Large buoyant particles dominated by cyanobacterial colonies harbor distinct bacterial communities from small suspended particles and free‐living bacteria in the water column

Abstract Worldwide cyanobacterial blooms greatly impair ecosystems in many eutrophic lakes and impact the microbial environment. In particular, large cyanobacterial colonies that are buoyant on the water surface may provide a distinct habitat for bacteria from other small particles that are suspended stably in the water column. To test this hypothesis, bacterial communities (excluding cyanobacteria) attached to large particles dominated by cyanobacterial colonies (>120 μm, LA), small particles (3–36 μm, SA), and free‐living bacteria (0.2–3 μm, FL) were investigated monthly for a year in Lake Taihu, China. Results confirmed that the Shannon diversity index of LA was significantly lower than that of FL, which was lower than that of SA. Cytophagia and Alphaproteobacteria were specially enriched in LA. Although samples in each habitat collected during high‐ (May to November) and low‐bloom seasons (December to April) were separated, all samples in LA were clustered and separated from SA and FL, which were also clustered during the same sampling seasons. In addition, the bacterial communities in LA were correlated with nitrate level, whereas FL and SA were correlated with nitrate level and temperature. Mantel analysis revealed that bacterial composition significantly correlated with the cyanobacterial composition in LA and FL but not in SA. These results indicate that LA provides distinct niches to bacteria, whereas the differentiation of bacterial communities in FL and SA is seasonally dependent.

biomass of cyanobacterial blooms result in the domination of cyanobacterial colonies in the lake's particle composition (Shi et al., 2017). Compared with other abiotic particles and the water column, these colonies are rich in organic matter produced by cyanobacterial cells and offer a "hotspot" for bacterial colonization (Worm & Sondergaard, 1998). Interactions such as nutrient transference (Yuan, Zhu, Xiao, & Yang, 2008), growth promotion or inhibition (Berg et al., 2009;Xie et al., 2016), and toxin production and degradation between bacteria and cyanobacteria are intense in these colonies (Maruyama et al., 2003). Furthermore, these bacteria are closely associated with cyanobacterial bloom formation and nutrient cycling, which greatly affect the development and decline of cyanobacterial blooms (Rashidan & Bird, 2001;Wang, Zhang, Shen, Xie, & Yu, 2015;Wang et al., 2016).
In particular, large cyanobacterial colonies exhibit different physiological and ecological characteristics from small particles composed of eukaryote algae, single-cell/small cyanobacterial colonies, algal detritus, and abiotic particles. First, cyanobacterial colonies larger than 120 μm are often buoyant on water surface (Wu et al., 2010), whereas small particles encounter more difficulty rising onto the water surface and are mainly suspended in the water column (Wu & Kong, 2009;Zhu et al., 2014). Second, large Microcystis colonies (>100 μm) have higher proportions of microcystin-producing genotypes, whereas the smallest size class of Microcystis colonies (<50 μm) has a low proportion of microcystin-producing genotypes (Kurmayer, Christiansen, & Chorus, 2003;Wang et al., 2013). This case is similar to that for other small particles. Third, compared with small particles suspended in the water column, large cyanobacterial colonies have higher polysaccharide content, are more effectively resistant to high-light inhibition (Zhang, Shi, Yu, & Kong, 2011), have higher affinity for low levels of phosphorus (Shen & Song, 2007), and have stronger defense against grazing (Nielsen, 2006;Yang, Kong, Shi, & Cao, 2006). Small particle-attached bacteria may face a considerably different environment from that embedded in large buoyant cyanobacterial colonies. Furthermore, bacteria free living in the water column may be affected by dissolved substances, such as dissolved organic matter and toxins. Therefore, the compositions of bacterial community attached to large cyanobacterial colonies and small suspended particles may differ due to different characteristics, compositions, and positions in the water column; they may also be different from free-living bacteria in the water column (Schmidt, White, & Denef, 2016).
Many works have focused on bacterial communities associated with cyanobacterial blooms (Berg et al., 2009;Berry et al., 2017;Dziallas & Grossart, 2011;Eiler & Bertilsson, 2004;Niu et al., 2011;Tang et al., 2010;Woodhouse et al., 2016). Some works revealed the transition of particle-attached bacteria to free-living bacteria during a 4-month investigation (Tang et al., 2015(Tang et al., , 2017, and others investigated the difference between particle-attached bacteria and free-living bacteria (Zhao et al., 2017). The bacteria attached to medium-and small-sized cyanobacterial aggregates during August and September were clustered, whereas large-and medium-sized aggregate communities in the October sample were grouped together and appeared distinct from the small-sized aggregate community (Cai, Jiang, Krumholz, & Yang, 2014;Cai, Yan, Wang, Krumholz, & Jiang, 2013). Different bacteria attached to Microcystis colonies from freeliving bacteria for several months were also documented (Parveen et al., 2013;Shi, Cai, Kong, & Yu, 2012). However, bacteria involved in three habitats, including large buoyant cyanobacterial colonies, small particles, and free-living bacteria in the water column, have yet to be compared in a duration of 1 year. Therefore, whether this controversial conclusion results from the lack of differentiation among these habitats or sampling season still needs to be determined. In addition, the detailed response of bacterial communities from different habitats to cyanobacterial blooms remains unknown.
In this study, we resolved this problem in Lake Taihu, which is shallow eutrophic, well-mixed, and known for having a long history of cyanobacterial blooms . Satellite images revealed that cyanobacterial blooms are even more intense and occur yearround in Lake Taihu, especially in Meiliang Bay, which is a hypertrophic area of the lake (Ma et al., 2016). Thus, we can obtain large cyanobacterial colonies even in winter. Our goal is to determine whether large buoyant cyanobacterial colonies and small particles provide different habitats for bacterial communities and how they are different from free-living bacteria.

