Prokaryotic Community Distribution along an Ecological Gradient of Salinity in Surface and Subsurface Saline Soils

Salinity effects on microbial communities in saline soils is still unclear, and little is known about subsurface soil microbial communities especially in saline or hypersaline ecosystems. Here we presented the survey of the prokaryotic community in saline soils along a salinity gradient (17.3–148.3 dS/m) in surface (0–10 cm) and subsurface (15–30 cm) saline soils of Qarhan Salt Lake, China. Moreover, we compared them with three paired nonsaline normal soils. Using the high-throughput sequencing technology and several statistical methods, we observed no significant community difference between surface soils and subsurface soils. For environmental factors, we found that TOC was the primary driver of the prokaryotic community distribution in surface saline soils, so was pH in subsurface saline soils. Salinity had more effects on the prokaryotic community in subsurface saline soils than in surface saline soils and played a less important role in saline soils than in saline waters or saline sediments. Our research provided references for the prokaryotic community distribution along a salinity gradient in both surface and subsurface saline soils of arid playa areas.

in surface soil layers (0-10 cm). No previous reports of microbial communities in subsurface saline soil layers (15-30 cm) seems to exist.
The Qaidam Basin is located in the northeast of Qinghai-Tibetan Plateau, China. There are twenty seven salt lakes in the basin; the playas and salt lakes take up about a quarter of the whole basin area 20 . Among all 27 lakes, Qarhan Salt Lake is the largest playa, where 70% of the total thickness of the upper salt-bearing deposits is composed of the halite layers 20 . It is the largest industrial base for potassium fertilizer in China, and it is also rich in Mg, Li, and many other salt resources. Its unique significances in scientific research, resource exploitation, and environmental protection are increasingly being recognized 21 . Many previous studies have focused on the prokaryotic communities in waters [22][23][24][25] or sediments 26,27 of the lakes in Qinghai-Tibetan Plateau, and they all observed that salinity was the primary driver of the microbial community distribution. However, the prokaryotic community distribution in saline soils and its correlations with environmental factors, especially the salinity, are still uncharacterized in this area.
The purposes of this research are: 1) to clarify the prokaryotic community distribution along several environmental gradients, especially the salinity gradient, in both surface and subsurface saline soils; 2) to characterize the prokaryotic community composition in saline soils and compare with that in nonsaline normal soils; 3) to ascertain the microbial resources for further exploitations.

Results
Physical and chemical properties of soils. The 23 saline soils from Qarhan Salt Lake showed a wide salinity gradient from 17.3 dS/m to 148.3 dS/m, a pH gradient from 7.36 to 8.92, a TOC (total organic carbon) gradient from 4.19% to 21.48%, and a water content gradient from 1.1% to 32.3% (Table 1).The data indicated a wide environmental range, especially the salinity range. In eight paired surface-subsurface saline soils, surface soils had higher salinity (except SS1 and SD1, SS8 and SD8), lower water contents (except SS4 and SD4), and lower pH values (except SS7 and SD7, SS8 and SD8) than subsurface soils ( Table 1). The six nonsaline normal soils encompassed few variations of environmental factors compared with saline soils. Spearman's rank correlation was conducted between EC (electrical conductivity) and other environmental factors (Supplementary Sample ID pH EC (dS/m) TOC (%) WC (%) K (g/kg) Ca (g/kg) Mg (g/kg) Na (g/kg) locations (N,E) Altitude (m)  Table 1. Geographical, physical, and chemical properties of soils. ND: not detected. Abbreviations: "SS", "SD", "US", "UD", "NS", and "ND" represent shallow saline soil (surface saline soil), deep saline soil (subsurface saline soil), unpaired surface saline soil, unpaired subsurface saline soil, surface normal soil and subsurface normal soil, respectively. EC: electrical conductivity, TOC: total organic carbon, WC: water content.
Scientific REPoRtS | 7: 13332 | DOI:10.1038/s41598-017-13608-5  Table S1). We also summarized Q20, Q30, GC%, and the number of effective tags (Supplementary Table S1).  Table S1). The mean of OTUs in normal soils was more than twice as many as that in saline soils, which indicated low microbial diversity in saline soils. The sum of OTUs in saline soil samples (surface saline soils: 12075; subsurface saline soils: 11689) was higher than that in normal soils (surface normal soils: 10804; subsurface normal soils: 9683), and the sum of OTUs in surface soils was higher than in subsurface soils of both saline soils and normal soils ( Fig. 1). Moreover, shared OTUs took up more than 70% between surface normal soils and subsurface normal soils (7612/10804), surface saline soils and subsurface saline soils (8591/12075). And shared OTUs took up about 50% between surface normal soils and surface saline soils (6295/12075), subsurface normal soils and subsurface saline soils (5329/11689) (Fig. 1). The number of OTUs shared by all four groups was 3550.
