Algal blooms could cause severe ecological problems in both marine and fresh water and have both a high frequency and large distribution (Yang C. et al., 2012; Woodhouse et al., 2016). Blooms reduce the water quality (Cai et al., 2013a,b) (e.g., restrict light penetration and increase water viscosity), increase the economic costs of water treatment, change food web dynamics (Turner and Chislock, 2010) and threaten human health (Paerl and Otten, 2013). The relationship between bloom-forming algae and other water microbes has received great attention because they are reported to closely interact; the diversity and community composition of water microbes can be significantly influenced by blooms (Yang C. et al., 2012, 2015; Woodhouse et al., 2016) but can also affect bloom development (Grossart et al., 2005, 2006; Yang C. et al., 2012; Cai et al., 2013a).

Water microbes consist of numerous diverse species (Cai et al., 2013a) and play key roles in biospheric biogeochemical cycling (Woese, 1994). In water ecosystem, bacteria can live either freely or attached to particle (e.g., algae) surfaces (Jasti et al., 2005), with obvious differences in community composition, phylogeny and metabolism (Riemann and Winding, 2001; Grossart et al., 2005; Cao et al., 2015). Free-living bacteria (FLB) are the primary degraders of dissolved organic matter (DOM) in aquatic ecosystems (Karner and Herndl, 1992), and particle-associated bacteria (PAB) are important for the remineralization of both dissolved and particulate organic matter (Tang et al., 2009). PAB are usually associated with colloids and algae (Shao et al., 2014) and are considered algae-associated microbes, particularly during algal blooms (Riemann and Winding, 2001; Cai et al., 2013b), that were shown to have positive or negative effects on algal growth (e.g., algicidal bacteria) in a previous study (Berg et al., 2009). Microbes do not live separately but rather interact with each other to form ecological networks and accomplish ecosystem functions (e.g., material and energy flows, ecosystem stability) (Deng et al., 2015; Shi et al., 2016). Therefore, explaining and predicting the variation in network structure are essential parts of microbial ecology (Zhou et al., 2010) and are useful for guiding practical applications, including biodiversity preservation and ecosystem management. Although the importance of FLB and PAB during algal blooms has been widely addressed (Tuomainen et al., 2006; Cai et al., 2013b; Shao et al., 2014; Cao et al., 2015; Yang C. et al., 2015), the bacterial interactions within the two fractions are still unknown.

Our understanding of microbial interactions in natural environments is limited because most of these microbes are unculturable (Zhou et al., 2010). Culture-independent molecular technology has developed rapidly in recent years and has been acknowledged as a feasible approach to study complex microbial communities. Emerging high-throughput sequencing technologies can provide a comprehensive understanding of the specific microbial species that live and thrive in various environments (Logares et al., 2014), and network-based approaches have been successfully used to evaluate the influence of environmental disturbances on the whole community (Raes and Bork, 2008; Zhou et al., 2010) and analyze microbial interactions in diverse habitats, including marine (Wang Y. et al., 2015), soil (Tu et al., 2015), acid mine drainage (Kuang et al., 2016) and groundwater environments (Deng et al., 2015). However, relevant research on microbial interactions during algal blooms is still limited, especially for the relationship between algae and associated bacteria. Woodhouse et al. (2016) found that microbial communities were closely associated with cyanobacterial composition by using datasets of cyanobacterial cell densities, bacterial community composition and abiotic parameters to demonstrate the interactions of abiotic and biotic factors during cyanobacterial blooms in a shallow ephemeral lake. Nevertheless, the network interaction of PAB or FLB with algae during blooms is still largely unknown.

To fill this gap, in this study, we focused on the following questions during a cyanobacterial bloom in a freshwater lake in the Yunnan-Guizhou plateau: (1) the differences in community composition between PAB and FLB during cyanobacterial blooms; (2) how they respond to environmental fluctuation; and most of all, (3) how the microbes interact in PAB and FLB, particularly considering the role of cyanobacteria in the network interactions.

In this study, we found significant differences between PAB and FLB communities in terms of diversity, taxonomic composition and microbial network interactions. The greater difference observed between fractions (i.e., PAB vs. FLB) rather than sites was consistent with previous bacterial community studies during a diatom bloom (Riemann and Winding, 2001; Rieck et al., 2015). Although PAB community structure was quite conserved across the environmental changes in surrounding water, the FLB community was significantly correlated with the measured environmental variables (Table 2).

