Sampling cruise was carried out in May 2016. Six Qinghai-Tibetan lakes (Supplementary Figure S1) were selected for this study: Erhai Lake (EHL) is a freshwater lake; Qinghai Lake (QHL) and Tuosu Lake (TSL) are saline lakes; Gahai Lake (GHL), Xiaochaidan Lake (XCDL) and Chaka Lake (CKL) are hypersaline lakes (Yang et al., 2013). The TSL, GHL, and XCDL are located in the Qaidam Basin (QB) hinterland, while the CKL is on the QB fringe and the other lakes (i.e., EHL, QHL) are situated out of the QB (Zheng, 1997) (Supplementary Figure S1). In this study, a total of 18 sampling sites (triplicates from each of the studied lakes) were sampled. At each sampling site, the pH and temperature of lake surface water were measured with a portable SX711 pH meter (SANXIN, Shanghai, China); water turbidity was analyzed using a turbidity meter (HANNA, Romania); In-situ illumination intensity was determined by using a TES1335 light meter (TES, Taiwan, China). Water samples (∼20 mL) for measurements of major ions were collected after filtrating through 0.2-μm Nuclepore filters (Whatman, United Kingdom); water samples (∼20 mL) for measurements of dissolved organic carbon (DOC) were filtered through combusted 0.7-μm Whatman GF/F filters and the resulting filtrate was collected into a dark glass vial pre-acidified with concentrated phosphoric acid (∼40 μL); water samples (∼40 mL each) for measurements of dissolved inorganic carbon (DIC), water samples for total nitrogen (TN) and total phosphorus (TP) were collected into 40 mL dark glass vials without air bubbles, supplemented with saturated mercury chloride (∼40 μL) before covering lid. Additionally, 500 mL water samples were filtered through 0.7-μm glass fiber filters (Whatman, United Kingdom), and the filters were stored in dry ice collected for the measurements of chlorophyll a (Chl-a) concentration. Surface sediments (∼0–1 cm) were collected using a grab-bucket collection sampler in the littoral zones of lakes with water depth of ∼1 meter. The surface sediments were then collected into 50 mL sterilized tubes using sterile spoons for DNA samples. The DNA and Chl-a samples were stored in dry ice in the field and during transportation and then were transferred to a -80°C freezer in the laboratory until further analyses. Other samples (e.g., water samples for major ions, DOC, DIC, TN, and TP) were stored at 4°C during transportation and were analyzed immediately after arrival in laboratory.

Cation and anion concentrations (e.g., K⁺, Na⁺, Ca²⁺, Mg²⁺, SO₄^2-, Cl^-) of the lake waters were measured by using ion chromatography (Dionex DX-600, United States). Salinity was calculated by summing up the concentrations of six major ions including K⁺, Na⁺, Ca²⁺, Mg²⁺, SO₄^2-, and Cl^-. DOC and TN concentrations were measured on a multi N/C 2100S analyzer (Analytik Jena, Germany). DIC was measured by mean of the potentiometric acid titration method (Bradshaw et al., 1981). TP was analyzed by phosphormolybdic acid colorimetry (Neal et al., 2000). Chl-a was measured using a fluorospectrophotometer (Shimadzu Corp., Japan) following an overnight freeze-thaw extraction in 90% acetone (Liu et al., 2006).

Total DNA was extracted from 0.5 g sediment samples using the Fast DNA SPIN Kit for Soil (MP Biomedical, United States). The extracted DNA was amplified with a universal algal 23S rRNA gene primer set p23SrV_f1 and p23SrV_r1, and the detailed PCR conditions were described in a previous study (Sherwood and Presting, 2007). Briefly, a unique 12-bp barcode sequence was added between the sequencing adapter and reverse primer to differentiate among samples. Triplicate PCR reactions for each sample were conducted and the resulting successful PCR products were purified using a DNA Gel Extraction Kit (Axygen, United States). The PCR amplicons (∼400 bp) from each sample were pooled with equimolar concentrations and then were sequenced by using an Illumina-Miseq platform (paired-ends sequencing of 2 × 250 bp) (Caporaso et al., 2012).