Prokaryotic community composition and alpha diversity analysis. Archaea and Bacteria accounted for an average percentage of 52.8% (12.3-86.7%) and 47.2% (13.3-87.7%) of all 1,335,650 effective tags respectively in saline soils, with a huge difference compared with 6.8% (2.9-12.1%), 93.2% (87.9-97.2%) in normal soils. Detailed relative abundance of each phylum/class were summarized in Supplementary Table S3. Saline soils showed low microbial complexity compared with normal soils, and there were high percentages of Euryarchaeota in most saline soil samples (Fig. 2).
For Archaea, almost all effective tags were assigned to Halobacteria of Euryarchaeota in saline soils. But in normal soils, about 25% of Archaea tags were assigned to Thaumarchaeota except Halobacteria of Euryarchaeota (Fig. 2). ANOSIM results showed significant prokaryotic community differences between surface saline soils and surface normal soil (R = 0.37, P < 0.05), subsurface saline soils and subsurface normal soils (R = 0.36, P < 0.05). To further identify the phyla/classes which contributed primarily to the community variations, we conducted SIMPER analysis between surface normal soils and surface saline soils, subsurface normal soils and subsurface saline soils (Supplementary Table S4). The overall community dissimilarity (the sum of contribution) between surface normal soils and surface saline soils was 68.9%, and the value was 72.1% between subsurface normal soils and subsurface saline soils at the phylum/class level. Euryarchaeota made the greatest contribution to the community variations in both surface and subsurface soils between saline soils and normal soils. Gammaproteobacteria and Deltaproteobacteria played a more important role in differentiating the surface soil community, but Alphaproteobacteria contributed more in subsurface soil community (Supplementary Table S4).
As it is shown in Fig. 3, microorganisms in surface soils are richer than those in subsurface soils for both saline soils and normal soils. Surface soils had more OTUs than their corresponding subsurface soils except SS1 and SD1, SS2 and SD2 (Supplementary Table S1). Though we observed more Gammaproteobacteria in surface saline soils than in subsurface saline soils, and Bacteroidetes and Gemmatimmonadetes in surface saline soils were less  than those in subsurface saline soils by summarizing prokaryotic community compositions in all paired saline soils ( Supplementary Fig. S1). ANOSIM results showed no significant (P = 0.8 for normal soils and P = 0.26 for paired saline soils) prokaryotic community difference between surface soils and subsurface soils, and Student's t-test results of each phylum/class were also unsignificant (P > 0.05) between the two groups in both saline soils and normal soils.
Rarefaction curves of most samples (except UD3, SS8, and SD8) did not reach an obvious asymptote, indicating that there were still many undetermined species (Supplementary Fig. S2). Chao1 estimators suggested that our sequencing efforts contained about 54.8% ± 7.8% of the estimated diversity for saline soils and about 64.8% ± 1.3% for normal soils (Supplementary Table S1), which indicated larger proportions of unknown species in saline soils than in normal soils. The curves of six normal soils were obviously above curves of 23 saline soils, demonstrating low microbial diversities in saline soils ( Supplementary Fig. S2). Shannon diversity index (5.9 ± 1.5 in saline soils and 10.2 ± 0.3 in normal soils) also verified that (Supplementary Table S1).
Correlations between environmental factors and the prokaryotic community distribution. Two main axes explain 77.2% of the soil community variations, and soil samples can be clustered together by both the sizes and colors of points, indicating salinity, pH, and TOC can all affect the prokaryotic community distribution (Fig. 4). CCA biplots showed that pH, TOC, and EC were main factors that affected the prokaryotic community in saline soils: TOC was the strongest factor in surface saline soils (Fig. 5a), so was pH in subsurface saline soils (Fig. 5b).
