Geographic Patterns of Bacterioplankton among Lakes of the Middle and Lower Reaches of the Yangtze River Basin, China

The middle and lower Yangtze Plain is a typical floodplain in which many lakes connect with each other, especially in the wet season. More importantly, with the frequent change of regional water level in the wet season, there is a mutual hydrodynamic exchange among these lakes. The microbial biogeography among these interconnected lakes is still poorly understood. This study aims to unravel the mechanisms underlying the assembly process of abundant and rare bacteria among the interconnected lakes in the middle and lower Yangtze Plain. Our findings will provide a deeper understanding of the biogeographic patterns of rare and abundant bacterial taxa and their determined processes among interconnected aquatic habitats.

ABSTRACT

The revolution of molecular techniques has revealed that the composition of natural bacterial communities normally includes a few abundant taxa and many rare taxa. Unraveling the mechanisms underlying the spatial assembly process of both abundant and rare bacterial taxa has become a central goal in microbial ecology. Here, we used high-throughput sequencing to explore geographic patterns and the relative importance of ecological processes in the assembly of abundant and rare bacterial subcommunities from 25 lakes across the middle and lower reaches of Yangtze River basin (MLYB), located in Southeast China, where most of the lakes are interconnected by river networks. We found similar biogeographic patterns of abundant and rare subcommunities which could significantly distinguish the community compositions of the two lake groups that were far from each other but which could not distinguish the community compositions of the nearby lakes. Both abundant and rare bacteria followed a strong distance-decay relationship. These findings suggest that the interconnectivity between lakes homogenizes the bacterial communities in local areas, and the abundant and rare taxa therein may be affected by the same ecological process. In addition, based on the measured environmental variables, the deterministic processes explain a small fraction of variation within both abundant and rare subcommunities, while both neutral and null models revealed a high stochasticity ratio for the spatial distribution patterns of both abundant and rare taxa. These findings indicate that the stochastic processes exhibited a greater influence on both abundant and rare bacterial subcommunity assemblies among interconnected lakes.

IMPORTANCE The middle and lower Yangtze Plain is a typical floodplain in which many lakes connect with each other, especially in the wet season. More importantly, with the frequent change of regional water level in the wet season, there is a mutual hydrodynamic exchange among these lakes. The microbial biogeography among these interconnected lakes is still poorly understood. This study aims to unravel the mechanisms underlying the assembly process of abundant and rare bacteria among the interconnected lakes in the middle and lower Yangtze Plain. Our findings will provide a deeper understanding of the biogeographic patterns of rare and abundant bacterial taxa and their determined processes among interconnected aquatic habitats.

INTRODUCTION

In aquatic systems, bacteria are the major drivers of nutrient regeneration and energy flow (1, 2), and understanding their spatial diversity and biogeography is critical for understanding their relationship with ecosystem functioning (3–5). A typical phenomenon of bacterial community compositions (BCCs) in natural habitats is that a few species have numerous individuals, while numerous species have relatively few individuals; the former are often called abundant bacterial taxa, and the latter are often called rare taxa (6). Abundant taxa are traditionally considered to be the core groups that perform major ecological functions, and these taxa were therefore intensively studied (7, 8). However, more recent studies have suggested that rare bacterial taxa also play a key role in global biogeochemical cycles and ecosystem functions (6, 9–12). Therefore, the biogeography of both the abundant and rare bacteria needs to be explored rigorously (3, 13). However, it has remained unclear whether rare bacterial taxa share similar biogeography with abundant ones between inland interconnected lakes.

A central goal in microbial biogeography research is to reveal the mechanisms underlying microbial geographic patterns. In general, microbial geographic distribution patterns are determined by two ecological processes. One is deterministic (environmental and selection related), and the other is stochastic (dispersal related) (3, 14, 15). In aquatic ecosystems, deterministic variables, including water environmental properties (e.g., temperature, salinity, pH) (1, 16, 17), aquatic morphology (e.g., elevation gradient) (18), and watershed land use (8), are regarded as important deterministic factors affecting the microbial structure. Meanwhile, a recent study suggested that the microbial biogeography in some open-water areas (e.g., subtropical bays of China) (19) is determined by stochastic processes. Even though the two ecological processes have been explored in many studies of microbial communities, until now, only a few studies compared the effects of both deterministic and stochastic processes on abundant and rare bacteria (8, 12, 13, 20), especially among inland interconnected lakes.