| Sample site description
Lake Taihu is located in the Yangtze Delta in Eastern China (30°55′40″-31°32′58″N, 119°52′32″-120°36′10″E). With a total area of 2338 km 2 and an average depth of 2 m, Lake Taihu is the third largest freshwater lake in China. With increased nutrient inputs into the lake during the past decades, Lake Taihu is eutrophic with blooms of Microcystis (cyanobacteria) occurring annually during warm seasons (Chen, Qin, Teubner, & Dokulil, 2003). Meiliang Bay, located in the northern part of Lake Taihu ( Figure S1), is the most eutrophic area where cyanobacterial blooms break out and last for almost a whole year (Ma et al., 2016;Qin et al., 2007).

| Sampling procedure
Sampling was performed monthly from July 2014 to July 2015 near the regular monitoring stations of the Taihu Laboratory for Lake Ecosystem Research in Meiliang Bay of Lake Taihu. Because Taihu Lake is shallow and well-mixed (McCarthy et al., 2007), sampling of the following two parts was considered as originated from the same environment. For bacteria attached to large cyanobacterial colonies (LA, >120 μm), samples were collected by towing a phytoplankton net (64 μm mesh) through water surface, then the top buoyant cyanobacterial colonies were pipetted and filtered through 120 μm mesh net, and then collected into sterile polypropylene tubes. For bacteria attached to small particles (SA, 3-36 μm) and free-living bacteria (FL, 0.2-3 μm), lake water at 0-0.5 m depth was collected with a water sampler, and was sequentially filtered through 36 μm mesh, 3 and 0.2 μm polycarbonate filters (GTTP, 47 mm diameter; Millipore ), water samples were filtered through precombusted GF/F glassfilbers (47 mm diameter; Whatman), and then the filtrates were analyzed with a continuous flow analyzer (Skalar San++, Netherlands). Total phosphorus (TP) and total nitrogen (TN) were quantified using standard methods (Jin & Tu, 1990). Chlorophyll a (Chl a) was extracted from the GF/F filters with 90% acetone and analyzed with a spectrofluorophotometer (RF-5301PC; Shimadzu, Japan).

| Microscopic examination of LA and SA
Particle composition of LA and SA, representative samples were examined under a JSM-5610LV/Vantage IV scanning electronic microscope (SEM). All samples for SEM were prepared with direct freeze-drying to minimize changes on the cell surfaces induced by chemical reactions such as the chemical fixation and ethanol dehydration, and minimize the in-hand manipulation time (Lee & Chow, 2012). The dried samples were then coated with gold and viewed on the SEM.