To further verify the main drivers of the prokaryotic community distribution, we conducted Mantel test between the prokaryotic community and each environment factor of saline soils in Qarhan Salt Lake ( Table 2). Only EC, pH, and TOC had significant (P < 0.05) correlations with the prokaryotic community distribution in all saline soils. TOC and pH affected the prokaryotic communities in both surface saline soils and subsurface saline soils significantly, but EC only had significant effects on prokaryotic communities in subsurface saline soils (Table 2). Moreover, the r value of TOC (r = 0.54) were higher than that of pH (r = 0.49), suggesting TOC was the main driver of the prokaryotic community distribution in surface saline soils, so was pH in subsurface saline soils. In summary, both CCA and Mantel test proved that TOC was the main driver of the prokaryotic community distribution in surface saline soils, and pH was the main driver of that in subsurface saline soils of Qarhan Salt Lake, China.
Salinity effects on the prokaryotic community. Most samples can be clustered together according to their saline level ( Fig. 4 and Supplementary Fig. S4). ANOSIM results further verified the significant prokaryotic community difference across different saline levels (R = 0.69, P < 0.01). Halobacteriaceae (belonging to Euryarchaeota of Archaea) and Moraxellaceae (belonging to Gammaproteobacteria of Bacteria) have deeper red grids than other families, suggesting they are two most abundant families in saline soils ( Supplementary  Fig. S4). Half of top 40 most abundant families belongs to Proteobacteria, indicating the widespread distribution of Proteobacteria in soils. Moreover, normal soils apparently had more complex communities and higher evenness than saline soils (Supplementary Fig. S4).
Though TOC and pH were the main drivers, salinity can still strongly affect the prokaryotic community in saline soils. EC have significant correlations with the prokaryotic community distribution in subsurface saline soils but not with that in surface saline soils ( Table 2), suggesting that salinity has more effects on subsurface  Fig. S3): medium saline soils (group S2 and D2) had higher percentages of Euryarchaeota and lower percentages of Proteobacteria, Bacteroidetes than low saline soils (group S1 and D1) or high saline soils (group S3 and D3) in both surface and subsurface soils. Within Proteobacteria, lower percentages of Gammaproteobacteria were in medium saline soils than in low saline soils or high saline soils for subsurface soils, and the same trend was observed between medium saline soils and high saline soils but not low saline soils for surface soils (Supplementary Fig. S3). That was probably because group S1 of surface soils contained only one soil sample (sample US2) with relatively high salinity (52.1 dS/m), suggesting that it was probably unrepresentative. We also observed significant correlations between salinity and percentages of several taxa across all saline soil samples ( Supplementary Fig. S5). Salinity significantly (P < 0.05) affects the percentages of Euryarchaeota and Proteobacteria, though R 2 values were relatively low: percentage of Euryarchaeota peaks at about 85 dS/m; Proteobacteria reaches the lowest at also about 85 dS/m (Supplementary Fig. S5a). Gammaproteobacteria showed the same trend with Proteobacteria ( Supplementary Fig. S5b).

Discussion
The extremely arid desert climate with high evaporation-precipitation ratio and strong sun exposure is observed in the area of Qarhan Salt Lake. These characters made the unique prokaryotic community in soils of Qarhan Salt Lake. Several reports showed that Bacteria were numerically dominant relative to Archaea in saline soils of the areas they explored 10,28,29 , and there were also reports showing that Archaea were dominant 30,31 . In the present study, we found that Archaea were dominant in saline soil samples with EC about 58 dS/m to 93 dS/m; Bacteria were dominant in soil samples with lower or higher salinity except three outliers (SD6, SD3, SS3) in Qarhan Salt Lake (Fig. 2). Almost all archaeal tags were assigned to Halobacteria of Euryarchaeota, which agreed with several previous microbial community studies of hypersaline waters 22,32 , saline soils or sediments 10,26,33,34 . Also, saline soil samples in which Archaea were dominant has low alkalinity (pH: 7.36-8.02) and low TOC contents (4.19-6.22%) except SS8 (Fig. 2 and Table 1). That agreed with the fact that most halophilic archaeal representatives (mainly Halobacteria) isolated from hypersaline lakes are aerobic, neutrophilic, and some are alkaliphilic [35][36][37] .   Table 2. Mantel test results between the prokaryotic community and environmental factors in saline soils of Qarhan Salt Lake. Significant (P < 0.05) correlation was labeled as boldface.