The raw 23S rRNA gene sequences were processed following the pipeline coupling USEARCH (Edgar, 2013) and QIIME (Caporaso et al., 2010) software. The paired reads were joined with FLASH (fast length adjustment of short reads) using default setting (Magoč and Salzberg, 2011). Forward and reverse primers were removed from the joined reads. The remaining reads were then de-multiplexed and quality filtered using QIIME v1.9.0 with split_libraries_fastq.py script (Caporaso et al., 2010). Briefly, reads having more than three consecutive low quality (Phred quality score <30) bases were removed, and reads containing ambiguous base were discarded, as well as reads comprising consecutive high quality bases less than 75% of the total read length were culled out. Chimera checking was performed using the UCHIME module with de novo method in USEARCH (Edgar et al., 2011). Singleton and read length less than 200 were discarded, and operational taxonomic units (OTUs ) were defined at the 97% cutoff (Steven et al., 2012) by using the UCLUST algorithm (Edgar, 2010). OTU representative sequences were then selected and their taxonomy were assigned using parallel_assign_taxonomy_blast.py with default set (sequences similarity >90% and blasted exception value<10^-3) against the SILVA 128 LSU database in the QIIME program. Sequences failing to be assigned into Cyanobacteria and eukaryotic algae were removed. In order to validate these assignments of taxonomy, OTU representative sequences were locally BLASTed in NCBI database¹. The BLASTed results were provided in Supplementary Table S1. The final OTU table was rarefied to equal sequence number (n = 8843) for each sample with 1000 times, and then alpha diversity was calculated at the 97% identity level in QIIME. A variety of alpha diversity indices were calculated including Simpson, Shannon, Equitability and Chao1.

All environmental variables in this study were normalized to values ranged between 1 and 100 as described previously (Yang et al., 2016a). The non-metric dimensional scaling (NMDS) ordination with 500 random starts were performed to depict the difference of algal community compositions among lakes based on the Bray-Curtis dissimilarity using the package “vegan.” Cluster analysis was performed according to the Bray-Curtis dissimilarity among samples using PAST software². Simple Mantel tests were performed to assess the Spearman’s correlations between algal community compositions and geographic distance/environmental variables by using the “vegan” package. Geographical distances among sampling sites were calculated based on the GPS locations of each sites using Euclidean method in PAST software (Supplementary Table S2). Canonical correspondence analysis (CCA) was also performed to explore the relationships between algal communities and environmental and spatial variables. Before the CCA, a set of spatial variables were generated through the method of principal coordinates of neighbor matrices (PCNM) analysis according to the longitude and latitude coordinates of the sampling sites (Borcard and Legendre, 2002). Subsequently, we used a forward selection procedure to select environmental and spatial variables through the ‘ordiR2step’ function in R package “vegan” (Blanchet et al., 2008). Only significant (p < 0.05) environmental and spatial variables were shown in the CCA ordination.

In order to discern the difference between benthic and planktonic algal community composition in lakes, planktonic algal 23S rRNA gene sequences were collected from the two published studies (Steven et al., 2012; Liu X. et al., 2016). To avoid any bias resulting from different primers, only 23S rRNA gene sequences derived from the same primer set (p23SrV_f1 and p23SrV_r1) and the same PCR protocol were included in this analysis. Sequences were processed according to the procedures described above. NMDS ordination with 500 random starts were conducted to discern the difference between benthic (this study) and planktonic (previous studies) algal community compositions in lakes according to the Bray-Curtis dissimilarity. In addition, the dominant OTU representative sequences of Cyanobacteria (average relative abundance >0.1%) were selected to perform BLAST³ against available 23S rRNA genes in the GenBank. Meanwhile, their closest references were retrieved for constructing phylogenic tree. All the OTU representative sequences were aligned with their references by using Clustal W implemented in the Bioedit program. Maximum-likelihood tree was constructed from the representative cyanobacterial 23S rRNA sequences and their references by using the MEGA 6.0.

The sequence data generated in this study were deposited at the Sequence Read Archive (SRA) in the National Center for Biotechnology Information (NCBI) under the BioProject PRJNA376846 with accession no. SRP101378.