| Statistical analyses
The raw data were first quality filtered with QIIME (Caporaso et al., 2010) to remove reads that did not meet the quality control standards. Any chimeric sequence was identified and removed with UCHIME (Edgar, Haas, Clemente, Quince, & Knight, 2011).
To compare the samples, the dataset was randomly subsampled to an equal number of sequences. Nonmetric multidimensional scal- Mantel analysis with Bray-Curtis dissimilarity matrix was used to analyze correlation between cyanobacterial and bacterial composition in each habitat. All these analyses were performed with the "vegan" package (Oksanen et al., 2008) of R software (R Development Core Team 2012)A. Furthermore, to identify characteristic community members on genus level in the three different habitats (LA, SA, FL), we applied the linear discriminant analysis (LDA) coupled with effect size measurements (LEfSe) method (Segata et al., 2011).

| Temporal dynamics of environmental factors and particle composition
From July 2014 to July 2015, water temperature decreased gradually from 32°C in July to 5°C in December, and then gradually increased to 26°C (Figure 1a). Pearson correlation analyses revealed that pH was positively correlated with temperature (R = .864, p < .001), whereas DO was negatively correlated with temperature (R = −.686, p = .01). Chl a varied from 9.7 μg/L in December 2014 to 332.6 μg/L in July 2015 (Figure 1a), and was significantly correlated with TN (R = .735, p = .004) and TP (R = .788, p = .001). So, it was low-bloom seasons from December to April, and was high-bloom seasons from May to November in this lake. SEM analysis confirmed that LA was overwhelmingly dominated by cyanobacterial colonies, whereas SA was dominated by diatoms, single-cell/small cyanobacterial colonies, chlorophyta, algal detritus, and sediment particles. Bacteria attached to these particles can also be viewed via SEM (Figure 2).

| Sequence analyses and cyanobacterial composition
We analyzed the sequence data at two levels and normalized the analysis before and after elimination of the cyanobacterial sequences. The primary result was normalized to 25,123 sequences in each sample. In LA, the relative abundances of cyanobacterial sequences in the total sequences varied from 43.2% to 96.7%, and the proportion tended to be higher from December to April (lowbloom seasons) compared with other months (Figure 1b). In SA, the relative abundances of cyanobacterial sequences in the total sequences varied from 7.4% to 89.6%, which is a wider range than that in LA (Figure 1b). The proportion also tended to be higher from December to April (low-bloom seasons) compared with other months; this result is similar to that in LA. Only 0.1%-13.8% of the total sequences corresponded to cyanobacteria in FL (Figure 1b   Mann-Whitney U test, p < .001), which was higher than that of LA (using nonparametric Mann-Whitney U test, p = .003) ( Figure 3b).

| Diversity of bacterial communities in
NMDS analysis of the 16S rRNA gene sequence frequency data clustered on the OTU level showed a clear separation between LA and both SA and FL (Figure 4). Cluster analysis based on Jaccard's coefficient also revealed that LA was separated from SA and FL ( Figure S2). Furthermore, most samples from May to November (high-bloom seasons) were clustered and separated from that from December to April (low-bloom seasons) in all three habitats ( Figure S2). Moreover, samples collected during the same season (either high-bloom or low-bloom seasons) in SA and FL were much more similar to each other than among samples of the same habitat ( Figure S2). ANOSIM results on the same data showed a  The cladogram that visualizes the output of the linear discriminant analysis coupled with effect size (LEfSe) algorithm, which identifies taxonomical differences between LA, SA, and FL community members, is shown in Figure S4 observed in LA. This observation may be explained by the fact that Dolichospermum is more favorable in lower water temperature than Microcystis (Robarts & Zohary, 1987). However, in SA, uncultured cyanobacteria 1 were predominant in addition to Microcystis in November and December, and to Dolichospermum in April. This finding indicated the different compositions of cyanobacterial genera in LA and SA during the most investigated months. Phytoplankton composition can shape bacterial communities (Niu et al., 2011), and high velocity of large colonies enables them to rise readily onto the surface, whereas small particles are mainly suspended in the water column (Wu & Kong, 2009;Zhu et al., 2014). Therefore, large cyanobacterial colonies and small particles would have different physiological characteristics, thereby providing different habitats for microbes.

