Bacterial abundance and diversity in pond water supplied with different feeds

The abundance and diversity of bacteria in two types of ponds were investigated by quantitative PCR and Illumina MiSeq sequencing. The results revealed that the abundance of bacterial 16S rRNA genes in D ponds (with grass carp fed sudan grass) was significantly lower than that in E ponds (with grass carp fed commercial feed). The microbial communities were dominated by Proteobacteria, Cyanobacteria, Bacteroidetes, and Actinobacteria in both E and D ponds, while the abundance of some genera was significantly different between the two types of ponds. Specifically, some potential pathogens such as Acinetobacter and Aeromonas were found to be significantly decreased, while some probiotics such as Comamonadaceae unclassified and Bacillales unclassified were significantly increased in D ponds. In addition, water quality of D ponds was better than that of E ponds. Temperature, dissolved oxygen and nutrients had significant influence on bacterial communities. The differences in bacterial community compositions between the two types of ponds could be partially explained by the different water conditions.

Scientific RepoRts | 6:35232 | DOI: 10.1038/srep35232 high-throughput sequencing, such as in marine water 19 , natural wetland 20 , soil 21 , drinking water 22 , activated sludge 23 and hot spring 24 . Therefore, these recent advances in DNA sequencing technologies allow us to develop a broader understanding of complex microbial communities affected by adding carbon sources into aquaculture system.
In this study, we hypothesized that adding carbon sources into aquaculture system will not only improve water quality but also change microbial communities in water column. To validate this hypothesis, we compared three typical ponds using commercial feed (control) with two ponds using sudan grass as a carbon source in the Gong'an county of Jingzhou, Hubei Province, China. Water samples were collected every month from June to October 2014 to explore the microbial communities using MiSeq Illumina sequencing technique. The findings of this study will help to understand the impact of the additional carbon sources on microbial communities in aquaculture ponds, and evaluated the impact of adding carbon sources on culture environment.

Results
Environmental description. The characteristics of the surface and bottom water samples from two different feeding types of ponds within five months are summarized in Table 1. As observed, there were no significant differences in the pH values and DO concentrations no matter between surface and bottom or between two types of ponds (p > 0.05). The NH 4 + -N and NO 2 − -N concentration were also found no significant differences between the surface and bottom from the two type of ponds (p > 0.05). However, both NH 4 + -N and NO 2 − -N concentrations were found significantly lower in the ponds D than that in the ponds E (p < 0.05). Variations of water temperature, and the NH 4  Fig. 1. Temperature of the pond water ranged from16.9 ± 0.25 to 32.4 ± 0.06 °C, with the highest record in August. Concentrations of ammonia and nitrite exhibited the same variations in the grass carp fed with sudan grass (D) and those fed with commercial feeds (E). Ammonia and nitrite concentrations were higher in Ponds E than those in Ponds D.
Abundance of bacterial 16S rRNA genes. The abundance of bacterial 16S rRNA genes is shown in Fig. 2 as determined by qPCR from two types of ponds. The abundance of bacterial 16S rRNA genes in all samples had the same changing trend: it showed a decrease first, followed by an increase, a decrease and then an increase again (Fig. 2). The bacterial abundance ranged from3.53 ± 0.62 × 10 9 to 8.63 ± 1.01 × 10 10 copies/mL and from2.81 ± 0.64 × 10 9 to 7.42 ± 0.87 × 10 10 copies/mL for D ponds and E ponds, respectively.
In D ponds, the abundance of bacterial 16S rRNA genes was significantly higher in the surface water layer than in the bottom layer in August (p < 0.05), while no significant differences were observed between them in other months (p > 0.05). Similar results were obtained for E ponds.
The grass carp were fed with sudan grass in July and August in D ponds, and there were no significant differences in the abundance of bacterial 16S rRNA genes between D ponds and E ponds in these two months (p > 0.05). However, in September when the feed was changed to commercial feed again, the abundance of bacterial 16S rRNA genes in D ponds was significantly lower than that of E ponds (p < 0.05), and no differences were observed between the two types of ponds in October with further supplement of commercial feed (p > 0.05).
