Prokaryotic Community Structure Driven by Salinity and Ionic Concentrations in Plateau Lakes of the Tibetan Plateau

Abstract

The prokaryotic community composition and diversity and the distribution patterns at various taxonomic levels across gradients of salinity and physiochemical properties in the surface waters of seven plateau lakes in the Qaidam Basin, Tibetan Plateau, were evaluated using Illumina MiSeq sequencing. These lakes included Lakes Keluke (salinity, <1 g/liter), Qing (salinity, 5.5 to 6.6 g/liter), Tuosu (salinity, 24 to 35 g/liter), Dasugan (salinity, 30 to 33 g/liter), Gahai (salinity, 92 to 96 g/liter), Xiaochaidan (salinity, 94 to 99 g/liter), and Gasikule (salinity, 317 to 344 g/liter). The communities were dominated by Bacteria in lakes with salinities of <100 g/liter and by Archaea in Lake Gasikule. The clades At12OctB3 and Salinibacter, previously reported only in hypersaline environments, were found in a hyposaline lake (salinity, 5.5 to 6.6 g/liter) at an abundance of ∼1.0%, indicating their ecological plasticity. Salinity and the concentrations of the chemical ions whose concentrations covary with salinity (Mg²⁺, K⁺, Cl⁻, Na⁺, SO₄²⁻, and Ca²⁺) were found to be the primary environmental factors that directly or indirectly determined the composition and diversity at the level of individual clades as well as entire prokaryotic communities. The distribution patterns of two phyla, five classes, five orders, five families, and three genera were well predicted by salinity. The variation of the prokaryotic community structure also significantly correlated with the dissolved oxygen concentration, pH, the total nitrogen concentration, and the PO₄³⁻ concentration. Such correlations varied depending on the taxonomic level, demonstrating the importance of comprehensive correlation analyses at various taxonomic levels in evaluating the effects of environmental variable factors on prokaryotic community structures. Our findings clarify the distribution patterns of the prokaryotic community composition in plateau lakes at the levels of individual clades as well as whole communities along gradients of salinity and ionic concentrations.

INTRODUCTION

Salinity has a significant effect on the microbial community composition (1, 2), and microbial diversity decreases as salinity increases (3, 4). Increasing salinity was found to be correlated with a reduced richness of plant and animal species (5, 6) and of microbial community diversity in some cases (7, 8). However, this pattern is not consistently observed for all microbial communities (9–14). A better understanding of the taxonomic composition and diversity of microbial communities in saline lakes and the mechanisms that govern the community structure is a long-standing goal and challenge for microbial community ecologists (15–17). Many previous studies have addressed the effect of salinity on microbial community composition and diversity; however, such studies have generally been limited to the salinity gradients of dynamic estuaries (18–22), solar saltern ponds (4, 7, 9, 23–25), and the vertical water columns or sediments of salt lakes (26–28). Only three studies, based on limited numbers of water samples, have focused on natural inland lakes (8, 11, 12). We therefore have very limited knowledge regarding the patterns of the prokaryotic community composition and diversity across the gradients of salinity and other physiochemical parameters in inland lakes, particularly plateau lakes. In addition, these studies have typically described the community composition at a high taxonomic level (the phylum or class level) (8, 11, 12). The failure to analyze communities at lower taxonomic levels can easily result in the overlooking of specific taxa that account for the variations in the community composition (29). Therefore, little is known regarding the relationships of salinity and other physiochemical factors with variations in the community composition at the lower taxonomic levels, particularly on a broad sampling scale, in inland lakes. Salinity was found to be the key environmental selective force controlling microbial communities in the Baltic Sea (10) and even on a global scale (30–33). A better understanding of the distribution pattern of the prokaryotic community composition and diversity along a salinity gradient in inland lakes might look into this distribution pattern on a wider scale.

The Qaidam Basin, a large intermountain depression with an arid to semiarid continental climate, is located in the northeastern margin of the Tibetan Plateau, China (34), and is surrounded by the Qilian, Kunlun, and Aerjin Mountains. This basin contains dozens of lakes, from freshwater to hypersaline lakes, at high elevations ranging from ∼2,700 to 3,200 m above sea level. Bacteria and archaea are key components of microbial food webs and play key roles in the biogeochemical cycling of essential elements (C, N, P, S) in lakes (35, 36). The microorganisms that inhabit these lakes are potential valuable gene reservoirs for future biotechnological applications, particularly those involving saline conditions (e.g., microbial treatment of saline or high-salt wastewater). We previously described eukaryotic plankton assemblages in 13 water samples from six lakes in the Qaidam Basin (37). In this study, 46 surface water samples from seven lakes in this basin, including Lakes Keluke (salinity, <1 g/liter), Qing (salinity, 5.5 to 6.6 g/liter), Tuosu (salinity, 24 to 35 g/liter), Dasugan (salinity, 30 to 33 g/liter), Gahai (salinity, 92 to 96 g/liter), Xiaochaidan (salinity, 94 to 99 g/liter), and Gasikule (salinity, 317 to 344 g/liter), were collected for analysis for their prokaryotic community compositions. These lakes were largely separated from each other (50 to 600 km), except that the freshwater supply to Lake Tuosu is from Lake Keluke, since these two lakes are connected (Fig. 1). Lake Qing is a hyposaline lake which originated from a hypersaline salt flat and was formed after being supplied with freshwater. These lakes have a wide range of salinities (0.8 to 344 g/liter) and physicochemical factors, providing a useful model to obtain better insight into the relationship between the prokaryotic community composition and various environmental factors in plateau lakes.

FIG 1.

Locations of the sampling sites in seven lakes in the Qaidam Basin. (Maps were produced using ArcGIS version 9.3 [ESRI, Redlands, CA].)

The objectives of this study were to (i) compare the prokaryotic community composition and diversity in the seven lakes using the Illumina MiSeq sequencing method with 16S rRNA gene amplicons and (ii) evaluate the distribution patterns of the prokaryotic community composition within individual clades and across entire prokaryotic communities along gradients of salinity and other physiochemical parameters.

