Coral and seawater samples were collected from three locations, Kenting, Yehliu, and Lyudao, in Taiwan (Figure 1) from February 2008 to January 2010. In each location and at each sampling time, three colonies (replicates) of S. pistillata at depths of 3–7 m were selected and tagged for sampling. For each colony, a coral fragment approximately 2 cm long and 2 cm in diameter was collected. Samples were washed twice with filtered seawater and stored at 4°C until coral tissue was processed. In addition, 1 L of seawater surrounding the coral colony was simultaneously collected to analyze the microbial community in reef water. A total of 88 S. pistillata and seawater samples were collected (Supplementary Table S1). Environmental data (precipitation, daylight duration and air and water temperatures) were obtained from the Central Weather Bureau in Taiwan (Supplementary Table S1).

Coral tissues were sprayed (using an airbrush) with 2–5 mL 10× TE buffer (10 mM Tris–HCl, pH 8.5, 1 mM EDTA, pH 8.0) and collected in a sterilized bag. Coral tissues were transferred to a 15 mL sterilized tube and stored at -20°C pending DNA extraction.

Coral tissue was recovered for genomic DNA extraction by centrifugation (10,000 × g, 15 min). After discarding the supernatant, tissue pellets were put into 600 μL of 10× TE buffer and liquid nitrogen, homogenized (mortar and pestle), and the resulting solution transferred to a clean tube. Seawater samples were filtered through membrane filters (0.2 μm pore size; Adventec, Tokyo, Japan) to collect microbial particulate. Total genomic DNA of coral tissue and seawater were extracted as outlined in the UltraClean Soil DNA Kit (MioBio, Solana Beach, CA, United States).

The PCR was performed using two bacterial universal primers: 27F (5′-AGAGTTTGATCMTGGCTCAG-3′) and 341R (5′-CTGCTGCCTCCCGTAGG-3′), which were designed for the V1–V2 hypervariable region of the bacterial 16S rRNA gene (Hamp et al., 2009). The reaction mixture contained 1 μL of 5 U TaKaRa Ex Taq HS (Takara Bio, Otsu, Japan), 5 μL of 10×Ex Taq buffer, 4 μL of 2.5 mM deoxynucleotide triphosphate mixture, 1 μL of each primer (10 μM), and 1–5 μL (10–20 ng) template DNA in a volume of 50 μL. The temperature program for 30 PCR cycles was the initial step of 94°C for 3 min, 94°C for 20 s, 52°C for 20 s, 72°C for 20 s, and 72°C for 2 min (final extension after the last cycle).

The PCR amplicons of the bacterial 16S rRNA gene V1–V2 region were verified by DNA agarose gel electrophoresis with 1.5% agarose gel and 10× TE buffer. The expected DNA band (∼320 bp) was cut from the gel and DNA was recovered with a QIAEX II Agarose Gel Extraction kit (QIAGEN, Hilden Germany) and quality verified with a Scandrop spectrophotometer (Thermo Scientific, Vantaa, Finland).

A DNA tagging PCR was used to tag each PCR product of the bacterial 16S rRNA gene V1–V2 region (Chen et al., 2011). According to Chen et al. (2011), the tag primer was designed with four overhanging nucleotides; this arrangement ensured 256 distinct tags, at the 5′ end of the 27F and 341R primers for bacterial DNA. The tagging PCR conditions consisted of an initial step of 94°C for 3 min, 5 cycles of 94°C for 20 s, 60°C for 15 s, 72°C for 20 s, and a final step at 72°C for 2 min.

Pooled 40 ng lots of each tagged bacterial 16S rRNA gene V1–V2 region DNA samples (88 samples in total) were used for massively parallel pyrosequencing using a Roche 454 Genome Sequencer FLX System (Mission Biotech, Taipei, Taiwan). Raw sequencing reads were sorted according to unique tags using an in-house script, and deposited in NCBI Sequence Read Archive database, under accession number SRP098878, after the primer was removed. Reads from replicate colonies, sampled at the same time and location, were combined into one sample.

On a per-sample basis, bacterial reads were quality-filtered using Mothur v1.36.1 (Schloss et al., 2009) to retain reads of lengths from 290 to 340 base pairs (bp), and an average quality score >27. Reads containing any ambiguous base or homopolymer >8 bp were excluded. Chimeric reads were detected and removed using USEARCH v8.1.1861 (Edgar, 2010) with reference mode (3% minimum divergence). Quality-filtered and non-chimeric reads were analyzed using UPARSE pipeline (Edgar, 2013) to generate OTUs per sample (97% identity level), which were classified with RDP classifier (v2.12) (Wang et al., 2007) with a pseudo-bootstrap threshold of 80%.

The rarefied diversity measures (e.g., Shannon, Simpson, Chao 1, ACE, and Good’s coverage) were estimated using Mothur with 1000 iterations. The sample size (N) varied from 567 to 29819; therefore, 567 was chosen for rarefaction analysis.