The studied lakes have a large range of environmental parameters (Supplementary Table S3). For example, the salinity was 0.8–365.6 g/L and the pH was 7.4–9.2; DOC was 1.9–26.3 mM and DIC was 6.2–29.8 mM; turbidity was 0.6–17.4 NTU (Nephelometric Turbidity Unit) and the concentration of Chl-a was 0.1–21.4 μg/L; TN and TP ranged 97.7–441.1 and 3.8–13.7 μM, respectively.

A total of 2, 244, 853 qualified sequence reads were obtained after data processing, and 89.1% of total qualified sequence reads (1, 999, 487 sequence reads) were assigned to Cyanobacteria and eukaryotic algae. The algal sequences per sample ranged from 8,843 to 764,193 with an average of 101, 082. Alpha diversity indices of the studied samples were summarized in Table 1. The number of the observed algal OTUs of the studied samples ranged 9.0–65.2 with Shannon indices and Chao 1 being 0.6–3.9 and 9.0–83.4, respectively (Table 1). These alpha-diversity indices were not significantly correlated with any environmental parameters of the studied lakes (data not shown). The relative abundances of Cyanobacteria and eukaryotic algae ranged 0.2–99.2 and 0.8–99.8% among the samples, respectively (Figure 1). Cyanobacteria was dominant (relative abundance >45%) in the samples of EHL, QHL, and CKL, whereas eukaryotic algae largely dominated (relative abundance >60%) in the samples of TSL, GHL and XCDL (Figure 1). The algal communities of the studied samples were composed of seven dominant (relative abundance >5% at least in one sample) orders (i.e., Bacillariales, Ceramiales, Naviculales, Oscillatoriales, Spirulinales, Synechococcales, and Vaucheriales) (Figure 2). Algal sequences belonging to Synechococcales were dominant in the samples of EHL, QHL, and CKL, whereas Ceramiales sequences dominated in the samples of TSL, GHL, and XCDL (Figure 2).

The clear geographic patters of benthic algal community were observed among the studied lakes (Figure 3): the benthic algal communities in the lakes (i.e., TSL, GHL, XCDL) within the Qaidam Basin (QB) hinterland were grouped into the G1 cluster, and those from EHL, QHL, and CKL were grouped into the G2 cluster (Figure 3A). Similar grouping patterns were also observed in the NMDS and CCA ordination, which showed that the distributions of algal communities in TSL, GHL, and XCDL (G1) were separated from those of EHL, QHL and CKL (G2) along the axis NMDS1 (Figures 3B, 4). The CCA result also indicated that many local environmental (i.e., DOC, DIC, illumination intensity, TN, TP, turbidity, water temperature) and spatial (i.e., PCNM1, PCNM3, and PCNM4) variables significantly (p < 0.05) affect the benthic algal distribution in the studied lakes (Figure 4). Furthermore, Mantel tests showed that the algal community compositions of the studied samples were significantly (p < 0.05) correlated with DIC, DOC, geographic distance, pH, salinity and water temperature (Table 2). Geographic distance possessed higher correlation coefficient than other environmental variables (e.g., salinity, pH, etc.). Linear analysis gave a R² of 0.341 (p < 0.001) between the dissimilarities of benthic algal community among lakes and geographic distances (Figure 5).

Non-metric dimensional scaling (NMDS) ordinations on the basis of both presence-absence and abundance data showed that benthic (this study) and planktonic (previous studies) algal community compositions were distinctly different (Figure 6). Moreover, the algal community composition in American lakes were different from that in Chinese lakes (Figure 6). Additionally, phylogenetic analysis indicated that the dominant cyanobacterial OTUs (average relative abundance >0.1%) were mainly affiliated with Synechococcales and Oscillatoriophycideae (Supplementary Figure S2A) and the relative abundances of those OTUs ranged 0–74.5% (Supplementary Figure S2B) in the studied lakes. Some cyanobacterial OTUs (e.g., OTU98, OTU248, OTU134) occurred in both water and sediment samples, whereas other OTUs only occurred in either water (e.g., OTU80, OTU135) or sediment (e.g., OTU13, OTU14) samples (Supplementary Figure S2B).