MATERIALS AND METHODS

Sample collection.

In September 2013, 46 water samples were collected, using a 5-liter Schindler sampler, from 0.3 m below the surface of seven lakes in the Qaidam Basin with salinities ranging from 0.8 to 344 g/liter. The seven lakes consisted of a freshwater lake with a salinity of <1 g/liter (Lake Keluke with six samples), a hyposaline lake with a salinity of ∼5 g/liter (Lake Qing with four samples), two mesosaline lakes with salinities of ∼30 g/liter (Lakes Tuosu and Dasugan with eight samples each), two hypersaline lakes with salinities of ∼100 g/liter (Lakes Gahai and Xiaochaidan with seven samples each), and a hypersaline lake with a salinity of ∼330 g/liter (Lake Gasikule with six samples) (Fig. 1). For each lake, 3 liters of water was collected and concentrated to 500 ml with a hollow-fiber membrane module (pore size, 0.22 μm; Tianjin Aisheng Membrane Filtration Technology Co., Tianjin, China), and the concentrated sample was placed in a 500-ml sterile opaque polypropylene bottle for DNA extraction. An additional 100 ml of water was collected for physicochemical studies. All water samples were stored at 4°C (<3 days) until they were transported to the laboratory.

Analysis of limnological properties.

Temperature, electrical conductivity, pH, the dissolved oxygen (DO) concentration, altitude, latitude, and longitude were measured in situ using a Hydrolab sensor (Austin, TX, USA). Water samples for chemical studies were filtered with cellulose acetate membranes (pore size, 0.45 μm). Salinity was determined by drying each sample at 105°C, calcinating the sample at 600°C to a constant weight, and analyzing the resulting residue (38). The concentrations of potassium (K⁺), sodium (Na⁺), calcium (Ca²⁺), magnesium (Mg²⁺), chloride (Cl⁻), sulfate (SO₄²⁻), phosphorus (PO₄³⁻), total nitrogen (TN), and ammonia (NH₄⁻-N) were measured by standard methods (39).

Genomic DNA extraction.

Water samples concentrated to 500 ml as described above were filtered through polyether sulfone membranes (pore size, 0.22 μm; Jinteng, Beijing, China) to obtain organisms, including all bacterial and archaeal cells, whose cell size was more than 0.22 μm. These filters were stored at −80°C until extraction of genomic DNA. Community DNA was extracted from the water samples with an E.Z.N.A water DNA extraction kit (catalog number D5525-1; Omega Bio-Tek, USA) according to the manufacturer's instructions and stored at −80°C.

Tag-encoded amplicon pyrosequencing of bacteria and archaea.

The V4 hypervariable regions of the 16S rRNA genes of bacteria and archaea were amplified by PCR in triplicate with primers 515F/806R (40, 41). A barcode and Illumina adaptor were fused to the primers. The replicates were pooled. The resulting amplicons were sequenced using the Illumina MiSeq platform (paired-end reads), as described previously (40, 41), by Novogen Bioinformatics Technology Co. (Beijing, China).

Sequence analysis.

Sequences with an expected error of >1.0 or a length of <240 nucleotides (nt) were excluded (42). The remaining reads were truncated to a constant length (240 nt). Various analyses, as described in the following, were performed, and the results were analyzed using the QIIME (Quantitative Insights into Microbial Ecology, version 1.9.0) software package (43) with default parameters, except that chimera filtering, clustering of operational taxonomic units (OTUs), and exclusion of singletons were performed within QIIME through the UPARSE pipeline (42). A phylogenetic tree relating the OTUs was constructed with a set of sequences representative of the OTUs using the method with FastTree software (44). Chimeras were identified and filtered by the use of UPARSE with the UCHIME algorithm and the ChimeraSlayer reference database (45), which is considered to be sensitive and quick (46). Reads with 97% sequence similarity were clustered into OTUs by UPARSE for alpha-diversity analysis. A representative sequence from each OTU was selected for taxonomic annotation using the Ribosomal Database Project (RDP) classifier (47) and the RDP (release 11.4) database. Taxonomic assignments with <80% confidence were marked as unclassified taxa. Sequences assigned to be mitochondrial or chloroplast were excluded from further analysis.

Diversity and statistical analysis.

Prior to statistical analysis, the number of sequences was normalized by random resampling of the reads in each sample so that the number of reads in each sample was the same, on the basis of the number of reads in the sample with the smallest sample size (n = 15,900 sequences in this study). The relative abundances of the prokaryotic community composition at various taxonomic levels (phylum, class, order, family, genus) were summarized for each sample, and distance matrixes (weighted UniFrac) among samples were constructed. Alpha-diversity measures, i.e., the Chao1 richness estimator (48), the Shannon diversity indexf (49), the phylogenetic diversity (PD) index (50), and Good's coverage (51), were calculated. Samples were clustered by the unweighted pair group method with arithmetic mean (UPGMA) on the basis of weighted UniFrac distances, which account for changes in relative taxon abundance (43). Principal coordinates analysis (PCoA) using weighted UniFrac metrics was performed to distinguish the general distribution patterns of the prokaryotic community composition among the samples.

Environmental parameters were compiled and tested for normality (one-sample Kolmogorov-Smirnov test; P > 0.05) using the SPSS software program for Windows (version 18.0). Parameters found to have a nonnormal distribution were transformed as close to normality as possible. Pearson's correlation analysis and curve estimation were also performed using the SPSS program. The Mantel test, redundancy analysis (RDA), and variation partitioning analysis (VPA) were used to evaluate the linkages between the prokaryotic community structure and environmental parameters. The Canoco software program (version 4.5) (52) was used for RDA with forward selection using the Monte Carlo test (n = 999) on the basis of the results of pretested detrended correspondence analysis (DCA). The variables with variance inflation factors of greater than 20 were sequentially removed from the RDA model (53). The significance of the differences among the prokaryotic community compositions grouped by lakes was evaluated by the use of analysis of similarity (ANOSIM) statistics (54). VPA and ANOSIM were performed using functions in the Vegan package (version 2.3-0) (55) in R (version 3.2.2) (56).