Illumina MiSeq sequencing results. After the removal of unqualified reads, the Illumina MiSeq sequencing analysis of fifty water samples yielded 745,337 sequences, with an average length of about 396 bp in the V4-V5 hypervariable regions of the 16S rRNA gene. As summarized in Table 2, the number of operational taxonomy units (OTUs) in the E ponds was larger than that in D ponds no matter in the surface or bottom. By contrast, the Shannon indices showed higher bacterial diversity in D ponds than those in E ponds both on the surface and bottom. It was also noted that bacterial diversity in D ponds was greater than that in E ponds at the bottom layer in August. The rarefaction curves of the twenty samples at the 3% distance cutoff level revealed that the bacterial phylotype richness of sample 7Es was considerably higher than that of other samples (Fig. 3).
Bacterial community composition. RDP Classifier was used to assign the effective sequence tags into different phylogenetic bacterial taxa. Figure 4 shows the relative abundance of bacterial community at the level  Table 1. Environmental parameters of the surface and bottom water samples. All data are presented as means ± standard deviation (SD); n = 10, 10, 10, and 15 for Ds, Db, Es, and Eb, respectively. In the sample name, the capital letters D and E represent the ponds in which grass carp were fed with sudan grass and commercial feed, respectively; the lower case s and b represent the surface and bottom water layer, respectively. Lowercase letters (a-d) indicate significant differences between two groups (p < 0.05).     Table 2. Diversity indices from 20 water samples. The numbers of reads after quality control and noise clearance are shown. The number before the sample name represent the sampling month. In sample name, the capital letters D and E represent the ponds in which grass carp were fed with sudan grass and commercial feed, respectively; the lower case s and b represent the surface and bottom water layer, respectively. layer of D ponds, Cyanobacteria were first increased and then decreased during August to October, while in other three sampling sites it had the opposite varying tendency.
To further understand the differences in dominant phyla between D and E ponds, the mean relative abundance of the four dominant OTUs at each sampling site in the five months was analyzed ( Supplementary Fig. S1). Proteobacteria were found to be the most dominant phylum in all samples (except for the samples in July) followed by Actinobacteria, Bacteroidetes and Cyanobacteria. As observed, the relative abundance of Proteobacteria and Actinobacteria in D ponds were both higher than those in E ponds. However, the relative abundance of Bacteroidetes in D ponds was lower than that in E ponds. The relative abundance of Cyanobacteria fluctuated in D and E ponds, and showed no particular trends. For the rest, Firmicutes and Chlorobi were more abundant in E ponds than that in D ponds (4.0% and 1.1%, respectively), whereas Chloroflexi and Planctomycetes were less abundant.
To conduct a more detailed analysis of the composition of the communities in the water layers, all the reads had been assigned to a phylum into classes (Fig. 5). As observed, Betaproteobacteria were the dominant class among Proteobacteria in all water samples, accounting for 15.7%, followed by Alphaproteobacteria, Gammaproteobacteria and Deltaproteobacteria. Among these classes, Gammaproteobacteria were more abundant in D ponds (9.4% in surface layer, 8.5% in bottom layer) than in E ponds (5.9% in surface layer, 7.6% in bottom layer). However, the mean relative abundance of the five subdivisions within Proteobacteria in the two types of ponds was not significantly different over the five months (p > 0.05). Interestingly, in D ponds, Epsilonproteobacteria were only detected in June and July, averaging at 0.006% in total effective sequences, whereas in E ponds, they were detected in June, July and August, averaging at 0.003%. In addition, Deferribacteres and Thermotogae were only detected in the surface water of E ponds in June.
Hierarchically clustered heatmap showed the similarities and differences of these twenty bacterial communities at genus level (Fig. 6). Cluster analysis classified the samples into five clusters. Except for the samples from September, other samples from the same types of the pond were grouped together first, and then the samples in the same month from different types of the pond were clustered together.
At genus level, some interesting differences between the two types of ponds during the five months were observed. Some probiotics, such as Comamonadaceae unclassified and Bacillales unclassified were found to be significantly higher in D ponds than those in E ponds (p < 0.05) in July and August when the feed was changed to sudan grass feed. On the other hand, some potential pathogens such as Acinetobacter and Aeromonas were found to be significantly lower in D ponds than those in E ponds (p < 0.05) in August.
Similarity analysis of the twenty water samples. The weighted UniFrac clustering method was used to calculate the similarity or dissimilarity of the obtained sequences among different samples 25 . As shown in Fig. 7, based on abundances of orders, the bacterial communities in the twenty samples could be clustered into five groups, which included all four samples from one of the five months respectively: Group I (September), Group II (August), Group III (October), Group IV (July), Group V (June). Except for those from the samples in September, the communities from the same type of ponds were clustered together. The capital letters D and E represent the ponds in which grass carp were fed with sudan grass and commercial feed, respectively. The lower case s and bin the sample name represent the surface and bottom water layer, respectively). Taxa represented occurred at >1% abundance in at least one sample.