Cyanobacteria release considerable amounts of organic matters to surroundings and these organic substrates can attract a lot of heterotrophic bacteria into the phycosphere (Cai et al., 2013a). Because the compositions of these exudates from different algal species are distinct (Sapp et al., 2007; Paver et al., 2013), the divergent availability and physicochemical characteristics of these substrates result in different bacterial communities in the specific algal phycosphere (Louati et al., 2015). Only bacteria from the aquatic system adapting to this microenvironment live and persist in the phycosphere. When M. aeruginosa dominated the phytoplankton of Dianchi Lake, phycosphere selection theory could well explain the stability and lower diversity of PAB in this conserved phycosphere (Louati et al., 2015). The regional water environment fluctuated in different sites, including organic and inorganic matter (Table 1). This could lead to significant differences in FLB communities from the four sites.

For those genera whose relative abundance significantly changed between PAB and FLB communities, their most likely ecological physiological functions were investigated based on taxonomy according to the literature. The relative abundance of bacteria affiliating with nitrogen-fixing taxa Candidatus Competibacter (Mcilroy et al., 2014), Pleomorphomonas (Xie and Yokota, 2005; Im et al., 2006) and Bradyrhizobium (Kaneko et al., 2002) significantly increased in the PAB community, while the abundance of a potential nitrogen fixer affiliating with Paenibacillus (von der Weid et al., 2002) significantly increased in FLB community. Inorganic N is essential and beneficial for cyanobacterial growth; also algal density was found significantly positive correlations with the concentrations of NO₂-N and NO₃-N in this study. In our previous study, the nifH gene (nitrogen fixation) intensity was also found increased when algal bloom of Akashiwo sanguinea occurred (Yang et al., 2016). Nitrogen-fixing microorganisms and algae can form symbiotic relationships, e.g., azotobacteria provided cyanobacteria with inorganic N in exchange for organic exudates from algae (Lili et al., 2014) and Xu et al. (2016) found the cell density of Bradyrhizobium japonicum significantly increased when co-cultured with transgenic Chlamydomonas reinhardtii hemHc-lbac. This mutually beneficial relationship may have resulted in a significantly higher relative abundance of nitrogen-fixing bacteria in the PAB community than in FLB.

For the bacteria known for methane-utilizing, methanotrophy genera, the relative abundance of bacteria affiliating with methanotrophy genera Methylomonas (Koh et al., 1993), LD28 freshwater group (Salcher et al., 2015) and Candidatus Methylacidiphilum (Khadem et al., 2012), and methanotrophic genus Methylocaldum (Murrell, 2010) significantly increased in the FLB community, while the abundance of one filamentous methane oxidizer affiliating with Crenothrix (Stoecker et al., 2006) significantly increased in the PAB community. The relative abundance of some potentially algicidal bacteria of M. aeruginosa also significantly changed, e.g., Aquimonas (Cui et al., 2012) was more abundant in PAB community, and Pedobacter (Yang et al., 2012) and Aeromonas (Liu et al., 2012) were more so in FLB community. For potentially organic matter-degrading bacteria, Flavobacterium (O’Reilly and Crawford, 1989), Novosphingobium (Tiirola et al., 2002), Sphingopyxis (Xu et al., 2015), their relative abundances significantly increased in FLB community, while Labrys (Carvalho et al., 2005) was more abundant in PAB community. Many strains from Sphingopyxis have microcystin-degrading ability (Shimizu et al., 2012; Xu et al., 2015). As a water-soluble compound, microcystin released by Microcystis can easily diffuse into surrounding water and could be utilized by microcystin-degraders, e.g., Sphingopyxis. In PAB community, the relative abundance of some potentially methanol-utilizing [i.e., Phreatobacter (Tóth et al., 2014)] and ferrous iron-oxidizing genera [i.e., Ferroviario (Dubinina and Sorokina, 2014)] also significantly increased. In FLB community, the relative abundance of genera known for algal-derived DOC-utilizing (e.g., Limnohabitans and Limnobacter), eutrophic water-affiliating [e.g., CL500-29 marine group and hgcI clade (Lindh et al., 2015)] and pathogenic microorganism [e.g., Flavobacterium (Nicolas et al., 2008), Legionella (Hägele et al., 2000), Mycobacterium (Cosma et al., 2003)] significantly increased. The physiology of many other genera remains unclear (e.g., Hirschia, Candidatus Trichorickettsia and so on).