Nucleotide sequence accession numbers.

The nucleotide sequences discovered during this study have been deposited in the NCBI Sequence Read Archive under accession number SRP063473. The accession number for each sample is listed in Table S1 in the supplemental material.

RESULTS

Environmental data.

The geographical and physiochemical parameters of the 46 water samples from the seven lakes are summarized in Table S1 in the supplemental material. The lakes were characterized as a freshwater lake (Lake Keluke; salinity, 0.8 g/liter), a hyposaline lake (Lake Qing; salinity, ∼5 g/liter), two mesosaline lakes (Lakes Tuosu and Dasugan; salinity, ∼30 g/liter), and three hypersaline lakes (Lakes Gahai and Xiaochaidan [salinity, ∼100 g/liter] and Lake Gasikule [salinity, ∼330 g/liter]), and the lakes were at similar elevations (2,678 to 3,173 m). Within a given lake, differences among samples were not significant (P > 0.05), except for sample T13 from Lake Tuosu, which was collected very close to the inlet and had a salinity and ionic concentrations significantly lower than those of the other samples in Lake Tuosu (see Table S1 in the supplemental material). Salinity and certain physiochemical parameters varied widely along the spectrum from freshwater to hypersaline water. In some lakes, the PO₄³⁻ concentration was extremely low or PO₄³⁻ was undetectable (see Table S1 in the supplemental material). Significant Pearson's correlation values between salinity and many physiochemical parameters were found: conductivity, pH, and the concentrations of TN, DO, Cl⁻, Na⁺, K⁺, Mg²⁺, SO₄²⁻, and Ca²⁺ (see Table S2 in the supplemental material).

General statistics for 16S rRNA gene sequences and the taxonomic compositions of the prokaryotic communities.

Clean data for a total of 2,541,568 sequences (each truncated to a length of 240 nt, as explained above) were obtained from the raw data for 3,211,120 total sequences after quality filtering (see Table S3 in the supplemental material). The clean data for each sample (i.e., each MiSeq sequencing library) ranged in size from 15,971 sequences for sample Q4 to 76,644 sequences for sample K15. After normalization to 15,900 sequences for each sample, 731,400 rarefied sequences for the 46 samples remained and were subjected to further statistical analyses.

These sequences were widely distributed across the domains Bacteria and Archaea, wherein sequences classified as Bacteria, Archaea, and unassigned domains accounted for 87.8%, 11.7%, and 0.5% of all sequences in the sequencing libraries, respectively. Most of the archaeal sequences (10.6% of all sequences) were detected in Lake Gasikule, which had the highest salinity (317 to 344 g/liter). In contrast, the archaeal sequences detected in Lakes Keluke, Qing, Tuosu, Dasugan, Gahai, and Xiaochaidan comprised only <0.1%, <0.1%, <0.1%, 0.2%, 0.5%, and 0.1% of all sequences, respectively. PCoA showed that prokaryotic communities varied for different lakes and apparently clustered by salinity (Fig. 2a and b). The first, second, and third dimensions of PCoA showed that the distribution of all samples accounted for 83.0%, 4.7%, and 4.2% of the community variability, respectively. The weighted UniFrac tree (UPGMA) showed that samples from a given lake formed an independent lineage and that samples from lakes with similar salinities clustered together (Fig. 2c). Interestingly, sample T13, taken near the Lake Tuosu inlet (Fig. 1), clustered with samples from freshwater Lake Keluke but not with other samples from Lake Tuosu (Fig. 2). Samples from Lake Gasikule formed a separate and distant cluster outside the other samples, indicating a more distant phylogenetic relationship. The results seen on the weighted UniFrac tree confirmed that the prokaryotic communities varied depending on the lake and salinity. In addition, the results from ANOSIM shown that the prokaryotic community compositions were significantly different among lakes (P < 0.01 for all comparisons, n = 999; data not shown).

FIG 2.

Relationships between individual samples illustrated by PCoA plots (a, b) and a UniFrac tree (UPGMA) (c). Both analyses were performed on the basis of the weighted UniFrac metric. Symbols of the same color indicate samples from the same lake. The distance between symbols (a, b) reflects their dissimilarity. The values at the right of the symbols are the approximate salinity (in grams per liter) of the samples (a, b). Approximate salinity is also indicated below the name of each lake (c).

Rarefied sequences of all libraries were affiliated with 616 genera, 397 families, 231 orders, 120 classes, and 38 phyla, of which 344 genera, 253 families, 173 orders, 106 classes, and 38 phyla had recognized names. The 18 most abundant phyla, each of which accounted for >0.03% of all sequences (across all samples), comprised >99.8% of the communities and were designated major phyla. The prokaryotic community structures of these 18 phyla for all samples as a function of increasing salinity gradient are shown in Fig. 3. Microorganisms assigned to unclassified prokaryotes or Archaea represented only a small part of each sample, with their relative abundances being 1.2 to 6.3% and 0.1 to 1.8%, respectively, for Lake Xiaochaidan and 0 to 0.6% and 0 to 0.03%, respectively, for the other lakes. However, unclassified bacteria were significant components of certain samples, particularly samples G7, G1, G6, and X11, in which their relative abundances were 30.3, 16.2, 8.6, and 8.1%, respectively (Fig. 3). In contrast to the other lakes, which had much lower percentages (<4.0%) of archaea, Lake Gasikule had a highly distinctive prokaryotic community, in which archaea had a relative abundance of 79.1 to 89.4% of all prokaryotes across six samples, again illustrating the distant phylogenetic relationship between Lake Gasikule and the other lakes. Essentially all (>99.5%) archaeal sequences of Lake Gasikule were assigned to members of the family Halobacteriaceae within the phylum Euryarchaeota. The other two most abundant phyla in Lake Gasikule were Bacteroidetes (6.5 to 18.2%) and Proteobacteria (1.7 to 2.8%), which were also abundant in the other lakes. The 56 most abundant families, each of which accounted for >0.1% of all sequences (across all samples), comprised >97.0% of the communities and were designated major families. The prokaryotic community structures of these 56 families as a function of increasing salinity gradient are shown for all samples in Fig. S1 in the supplemental material. Eight families (including the Microbacteriaceae within the Actinobacteria; the Balneolaceae and Microbacteriaceae within the Bacteroidetes; the Pseudomonadaceae, Chromatiaceae, Comamonadaceae, and Rhodobacteraceae within the Proteobacteria; and the Bacillaceae within the Firmicutes) were detected in all samples, indicating an adaption to a wide range of salinity for these clades. At the phylum level, the Bacteroidetes were abundant (relative abundance, >12.0% for all samples except sample M6, in which the abundance was 6.5%) in all lakes. At the family level within the Bacteroidetes, however, Lake Gasikule was clearly distinct from the other six lakes by the presence of an unassigned family within the class At12OctB3 (3.8 to 8.3%) and the family Rhodothermaceae (4.9 to 9.0% for all samples except sample M6, in which the abundance was 1.3%) as the major families, which were rare (<0.2%) in the other lakes, except Lake Qing (∼1.0%). Other significant clades (average relative abundance, >2% in at least one lake) at the phylum and family levels found in seven lakes are presented in Table S4 in the supplemental material.

