirmed by using the analysis of similarity (ANOSIM) statistic test of pairwise Bray-Curtis dissimilarities (Additional file 6: Figure S5). No significant difference was found between spring and autumn (global r = 0.122, P = 0.001) for sedimentary bacterial communities, whereas two seasonally distinct groups (global r = 0.525, P = 0.001) were observed for the water samples.

Fig. 2
figure 2

Non-metric multidimensional scaling diagram showing bacterial composition differences obtained among the 280 sampling sites

Full size image

Overall, the spatial similarity of bacterial communities is better described by mean dendritic distance rather than cumulative dendritic distance or site catchment area (Additional file 7: Table S2). Closer correlation was demonstrated between the bacterial community similarity matrix and mean dendritic distance (Fig. 3b (mantel r = 0.3324, P = 0.001), d (mantel r = 0.3480, P = 0.001)) in terms of sediment samples than those based on water samples (Fig. 3a (mantel r = 0.2843, P = 0.001), c (mantel r = 0.2415, P = 0.001)). From the results, the mean dendritic distance appears more appropriate to describe bacterial community similarity in autumn sediment samples (Fig. 3d) and spring water samples (Fig. 3a). The distance decay analysis suggests that geographical distance could be of importance in structuring the bacterial assembly and determining the spatial similarity between different sites along the Yangtze River.

Fig. 3
figure 3

Relationship between mean dendritic distance and Bray-Curtis similarity of bacterial communities in a spring-water, b spring-sediment, c autumn-water, and d autumn-sediment samples. Mantel Spearman correlations (r) and probabilities (P) are stated

Full size image

The Yangtze River flows through various landform types, including plateau, mountain, basin, foothill, and plain. Mainstream samples were used to study the effects of spatial variation, such as the river continuum and landform on bacterial communities. For water bacterial communities, NMDS (Additional file 8: Figure S6 (a) and (b)) gave similar results in both spring and autumn, with five separate groups: group 1 (between stations 1 and 2), group 2 (between stations 3 and 5), group 3 (from stations 6 to 9), group 4 (from stations 10 to 13), and group 5 (from stations 14 to 24), corresponding to the local landforms: mountain, foothill, basin, foothill-mountain, and plain. Moving window analysis (Additional file 9: Figure S7) was used to characterize the change rate of bacterial communities along the mainstream based on a comparison of results between two consecutive sampling sites. Higher change rates were always found at sampling sites where the landform type changed. Meanwhile, ANOSIM analysis (Additional file 10: Figure S8) further illustrated that taxonomic compositions of microbial communities significantly varied by landform type (P = 0.001). For the sediment samples, a similar clustering result (Additional file 8: Figure S6 (c) and (d)) was also obtained; this indicated that bacterial communities from the same landform tended to be similar to each other. The results of these analyses revealed that spatial variation in bacterial compositions across the samples could be partially attributed to the landform.

To further investigate the taxonomic distribution and differentially dominant clades of diverse landform ecosystems in water and sediment, we used the LEfSe biomarker discovery suite [31] to compare the abundance of bacterial compositions at each taxonomic level and determine taxa differentially abundant in at least one landform. Figure 4 depicts cladograms that visualize all detected bacterial compositions (relative abundance > 0.5%) from domain to genus level. Thirty and 70 differentially abundant taxa (i.e., colored circles in Fig. 4) were detected in water and sediment, respectively. These significantly enriched taxa provide a good indication of the primary characteristics of bacterial community structures in the Yangtze River, corresponding to the five landform types.

Fig. 4
figure 4

LEfSe cladogram of microbial community obtained for five landform types in water (a) and sediment (b). All detected taxa, with relative abundance ≥ 0.5% in at least one sample, assigned to domain (innermost), phylum, class, order, family, and genus (outermost), are used to determine the taxa or clades most likely to explain differences between landform types. Differentially abundant taxa (biomarkers) are colored according to their most abundant landform habitats; red, green, orange, purple, and blue circles stand for taxa that are abundant in plains, mountains, foothill-mountains, basins, and foothills, respectively. The color intensity of the outmost ring is proportional to the taxa abundance (genus level) at the landform type of greatest prevalence

Full size image

Influential factors on bacterial community compositions

Even though canonical correspondence analysis (CCA) of bacterial communities in water and sediment indicated weak correlation to environmental factors (Additional file 11: Table S3, and Additional file 12: Figure S9), water temperature was found to be the primary factor in structuring bacterial community assemblages in both water and sediment of the Yangtze River. In addition, DO (dissolved oxygen) influenced the bacterial community in water, and pH, Mdd (mean dendritic distance), and TN (total nitrogen) influenced the bacterial community in sediment.