PCoA was used to estimate the similarities among different water samples using three different approaches: RDP Classifier taxa, OTUs and UniFrac, and PCoA (Fig. 8). The first principal coordinate of the weighted analysis accounted for 47.19% of the variation in the data. It clearly separated the samples in June from those in other four months. PC2 accounted for 20.79% of the variance in the bacterial communities.
Microbial community composition in relation to environmental variables. CCA was used to establish the relationships between the environmental factors and the bacterial community (Fig. 9). CCA plot was carried out using OUTs data together with environmental data (ammonia, nitrite, nitrate, total phosphorus, total organic carbon, temperature, pH, and dissolved oxygen). According to Monte Carlo permutation test (499 permutations), the significant relationships between environmental variables and canonical axes were analyzed by using Canoco program.
Based on the 5% level in a partial Monte Carlo permutation test, the bacterial community and structure were significantly linked (p < 0.05) to the water environment factors. As shown in Fig. 9, the water samples were clearly clustered according to sampling time rather than sampling water layer. CCA results explained 36.8% and 18.9% of the variation in the first two axes, respectively (Fig. 9). Dissolved oxygen, temperature and total organic carbon were the most important environmental factors to influence the water community composition, and were positively correlated with Axis 1 (p < 0.05) ( Table 3). Axis 2 had a positive correlation with temperature, nitrite and nitrate (p < 0.05), but was negatively correlated with ammonia (p < 0.01) ( Table 3), suggesting that Axis 2 had a gradient in temperature, nitrite, nitrate and ammonia. By contrast, other nutrient factors (pH and total phosphorus) had no significant correlation to bacterial community (p > 0.05).

Discussion
In this study, we showed the vertical stratification of the 16S rRNA bacterial abundance in the water layers of two different types of aquaculture ponds. The bacterial 16S rRNA gene abundance was significantly higher in August than in other four months in both types of ponds. And the highest temperature was observed in August. These results indicate that temperature may play an important role in the seasonal dynamics of bacterial abundance. Similar phenomenon was observed in other aquaculture ponds, where total bacterial abundance varied with temperature 26 .
After changing the types of feeds, the bacterial abundance in E ponds was higher than that in D ponds (with sudan grass supplied in July and August) in September (p < 0.05) and October (p < 0.05), respectively. But there was no significant difference in bacterial abundance between D and E ponds in July and August. These results reveal that different types of feeds can influence the bacterial abundance, and possibly the effect of supplying sudan grass on the bacterial abundance was slightly delayed. The higher bacterial abundance in E ponds in September and October may be due to the trophic state of the ponds, as previous studies have demonstrated that the total number of bacterial cells observed in three different sediments may be influenced by the reservoir trophic state 27 .  The number of OTUs in each sample was estimated with the largest 16S rRNA libraries sequenced to date. However, a lot of "Unclassified" sequences were also detected in the samples, suggesting that there might be a high abundance of unknown microbial lineages in the aquaculture pond environment, which should be studied in detail in the future. At the 3% distance cutoff level, the OTUs ranged from 699 to 1153 in this study. The Shannon diversity values ranged from 3.97 to 5.31 and the highest was observed in 8Ds. These values are in accordance with those reported for some other aquaculture systems 28,29 . However, according to the study of the microbial diversity of seawater in the East China Sea, the Shannon diversity values were found in the range of 3.42 to 5.65 30 . Another study showed low diversity values, ranged from 2.4 to 3.4 when studying the microbial community of a warm monomictic tropical freshwater lake 31 . These results indicate that aquaculture system may harbor higher microbial species richness than other aquatic systems.