FIG 3.

Prokaryotic community structure as the 18 most abundant phyla (and four classes within the Proteobacteria) in the 46 samples along a gradient of increasing salinity. All phyla are within the domain Bacteria, except for the Euryarchaeota and Parvarchaeota, which are within the Archaea.

Environmental factors that have significant effects on prokaryotic communities.

At the beginning, all environmental variables were included in the RDA model used for explaining the variation of the prokaryotic community structure at the phylum, class, order, family, genus, and OTU levels. Correlations between environmental factors and prokaryotic community composition varied depending on the taxonomic level (see Table S5 in the supplemental material). Among the explanatory variables, salinity was the most important factor for species composition (see Table S5 in the supplemental material). The ionic concentrations, which were correlated with salinity and included those of K⁺, Mg²⁺, Cl⁻, Na⁺, SO₄²⁻, and Ca²⁺ (see Table S2 in the supplemental material), also significantly affected the prokaryotic community composition at certain taxonomic levels. To reduce the collinearities among the explanatory variables, variables with variance inflation factors of greater than 20 were dropped from the RDA model. Seven variables, including salinity, pH, DO and PO₄³⁻ concentrations, altitude, NH₄⁻-N concentration, and temperature, were retained for further RDA analysis. Salinity, pH, and the DO concentration were significant (P < 0.05) at all taxonomic levels except for the DO concentration at the OTU level (P = 0.09). However, altitude, temperature, and the NH₄⁻-N and PO₄³⁻ concentrations did not significantly affect the community structure (P > 0.05), except in the case of the PO₄³⁻ concentration at the OTU level (see Table S5 in the supplemental material). RDA of the relationships between the seven environmental factors and the prokaryotic community composition at various taxonomic levels showed that community structures were most strongly affected by salinity; pH and the DO concentration also significantly contributed to controlling the prokaryotic community composition (Fig. 4). Altitude was the major factor for Lake Xiaochaidan, while the PO₄³⁻ concentration was the major factor for Lake Dasugan. Temperature and the NH₄⁻-N concentration were not major factors at any taxonomic level for any of the lakes (Fig. 4).

FIG 4.

Sample/environmental parameter biplots from RDA, illustrating correlations at the phylum (a), class (b), order (c), family (d) genus (e), and OTU (f) levels. The same symbols and colors represent samples from the same lake. Environmental parameters are indicated by black arrows. The axes show the percentages of variation in the distribution of prokaryotic communities.

Water salinity is approximately the sum of the concentrations of eight major ions, including K⁺, Na⁺, Ca²⁺, Mg²⁺, Cl⁻, SO₄²⁻, CO₃²⁻, and HCO₃⁻ (57). Salinity was significantly (P < 0.01) correlated with the concentrations of K⁺, Na⁺, Ca²⁺, Mg²⁺, Cl⁻, and SO₄²⁻ in 46 water samples from the seven plateau lakes (see Table S2 in the supplemental material). To quantify the relative contributions of salinity, as well as the concentrations of the ions whose concentrations covary with salinity, and other environmental factors to the prokaryotic community composition on the basis of the OTU composition, VPA was conducted by partitioning all variables into two groups: group 1 included salinity and the concentrations of the various ions whose concentrations covary with salinity (K⁺, Na⁺, Ca²⁺, Mg²⁺, Cl⁻, and SO₄²⁻), and group 2 included other factors, including pH, the DO concentration, temperature, the PO₄³⁻ and NH₄⁻-N concentrations, altitude, and the TN concentration. The combination of all variables (in both groups) showed a highly significant (P < 0.01) correlation with the variation of the prokaryotic community structure. These variables explained 77.9% of the observed variation, leaving 22.1% of the variation unexplained (Fig. 5). Group 1 (i.e., salinity and the concentrations of the six ions) individually could explain 20.6% of the variation (P < 0.01), while group 2 (i.e., other factors) individually explained only 4.2% of the variation (P < 0.01). The interaction between the two groups was 53.1% (Fig. 5). Thus, salinity and the concentrations of the six major ions accounted for most of the observed variation in the total (73.7% of 77.9%) on the basis of the OTU composition. Similar results were observed at the phylum, class, order, family, and genus levels (data not shown).

FIG 5.

VPA showing the effects of group 1 and group 2 factors and their interaction on the variance of the prokaryotic community structure on the basis of the OTU composition. Group 1 consists of salinity and the concentrations of six ions, including K⁺, Na⁺, Ca²⁺, Mg²⁺, Cl⁻, and SO₄²⁻; group 2 consists of the other factors measured, including pH, the DO concentration, temperature, the PO₄³⁻ and NH₄⁻-N concentrations, altitude, and the TN concentration.