Ecosystems are very complex, with complicated networks that form between considerably interacting organisms (Montoya et al., 2006). In this study, a network-based approach using RMT was used to show the co-occurrence pattern and delineated interactions of bacterial communities during cyanobacterial blooms. The specific microbial interaction networks of PAB and FLB were established, and the role of cyanobacteria (especially for M. aeruginosa) in the network was proposed.

The stability of an ecosystem function lies in species interactions (Zhou et al., 2010). Lower connectivity in a community leads to higher functional stability of the system (e.g., scale-free network) because the whole network module will be less affected by the loss of nodes. (Zhou et al., 2010). For PAB, the higher connectively (avgK), highly concentrated modularity and fewer nodes could result in greater vulnerability than that of FLB. For FLB, the lower connectivity (avgK), higher network size and higher bacterial diversity could result in greater resilience to disturbances than that of PAB. The differences in topological features (e.g., connectivity and geodesic distance) of PAB and FLB networks could be associated with the spatial distances among bacteria, i.e., bacteria on particles are overall closer to each other than FLB. Accordingly, direct interactions, such as mutualism, competition or predation are more likely to occur in the PAB community. Taxa tended to covary in both PAB and FLB networks with predominantly positive correlations (>88%), indicting similarly preferred conditions or cooperative behaviors (e.g., cross feeding, syntrophic interactions and mutualistic interactions) of most bacteria during M. aeruginosa bloom (Raes and Bork, 2008).

In an ecological network, module hubs and connectors play crucial roles in the maintenance of community stability and are considered keystone species in the system (Zhou et al., 2010; Shi et al., 2016). The putative keystone nodes in PAB and FLB networks were lowly abundant, which indicates that low abundance taxa may play important roles in maintaining network structures in freshwater microbial communities. This finding is consistent with studies of soil microbial communities (Lupatini et al., 2014; Shi et al., 2016). Alphaproteobacteria was the most important group in the PAB network and occupied the three module centers: Meganema, Pleomorphomonas and Brevundimonas sp.. For the Meganema genus, M. perideroedes is the sole described species, which was found to be a versatile facultative anaerobe in activated sludge with ability to utilize various organic substrates in aerobic conditions and use nitrate and nitrite as electron acceptors (Kragelund et al., 2005), and McIlroy et al. (2015) thought it is a potentially facultative methylotrophy based on the genome of stain Gr1; for Pleomorphomonas, almost all the species of this genus are N-fixing bacteria, i.e., P. diazotrophica, P. diazotrophica, and P. oryzae (Xie and Yokota, 2005; Im et al., 2006; Madhaiyan et al., 2013), and Pleomorphomonas sp. were found can methylate Hg (II) (Achá et al., 2012); Brevundimonas sp. have been found the abilities of algicidal effect on M. aeruginosa, growth promotion on Chlorella ellipsoide, cadmium biosorption, lactofen-hydrolyzing, quinoline-degrading and antimicrobial silver nanoparticles producing (Park et al., 2008; Masoudzadeh et al., 2011; Rajamanickam et al., 2013; Wang C. et al., 2015).

For the five identified genera of module hubs and connectors in FLB network, Nitrospira are known as the most diverse and widespread nitrifiers in various ecosystems, e.g., natural environments and biological wastewater treatment plants (Lücker et al., 2010; Feng et al., 2016; Pinto et al., 2016). Ecological function or relationships with algal blooms of the remaining four genera are unclear except for the following information: for Fluviicola, only novel species descriptions have been reported and they were mainly isolated from fresh water and chemical industrial wastewater (O’Sullivan et al., 2005; Yang et al., 2014); CL500-29 marine group is widespread in waters, particularly in copiotrophic environments (Lindh et al., 2015); Legionella are wide-spread pathogens of human, animals, protozoa and aquatic eukaryotes (Golovlev, 2000; Hägele et al., 2000), and most species are aquatic origin; C. catalasitica is the sole species of Crocinitomix, which was isolated from algal-rich littoral zone of Auke Bay, AK, United States (Bowman et al., 2003). In general, module hubs of both networks were potentially involved in biogeochemical cycling of nitrogen taxa. This finding indicates that bacteria with ecological physiological functions of N metabolism might be quite important in the two communities. Moreover, module I of PAB network was centered with a potentially algicidal bacteria of M. aeruginosa (OTU1156, Brevundimonas), which indicates that algicidal bacteria might have important roles in PAB communities.

