The coral holobiont is a multi-partner symbiotic system that forms associations with both external and internal microbiota (Mouchka et al., 2010). There is increasing evidence that coral-associated microbial communities are crucial for biogeochemistry and control the health and resilience of coral reef ecosystems (Mouchka et al., 2010; Thompson et al., 2014). Due to their relative size and high per cell activity, a small change in microbial biomass may signal a large reallocation of available energy in the ecosystem (McDole et al., 2012; Haas et al., 2016). Corals often reside in oligotrophic waters. Nitrogen fixation, one of the new fixed nitrogen input into the ecosystem, has long been thought to be an important linchpin to sustain biological productivity in coral reef areas (Wiebe et al., 1975; Capone et al., 1977; Shashar et al., 1994b; Cardini et al., 2014). Nitrogen fixation activity within the coral holobiont has been detected using acetylene reduction and isotopic assays in several coral species (Williams et al., 1987; Shashar et al., 1994a; Lesser et al., 2007; Chimetto et al., 2008; Grover et al., 2014; Benavides et al., 2016). Coral-associated nitrogen-fixing bacteria, capable of fixing and converting gaseous nitrogen (N₂) to biologically available nitrogen forms, contribute an important source of nitrogen to the coral holobiont (Lesser et al., 2007; Cardini et al., 2014). Both coral cells and their symbiotic dinoflagellates (Symbiodinium) have the capacity to assimilate ammonium (Pernice et al., 2012) and benefit from nitrogen-fixing bacteria (Lesser et al., 2007; Cardini et al., 2014; Santos et al., 2014). Additionally, a close relationship was found between the abundance of coral-associated diazotrophs and symbiotic dinoflagellates (Lesser et al., 2007; Olson et al., 2009; Santos et al., 2014), and between coral-associated diazotrophs and coral productivity (Cardini et al., 2014).

Accumulating evidence shows that diazotrophic organisms are ubiquitous members of coral-associated microbial communities and form species-specific associations with their hosts (Lema et al., 2012, 2014b). However, the degree to which nitrogen-fixing bacterial communities are specific to their coral hosts can vary, as both species-specific (Lema et al., 2012) and site-specific (Lema et al., 2014b) nitrogen-fixing microbial communities have been reported. Nitrogen fixation activity in corals is also highly dynamic and can be rapidly affected by changes in environmental conditions (Lesser et al., 2007; Rädecker et al., 2014). It is suggested that coral holobionts harbor both a core microbiome determined by holobiont macroorganisms, and a variable microbiome to adapt to local conditions (Kelly et al., 2014; Ainsworth et al., 2015). As the disturbance of microbial nitrogen-cycling may be tightly linked to coral bleaching and disease (Rädecker et al., 2015), knowledge of the variable diazotrophic groups would help to further evaluate the importance of these communities to the coral host; however, such knowledge is currently lacking.

The nifH gene encodes a conserved subunit of the dinitrogenase iron protein responsible for nitrogen fixation, and is conserved in all known diazotrophs (Zehr et al., 2003). As the agreement with 16S rRNA gene-based phylogeny, nifH is an ideal molecular target and is widely used for gene-based phylogenetic characterization of diazotrophs (Gaby and Buckley, 2012). In this study, high-throughput Illumina sequencing was used to investigate seasonal and species-specific patterns in diazotrophic communities associated with the corals Galaxea astreata, Pavona decussata, and Porites lutea. The core and variable diazotrophic microbiome were tested through comparative analysis. The aims were to: (i) investigate the diversity and abundance of diazotrophic communities associated with three coral species; (ii) determine the seasonal shifts of diazotrophic communities and the variable diazotrophic species associated with different coral species; (iii) explore the possible relationship between the coral-associated seasonal-variable diazotroph groups and environmental variables, given the important role of diazotrophic microbes in driving biogeochemical cycles.