Alpha diversity of prokaryotic communities and correlation with salinity.

Alpha-diversity indices (OTU number, the Chao1 richness estimator, the Shannon diversity index, the PD index, Good's coverage) were applied to all sequencing libraries normalized to a common size (15,900 sequences) and calculated. OTUs were defined as reads with 97% sequence similarity. The rarefied sequences for all MiSeq sequencing libraries contained 6,150 unique OTUs. Chao1 richness estimates suggested that >50% of the estimated prokaryotic diversity in these lakes was captured by our sequencing methods and that further sequencing would yield more unique OTUs (Table 1). OTU number, the Chao1 richness estimator, the Shannon diversity index, and overall PD indices revealed the highest alpha diversity to be in Lake Keluke (lowest salinity), followed by Lakes Gahai, Tuosu, Xiaochaidan, Dasugan, and Qing, and the lowest alpha diversity to be in Lake Gasikule, which had the highest salinity (Table 1). This finding was supported by rarefaction curves based on OTU number and the Shannon index (see Fig. S2 in the supplemental material). Good's coverage values were 96.7%, 98.5%, 97.0%, 97.9%, 96.8%, 98.0%, and 98.9% for Lakes Keluke, Qing, Tuosu, Dasugan, Gahai, Xiaochaidan, and Gasikule, respectively (Table 1). Pearson correlation analysis showed a highly significant negative correlation between OTU number (r = −0.65, P < 0.01), the Chao1 richness estimator (r = −0.58, P < 0.01), the Shannon diversity index (r = −0.90, P < 0.01), and the PD index (r = −0.63, P < 0.01) and salinity. In summary, prokaryotic diversity in the seven plateau lakes decreased as salinity increased.

TABLE 1.

Mean planktonic diversity of the seven lakes^a

Lake	Salinity (g/liter)	No. of OTUsb	Chao1 richness estimator	Shannon diversity index	PD index	Good's coverage (%)
All lakes		6,150
Keluke	<1	1,240	2,009	7.90	84.73	96.69
Qing	5.5–6.6	633	972	6.91	43.37	98.52
Tuosu	24–35	1,036	1,823	7.33	76.73	96.97
Dasugan	30–33	779	1,296	6.82	59.47	97.90
Gahai	92–96	1,130	1,958	7.38	80.85	96.81
Xiaochaidan	94–99	919	1,429	7.04	62.87	97.79
Gasikule	317–344	347	656	3.45	30.55	98.88

^aThe MiSeq sequencing libraries for each sample were normalized to 15,900 sequences.

^bAn OTU was defined as reads with 97% sequence similarity.

Relationship between salinity and relative abundance of major bacterial clades.

The covariation between salinity (the major factor affecting the distribution of the prokaryotic community composition in these lakes) and the relative abundance of prokaryotic clades at the phylum, class, order, family, and genus levels was evaluated by the use of Pearson correlation tests. Significant correlations were detected for 20 bacterial taxa (2 phyla, 6 classes, 4 orders, 5 families, 3 genera) and for the Archaea domain (see Fig. S3 in the supplemental material). Salinity was positively correlated with the relative abundance of the genus Rhodothermus (see Fig. S3a in the supplemental material). The abundances of the members of the class Gammaproteobacteria, the family Rhodobacteraceae, and the genus Marinobacter increased significantly with increasing salinity over the range of 0 to 100 g/liter but were minimal at salinities of 317 to 344 g/liter in Lake Gasikule (see Fig. S3b to d in the supplemental material). Salinity had significant negative correlations with the relative abundance of most of the clades evaluated, including the Verrucomicrobia, Planctomycetes, Planctomycetia, Sphingobacteriia, Betaproteobacteria, Deltaproteobacteria, Pirellulales, Sphingobacteriales, Verrucomicrobiales, Chthoniobacterales, Pirellulaceae, Planctomycetaceae, Verrucomicrobiaceae, Chthoniobacteraceae, Xiphinematobacter, and Planctomyces (see Fig. S3e to t in the supplemental material). The Archaea were more abundant in lakes with higher salinities (see Fig. S3u in the supplemental material). These clades showed significant correlations with certain other physiochemical parameters (see Table S6 in the supplemental material) that were correlated with salinity (see Table S2 in the supplemental material).

DISCUSSION

Research on the effects of salinity on prokaryotic community structures has largely been restricted to dynamic estuaries, solar saltern ponds, vertical water columns, and sediments of salt lakes. Only three studies have focused on multiple inland lakes with different salinity levels (8, 11, 12). However, these studies addressed the effects of salinity on the bacterial community composition at the phylum and class levels using traditional molecular methods and provided limited taxonomic information. Because of the great difficulties associated with the large-scale sampling of inland lakes, the number of water samples in these studies was small. The present study utilized the Illumina MiSeq sequencing method to evaluate the effects of numerous environmental parameters on the variation of the prokaryotic community structure in 46 water samples from seven plateau lakes with different salinities. The distribution patterns of the prokaryotic community composition across a gradient of salinity and the concentrations of the various ions whose concentrations covary with salinity were evaluated for entire communities as well as individual clades at different taxonomic levels. Clearly, our understanding of the precise microbial distribution patterns in response to gradients of salinity and other physiochemical variables in the different lakes would have been improved in this study with a higher-resolution analysis involving lower taxonomic levels, higher numbers of samples, and a greater number of environmental parameters. This study examined correlations between specific ionic species and prokaryotic community structures at many taxonomic levels along a salinity gradient of inland lakes.

Diversity of prokaryotic communities in plateau lakes.