Bacterial OTU representative sequences of all samples affiliated with a given taxonomy were collectively clustered using USEARCH, with options –cluster_smallmem and –id 0.97. Abundance of each centroid (or cross-sample OTU) was represented by average relative abundance of its cluster members. Phylogenetic analysis was conducted in MEGA7 (Kumar et al., 2016), with maximum-likelihood trees generated with 1000 bootstraps for aligned representative sequences.

The present study defined six resident types of bacteria in S. pistillata and seawater, based on relative abundance and detection frequency. Type I (prevalent abundant group), type II (moderately prevalent abundant group), and type III (non-prevalent abundant group) each had ≥1% average relative abundance and were detected in ≥80, 80–60, and <60% samples, respectively. Type IV (prevalent minor group), type V (moderately prevalent minor group), and type VI (non-prevalent minor group) had <1% average relative abundance and were detected in ≥80, 80–60, and <60% samples, respectively. Coral and seawater samples were separately considered (in terms of detection frequency).

Relative abundance of each classified bacterial genera in individual samples was incorporated into a matrix to estimate the Bray–Curtis distance. Results were presented by non-metric multidimensional scaling (nMDS) analysis using Primer 6 (PRIMER-E, Lutton, Lvybridge, United Kingdom; Clarke and Warwick, 2001) to determine relationships of bacterial communities among times and locations. Analysis of similarity (ANOSIM) via Primer 6 was used to test bacterial diversity indices and compositions in spatiotemporal variations.

For analysis of abundant and minor bacterial members, abundant groups (type I–III) and minor groups (type IV–VI) were separately incorporated into two matrixes. These matrices were used to perform ANOSIM in a Primer 6 based Bray–Curtis distance derived from relative abundance transformed by three methods, including no transformation, square root transformation, and binary transformation (i.e., presence and absence analysis).

Bacterial compositions between sampling times in each location were compared with Pearson correlations using R package Hmisc¹. In addition, Pearson correlations were used to compare fluctuations of Endozoicomonas and Acinetobacter.

Hierarchical clustering analysis was performed using the average linkage via R package pheatmap (Kolde, 2015) and visualized on a heat map. The distance matrix used for clustering was calculated using the R package vegan (Dixon, 2003) based on a base-10 logarithmic transformation of relative abundance (in percentage), plus a pseudo-count of 1 × 10^-8 (Costea et al., 2014). Genera of unknown taxonomy or <0.2% relative abundance were excluded.

Seawater sample K1001S yielded the most operational taxonomic units (OTUs), 406.15 (in rarefaction), whereas coral sample K0808C had the fewest (58.68 OTUs; Supplementary Table S2). According to ANOSIM, bacterial diversity indices in corals and seawater samples differed (richness: R = 0.634, p = 0.001; Shannon: R = 0.298, p = 0.001) and bacterial community in seawater samples had greater richness and Shannon diversity than those in corals (Supplementary Table S2).

Overall, bacterial compositions in S. pistillata from Kenting, Yehliu, and Lyudao were similar at a class level, similar to seawater samples (Supplementary Figure S1). However, bacterial communities in S. pistillata and seawater differed from each other. Dominant bacterial classes in seawater were Flavobacteria, unclassified Marinimicrobia, Gammaproteobacteria, and Alphaproteobacteria. Among these, most dominant genera were unclassified, such as unclassified Marinimicrobia, unclassified Gammaproteobacteria, and unclassified Flavobacteriaceae. In addition, the genus Vibrio, which belongs to Gammaproteobacteria, was higher than those in corals.

In S. pistillata, Actinobacteria, Betaproteobacteria, and Gammaproteobacteria were dominant, and their relative abundances higher than those in seawater. In Gammaproteobacteria, dominant genera were Endozoicomonas, Acinetobacter, Pseudomonas, and Serratia (Supplementary Figure S1), among which Endozoicomonas was particularly more dominant in Kenting and Lyudao compared to Yehliu. Propionibacterium and Actinophytocola were the two dominant Actinobacteria genera, and Massilia Betaproteobacteria in coral.

Bacterial community structure in S. pistillata and seawater at various sampling times is shown (Figure 1). Some genera, such as Ralstonia, Cupriavidus, and Aquabacterium, were only detected in coral samples, whereas Pseudoalteromonas was common only in seawater, and Gramella was exclusively retrieved from seawater samples. In addition, among coral samples, although some bacterial taxa were prevalent at every collection, their relative abundance varied from time to time (Figure 1). For example, Endozoicimonas and Acinetobacter fluctuated in abundance throughout this survey.

There were differences among sampling years in bacterial community structure (ANOSIM: R = 0.301, p = 0.002; Supplementary Figure S2). Among all sites, based on nMDS, the bacterial community changed along the sampling time (Figure 2). Within each location, bacterial composition also differed among sampling years or times (Supplementary Figure S3), especially in samples from Kenting (ANOSIM: R = 0.370, p = 0.028). In Lyudao and Yehliu, bacterial compositions were clustered according to sampling years (difference not significant), but the Pearson correlation value between each sampling time in Lyudao and Yehliu decreased significantly (in Lyudao, from 1.00 to 0.24; in Yehliu, from 1.00 to 0.16) with increasing time span (Supplementary Figure S4), which indicated changes in bacterial composition during sampling times. In addition, there was no difference between seasons (ANOSIM: R = -0.072, p = 0.791).