Inland lakes are often separated by land, and even when they are connected by rivers, there are hydrodynamic gradients. The environmental conditions and diversity of microbial communities in these habitats often have significant spatial heterogeneity (13, 17, 21–24). Abundant and rare bacteria differ significantly in their biodiversities (6) and may have discrepant ecological niches and different ecological responses to environmental changes (6, 13, 20). Some recent studies suggest that the geographic patterns of both abundant and rare bacterial taxa are governed by local environmental variables across some isolated lakes (8, 17), and abundant and rare bacterial taxa between some disconnected lakes have contrasting patterns (17). Notably, the result of a recent simulation experiment indicated that local diffusion in interconnected habitats (e.g., isotropic grid landscapes) homogenizes local species richness (25). In this scenario, the spatial relationship between abundant and rare bacterial taxa between interconnected habitats may be different from that between isolated habitats. In addition, unlike the isolated lakes, the spatial heterogeneity of environmental factors will be reduced due to hydraulic connections in open water (26, 27), and the assemblies of bacteria may be dominated by stochastic processes (27). These interesting findings are probably related to the connectivity between habitats, but direct evidence is still lacking. The geographic patterns and the relative importance of ecological processes in the assembly of abundant and rare bacterial subcommunities among interconnected habitats, especially in inland interconnected waters, need to be further explored.

An intriguing area for study is in the east of China, namely, the middle and lower reaches of the Yangtze River basin (MLYB), which is low and flat and covered by hundreds of shallow lakes and intricate river networks (28). This lake-river ecosystem stretches for more than 1,000 kilometers, and some of the nearby lakes are connected by river networks, especially in the wet season. More importantly, with the frequent change of regional water level in the wet season, there is a mutual hydrodynamic exchange among these lakes. In the present study, we first hypothesize that the interconnectivity between lakes homogenizes the bacterial communities in local areas and creates abundant and rare shared similar geographic patterns. Our second hypothesis is that the stochastic processes explain more of the differences in bacterial assembly along interconnected lakes in MLYB than does environmental selection.

RESULTS

Bacterial abundance and multivariate abundance cutoff.

In total, 720,326 high-quality sequences were recovered, with a mean of 28,813 ± 4,218 per sample, which clustered into 9,847 operational taxonomic units (OTUs). The definition of abundant or rare OTUs combined their local and regional relative abundance (see Materials and Methods). At the regional and local levels, 468,822 sequences (65.1%) representing 81 (0.8%) OTUs were classified as abundant, while 21,083 (2.9%) sequences, representing 6,788 (68.9%) OTUs were classified as rare (see Table S1 in the supplemental material).

Among all of the lake samples, no single OTU was abundant (>1%); only 2 OTUs (containing 97,781 sequences) with >1% abundance were present in >70% of the samples, and 49 OTUs (containing 26,319 sequences) were classified as being abundant in only 1 lake. In addition, 6,238 OTUs (91.9% of the total) were classified as always rare, and 3,086 rare OTUs (containing 95,443 sequences) were present in >70% of the samples (Fig. S1).

The result of multivariate cutoff level analysis (MultiCoLA) showed that both the Spearman and Procrustes correlations remained at around 100% even after removal of more than 40% of the rare part of the original data and the truncated data sets. In addition, when the number of rare types increased to 5%, the data structure of the truncated matrices exhibited little variation (Fig. S2). This result indicated that our definitions of abundant (34.5%) and rare (2.9%) bacteria are reasonable.

Geographic patterns of the bacterioplankton community.

Geographic patterns of abundant and rare bacterial communities had a significant positive correlation, both dividing two similar clusters (Fig. S3; P < 0.01). We found that seven lakes around Lake Taihu (TH) were clustered together (named cluster 1), and the remaining lakes were divided into another cluster (named cluster 2). Notably, there is a relatively larger geographic distance among the lakes between cluster 1 and cluster 2 than that among the lakes within each cluster (Fig. 1 and 2a). In addition, linear regression analysis showed that with increasing geographic distance, the dissimilarities in both abundant and rare communities were increased. Spearman correlation analysis gave a correlation coefficient of 0.587 (Mantel test, P < 0.01) between geographical distance and the dissimilarity of all bacterial taxa. Correspondingly, the Spearman correlation coefficients for abundant and rare bacterial taxa were 0.586 (P < 0.01) and 0.438 (P < 0.01), respectively (Fig. 2b). This phenomenon also suggested that lakes that were closer to each other presented more similar BCCs than lakes that were more distant from each other.