Seasonal pattern of bacterioplankton community composition has been reported for aquaculture systems 32 , river 33 , lake 34 , coastal water 35 and the open ocean 36 . However, these previous studies were mostly based on fingerprinting methods, which only include the most abundant members of the community. The present study is the first attempt to apply Illumina MiSeq sequencing to study the seasonal dynamics, including both abundant  and rare populations of the observed OTUs in two different types of aquaculture ponds. We showed the vertical stratification of the microbial communities in the pond water layers over five months, and two water layers corresponding to the surface and bottom water bodies were investigated. We hypothesized that there would be significant differences among the water samples from different months. It was shown that the studied microbial community was featured by strong temporal shifts and seasonal clustering. As demonstrated by the clustering, PCoA, and CCA analyses, the water samples from the same sampling month were fairly similar, while the samples from different months were different. This difference may be due to the difference in water conditions in different months, which determine the microbial community distribution. Previous studies have demonstrated that different temperatures and oxygen concentrations significantly alter the microbial community composition in freshwater sediment 37,38 . One potential driver of seasonal change in community composition is temperature 39 , and this can also be the case for pond water, as the growth of microbes in this study was at least seasonally affected by temperature. This result is consistent with the results of studies on phytoplankton communities, which showed that pelagic bacterial communities experienced climate-driven seasonal environmental changes 40 . The CCA result in our study also demonstrates that the temperature is positively correlated with the water community composition. Thus, temperature seems to be an important factor that affects microbial communities.
Many previous studies on the composition of microbial communities in the water have relied heavily on clone library analysis, which only sequences rather few16S rRNA gene fragments 11,41,42 . A previous study analyzed the bacterial diversity in the influent from a municipal waste water treatment plant using high-throughput sequencing method 43 . Their results showed that Proteobacteria were the most abundant phylum in the influent sample, which is consistent with our study. However, the clustering of some phyla in the present study is consistent with the findings of the study of the surface water samples in sewage treatment plants 44 , which revealed that the main phyla were Proteobacteria, Actinobacteria, and Bacteroidetes based on an examination of bacterial community structures by 454 sequencing, but the relative abundance differed from that in the present study. Their results showed that the dominant phyla were Proteobacteria (20.28-67.89%), Bacteroidetes (3.85-16.14%), Acidobacteria (19.78-53.59%), and Cyanobacteria (0.68 to 2.6%) in surface water samples, while our results revealed that Proteobacteria were the most abundant phylum in all twenty water samples and accounted for 24.28-48.22%, followed by Cyanobacteria (5.9-37.61%), Bacteroidetes (5.4-34.82%), and Actinobacteria (8.2-22.37%). Another study reported the bacterial community composition in the Red Sea based on 454 pyrosequencing, and revealed the vertical stratification of the microbial communities in the water layers above the Atlantis II and Discovery  Deeps. Their study showed that the classified bacterial reads from the upper layers (20 and 50 m) were dominated by Cyanobacteria, whereas in the deeper layers (200 and 1500 m), the largest group was Proteobacteria 19 .
The microbial compositions at different depths seemed to be very similar in both types of ponds, and no significant differences of the four dominant phyla (Proteobacteria, Cyanobacteria, Bacteroidetes, and Actinobacteria) were observed between the surface and bottom layers. These results are similar to those of a study using 16S rRNA pyrosequencing in seawater 19 , which showed that the microbial composition in the surface (20 and 50 m) seems to be very similar. This phenomenon may be due to the similarity of water conditions in the surface and bottom water layers. Freshwater pond is a shallow water body, in which the nutrient concentrations showed no significant difference between the surface and bottom water layers as shown in Table 1. Similar results were also reported in intensive GIFT tilapia (Oreochromisniloticus) ponds 45 , which demonstrated that the microbial compositions of four different water layers were similar.
There were no significant differences in the four dominant phyla (Proteobacteria, Cyanobacteria, Bacteroidetes and Actinobacteria) between the two types of ponds over the five months. Similar phenomenon was also observed in two rivers, where two clone libraries of 16S rDNA constructed with summer samples from each river were not significantly different and contained typical freshwater bacterioplankton of Betaproteobacteria, Bacteroidetes, and Actinobacteria 33 . However, as demonstrated by the clustering and PCoA analyses, the communities from the same type of ponds were clustered together in this study, except for the samples in September. Besides, except for in the water samples in June, Proteobacteria were found to be the dominant phylum in the samples from other months, which is similar to the analytical results of bacterial communities in drinking water 46 . The classified bacterial reads of the samples from D ponds were dominated by Proteobacteria, which were decreased in E ponds in each depth (Fig. 5). Proteobacteria contain a very high level of bacterial metabolic diversity related to global carbon, nitrogen and sulfur cycling 47 . The effects of different types of feeds on the distributions of the proteobacterial subdivisions in the ponds were also studied. Here, Betaproteobacteria were the most abundant class in the Proteobacteria phylum. This is different from the results of a study using 454 pyrosequencing 19 , which showed that Gammaproteobacteria were the most abundant Proteobacteria in sea water. However, our finding is similar to the analytical results of bacterial communities in soil 48 and activated sludge 18 , which demonstrated that Betaproteobacteria were the most abundant in Proteobacteria. The classified bacterial reads of the samples from D ponds were dominated by Gammaproteobacteria, which were decreased in samples from E ponds (Fig. 5). A previous study showed the shift of bacterial community structure to a relatively higher abundance of Gammaproteobacteria due to the addition of organic carbon substances 49 . In our study, the TOC concentration (Table 1) in D ponds was slightly higher than that in E ponds, which might result from the supply of sudan grass for D ponds in July and August. A decrease of Actinobacteria in the samples of E ponds was observed compared within the samples of D ponds. Here, both the nitrate and ammonium concentrations were higher in E ponds than in D ponds, which is supported by a study which demonstrated that both the nitrate and ammonium concentrations negatively affect the abundance of Actinobacteria 50 . Bacteroidetes were enriched in water samples of E ponds. This might be explained by the reason that the ammonia concentration was higher in E ponds than in D ponds, which is supported by a study showing that Bacteroidetes are positively correlated with ammonia concentration in a hot spring 51 .