| Comparison of bacterial diversity in the three habitats (LA, SA, and FL)
Interestingly, all the three habitats harbored specific bacterial taxa that were not observed in the other two; this result indicates that some bacterial species have high niche specificities (Bertos-Fortis et al., 2016). However, the number of overall OTUs that overlapped between SA and FL was higher than that between LA and FL and between LA and SA. This result indicated that the exchanges between FL and SA were relatively easy and intense, whereas the bacteria in LA were relatively isolated. Although the sampling method of sequential filtering for SA and FL and the specific colony isolation method for LA used in this study may increase the overlaps between SA and FL, significant differences between them were also observed. These results can help exclude the effect of sequential filtering to some extent and assume biological reasons for this overlap, indicating that a narrower spectrum of bacteria can thrive on large colonies. Moreover, a lower diversity of bacterial communities was observed in LA than in FL, which has lower diversity than SA. Consistent with our previous observation, the bacterial communities attached to buoyant Microcystis colonies had lower diversity compared with other bulk bacteria (Shi et al., 2012). These results indicated that LA was different from regular large particles, which may also harbor different communities compared with FL but had selectivity for the attached bacteria. Cyanobacteria can release antimicrobial substances surrounding algal cells (Casamatta & Wickstrom, 2000;Ostensvik, Skulberg, Underdal, & Hormazabal, 1998); thus, some bacteria may not survive in LA. These results indicated that SA composed of diverse particles harbor broader bacterial communities, whereas LA can be assumed to be enriched in selected bacterial communities to some extent.
NMDS result further suggested the distinctive microenvironment provided by LA. This result is consistent with the observation from the comparison of bacterial communities attached to size-fractioned Microcystis colonies collected from August to October 2012 (Cai et al., 2014). Furthermore, LA samples dominated by Microcytsis were separated from the samples dominated by Dolichospermum, indicating that different cyanobacterial genera may also lead to separated bacterial communities, which is similar to the evidence from a previous experiment (Zhu et al., 2016).
These results are consistent with our previous observation that specific bacterial communities are attached to Microcystis spp. ). Moreover, a higher similarity between SA and FL was noted during the same season (high-bloom season or low-bloom season) than that of the same habitat. This result indicated that sampling season also affected the distribution of these two communities. Many studies compared particle-attached bacterial communities and free-living bacterial communities, but controversial observations were also drawn. Some studies observed that particle-attached bacterial communities are phylogenetically distinct from free-living bacterial communities (Allgaier & Grossart, 2006;Zhao et al., 2017), whereas some studies observed that these two communities are similar and may exchange in freshwater mesocosms (Riemann & Winding, 2001;Tang et al., 2015Tang et al., , 2017Worm, Gustavson, Garde, Borch, & Sondergaard, 2001). However, this study further concluded that particles collected during different season may be a major reason for these controversial observations.
The study also evidenced the influence of cyanobacterial compositions of particles on bacterial communities. This notion is similar to the conclusion that different sizes, origins, and phytoplankton compositions may account are responsible for the discrepancies (Schmidt et al., 2016). Furthermore, particles mainly composed of LA may be more distinctive compared with general particles, which may favor distinct bacterial communities.
In addition, dbRDA revealed that NO 3 − significantly correlated with the distribution pattern of bacterial community in all three habitats. The effect of NO 3 − on bacterial community structure was also shown in previous studies of Lake Taihu (Tang et al., 2017), Lake Tanganyika (De Wever et al., 2005), and the mesotrophic Lake Tiefwaren (Roesel, Allgaier, & Grossart, 2012). NO 3 − may directly or indirectly affect bacterial proliferation through cyanobacterial abundance and composition, which are greatly affected by nutrient concentration (Xu, Paerl, Qin, Zhu, & Gaoa, 2010). Moreover, the bacterial communities in SA and FL rely more on the temperature in the water column compared with those in LA. In addition, the samples collected during low-bloom seasons (from December to April) formed a separate cluster from others during high-bloom seasons (from May to November) in all the three habitats ( Figure S2). A significant correlation was noted between bacterial composition and cyanobacterial composition in LA and FL but not in SA. These results indicated the seasonal dynamics of bacteria in all the three habitats and the close association of bacterial communities in LA and FL with cyanobacterial blooms. In contrast, the bacteria in SA faced a much more complex microenvironment.