Impacts of the large dams

The Yangtze River contains a cascade of large dams, including two of the world’s largest dams, the Three Gorges Dam and the ** prokaryote community structure. Environ Microbiol. 2006;8:732–40." href="/article/10.1186/s40168-017-0388-x#ref-CR35" id="ref-link-section-d167081282e1150">35] neutral model was used to interpret the biogeographic distribution of bacterial communities in both water and sediment of the Yangtze River. The results showed that the neutral interpretation gave an excellent fit to the bacterial community distribution in the large riverine system considered (R2 > 0.7756), and an even higher correlation than previously achieved for smaller systems, such as coastal lakes in Antarctic (R2 ≤ 0.50) [36], and zebrafish (R2 ∈ [0.39, 0.81]) [37] (Fig. 6). Furthermore, the estimated immigration rate (m) originating from sediment communities (spring sediment 0.1640, autumn sediment 0.1463) was much lower than from water communities (spring water 0.1997, autumn water 0.1814), suggesting there were much more serious dispersal limits experienced by sediment communities. In general, the Sloan et al.’s [35] neutral model predicts the occurrence frequency to be > 85.44% in water communities, whereas only 66.08 and 61.96% taxa could be described in spring sediment and autumn sediment samples, respectively (Fig. 6, Additional file 14: Table S4).

Fig. 6
figure 6

Fit the occurrence frequency of different OTUs as a function of mean relative abundance using Sloan et al.’s [35] neutral model, for a spring-water, b spring-sediment, c autumn-water, and d autumn-sediment communities. Orange and green dots indicate the OTUs that occur more and less frequently than given by the model. Dashed lines represent 95% confidence intervals around the model prediction (red line)

Full size image

The neutral model did not interpret 100% of the community species distribution, indicating that other community assembly mechanisms were perhaps operating at the same time. There is some evidence of species sorting in the present results. For example, the LEfSe analysis indicated that different abundant species or clades (biomarkers) occurred in five landform types (Fig. 4). Canonical correspondence analysis (CCA) of the bacterial communities in water and sediment showed a weak dependence on environmental factors. A significant discrepancy in the abundance of OTUs (Fig. 5) was observed upstream and downstream of the Three Gorges Dam, with most OTUs having higher abundance upstream of the dam except for several OTUs belonging to the genera Anaerolinea and Flavobacterium (Fig. 5). This implies that each sampling site possesses specific species due to its particular environment. The non-random distributions of bacteria are ascribed to the heterogeneous environment that affects their natural habitats and nutrients, influencing selection.

Discussion

A rapidly increasing number of studies based on high-throughput sequencing technologies have revealed a tremendous diversity of bacterial communities residing in the aquatic environment [38,39,40]. Most previous investigations on the variability and diversity of bacterial communities in rivers focused on a single dimension, i.e., either a long-term time series or a small-scale spatial dimension across environmental gradients [4, 19]. Here, we describe spatiotemporal patterns of lotic bacterial communities over a 4300 km river continuum for both water and sediment during spring and autumn seasons. Although changing slightly across the seasons, bacterial communities in sediment provide the main contribution to the bacterial diversity of the Yangtze River, and only 1.2% of the total OTUs are unique to water samples. The bacterial population fluctuation in water samples is higher than in sediment samples, as expected. Few previous studies have compared planktonic and sedimentary bacterial diversity in river reaches and coastal areas [35] neutral model was fitting to describe the relationship between the observed occurrence frequency of OTUs (the proportion of local communities in which each OTU is detected) and their abundance (the mean relative abundance across all local communities) [66]. The model is an adaptation of Hubbell’s neutral community model adjusted to bacterial populations analyzed with molecular tools [24]. This model emphasizes the effects of stochastic dispersal and drift (birth-death immigration process) but ignores the ecological difference between species and their response to the surrounding environment. In this model, the random loss of an individual is immediately replaced by immigration from the meta-community, with probability m, or reproduction within the local community, with probability 1-m [35]. The immigration rate was determined using non-linear least squares fitting in minpack.lm package of R-3.2. To further assess the deviations from the neutral model fitting, OTUs were subsequently sorted into three partitions depending on whether they occurred more frequently than “above” partition, less frequently than “below” partition, or within “neutral” partition the 95% confidence interval of the neutral model predictions.

Season-associated taxa analysis

OTUs with occurrence in more than 30% of all sediment or water samples were defined as persistent bacterial OTUs. An occupancy criterion was employed in order to generate the overall trend for taxonomic dendrograms (Fig. 1a, c) in water and sediment sample. Persistent bacterial OTUs that differed significantly between spring and autumn (P < 0.05) were further characterized as season-associated OTUs. According to their abundance in two seasons, season-associated OTUs were classified as autumn-associated OTUs (with significantly higher abundance in autumn) or spring-associated OTUs (with significantly higher abundance in spring). The network density value (d) was determined as the number of significant co-correlations divided by the number of all nodes, that is, a higher value represents a more intensive or dense response. Only persistent bacterial OTUs were displayed in taxonomic dendrograms (Fig. 1a, c) to avoid unstable associations and inconsistent trend caused by transient OTUs. To obtain comprehensive correlation among season-associated OTUs and non-associated OTUs, association networks were applied to both persistent and transient OTUs for each dominant phylum using edge-weighted spring-embedded layout algorithm (Fig. 1b, d).