As the dominant phytoplankton phylum in Dianchi Lake, cyanobacteria (green nodes in Figure 5) were mostly connected with negative links (especially for the node OTU359), representing antagonistic relationships such as predator-prey, competition or variable fitness (Chow et al., 2014). As the photosynthetic organism, phytoplankton is often the most important biotic importer of organic substrates (e.g., DOM) in water systems. Cyanobacteria could act as a nutrient contributor by supplying heterotrophic bacteria with organic substrates from photosynthesis or by acting as prey (Paver et al., 2013).

The cyanobacteria node OTU359 represents the bloom-causing species M. aeruginosa, which is the most common colonial cyanobacteria in fresh water. As an oxygenic and photoautotrophic species, M. aeruginosa often form blooms in eutrophic waters and produce toxic microcystins. OTU359 was the only node with purely negative connections in module I, indicating its uniquely antagonistic relationships with neighboring nodes. It had 18 neighboring nodes in the PAB network, including the potentially algicial bacteria OTU1156 (Brevundimonas), which was the hub of module I. Among other neighbors of OTU359, OTU423-CL500-29 marine group are widespreadly aquatic bacteria; OTU1032-Gemmatimonas strains have photosynthetic pigments [i.e., bacteriochlorophyll a (Zeng et al., 2015) and carotenoids (Takaichi et al., 2009)] and polyphosphate-accumulating functions (Zhang et al., 2003). It is difficult to understand the connection of other neighbors with OTU359 because their ecophysiology is unclear (e.g., SM1A02, I-10) or appears irrelevant to cyanobacteria [e.g., Phenylobacterium; the physiology of reported strains are mainly involved in organic contaminant biodegradation (Tiago et al., 2005; Eberspächer and Lingens, 2006)]. The relationships between OTU359 and the related two neighbors (OTU74 and OTU 190) in the FLB network were also unclear.

The bloom-causing species M. aeruginosa (OTU359) was assumed to play different roles in PAB and FLB communities according to its network topological characteristics. Although node OTU 359 has few connections (two links) within the major module of the FLB network, M. aeruginosa seems more important because it connected to the hub of the largest module (module I) and occupied a higher hierarchical position with the 18 links in the PAB network. Therefore, M. aeruginosa possibly played a more important role in PAB communities than that of FLB communities. Additionally, the tight connections between nodes in the PAB network could indicate quick and effective metabolic interactions (Zhou et al., 2010) among species by transferring considerable organic matter from algae quickly in this phycosphere, thus indicating their potentially important roles in material and energy flows in this aquatic ecosystem.

In general, the communities of PAB and FLB during a M. aeruginosa bloom were significantly different in their structure, diversity, microbial interaction network and the topological role of M. aeruginosa. For PAB network, bacterial connections were tighter, potentially indicating more efficient metabolism interactions than that of FLB network; the higher importance of M. aeruginosa (OTU359) was indicated by high connectivity in the core region of the PAB network; Alphaproteobacteria were predominant, among which potentially nitrogen-fixing and algicidal bacteria occupied the module centers. FLB communities had significantly higher diversity and were sensitive to the detected environmental factors; genera related to methane-utilizing and pathogenic bacteria had significantly higher relative abundance; M. aeruginosa was marginal in the network, with less connectivity and more modules and nodes, possibly leading to the higher resilience of the ecological function than that of PAB. In both PAB and FLB interaction networks, a co-occurrence pattern was predominant, with positive connections. Bacteria affiliated with N metabolism were important by occupying the module hubs. Here, we have made an important attempt to explore the differential response of PAB and FLB communities to a cyanobacterial bloom by applying network-based analyses. Our findings provide ecological insights into the different interactions among PAB or FLB communities during algal blooms.

YI and FL designed the research and revised the manuscript. CY, QW, PS, JL, and JK analyzed the data and drafted the manuscript. CY, QW, PS, JL, LL, XD, XZ, and XP contributed to the sampling and experiments. LL, XD, XZ, and XP revised the manuscript. All authors approved the final manuscript and agreed to be responsible for all aspects of the work.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.