Although high-throughput sequencing approaches are now commonly applied to investigate coral-associated microbial communities (Ceh et al., 2011; Chen et al., 2011; Lee et al., 2012; Mckew et al., 2012; Sunagawa et al., 2009), it has not been widely used to target nitrogen-fixing functional genes (nifH) to explore the coral-associated diazotrophic communities (Lema et al., 2014b). Compared with previous high-throughput sequencing studies on diazotrophic communities associated with the common Great Barrier Reef (GBR) coral Acropora millepora (Lema et al., 2014b), a much greater diversity of bacteria having the potential to fix nitrogen (possess the nifH gene) associated with G. astreata, Pavona decussata, and Porites lutea from Luhuitou fringing reef, South China Sea (Table 1), and might be attributed to the differences in coral species. Alphaproteobacteria was reported as the dominant diazotrophic group associated with the common GBR coral Acropora (i.e., A. millepora and A. muricata) (Lema et al., 2012), while Gammaproteobacteria was the most commonly found class in the Hawaiian Montipora corals (M. capitata and M. flabellata) (Olson et al., 2009). In the present study, nifH sequences related to Alphaproteobacteria, Deltaproteobacteria, and Gammaproteobacteria were the ubiquitous and dominant groups of three coral species in two seasons. These three groups constitute from 46.65 to 91.1% of the sequences within individual coral samples (Figure 1), suggesting that those groups play an important functional role in the coral holobiont. The coral holobiont is a complex habitat, and microenvironmental variability can strongly influence the abundance of associated microbial communities (Rohwer et al., 2002; Ainsworth et al., 2010). Previous results reported that the coral-associated bacterial community showed rapid seasonal changes for several coral species distributed throughout different regions (Ceh et al., 2011; Chen et al., 2011). However, season did not influence the diazotrophic communities associated with common GBR coral A. millepora (Lema et al., 2014b). Significant differences were detected in the diazotrophic abundance of G. astreata and Porites lutea, indicating that for these two coral species the diazotrophic communities significantly changed according to the season.

Understanding the seasonal dynamics of this association is important, as they have ecological implications. In this study, seasonal variable diazotrophic sequences were detected based on the relative abundances at 95% confidence interval. The analyses showed that three corals have a low diversity group of seasonal variable species (range from 44 to 48 OTUs) whose abundance varies widely across individuals (Supplementary Table S3). These results suggested that there are core diazotrophic microbiomes associated with corals and these cores are complemented with seasonal variable diazotrophic microbiomes. Most of the variable nifH groups from Pavona decussata were species-specific. Only a few variable OTUs overlap with two other coral species. This may be an indicator of a specific association between coral and diazotrophic microorganisms. The majority of overlapping OTUs in G. astreata and Porites lutea showed the same seasonal variation suggesting that except host species, the variation of diazotrophic communities might be related to the biogeochemical cycling processes within the holobiont, given the important role of diazotrophs in driving biogeochemical cycles (Rädecker et al., 2014).