| Phylogenetic composition and dynamics of bacterial communities in three habitats (LA, SA, and FL)
The result that Alphaproteobacteria and Cytophagia dominated in LA was similar to those of studies on bacterial community attached with a diatom bloom (Riemann, Steward, & Azam, 2000). Some Proteobacteria and Bacteroidetes are well adapted to the phycosphere of phytoplankton and are specialized for successive decomposition of algal-derived organic matter (Teeling et al., 2012  . Many Cytophagia degrade macromolecules, such as proteins, chitin, pectin, agar, starch, or cellulose (Reichenbach, 2006). Most species of Burkholderiales utilize a variety of organic and amino acids as carbon sources (Garrity et al., 2005). Dominance of these bacteria, which vary in their ability to utilize different types of organic matters, suggested that LA may form a specific habitat, where intense bacteria algal interaction may facilitate organic matter cycling and nutrient generation, thereby benefitting algal growth.
In SA, Actinobacteria were predominant during most months, but Proteobacteria predominated from November to April when cyanobacterial blooms declined and began to form, indicating that the bacterial community in SA was also influenced by the different periods of cyanobacterial blooms. Moreover, the bacterial community in FL was predominated by Actinobacteria from July 2014 to May 2015 but shifted to Proteobacteria from June and July 2015 when cyanobacterial blooms were intense, indicating the influence of cyanobacterial blooms on bacterial community in FL. The result that Actinobacteria were dominant in FL was consistent with previous observations during the study of bacterial communities associated with organic aggregates in Lake Taihu (Tang et al., 2009). The result was also consistent with the results obtained from the comparison of free living and particle associated bacterial communities in the four lakes of Northeastern Germany (Allgaier & Grossart, 2006). However, our result further indicated that Actinobacteria are dominant in FL and in SA but not yet in LA. This result was different from some other data (Schmidt et al., 2016). Lake Taihu received a large amount of particles from surrounding terrestrial environment (Dokulil, Chen, & Cai, 2000), and these terrestrial particles may be involved in SA.
Thus, the presence of Actinobacteria, which is well-known from soil environments, may be one reason for its dominance in SA.
These results may facilitate further recognition of the ecotypes of Actinobacteria.
Particularly, Nitrospira was reported to complete nitrification process and is a key component of nitrogen-cycling microbial communities (Daims et al., 2015;van Kessel et al., 2015). Enrichment of these bacteria in SA indicated that the nitrification process may be involved in this microenvironment. Interestingly, proportions of Planctomycetaceae were significantly higher in SA than that in LA. This result was consistent with our previous observation that Planctomycetaceae were few in buoyant Microcystis colonies (Shi et al., 2012). Although close associations between Planctomycetes and cyanobacterial colonies were observed (Cai et al., 2013;Tang et al., 2010), free-living Planctomycetes were also observed in hypoxic zone induced by Microcystis blooms (Li, Xing, & Wu, 2012).
Actually, association of Planctomycetes and cyanobacterial blooms have only been observed in a lake in Sweden among the investigated 32 sites in three continents (North America, Europe and Asia) (Dziallas & Grossart, 2011), and also only in Lake Erken and Lake Limmaren among the four Swedish lakes with cyanobacterial blooms (Eiler & Bertilsson, 2004). These findings suggested that the association between Planctomycetes and cyanobacterial blooms is not so firm and stable.
In conclusion, this study highlights the dynamics of freshwater microbial communities in a eutrophic lake with cyanobacterial blooms during a whole-year investigation, with regard to both the cyanobacterial and bacterial species in LA, SA, and FL. The compositions of cyanobacterial genera were different among the three habitats. The diversity of bacterial communities in LA was lower than that in FL, which was lower than that in SA. Moreover, different and more narrow bacterial communities were present in LA relative to those in SA and FL. The samples in all three habitats were collected during high-(May to November) and low-bloom seasons (December to April) were separated. Furthermore, the effect of cyanobacterial composition on bacterial communities was observed in LA and FL but not in SA. This work further confirms that sampling season and particles with different characteristics may affect bacterial community composition, and that large buoyant cyanobacterial aggregates harbor specific bacterial communities. Therefore, if we take all the bacterial communities in the water column together to analyze bacterial communities associated with cyanobacterial blooms, then differences among large buoyant cyanobacterial colonies, other small particles, and free-living bacteria may be neglected. Thus, future studies should focus on bacterial communities attached to large buoyant cyanobacterial colonies to elucidate bacterial and cyanobacterial interactions.

ACK N OWLED G M ENTS
We sincerely appreciate two anonymous reviewers for their valuable comments. This work was jointly supported by the National Natural

CO N FLI C T O F I NTE R E S T
None declared.