Comamonadaceae unclassifiedand and Bacillales_unclassified were significantly enriched in D ponds compared with in E ponds when the feed was changed to sudan grass feed. Comamonadaceae unclassified was are likely to be nitrate-reducing bacteria 52 and could decrease organic carbon in the ponds. Their higher abundance in D ponds might be related to the higher TOC concentration in D ponds resulted from the addition of available organic carbon for denitrification. Some species of Bacillus are used as probiotics, such as Bacillus subtilis, Bacillus licheniformis, in the aquaculture system. Acinetobacter and Aeromonas, many species of which are pathogenic bacteria in aquaculture ponds 53,54 , were significantly abundant in E ponds in August. This result indicates that sudan grass is effective to decrease Acinetobacter and Aeromonas. The reasons for the higher abundance of Bacillus, Acinetobacter, and Aeromonas in E ponds still need further study.
CCA in this study shows that DO, temperature and nutrients (TOC, nitrate, ammonium and nitrite) significantly influence the composition of bacterial communities in pond water. A previous study showed that DO, temperature, pH and nutrients (total nitrogen, total phosphorus) are the key factors influencing the bacterioplankton diversity of Lake Taihu 55 . Another study showed that ammonium, chemical oxygen demand and total nitrogen have effects on the composition of bacterial communities in Litopeneaus vannamei aquaculture water 56 . CCA plots revealed that samples collected in the same month were almost clustered together. This result suggests that the bacterial communities might demonstrate a seasonal pattern, which is consistent with the previous studies 57, 58 .
In conclusion, the community structure differed between the ponds supplied with different types of feeds and the ponds supplied with sudan grass had better water conditions. The pathogenic bacteria genera Acinetobacter and Aeromonas were significantly decreased in the ponds supplied with sudan grass. In addition, the microbial communities in the ponds had a seasonal pattern in our study. These results suggest that the modulation of diets might influence the community structure and thus might effectively change water quality. Our findings could provide a promising direction for the healthy aquaculture of grass carp. grass carp (Ctenopharyngodon idellus) were raised as the major species. At the time of sampling, the ponds had been used for 3 years with stable production under intensive cultural management. Five ponds differed mainly on the supply of fish feeds: three of them were provided with commercial feed from June to October 2014, and the other two were provided with commercial feed in June, September, October 2014, and sudan grass (Sorghum sudanense) in July and August 2014, respectively.

Materials and Methods
The sampling was conducted in 2014 on June 16 th , July 16 th , August 28 th , September 21 st , and October 14 th . Samples were taken from the surface and bottom water column, which corresponded to 0.2 and 1.6 m of depth, respectively. Five samples (200 mL) were taken from five evenly distributed points within each pond then mixed for each depth. These water samples were collected, homogenized and sub-sampled for further analysis. One part of these water samples were used to measure the physicochemical factors immediately. The other part samples were placed in an incubator box with ice packs until further processing in the laboratory and these water samples were filtered through a 0.22-μ m nucleopore filter (diameter: 47 mm) within 12 h after water collection, then kept at − 80 °C until DNA extraction. The ponds supplied with sudan grass or commercial feed were designated as D ponds and E ponds, respectively. Thus, the water samples were designated as: Ds and Db (the surface and bottom water of D ponds, respectively), Es and Eb (the surface and bottom water of E ponds, respectively). The number before the sample name represents the sampling month (for example, 6Ds refers to the surface water sample of D ponds collected in June).