Alphaproteobacteria affiliated with the order Rhizobiales were reported as the continuous and dominant diazotrophic assemblages associated with the common GBR Acropora corals (Lema et al., 2012, 2014a,b). Additionally, the rhizobia found within pure mucus samples of New Caledonia reefs (predominantly composed of Acropora sp.) was much higher than in the surrounding seawater in the summer period and 400-fold higher in the winter (Camps et al., 2016). In the present study, the variable Alphaproteobacteria affiliated with Rhizobiales and Rhodospirillales. One rhizobial group (OTU3604) was most closely affiliated with purple non-sulfur phototrophic bacterium Rhodopseudomonas palustris. This group dominated all spring G. astreata and Porites lutea colonies, representing up to 6.57 and 14.51% of the total nifH sequences in spring G. astreata and Porites lutea samples, respectively (Figure 3). The Rhodospirillales groups closely related to Gluconacetobacter diazotrophicus and Azospirillum lipoferum were also the dominant variable groups found in all G. astreata and Porites lutea colonies. It is notable that the most dominant variable group of G. astreata was closely related to Gluconacetobacter diazotrophicus. Here Rhodopseudomonas palustris- and Gluconacetobacter diazotrophicus-affiliated OTUs were not correlated with investigated environmental factors, which suggested the seasonal variations of these diazotrophs are not affected by environmental factors. Rhodopseudomonas palustris is notable for its ability to flexibly switch between four different modes of metabolism: photoautotrophic, photoheterotrophic, chemoautotrophic, and chemoheterotrophic (Larimer et al., 2004), which might protect it from high concentrations of oxygen arising from dinoflagellate photosynthesis in coral tissues (Kuhl et al., 1995). The most common physiological characteristics of Gluconacetobacter diazotrophicus are its growth and nitrogen fixation at low pH and that nitrogen fixation is not affected by high concentrations of nitrate nitrogen (Pedraza, 2008; Eskin et al., 2014). Additionally, it is capable of undergoing nitrogen fixation with low amounts of ammonium-based nitrogen (Stephan et al., 1991; Fisher and Newton, 2005). In contrast with G. astreata and Porites lutea, the dominant Alphaproteobacteria variable groups in Pavona decussata showed different patterns. One rhizobial group affiliated with Hyphomicrobium sp. was the dominant group in spring colonies while another rhizobial group affiliated with Mesorhizobium sp. was the dominant group in summer colonies.

The second most seasonal variable group of diazotrophs detected by our nifH-based gene assay were all affiliated with anaerobic sulfate-reducing Deltaproteobacteria. Anaerobic sulfate-reducing bacteria were commonly detected in both 16S rRNA and nifH gene surveys of presumably healthy corals (Olson et al., 2009; Sunagawa et al., 2010 Lema et al., 2012, 2014b; Fernando et al., 2015). Although oxygenic photosynthesis renders most of the coral interior oxic during the day, coral microhabitats may host diazotrophic communities under oxygen-depleted conditions within the gastrodermis (Hoegh-Guldberg and Williamson, 1999) and microaerophilic regions in the gastrovascular cavities of coral polyps (Agostini et al., 2012). Additionally, coral tissues compromised by contact with stagnant water or sediment and coral surfaces (mucus layer) and skeletons enable anaerobic forms of bacterial respiration and fermentation to occur within the holobiont (Thompson et al., 2014). The present study results showed that the seasonal variations of Deltaproteobacteria diazotrophs differed among coral species. The Deltaproteobacteria groups were relatively abundant in summer G. astreata and Pavona decussata samples and spring Porites lutea samples. The dominant variable Deltaproteobacteria groups of G. astreata and Pavona decussata were positively correlated with nitrate and salinity; however, the dominant variable Deltaproteobacteria groups of Porites lutea were negatively correlated with nitrate, but positively correlated with ammonium (Figure 4). Additionally, the dominant variable Deltaproteobacteria groups also varied among coral host species, indicating that both host species and environmental factors cause seasonal shifts of Deltaproteobacteria diazotrophs.

Gammaproteobacteria genus Vibrio was reported as consistent members of diazotrophic assemblages associated with Hawaiian corals in the genus Montipora (Olson et al., 2009) and common GBR coral A. millepora (Lema et al., 2014b) based on nifH sequencing. Here, genus Vibrio were also consistent members of diazotrophic assemblages associated with three coral species, and no significant seasonal variation was detected based on nifH sequence. However, genus Vibrio is a variable bacterial group associated with coral G. astreata and Porites lutea and Vibrio-affiliated bacterial sequences represented 0–0.84% of the 16S rRNA gene sequences recovered. The dominant seasonal variable Gammaproteobacteria affiliated with Halorhodospira halophila, Thiorhodospira sibirica, and Pseudomonas stutzeri. Halorhodospira halophila, an obligately photosynthetic and extremely halophilic purple sulfur bacterium (Tsuihiji et al., 2006), showed relative abundance in spring coral G. astreata and Porites lutea, but was dominant in summer Pavona decussata samples. One group affiliated with Thiorhodospira sibirica, one of the alkaliphilic purple sulfur bacteria (Bryantseva et al., 1999), represented up to 7.2% of the total nifH sequences in spring G. astreata samples and positively correlated with ammonium and phosphate. Additionally, Pseudomonas stutzeri, a nitrogen-fixing and denitrifying bacterium (Lalucat et al., 2006), was relatively abundant in summer Porites lutea samples.