FIG 1.

Location of the 25 lakes in the middle and lower reaches of the Yangtze River Basin, China. The lake names in this figure are marked with the initials of their full name, which are shown in Table S1. Map generated with R v3.4.4 (available at https://www.r-project.org/) using a template from the Resource and Environment Data Cloud Platform (available at http://www.resdc.cn/).

FIG 2.

(a) The nonmetric multidimensional scaling (NMDS) for bacterial communities in the 25 lake sampling sites. (b) Spearman rank correlations between the Bray-Curtis dissimilarity of bacterial communities and geographic distance (n is the number of comparisons). The mean values of each bacterial community dissimilarity are marked with red numbers. All, all bacterial taxa; abundant, abundant taxa; rare, rare taxa.

Abundance-occupancy relationship of abundant and rare taxa.

The bacterial mean relative abundance, for both abundant and rare taxa, was significantly positively correlated with the sites occupied (abundant taxa, r = 0.409; rare taxa, r = 0.693; P < 0.01) (Fig. 3). The majority of abundant OTUs occupied >50% of the sampling sites, but all rare OTUs occupied <50% of sampling sites (Fig. 3).

FIG 3.

Abundance-occupancy relationship of bacterial taxa. Spearman rank correlation between the mean relative abundance of abundant bacteria (upper panel) and rare bacteria (lower panel) and number of samples occupied (n is the number of OTUs).

Taxonomic distribution of abundant and rare bacteria.

All reads were classified and grouped under 43 phylum-level taxonomic groups (including 2 unknown phyla; Table S2). Across all samples, the community structures of abundant and rare taxa at the phylum level showed the following differences. (i) The rare taxa have more species than the abundant taxa (Fig. 4). All of the rare taxa were present within each phylum, while abundant taxa were detected in only 11 phyla (Table S2). (ii) Abundant and rare bacterial taxa are dominated by different taxonomic groups. The most diverse and abundant groups in abundant subcommunities were Actinobacteria, Proteobacteria, and Verrucomicrobia, their average relative abundance representing 34.6% of the OTUs and 60.8% of the sequences, 21.0% of the OTUs and 10.7% of the sequences, and 6.2% of the OTUs and 10.2% of the sequences, respectively. In contrast, rare subcommunities were dominated by Proteobacteria, Actinobacteria, and Bacteroidetes, representing 34.1% of the OTUs and 34.0% of the sequences, 25.5% of the OTUs and 25.5% of the sequences, and 10.2% of the OTUs and 9.8% of the sequences, respectively. In addition, most abundant bacterial phyla were also most dominant in the entire community, but there were several phyla that showed high relative abundance (>1% of all sequences) in rare taxa only, such as Acidobacteria, Firmicutes, Chlamydiae, and TM6. Moreover, a total of 5.4% of the groups were occupied by large numbers of low-abundance phyla in rare bacterial taxa (Fig. 4 and Table S2).

FIG 4.

Heat map showing the community composition of abundant bacterial taxa compared with rare bacterial taxa in 25 lakes of the MLYB, China.

At the genus level, the main genera (relative abundance, >1%) of both abundant and rare bacteria also showed differences. Three unclassified genera that belong to the families C111 (Actinobacteria), ACK-M1 (Actinobacteria), and LD19 (Verrucomicrobia) and three unclassified genera that belong to the orders PHOS-HD29 (TA18), MWH-UniP1 (Betaproteobacteria), and Solirubrobacterales (Gammaproteobacteria) exhibited significant differences in abundant bacteria compared with rare bacteria. In addition, the three unclassified genera that belong to the families C111 (Actinobacteria), ACK-M1 (Actinobacteria), and LD19 (Verrucomicrobia) exhibited significantly higher abundance in the rare bacteria than in the abundant subcommunity (Fig. S4).