Scientific REPoRtS | 7: 13332 | DOI:10.1038/s41598-017-13608-5 And, specific membrane structures and catabolic pathways allows Archaea to out-compete Bacteria when faced with energy limitation, such as low nutrients in extreme conditions 38 . For Bacteria in saline soils of the present study, Proteobacteria were the most abundant phylum, and within Proteobacteria, Gammaproteobacteria were the most abundant class, followed by Alphaproteobacteria, Deltaproteobacteria, and Betaproteobacteria, which was consistent with a previous meta-analysis of sequences in saline soils 39 . Except Proteobacteria, Bacteroidetes and Gemmatimonadetes were the other two most abundant phyla ( Supplementary Fig. S1), agreed with previous reports 10, 28 .
For Archaea in nonsaline soils from the eastern Tibetan Plateau, the Halobacteria were dominant only in dry soils; the most abundant phylum was Thaumarchaeota 40 . Moreover, in nonsaline normal soils from Tianjin, China in the present study, Archaea only accounts for a small part, agreed with a previous report 41 . The most abundant archaeal phylum was Halobacteria, followed by Thaumarchaeota. Proteobacteria, Acidobacteria, and Actinobacteria were the three most abundant phyla of Bacteria ( Supplementary Fig. S1), consistent with the bacterial community composition in Tibetan permafrost soils 42 . For Proteobacteria in normal soils, Alphaproteobacteria were the most abundant phylum, followed by Gammaproteobacteria, Deltaproteobacteria, and Betaproteobacteria.
Though with relatively high percentage of shared species (about 50%) (Fig. 1), significant community difference was observed, and Euryarchaeota contributed the most to the community variations in both surface and subsurface soils between saline soils and nonsaline soils. Because almost all archaeal tags were assigned to Halobacteria of Euryarchaeota, Archaea and Halobacteria had similar contributions to Euryarchaeota. Nonsaline normal soils had more complex community and higher evenness than saline soils ( Fig. 2 and Supplementary  Fig. S4). Nonsaline normal soils were also more diverse than saline soils (Supplementary Table S1 and Fig. 3). Keshri J. et al. has also found that the community in nonsaline soils was significantly different from that in saline soils and more diverse than saline soils 18 . Those also agreed with and supported the general belief that relatively low microbial diversity existed in "extreme" environments 43 . Within Proteobacteria, most saline samples (16/23) were dominant by Gammaproteobacteria, and Arit S. de León-Lorenzana et al. 44 has observed a sharp relative abundance decrease of Gammaproteobacteria by flooding a saline soil, suggesting Gammaproteobacteria was more adaptive to saline environments than other proteobacterial classes. Moreover, it confirmed the widespread distribution of Halobacteria of Archaea, Proteobacteria of Bacteria in soils.
We observed trends of higher salinity, lower water contents, and lower pH in surface soils than in subsurface soils according to physical and chemical properties of eight paired surface-subsurface saline soils (Table 1), which was probably caused by strong sun exposure and high evaporation in Qarhan Salt Lake. In addition, we needed to point out that trends of salinity and pH between surface soils and subsurface soils were unsignificant (t-test, P > 0.05), and the water contents in subsurface soil were significantly (P = 0.048) higher than those in surface soils. Though with variations of environmental factors, microorganisms in surface soils were richer than those in their corresponding subsurface soils ( Fig. 3 and Table 1). According to Hollister E. B. et al., the water effect on microbial community was primarily due to its influence on soil oxygen concentrations 10 . And another reports suggested the decrease of biomass along the increasing depth was initially caused by redox state (depth of a few centimeters), and further caused by other variables, such as nutrients and stress (the anoxic zone) 45 . Therefore, we hypothesized that low oxygen concentrations, which might be affected by high water contents, caused low microbial richness in subsurface soils (15-30 cm). That also agreed with previous reports 12,13,15 microorganisms in surface soils were richer than those in their corresponding subsurface soils.
Different prokaryotic diversities were observed, but we observed no significant community difference between surface soils and subsurface soils in both saline soils and normal soils according to the results of ANOSIM and Student's t-test, contrary to a previous report 11 . Therefore, we hypothesized that the depth of 15-30 cm probably was not deep enough to generate significant community difference from the surface soil (0-10 cm).