This study examined the prokaryotic communities in 46 water samples from seven Qaidam Basin lakes using the Illumina MiSeq sequencing method. Although rarefaction curves based on OTU number did not reach the asymptote (see Fig. S2a in the supplemental material), Shannon diversity index curves clearly reached plateau levels (see Fig. S2b in the supplemental material). These results, as well as Good's coverage values (Table 1), indicate that most prokaryotic species in the lake habitats are well represented by these libraries. Lake Gasikule, which had the highest salinity (it is essentially salt saturated), had the lowest diversity, consistent with the general ecological principle that community diversity is low in extreme environments (7, 58). On the other hand, the diversity did not decline in association with increasing salinity for salinity values of <100 g/liter (Table 1). Similarly, in other studies comparing the prokaryotic community diversities in lakes with a range of salinities, the number of denaturing gradient gel electrophoresis bands did not decrease with increasing salinity (9, 11, 12). These findings reflect a higher microbial diversity in hypersaline environments (e.g., the hypersaline Lakes Gahai and Xiaochaidan) than in saline environments (e.g., Lake Dasugan) and a tendency for many closely related ribotypes to display only minor differences in 16S rRNA gene sequences (7). The diversity that we observed in the surface waters of our lakes was much higher than that observed in those of highland lakes, as determined by low-resolution/low-coverage profiling techniques, in previous studies (11, 12, 59, 60). Hayden and Beman reported similarly high levels of bacterial diversity in freshwater lakes using the Illumina MiSeq sequencing method (61). Thus, high-throughput sequencing may be more appropriate for evaluating overall microbial diversity in the surface waters of inland lakes.

The diversity in hyposaline Lake Qing (salinity, 5.5 to 6.6 g/liter) was much lower than that in mesosaline Lakes Tuosu and Dasugan and in hypersaline Lakes Gahai and Xiaochaidan (Table 1; see also Fig. S1 in the supplemental material). The low diversity may result from the unstable conditions of Lake Qing, which originated from a salt flat and was formed over a relatively short period by freshwater from a river. The evolutionary adaptation of autochthonous and allochthonous prokaryotic lake species to a changing environment may take a longer period of time.

Composition of prokaryotic communities in different lakes.

The prokaryotic communities in the lakes were dominated by bacteria, with the exception of hypersaline Lake Gasikule, which was dominated by archaea, with a proportion of 79 to 90% (Fig. 3; see also Fig. S3u in the supplemental material). These results indicate that archaea prefer to inhabit lakes saturated with salt (Lake Gasikule) rather than lakes with salinities of <100 g/liter (like the other six lakes studied). In agreement with this, upon electrophoresis samples from Lake Gasikule but no other lake showed clear bands for PCR production of archaeal 16S rRNA gene sequences amplified by universal primers 21F/958R (62) (data not shown). This finding is consistent with the findings described in previous reports of an increase in archaeal diversity with increasing salinity (2, 63). Numerous studies comparing the relative abundances of bacteria and archaea found very low levels of bacterial diversity in water bodies with salinities of >300 g/liter, such as natural inland lakes (59, 64) and solar saltern ponds (25, 65, 66). Another study demonstrated a higher abundance of archaea than of bacteria in water bodies with salinities of >240 g/liter (24). The competitive advantage of Archaea over Bacteria in extreme environments appears to be related to the degree of energetic stress (67) or salinity stress (3, 68) experienced by microbes in lakes and the physiological adaptations of the two groups for dealing with such stresses.

Members of the order Halobacteriales (comprising 78.8 to 89.5% of the prokaryotes in every sequencing library for Lake Gasikule) within the phylum Euryarchaeota and the genus Salinibacter (comprising 1.2 to 9.0% of the prokaryotes) within the phylum Bacteroidetes were important components in Lake Gasikule. They are distributed in hypersaline waters worldwide and are frequently dominant (69, 70). Unclassified class At12OctB3 within the Bacteroidetes, which was detected for the first time in a hypersaline lake (1), comprised 3.7 to 8.3% of the sequences in each sequencing library for Lake Gasikule. Halobacteriales, Salinibacter, and At12OctB3 comprised, respectively, 0.1 to 0.6%, 1.4 to 1.8%, and 0.7 to 1.0%, in each library of sequences for hyposaline Lake Qing, but they were not detected in hypersaline Lakes Gahai and Xiaochaidan (salinity, ∼100 g/liter). To our knowledge, Salinibacter and At12OctB3 are considered to live under hypersaline conditions and have not previously been reported from habitats with a salinity as low as that of Lake Qing (5.5 to 6.6 g/liter). In ecophysiological experiments, species of Salinibacter were unable to grow at salinities of <117 g/liter (71, 72). The presence of Halobacteriales, Salinibacter, and At12OctB3 in hyposaline Lake Qing may reflect the origin and history of this lake, which was originally a hypersaline salt flat possibly inhabited by these taxa. As freshwater entered, a few members of these groups displayed physiological adaptations and sufficient ecological plasticity to deal with low-salinity stress, while many other members disappeared because of the lack of such plasticity and were replaced by allochthonous freshwater taxa. This scenario may exemplify two common strategies that microbial communities use to respond to changes in salinity: slow adaptation of a given taxon or replacement of one taxon by another (12, 26). As freshwater was supplied to the hypersaline salt flat to form hyposaline Lake Qing, Halobacteriales, Salinibacter, and At12OctB3 might have slowly adapted to the decreased salinity and survived, whereas many other autochthonous members (unknown) were probably replaced by some allochthonous community from freshwater, such as some members within the Verrucomicrobia, Planctomycetes, Actinobacteria, Bacteroidetes, and Proteobacteria (Fig. 3).