The relative contribution of deterministic and stochastic processes in the assembly of abundant and rare taxa.

The spatial variables of each sampling site were calculated through the principal coordinates of neighbor matrices (PCNMs) analysis. After analysis, 14 physicochemical variables (Table S3) and 6 significant PCNMs (PCNM1 to 6, P < 0.05) were subject to redundancy analysis (RDA), canonical correspondence analysis (CCA), and variation partitioning analysis (VPA). The Mantel results revealed that both environmental variables and spatial variables had a significant effect on bacterial subcommunities. Abundant bacterial communities were primarily driven by environmental variables, contrasting with rare bacterial subcommunities, which were primarily driven by spatial variables (Table 1).

TABLE 1.

Mantel tests for the correlations between community similarity and local environmental and spatial variables using Spearman coefficient^a

Variables	Significanceb of effect on:
	All bacteria	Abundant bacteria	Rare bacteria
Environmental variables	0.420***	0.350***	0.349***
Spatial variables	0.385***	0.321***	0.464***

^aThe Euclidean distance was used to calculate the similarity matrix.

^bSignificance was tested based on 999 permutations. ***, P < 0.001.

The RDA ordination using forward selection showed that the changes of all, abundant, and rare bacterial subcommunities were significantly related to two environmental variables (water temperature [Temp] and chlorophyll a [Chl a]) and one spatial factor (PCNM1) (Fig. 5a). An additional spatial factor (PCNM2) contributed significantly to explaining the community composition of the rare bacterial subcommunity only.

FIG 5.

(a) Ordination plots showing all, abundant, and rare BCCs in relation to significant local environmental variables and spatial variables. Significance: *, P < 0.05; **, P < 0.01; ***, P < 0.001. (b) Variation in all, abundant, and rare bacterioplankton metacommunities explained by environmental and spatial variables. E|S, pure environmental variables; S|E, pure spatial variables; E&S, shared explained variation. Unexplained = 1-E|S-S|E-E&S. All, abundant, and rare represent entire taxa, abundant taxa, and rare taxa, respectively.

The result of VPA showed that environmental variables and spatial distance accounted for 35.7% of the abundant bacterial taxon variations; the part accounted for by these together was 27.2%. Based on the environmental variables measured, abundant bacterial taxa were determined by environmental variables, which accounted for 31.8% of the abundant taxon variation (Fig. 5b). For rare bacterial taxa, however, environmental and spatial variables together accounted for only 7.2% of the variations. The relative contribution of purely spatial variables was slightly higher than that of environmental variables, explaining 1.6% of the rare taxon variation (Fig. 5b).

The neutral community model (NCM) was used to assess the potential importance of neutral processes for the entire bacterial community. The result showed that the neutral model explained a large fraction (overall fit to the neutral model [R²] = 52%) of the variation of all bacterial taxa (Fig. S5a). In order to further quantify the relative contribution of the two processes, the null model analysis and beta nearest-taxa index (βNTI) were performed. We found that the mean relative contributions of the stochastic process for abundant and rare bacteria were up to 92.62% and 57.54%, respectively, indicating that the dispersal of bacteria is highly stochastic, especially for rare subcommunities (Table 2). In addition, we found that the mean value of βNTI and the value of Bray-Curtis-based Raup-Crick (RC_Bary) are –1.52 and 0.91 (Fig. S5b, Table S4), respectively.

TABLE 2.

Relative contribution of deterministic and stochastic processes for geographic patterns of all, abundant, and rare bacteria across all lakes

Bacterial group	Mean RCBary (%)	Mean deterministic (%)	Mean stochastic (%)	Ecological processes shaping biodiversity
All	0.91	24.09	69.87	Stochasticity; undominated
Abundant	0.94	46.72	57.54	Stochasticity; undominated
Rare	0.60	30.13	92.62	Stochasticity; undominated

DISCUSSION

Geographic patterns in abundant and rare bacteria among interconnected lakes.