In the present study, though saline soils had much fewer diversities than normal soils (Supplementary Table S1, Fig. 3), however we didn't observe decreasing microbial richness (observed OTUs) with increasing salinity (r = 0.015, P = 0.947) in saline soils, just as previous studies have observed 22,25,46 . Moreover, salinity had more effects on subsurface saline soils than surface saline soils according to the result of Mantel test (Table 2). That was probably because compared with subsurface soils, surface soils were more easily subject to other environmental factors 47 , such as strong sun exposure, precipitation or oxygen. The prokaryotic community composition in soils with different saline level differed from each other (Supplementary Fig. S3): higher percentage of Euryarchaeota and lower percentage of Proteobacteria, Gammaproteobacteria existed in medium saline soils (group S2 and D2) than in low saline soils (group S1 and D1) or high saline soils (group S3 and D3) for both surface and subsurface soils. Nonlinear fitting curves also verified the same trend: Proteobacteria reaches lowest at about EC 85 dS/m contrary to the percentage of Euryarchaeota (Supplementary Fig. S5a); Gammaproteobacteria showed the same trend with Proteobacteria ( Supplementary Fig. S5b), which was possibly caused by the large proportion of Gammaproteobacteria in Proteobacteria. That agreed with a previous study on soils from the former lake Texcoco, which showed more Gammaproteobacteria clones were in low (0.65 dS/m) and high (158 dS/m) saline soils than in medium (56 dS/m) saline soils 48 .
Soil samples were well clustered by pH, TOC, and salinity (Fig. 4); pH and TOC all had significant correlations with the prokaryotic community distribution in both surface saline soils and subsurface saline soils (Table 2). Therefore, we can infer that the prokaryotic community in both surface saline soils and subsurface saline soils were affected by similar environmental factors, agreed with a previous research on the bacterial community distribution in nonsaline soils of the western Tibetan Plateau 11 .
Most previous studies focused on the microbial community in saline waters and sediments, and salinity has been characterized as one of the primary factors of microbial community distribution in aquatic environments 7,8,22 . Many previous studies have observed that salinity had the strongest effect on the prokaryotic community distribution in waters 22,23,25 , sediments 26,27,49 . In the two previous reports on saline soils 9,10 , the former showed salinity had the strongest effect on the bacterial community structure, but the latter showed the microbial community distribution was correlated with other factors better, including organic carbon contents, water contents, pH, and phosphorus contents. Also, another report of archaeal communities in saline soils showed that archaeal community structures considering phylogenetic information were correlated well with pH 34 . In the present study, by summarizing the results of PCoA, CCA, and Mantel test, we found that TOC was the main driver of prokaryotic community distribution in surface saline soils; pH was main driver in subsurface saline soils of Qarhan Salt Lake. Therefore, we confirmed that salinity played a more important role on the microbial community distribution in saline waters or saline sediments than in saline soils; the microbial community in saline soils was more sensitive to other environmental factors, such as TOC, pH, or water contents.

Conclusion
We collected 23 saline soils from playa of Qarhan Salt Lake to clarify the prokaryotic community distribution along several environmental gradients in both surface (0-5 cm) and subsurface (15-30 cm) saline soils. We showed high-resolution differences between saline soils and nonsaline soils in both prokaryotic diversities and communities. And we found that TOC was the main driver of the prokaryotic community distribution in surface saline soils, so was pH in subsurface saline soils; salinity had more effects on subsurface saline soils than surface saline soils. Our finding can provide references for sampling strategy and prokaryotic community compositions in saline soils of arid areas. Also, the distribution patterns of the prokaryotic community and taxa, especially some halophilic taxa (mainly Halobacteria of Archaea), along the environmental gradients in saline soils provided references for further new gene exploitations, such as some halophilic or halotolerant enzyme genes. At last, further in-depth investigations on saline soil microbial communities in large scale or with greater depth were expected.

Methods
Site climate description and the sample collection. Qarhan Salt Lake has a mean annual temperature of 5.33 °C, a mean annual precipitation of 24 mm, and a mean annual evaporation of 3,564 mm, which indicates an extremely arid desert climate. The climate data were achieved from Qarhan meteorologic station.
In summer 2016, 15 playa sites of Qarhan Salt Lake were selected with a minimum distance of 100 m as sampling sites. The sampling strategy was similar to a previous reports 28 : a) maximum reachable sites in the interior playa even without roads; b) different locations with various salinity levels. Most locations of the playa were bare zones dotted with scarce vegetations, including Lycium ruthenicum and Phragmites australis. For each site, surface soils (0-10 cm) and subsurface soils (15-30 cm) were collected using a metal auger with a 4-cm diameter. For each soil sample, four samples collected in the vertices of 1-meter side square were mixed into a representative sample. If present, stones, solid salt crust, and roots were removed before sampling. Three paired nonsaline normal soils were collected with the same sampling strategy in Water Park, Tianjin, China. The vegetation of the normal soil locations included Cynodon dactylon, Ophiopogon japonicas, Setaria leucopila, etc, and we selected the locations with few grass to avoid the influence of grassroots. Locations and altitudes were recorded using a GPS. Collected soils were placed into 50 ml sterile plastic centrifuge tubes and stored at 4 °C during transportation. For each sample, one portion was air dried, 2 mm sieved, then physically and chemically analyzed; the other portion was immediately stored at −80 °C for further DNA extraction.