Several lake-specific phyla were found to be highly abundant. Tenericutes and Cyanobacteria were the major components of mesosaline Lakes Tuosu and Dasugan but were absent from or rarely detected in the other lakes (Fig. 3), indicating a preference for mesosaline conditions by members of these two clades in Qaidam Basin. Mesosaline lakes may provide relatively stable osmotic conditions for Tenericutes, which have no cell wall and are sensitive to osmotic stress (73). The PO₄³⁻ concentration was significantly correlated with the relative abundance of Tenericutes and Cyanobacteria, suggesting that these clades are dependent on phosphorus and/or involved in its cycling. Further study of the degree to which the relative abundance of these clades in mesosaline lakes is driven by the PO₄³⁻ concentration or other nutrient parameters is needed. Although Cyanobacteria was the major phylum in both mesosaline lakes, the finer taxonomic composition of the Cyanobacteria in the two lakes was clearly different. Almost all members of the Cyanobacteria were assigned to the Synechococcaceae in Lake Tuosu, while Lake Dasugan contained the Chroococcaceae as the major part of the Cyanobacteria (see Fig. S1 in the supplemental material). This difference might result from the distinct conditions in the two lakes, such as significantly different Ca²⁺ and SO₄²⁻ concentrations (see Table S1 in the supplemental material). Parvarchaeota, for which no type strain has been isolated, was abundant in Lake Gahai (Fig. 3). This lake is therefore a good candidate for studies involving the isolation of Parvarchaeota, which was proposed (on the basis of phylogenetic analysis of microbial genomes) to be a novel phylum of archaea within the Diapherotrites-Parvarchaeota-Aenigmarchaeota-Nanohaloarchaeota-Nanoarchaeota (DPANN) superphylum having a small genome size and cell size (74). Lake Gahai yielded a large proportion of bacterial sequences that could not be assigned at the phylum level, particularly in samples G7 (30.3%) and G1 (16.2%) (Fig. 3). This lake appears to contain a large proportion of bacterial community members belonging to novel phyla and is a good potential source of novel prokaryotic resources.

The Bacteroidetes was the most abundant phylum in all the studied lakes and accounted for >12.0% of the sequences in each prokaryotic sequencing library, with the exception of that for sample M6 (relative abundance, 6.5%) (Fig. 3). Similarly, many previous studies found the relative abundance of Bacteroidetes to be fairly constant along salinity gradients in inland lakes (12, 60, 75) and estuaries (10, 18, 20, 76, 77). This phylum was reported to be associated with nutrient conversion in lake sediments (78, 79). Its abundance and wide distribution indicate its ecological importance in the lakes that we studied. At the family level, members of the Bacteroidetes in Lake Gasikule were distinct from those in the other six lakes. In this lake, almost all Bacteroidetes were of an unassigned family within the class At12OctB3 and the family Rhodothermaceae (all were members of the genus Salinibacter) (see Fig. S1 in the supplemental material). Perhaps this was because of their high degree of adaptability to hypersaline waters (1, 70). Other phyla abundant in six of the lakes (with Lake Gasikule being the exception) were Proteobacteria, Actinobacteria, Verrucomicrobia, and Planctomycetes. Their relative abundance varied with salinity (Fig. 3).

The results of PCoA and weighted UniFrac tree analysis (UPGMA) showed that Lake Tuosu sample T13 clustered with samples from freshwater Lake Keluke but not with other Lake Tuosu samples (Fig. 2). This interesting finding may reflect the unusual mixed prokaryotic community composition of sample T13, since this sample was taken near the Lake Tuosu inlet, whose water supply is from Lake Keluke. The prokaryotic community composition of sample T13 presumably was a mixture of that from both lakes obtained by water exchange. Unfortunately, only one sample was collected at this site in this study, which was not sufficient for further interpretation of statistically significant differences. The use of large numbers of samples from the Lake Tuosu inlet, as well as along a salinity gradient near the inlet, for investigating the relationship between the prokaryotic community composition of Lake Tuosu and that of Lake Keluke is warranted.

Major environmental factors that structure prokaryotic communities.

Salinity is a dominant environmental selective force governing prokaryotic communities in aquatic systems, such as natural inland lakes (8, 11, 12), solar saltern ponds (9, 24, 25), estuaries (19, 21, 22), and the Baltic Sea (10). Research conducted worldwide found that salinity, rather than extremes of temperature, pH, or other physical and chemical factors, is the major environmental determinant of the microbial community composition (30). Therefore, microbial studies along salinity gradients might shed additional light on the global distribution patterns of microbial community compositions according to changes of salinity.

In the present study, salinity was the most important environmental variable factor governing the variation of the prokaryotic community structures in 46 surface water samples from seven plateau lakes in Qaidam Basin across a salinity range of from 0.8 to 344 g/liter. Few previous studies have examined the effects of specific ionic species with a wide range of concentrations on prokaryotic communities, and even fewer have evaluated such effects at various taxonomic levels. We found significant correlations between the concentrations of individual ionic species (Mg²⁺, K⁺, Cl⁻, Na⁺, SO₄²⁻, Ca²⁺) and the distributions of the community composition (Fig. 4; see also Table S5 in the supplemental material). The effect of the degree of salinity on community structure was correlated with the Mg²⁺, K⁺, Cl⁻, Na⁺, SO₄²⁻, and Ca²⁺ concentrations, and these parameters showed strong positive correlations with each other and with salinity across all samples (see Table S2 in the supplemental material). Dillon et al. found that the concentrations of most major cations and anions in a series of solar saltern ponds increased consistently along a salinity gradient and were related to prokaryotic diversity (25). A study of microbial succession in a hypersaline lake with seasonally fluctuating ionic concentrations showed that the community composition was correlated with the K⁺, Mg⁺, and SO₄²⁻ concentrations, which covaried with each other (80). These results were not surprising, since water salinity was approximately the sum of the concentrations of eight major ions, including K⁺, Na⁺, Ca²⁺, Mg²⁺, Cl⁻, SO₄²⁻, CO₃²⁻, and HCO₃⁻ (57). Besides the individual contributions of salinity and six ions (K⁺, Na⁺, Ca²⁺, Mg²⁺, Cl⁻, and SO₄²⁻) to the community composition (see Table S5 in the supplemental material), VPA further indicated that a combination of salinity and the concentrations of six ions had a significantly stronger effect on the variation of the prokaryotic community structure than the combination of other factors, including pH, the DO concentration, temperature, the PO₄³⁻ and NH₄⁻-N concentrations, altitude, and the TN concentration (Fig. 5). The correlations among numerous environmental parameters complicate the search for the primary controlling factor, although salinity showed the strongest correlation to species variance (see Table S5 in the supplemental material). Overall, salinity, as well as the concentrations of the six major ions, acted as the key contributors affecting the variation of the prokaryotic community structure in 46 surface water samples from seven plateau lakes in Qaidam Basin.