Recent studies of different aquatic ecosystems (e.g., coastal bays and freshwater lakes) suggested that rare bacterial taxa had significant spatial distribution patterns and showed a similar geographic pattern to the abundant bacterial taxa (13, 20, 27). Our results, in particular, the correlation of geographic patterns between abundant and rare bacteria, supported this contention. We found that both rare and abundant bacterial taxa had similar biogeographic patterns across all lakes, although biodiversity patterns largely differed between them (Fig. 4). This result is consistent with our first hypothesis, which indicates that in interconnected lakes, rare bacteria and abundant bacteria might respond similarly to dispersal processes or environmental changes (8, 20). This phenomenon contradicts the conclusions about some isolated lakes (17).

In this study, the geographic distances among all lakes ranged from 5 kilometers to 700 kilometers (Fig. 1). Interestingly, bacterial assemblages (all, abundant, and rare communities) were not separate between those lakes (e.g., those lakes located within each cluster) with small geographic distances but were clearly separated between those lakes (e.g., those lakes located between the two clusters) with relatively larger geographic distances according to our nonmetric multidimensional scaling (NMDS) result (Fig. 2a and Fig. S3). Our observations reveal some important information about microbial geography among interconnected aquatic habitats.

On the one hand, as verified by previous simulation experiments (29), local microbial dispersal in isotropic aquatic habitats will significantly homogenize bacterial richness. The present study provides direct evidence to prove that the interconnectivity among natural aquatic habitats can reduce the spatial heterogeneity of bacteria therein. Our finding was consistent with two studies of interconnected lakes at the local scale that found small spatial differences in BCCs (1, 30). However, our observation was in contrast to Lear et al. (31), who proved that even within the highly continuous environment of lentic water, distinct bacterial communities are present across unexpectedly small spatial scales. The discrepancy between our present work and that of Lear et al. (31) lies in whether there is hydraulic exchange among interconnected habitats. Due to the special geographical condition, most lakes in the MLYB are connected by intricate river networks, which break the geographic isolation and provide pathways for bacterial dispersal. Furthermore, there was no significant hydrodynamic bias among these lakes, as they are located in plain areas. In this scenario, it may be the intricate river network that shapes the scattered geographical patterns of bacteria among lakes in each cluster (Fig. 2a) (32, 33).

On the other hand, even among well-interconnected lakes with hydraulic exchange, the bacterial assembly is still limited by spatial distance. The direct evidence was the significant distance-decay relationship of all, abundant, and rare bacterial community similarities (Fig. 2b). Studies have reported that bacterial taxa followed distance-decay relationships if they were filtered by local habitat conditions or if they were limited in dispersal (34). Interestingly, we found that there was no significant increase in environmental variability with geographic distance (Fig. S6). In this scenario, the assembly of bacteria would be directly related to the spatial scales among habitats. Also, among interconnected lakes, bacterial individuals may tend to colonize nearby sites (13, 34), or they will need more time to disperse to further habitats.

Further, although the mean community dissimilarity of abundant taxa was lower than that of rare taxa (Fig. 2b, abundant 63.7% versus rare 94.3%), it increased more significantly (r = 0.586) than that of rare bacteria (r = 0.438, Fig. 2b) with increasing geographic distance (Fig. 2b). This could be explained by the dissimilarity between abundant and rare bacterial subcommunity compositions. Even though they have a stronger distance-decay relationship, the abundant bacteria with high local abundance have a decreased probability of local extinction and an increased probability of dispersal (Fig. S1, Fig. 3), thereby resulting in a widespread and relatively lower community dissimilarity (Fig. 2b). In contrast, the rare bacteria with extremely low local abundance have high spatial community heterogeneity even in a very small spatial scale (Fig. 2b) (13, 20).

Controlling factors for geographical distributions of abundant and rare bacterial taxa.

Previous research has highlighted two key areas for which further study is needed (27, 35). First, the influence of both deterministic and stochastic processes on the biogeographic pattern of bacteria has seldom been evaluated (13, 36). Second is the unexplored topic of the influence of both deterministic and stochastic processes on the biogeographic pattern of bacteria in interconnected lakes. To address these deficits, we examined the relative contribution of deterministic and stochastic processes in the assembly of bacterial subcommunities. We found that although the deterministic processes and stochastic processes together govern the abundant and rare bacterial subcommunity assemblies, the stochastic processes exhibited a greater influence on bacterial subcommunities. Our findings support our second hypothesis. This finding could be attributed to the intricately interconnected relationship and the frequently hydraulic exchanges among these lakes.