Physicochemical analysis. Temperature (T), pH, and dissolved oxygen (DO) concentration were measured using an HQ 30d multi-parameter water quality analyzer (HACH, Loveland, CO). Total phosphorus (TP) was determined with ammonium molybdate spectrophotometric method. Total organic carbon (TOC) was determined on a Total Organic Carbon Analyzer (Elementar, Hanau, Germany). Ammonia (NH 4 + -N), nitrite (NO 2 − -N), and nitrate (NO 3 − -N) were determined by the methods described by Lu et al. 55 .
Quantitative PCR (qPCR). Total genomic DNA of each sample was extracted as previously described 59 .
The concentration of the extracted DNA was quantified with a ND-2000 UV-vis spectrophotometer (USA). Quantitative PCR was carried out using a Qiagen Q thermo cycler (Qiagen, Hilden, Germany). The reaction mixture (20 μ L) contained 10 μ L SYBR Premix Ex Taq II, 0.2 μ M of each primer and 1 μ L template DNA (10 ng). The specific primers for the amplification of bacterial 16SrRNA genes were 515F (5′ -GTGCCAGCMGCCGCGG-3′ ) and 907R (5′ -CCGTCAATTCMTTTRAGTTT-3′ ) as previously described 60 . The qPCR reactions were performed as follows: 95 °C for 30 s; 35 cycles of 95 °C for 5 s, 55 °C for 30 s, followed by 72 °C for 1 min. The specificity of the qPCR amplification was determined by melting curve and gel electrophoresis.
The standard curve was constructed using plasmid DNA as the standard sample. The plasmid DNA was extracted from the positive colony, which was confirmed by that bacterial 16S rRNA genes were successfully ligated into the pMD18-T Vector (Takara, Dalian, China). The concentration of the plasmid DNA was determined with a ND-2000 UV-vis Spectrophotometer and bacterial 16S rRNA gene copy numbers of plasmid DNA were calculated using the detected concentration. The standard samples were produced by ten-fold serial dilution of plasmid DNA. The standard curve was observed to have a correlation coefficient (R 2 ) of 0.995 and an efficiency of 93%.
Illumina MiSeq sequencing. In this study, we collected samples for five consecutive months (from June to October) to conduct Illumina MiSeq sequencing. All PCR products of 16S rRNA genes were performed in triplicate using Illumina MiSeq Sequencer (Illumina) by Majorbio Bio-pharm Biotechnology Co. Ltd. (Shanghai, China) as described previously 61 . PCR products were pooled and purified using an Axy Prep ™ DNA Gel Extraction Kit (Axygen). All the sequences used in this study are available from the NCBI Sequence Read Archive (SRA) under accession number SRA3016262.
Bioinformatics and statistical analysis. Sequencing data were processed using the Quantitative Insights Into Microbial Ecology (QIIME) pipeline (http://qiime.sourceforge.net/) 62 . The sequences with the same barcode were assigned to the same sample, and then the barcode and primer sequences were removed. Denoised sequences with one mismatched base in the barcode, overlapped shorter than 10 bp, containing ambiguous characters, with more than two mismatched bases in the primers, or with a sequence length shorter than 50 bp were eliminated. The chimeric sequences were removed from aligned sequences using the UCHIME method 63 . And the valid reads obtained from Illumina MiSeq sequencing were normalized to 20000 for comparison of community diversity. The reads were then clustered into operational taxonomic units (OTUs, 97% similarity) 64 . The Greengenes data base was used to determine the taxonomic identity of each phylotype 65 . The diversity indices ACE and Chao1 66 were estimated using Mothur 67 to indicate the community richness. Simpson and Shannon indices were estimated to indicate the community diversity. Coverage, which represents the sequencing depth, was calculated using custom R scripts. The differences in overall bacterial community structure between each pair of water samples were detected using the UniFrac metric 25 . Heatmap was constructed with R software. PCoA (principal coordinate analysis) was used to characterize the changes of the community composition by using R software.
All data are presented as the means ± standard deviation (SD); n refers to the number of samples. Statistical analysis was performed using the SPSS 21.0 software package. The differences in the abundance of bacterial 16S rRNA genes and in the four dominant phyla among the samples from the two types of ponds were evaluated by one-way ANOVA. The differences between the relative abundance of genus from the two types of ponds were also evaluated by one-way ANOVA. CCA (Canonical correspondence analysis) plot and Monte Carlo Permutation test were used to analyze the relationship between the pond water community composition and the environmental factors using CANOCO 4.5 program.