Aside from salinity and the concentrations of its associated ionic species, pH and the DO concentration, which significantly correlated with salinity (see Table S2 in the supplemental material), showed significant correlations with prokaryotic community structures (Fig. 4; see also Table S5 in the supplemental material). In agreement with our findings, many previous studies have shown that pH and the DO concentration are major drivers of microbial community distributions in various ecosystems (22, 81–86). pH has been described to be a master variable that integrates the physiochemical status of aquatic ecosystems (87). Ion concentrations affect oxygen solubility, which is lower in hypersaline waters (88). It is therefore reasonable that pH and the DO concentration showed significant correlations with salinity and with the distributions of the prokaryotic community composition. Taken together, the results indicated that salinity directly or indirectly determines the composition of prokaryotic communities.

We observed no significant effect of temperature or altitude on community structure (P > 0.05) (see Table S5 in the supplemental material). Although many previous studies have found temperature to be a significant determinant of community composition (89–92), we found that it had little effect on the community structures in the plateau lakes of the Qaidam Basin. This might be because all samples were taken simultaneously at a similar temperature. Consistent with our findings for altitude, Wu et al. reported that altitude had no significant effect on the bacterial community structures in inland lakes (12). Although the PO₄³⁻ concentration showed no significant (P > 0.13) correlation with the variation of the prokaryotic community structure at higher taxonomic levels, including the phylum, class, order, family, and genus levels, it significantly (P = 0.04) affected the variation of the OTU composition (see Table S5 in the supplemental material). Further analysis indicated that the PO₄³⁻ concentration also made a significant (P = 0.04) contribution to controlling the OTU composition when an OTU was defined as reads with 99% sequence similarity (data not shown). This result suggested that the PO₄³⁻ concentration tended to affect the compositions of some species or subspecies in samples rather than those of organisms at higher taxonomic levels (phylum, class, order, family, and genus).

VPA indicated that the measured parameters explained 77.9% of the observed variation on the basis of the OTU composition, leaving 22.1% of the variation unexplained (Fig. 5). The unexplained variation may result from ecologically neutral processes of diversification (93) and factors not measured in this study that are known to affect prokaryotic community structures, including the dissolved organic carbon concentration (86, 94, 95) and the chlorophyll a concentration (4, 94, 96).

Correlations between the prokaryotic community composition and certain environmental factors (Cl⁻, Na⁺, SO₄²⁻, and Ca²⁺ concentrations, conductivity, and DO and PO₄³⁻ concentrations) varied depending on the taxonomic level (see Table S5 in the supplemental material). In a study of the bacterioplankton community composition, the relative importance of the NH₄⁺, PO₄³⁻, NO₃⁻, and DO concentrations differed at the order level versus the genus level (22). Similarly, Mantel tests revealed that correlations between the prokaryotic community composition and certain parameters (HCO₃⁻ and Na⁺ concentrations, temperature, pH) differed at the genus (95%), species (97%), and subspecies (99%) levels (91). These results indicate that assessments of the effects of environmental factors on the variation of the prokaryotic community structure would be better with a comprehensive correlation analysis at various taxonomic levels.

Extreme physicochemical stresses shape microbial adaptive responses at the levels of individual clades and entire communities. We found that salinity was the best predictor of the prokaryotic community composition at the levels of entire communities and various taxonomic categories. The distribution patterns at certain levels were consistent with previous observations. The relative abundance of Betaproteobacteria was negatively correlated with salinity (see Fig. S3i in the supplemental material), in agreement with the findings of Wu et al. (12). An increased relative abundance of Gammaproteobacteria with increasing salinity was observed over the salinity range of 0 to 100 g/liter (12, 84), and the abundance of this class was the lowest in Lake Gasikule (salinity, ∼330 g/liter) (see Fig. S3b in the supplemental material). Previous studies have shown relationships between salinity and certain taxa, and the present results provide additional information on the relative abundance of other phyla and classes as a function of spatial salinity gradients. The relative abundance of the Verrucomicrobia (Fig. S3e in the supplemental material), Planctomycetes (see Fig. S3f), Planctomycetia (see Fig. S3g), Sphingobacteriia (see Fig. S3h), and Deltaproteobacteria (see Fig. S3j) decreased with increasing salinity, whereas the abundance of the Archaea was positively correlated with salinity (see Fig. S3u). Most previous studies focused on microbial communities at higher taxonomic levels (phylum or class), but few evaluated the relative abundance of clades at the order, family, and genus levels as a function of increasing salinity. The present study showed correlations between salinity and the abundance of bacterial clades at lower taxonomic levels in plateau lakes by determination of quantitative correlation values. Our quantitative analysis revealed significant negative correlations between salinity and the relative abundances of five orders, five families, and three genera (see Fig. S3k to t in the supplemental material). These clades presumably contribute significantly to the overall variability of entire prokaryotic communities. Experiments with pure isolates from lake samples under controlled laboratory conditions are under way and will help clarify the relationships between salinity and these various clades.

Conclusion.

We conducted a large-scale investigation of the prokaryotic community composition and diversity and the distribution patterns in response to various environmental parameters with 46 water samples from seven plateau lakes in the Qaidam Basin, Tibet Plateau. Salinity and chemical ionic concentrations were found to be the primary environmental factors that directly or indirectly determine the composition and diversity at the level of individual clades as well as entire prokaryotic communities. The effects of environmental factors clearly varied at different taxonomic levels, demonstrating the importance of comprehensive correlation analyses at various taxonomic levels in microbial ecological studies. Our findings clarify the distribution patterns of the prokaryotic community composition at the levels of individual clades as well as whole communities along gradients of salinity and ionic concentrations. This baseline information will be useful for predicting ecological responses to future environmental alterations, such as changes in salinity and ionic concentrations during commercial mineral mining and water resource utilization in plateau lakes, as well as for shedding light on the global distribution pattern of the microbial composition and diversity across a salinity gradient.