On the one hand, the strong interconnected relationship among nearby lakes reduces the spatial heterogeneity of environmental conditions (Fig. S5), which likely weakens the filtering effect of the deterministic processes. The result of VPA confirmed that a large proportion of the observed BCC variations could not be explained, especially for the rare taxa (Fig. 5b). In addition, although deterministic variables were significantly related to abundant and rare taxon subcommunities (Table 1), only two environmental factors (Temp and Chl a) were significantly related to the spatial distribution of abundant and rare taxon subcommunities (Fig. 5a). Our results are in contrast to previous studies that reported that numerous environmental and spatial variables could strongly affect (13, 25, 37) and explain a large fraction of the observed BCC variations among isolated aquatic habitats (9, 37).

On the other hand, the disordered interconnectivity among lakes increases the randomness of bacterial dispersal therein. In this study, the stochastic processes explained a large fraction (R² = 52%) of the variation of all bacterial taxa across all lakes according to our NCM analysis (Fig. S5a), indicating that neutral processes have a strong role in the entire bacterial community assembly. Further, the result of the null model showed that the geographic patterns of both abundant and rare bacterial taxa were dominated by stochastic processes (Table 2). Interestingly, our observations are consistent with similar studies on the bacterial subcommunities of different marine bays (19, 27, 38), which also suggested that stochastic processes explained a large fraction of the variation of bacteria. However, our findings contrast with the situation in some river-like landscapes, isolated lakes, and the deep aquifer (8, 12, 24), where deterministic processes dominate the assembly of bacteria. In addition, the mean values of |βNTI| < 2 and |RC_Bary| < 0.95 (Fig. S5b) indicated that the undominated stochastic processes mainly drive the assembly of bacterial communities across all lakes (39). Clearly, our findings revealed an important phenomenon in microbial ecology; that is, the bacterial dispersal among these aquatic habitats that are interconnected and not hydrodynamically biased is likely to be random.

Further, we found that the spatial variables explained more variation of the rare bacterial community than did the environmental variables (Fig. 5b), while the environmental variables explained more variation of the abundant bacterial community than did the spatial variables (Fig. 5b). This result was consistent with the Mantel tests for the correlation between community similarity and local environmental and regional factors using the Spearman coefficient (Table 1). Our results were different from recent studies of isolated aquatic abundant and rare bacterial subcommunities. For example, Logares et al. (20) suggested that both abundant and rare bacterial taxa were strongly influenced by purely environmental variables in isolated coastal Antarctic lakes. Similarly, Li et al. (17) found that environmental filtering played an important role in both abundant and rare bacterial assemblies between the lakes along various elevation gradients. The discrepancy between our present work and that of others may lie in different environmental gradients. In their studies, all lakes had a large environmental gradient (e.g., salinity gradient, pH), which is very different from the environmental gradients in our study. The large environmental gradients can affect the assembly of both abundant and rare bacterial subcommunities. In addition, the large ratio of unexplained variation for rare bacteria could be attributed to biotic or abiotic variables other than those mentioned in this paper. This result also indicated that more complex mechanisms may generate and maintain the rare biosphere diversity among interconnected lakes (27).

Conclusions.

In this study, we found that although the interconnectivity between lakes homogenizes the bacterial communities in local areas, the dispersal capability of bacteria in large spatial scales was still limited by geographic distance. This phenomenon was consistent with the result that both abundant and rare bacteria followed a strong distance-decay relationship. More importantly, the similar geographic patterns of both abundant and rare bacterial taxa indicated that they may be affected by the same ecological process among interconnected lakes. Furthermore, based on the environmental variables measured, deterministic processes and stochastic processes together drive the bacterial subcommunity assemblies, but stochastic processes exhibited a greater influence on the bacterial community assembly among the interconnected aquatic ecosystems. Our findings enhance the mechanistic understanding of the bacterial biogeography among interconnected lakes and have important implications for microbial ecology.

MATERIALS AND METHODS

Study area and sample collection.

With a total area of 18,400 km², the MLYB has more than 600 lakes larger than 1 km² (40). These lakes are vital to both regional economic development and ecological stability (41, 42). However, due to heavy external pollutant inputs from local and upstream areas, most of these lakes are polluted, and many are eutrophic (43). There is strong hydraulic disturbance and exchange among lakes due to the area’s location in the subtropical climatic zone, with distinct dry and wet seasons.