The most dominant variable group of Pavona decussata, closely related to Cyanobacteria Leptolyngbya boryana, represented up to 5.83% of the total nifH sequences in spring Pavona decussata samples, positively correlated with ammonium and phosphate. Another Cyanobacteria group closely related to Phormidium sp. was relatively abundant in summer Porites lutea samples. Both genera Leptolyngbya and Phormidium have previously been reported in a number of different coral species black band disease (BBD) samples where they have been hypothesized to play roles as pathogens (Frias-Lopez et al., 2003; Richardson and Kuta, 2003; Myers et al., 2007). The nifH phylotypes affiliated with Firmicutes Clostridium ljungdahlii were only detected showing significant seasonal variation in Porites lutea samples, comprising 2.77% of nifH phylotypes derived from spring Porites lutea and positively correlated with ammonium and phosphate.

The seasonal variations in coral-associated bacterial and diazotrophic communities may be influenced by the nutrient loads between spring and summer. The sample collection location Sanya Bay is affected by cold-water upwelling during the summer (Huang et al., 2003; Zhang et al., 2010; Wu et al., 2012). Cold-water upwelling affects the distribution of a variety of dissolved and particulate forms of nitrogen in Sanya Bay (Huang et al., 2003). Ammonium was the predominant dissolved nitrogen in spring, while nitrate was the predominant dissolved nitrogen in summer (Supplementary Table S1). In our previous study, the functional gene composition of the microbial community was significantly correlated with the concentrations of inorganic nitrogen and phosphate (Zhang et al., 2015). Here, all diazotrophic communities of the three coral species were significantly correlated with ammonium and nitrate (Table 3). In addition, the diazotrophic communities of coral G. astreata and Pavona decussata were significantly correlated with phosphorus (Table 3). This suggested that the apparent seasonal change in diazotrophic communities of corals could be linked to the seasonal shifts of nutrients. A priority for future studies should be to identify environmental variables contributing to these shifts in coral bacterial communities and to determine how they influence the health of the coral host.

This study revealed a much greater diversity of diazotrophs associated with G. astreata, Pavona decussata, and Porites lutea. Alphaproteobacteria, Deltaproteobacteria, and Gammaproteobacteria were the ubiquitous and dominant groups in all corals in two seasons. Seasonal factors did not cause shifts in diazotrophic richness and diversities of the three coral species; even no shifts in diazotrophic communities and abundance were observed in coral Pavona decussata. In contrast, the diazotrophic communities and nifH gene abundance of both G. astreata and Porites lutea showed significant seasonal changes. Most of the variable nifH groups from Pavona decussata were species-specific. The dominant overlap OTUs in G. astreata and Porites lutea showed the same seasonal variation. The seasonal variations of diazotrophic communities were significantly correlated with the seasonal shifts of nutrients. Variable diazotroph groups are widely distributed in the environment and may be of relevance to diverse metabolic potential, such as carbon fixation and sulfate reduction. This suggests that their potential to provide additional sources of fixed nitrogen to the coral holobiont may be functionally important. However, almost all metabolic potential of these diazotrophs was referred from crops and land plants. The physiological roles of these nitrogen-fixing symbionts in the nitrogen budget and cycling within corals need to be investigated in detail.

YZ and JD conceived the research. YZ and QY performed the experiments. YZ wrote the manuscript. JVN and JZ edited the manuscript. QY and ZS contributed sampling or data analysis pipeline. All authors reviewed and accepted the manuscript.

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.