Physical and chemical analysis. The pH was measured in a 1:2.5 (w/w) aqueous solution with a pH meter (PHS-3C; INESA, Shanghai, China). The electrical conductivity (EC) was measured in a 1:5 aqueous solution with a conductivity meter (FE-30; Mettler Toledo, Switzerland). The water content was detected by oven drying fresh soil to a constant weight at 105 °C. Potassium dichromate heating oxidation method 50 was used to determine the total organic carbon content (TOC). The concentrations of Mg, K, Na, Ca were measured with an atomic absorption spectroscopy (TAS-990; PERSEE, Beijing, China). DNA extraction and sequencing. PowerSoil DNA Extraction Kit (Mo Bio Laboratories, CA, USA) was used to extract DNA from soils (about 0.25 g) following the manufacturer's instructions. But we failed to extract enough DNA from seven samples among all 15 paired saline soil samples due to the low biomass in saline soils. And for the seven failures, we have tested three different kinds of soil weight (0.15 g, 0.25 g, and 0.5 g), all resulting in no success. Primers 515 F and 806R 41 , fused with a barcode and Illumina adaptor, were used to amplify V4 regions of prokaryotic 16S rRNA genes. The PCR products were sequenced using Illumina HiSeq. 2500 platform 51 (250 bp pair-end reads) by Novogene (Beijing, China).

Analysis of Illumina-sequencing data. Barcodes and primers of generated Illumina-sequencing reads
were removed, and the reads were merged using FLASH software 52 , generating raw tags. Raw tags were qualified using QIIME (version 1.9.1) 53 software package: a) raw tags, which had three or more consecutive low-quality bases (the threshold value was 19), will be truncated at the first low-quality base; b) some tags, whose consecutive high-quality base length is shorter than 75% of tags length, were further filtered. Then, chimeras were identified and filtered through blasting against the ChimeraSlayer reference database 54 by UCHIME Algorithm 55 . The effective tags were finally obtained. The obtained effective tags were used for further OTU (operational taxonomic unit) picking and diversity analysis with QIIME. OTUs were picked using the preferred open-reference 56 OTU picking strategy in QIIME, which includes both close-reference and de novo OTU picking strategy; USEARCH 57 was used to cluster sequences with ≥ 97% similarity together forming an OTU. A representative sequence was picked out from each OTU, and the taxonomy was assigned to each of the representative sequences with Ribosomal Database Project (RDP) classifier 58 against Greengenes 59 at a confidence threshold of 0.8 in QIIME. Representative sequences were aligned against the Greengenes core set with PyNAST 60 , then a phylogenetic tree was generated with FastTree 61 , which was used to calculated the weighted UniFrac distance 62 .
Before diversity analysis, the sequences of each sample were normalized to 47504 tags based on the number of tags in the sample with the fewest tags. Several alpha diversity indices, which includes observed OTUs, Chao1 richness estimator, and Shannon diversity index, were calculated with QIIME. Principal Coordinate Analysis (PCoA) based on weighted Unifrac distance matrix was conducted with the function "cmdscale" of package Stats in R (version 3.3.1). Heat map was drawn using the package Pheatmap in R. Canonical correlation analysis (CCA), analysis of similarity (ANOSIM), similarity percentage (SIMPER) analysis 63 , and Mantel test were conducted with the package Vegan 64 in R. We chose CCA according to detrended correspondence analysis (DCA) and removed environmental factors with variance inflation factors of more than ten from CCA to reduce effects of the collinearity. Mantel test was performed to analyze the correlation between the prokaryotic community composition (Bray-Curtis distance based on OTU-level communities) and each environmental factor (Euclidean distance). ANOSIM was used to evaluate significant differences of the prokaryotic community composition among different groups.
Data availability. The original sequences of this research were deposited at the NCBI Sequence Read Archive under the accession number SRP108198. The accession number of each sample was also listed (Supplementary Table S1).