The MLRB is divided into the middle reaches (MR) and the lower reaches (LR) by the city of Hukou (29). Field work was conducted in July 2016. We randomly selected 25 lakes in the MR and the LR from which to collect water samples (Fig. 1, and Table S1). At each lake, one surface water sample (from the top 50 cm of the water column) was collected in a 5-liter sterilized polypropylene bottle at the lake center. All water samples were transported to the laboratory in dark cooling boxes and processed for physicochemical analysis and bacterial community analysis as soon as possible.

Physicochemical analysis.

Water column depth (WD) and Secchi disk depth (SDD) were measured with a water depth gauge (Uwitec, Austria) and Secchi disk, respectively. Water temperature (Temp), pH, electrical conductivity (EC), dissolved oxygen (DO), salinity (Sal), and turbidity of the epilimnion layer were measured in situ with a multiparameter water quality sonde (YSI 6600 V2, Yellow Springs Instruments, USA). The following physicochemical parameters were determined using standard methods (44): total nitrogen (TN), ammonium (NH₄⁺), nitrate (NO₃⁻), nitrite (NO₂⁻), total phosphorus (TP), orthophosphate (PO₄^3–), dissolved inorganic carbon (DIC), dissolved organic carbon (DOC), total suspended solids (TSS), and chlorophyll a (Chl a).

DNA extraction, PCR, and high-throughput sequencing.

For bacterioplankton analyses, a 100- to 500-ml water sample was filtered through a 0.22-μm pore size polycarbonate filter (47 mm diameter; Millipore, Billerica, MA, USA) (45), and then the filter was put into a sterile centrifuge tube and immediately stored at –80°C. Total DNA was extracted directly from the filters using a FastDNA spin kit for soil (MP Biomedicals, Solon, OH, USA) according to the manufacturer’s instructions. After the quality and quantity of the amplified DNA fragment were checked (with a NanoDrop ND-1000 UV/Vis spectral photometer), the V4 region of the bacterial 16S rRNA gene was amplified using primers 515F (5′-GTGCCAGCM-GCCGCGGTAA-3′) and 806R (5′-GGACTACHVGGGTWTCTAAT-3′) (46) with 20 μl of DNA. The PCR amplifications for each DNA sample followed these steps: initial denaturation at 94°C for 5 min followed by 25 cycles of 30 s at 94°C, 30 s at 50°C, and 30 s at 72°C; the final step was an extension at 72°C for 7 min. Subsequently, the total nucleic acid was sent to BGI Co. Ltd. (Wuhan, China) for high-throughput sequencing on an Illumina MiSeq instrument (San Diego, CA, USA).

Bioinformatics analysis.

Raw sequence data were processed using the Quantitative Insights into Microbial Ecology (QIIME) v1.9.1 pipeline (47). After the quality of raw reads was estimated using FastQC v0.11.5, the Illumina paired-end reads were assembled using FLASH with a minimum overlap of 10 bases (48). Then, 16S reads were demultiplexed and quality-filtered following the pipeline and previous research results (49). Next, we checked and then removed potential chimera sequences with a reference base method using VSEARCH v1.11.1 (50), which implements the UCHIME algorithm (51).

Bacterial phylotypes were identified and assigned to operational taxonomic units (OTUs, 97% cutoff) based on the v13_8 release of the Greengenes database (52). After taxonomy classification, representative sequences were aligned and further refined using PyNAST (37, 52), and phylogenetic trees were built using FastTree (53). We removed all archaea, chloroplasts, mitochondria, unassigned sequences, and spurious singletons from the OTU table. Finally, to minimize the sequencing depth of different samples, the high-quality OTU table was randomly resampled based on the sample with the smallest sequence (18,753 reads) using the QIIME pipeline (47).

Definition of abundant taxa and rare taxa.

We defined abundant and rare OTUs based on their relative abundance, considering two thresholds, on the regional and local scales. At the regional scale, OTUs with a mean relative abundance of >0.1% across all samples were defined as abundant taxa, and those with a mean relative abundance of <0.001% were defined as rare taxa (4, 13). At the local scale, the thresholds were abundant if >1% within a sample and rare if <0.01% within a sample (6). In addition, in order to reveal more distribution information of abundant and rare OTUs in all sampling sites, we further divided them into four categories (normally abundant, oscillating, normally rare, and always rare) as described in reference 13. To reduce the effect of arbitrariness, we used the multivariate cutoff level analysis (MultiCoLA v1.4) as described by Gobet et al. (54).

Data analyses.

The β-diversity indices were analyzed using QIIME v1.9.1 as part of the OTUs’ downstream analysis (47). The geographic patterns of both abundant and rare taxa among the lakes were revealed by nonmetric multidimensional scaling ordination analysis (NMDS). In addition, the cluster analysis was used to test whether rare bacterial taxa share similar biogeography with abundant ones. For NMDS and cluster analysis, OTU data were initially subject to Hellinger’s transformation (55). The relationships between bacterial community dissimilarity and geographic distance of these lakes were calculated using Spearman rank correlations. The dissimilarity of the bacterial community was determined with the Bray-Curtis dissimilarity matrix, and the geographic distance was determined from geographic coordinates of each lake. To determine if there was a significant difference in bacterial communities with increasing geographic distance, the Mantel test was used.

Process analysis.

To explore the influence of ecological processes on BCCs, the canonical ordination analyses were used in the R environment using the Vegan package. The normal distribution of the environmental variables was checked using the Shapiro-Wilk test with SPSS v22 (IBM), and these factors were log(x + 1) transformed (with the exception of pH) to improve normality and homoscedasticity for multivariate statistical analyses. To avoid collinearity, environmental factors with variance inflation factors (VIF) of >20 were deleted. The spatial variables were generated with principal coordinates of neighborhood matrix (PCNM) analysis (56) based on the geographical coordinates of each sample.

The correlations between bacterial communities and environmental and spatial variables were analyzed using the suitable ordination analysis method determined by detrended correspondence analysis (DCA). The result of DCA revealed that the longest gradient lengths for both all and abundant bacteria were <3, indicating that redundancy analysis (RDA) is suitable for both communities, while the longest gradient length for rare bacteria was >4, indicating that canonical correspondence analysis (CCA) is suitable for rare bacterial communities (57).

Environmental variables and spatial variables that significantly explained parts of the variation in BCCs were selected by forward selection. Based on a Monte Carlo test with 999 permutations, only variables that explained a significant (P < 0.05) additional proportion of total variance were included in the subset of forward selected variables (58). In addition, the contributions of environmental variables and spatial variables to geographic patterns of BCCs were assessed with variation partitioning analysis (VPA).

To evaluate the potential importance of stochastic processes for BCCs, we used the neutral community model (NCM) (59). This model predicts the relationship between the frequency of detecting an OTU and its relative abundance along taxonomic ranks (38). In the NCM, the parameter Nm determines the correlation between occurrence frequency and regional relative abundance, with N describing the metacommunity size and m being the immigration rate. The parameter R² determines the overall fit to the neutral model (59).

In addition, there was still a large amount of unexplained variation that could be attributed to biotic or abiotic variables other than the ones mentioned in this paper. To quantify the contributions of various ecological processes to microbial community structure, we used a null-model-based statistical framework developed by Stegen et al. (60) to further verify our conclusion. These methods have been used to compare the roles of deterministic and stochastic processes in phylogenetic composition across spatial scales (61). The β nearest-taxa index (βNTI) and the value of Bray-Curtis-based Raup-Crick (RC_Bary) were calculated to identify processes driving microbial community composition within all lakes. Details for calculative processes are provided in Information S1 in the supplemental material. In addition, to investigate the relative importance of deterministic and stochastic processes underlying community assembly, we performed the null model analysis using abundance-based similarity metrics as described by Zhang et al. (62). The relevant R code is available at http://mem.rcees.ac.cn/download/ (see nullmodel_MEC.rar). In our present study, all computations for the models or methods were performed in R v3.4.4 (63).

Data availability.

The 16S gene sequences generated in the present study were submitted to the National Center for Biotechnology Information (NCBI) database (https://www.ncbi.nlm.nih.gov) under the SRA number SRP173893.