To determine differential effects of short- versus long-term observations, consecutive patterns of correlations between bacterial communities were analyzed (Supplementary Figure S5). Correlations fluctuated over five consecutive samplings (Supplementary Figures S5A–F). However, a longer time span revealed a more holistic view of variation in bacterial composition (Supplementary Figure S5G), not apparent in short-term observations. The longer time span also revealed that changes followed a pattern. For example, two peaks (times 4–6 and time 9) had different duration, indicating different resilience status (Supplementary Figure S5G). Both abundant and minor (<1% relative abundance) genera had a significant differences between sampling years, and only minor genera did so by considering presence/absence data (binary transformation; Supplementary Table S3). In Yehliu, these abundant genera (except Endozoicomonas) retained similar abundance and moderately increased in 2009, although they had more variable fluctuations in abundance in Lyudao and Kenting, especially Endozoicomonas and Acinetobacter (Figure 3). In Lyudao, Acinetobacter increased during July to November in 2008, and Endozoicomonas increased during May and July in 2008, but decreased thereafter. In Kenting, abundance of Acinetobacter was low during 2008, increased during May and July in 2009, and decreased from November 2009 to January 2010. However, Endozoicomonas bloomed in most sampling times during 2008 and decreased during May and July in 2009, followed by an increase in November 2009. Intriguingly, both in Lyudao and Kenting, the times of Endozoicomonas and Acinetobacter blooms were staggered (Figure 3). Furthermore, in Kenting, there was a negative correlation (Pearson correlation: R = -0.899, p = 0.038) between Endozoicomonas and Acinetobacter abundance fluctuations from November 2008 to November 2009.

Based on nMDS, there was distinct clustering of coral-associated bacterial composition in Yehliu, Lyudao, and Kenting (ANOSIM: R = 0.224, p = 0.014; Figure 2). Kenting and Yehliu had differences not only in bacterial composition (ANOSIM: R = 0.437, p = 0.004) but also in community richness (ANOSIM: R = 0.218, p = 0.045). There was a compositional difference among sampling sites in dominant genera (ANOSIM: R = 0.240, p = 0.008), but not in minor genera (Supplementary Table S3).

Endozoicomonas and Acinetobacter were the two frequently detected genera of high abundance (Figure 3), of which both had disparate inhabiting preferences on S. pistillata in different locations (Figure 4). Endozoicomonas were mostly present in corals from Lyudao and Kenting, and composed of distinct genotypes in each locations (i.e., OTU02, 03, 04, 07, 08, and 10 in Kenting; OTU05, 06, and 09 in Lyudao). These Endozoicomonas OTUs were merely detected in both Kenting and Lyudao, indicating that geographical setups determined their inhabiting specificity (Figure 4A). Based on phylogenetic analysis, these OTUs were highly associated with E. elysicola and E. atrinae (Figure 4C). In contrast to exclusive detection of Endozoicomonas in coral, Acinetobacter were present in both S. pistillata and seawater (Figure 4B). Furthermore, most Acinetobacter OTUs were shared by three sampling sites, with less geographical preference than Endozoicomonas (Figure 4B). These Acinetobacter OTUs were also more diverse compared to Endozoicomonas OTUs in terms of phylogenetic affiliation (Figure 4D).

According to relative abundance and detection frequency, bacteria in S. pistillata and seawater were categorized into six resident types (Section Resident Types of Bacterial Lineages in Materials and Methods and Table 1). In S. pistillata, type I group (i.e., with >1% relative abundance and detected in ≥80% samples) had eight genera, including Acinetobacter, Endozoicomonas, Propionibacterium, Corynebacterium, Staphylococcus, Arthrobacter, Actinophytocola, and Pseudomonas. However, there were only Vibrio and unclassified Marinimicrobia as type I in seawater. Among these abundant genera (resident type I–III), coral and seawater shared no genus.

Shifts in bacterial communities in healthy coral tissues were only apparent with a sufficient duration of survey (∼2 years). Although changes in bacterial composition in Lyudao and Yehliu were not confirmed (samples were lost due to typhoons and human activities), based on the correlation between sampling times, community resilience may be present over a longer sampling interval. In the present study, the long and complete observation in Kenting successfully captured compositional resilience that was not apparent in short-term surveys. A small group of bacteria (that comprised the core microbiome) were ubiquitously associated with coral, regardless of abiotic environmental parameters (Hernandez-Agreda et al., 2016). However, based on our results, prolonged observation enabled characterizing their succession in coral (Lynch and Neufeld, 2015). In this study, changes in prevalent abundant genera Acinetobacter and Endozoicomonas had disparate fluctuation patterns among locations, with abundances constant in Yehliu but variable in Kenting in 2009. Negative abundance correlations between these two genera were not detected before November 2008 but afterward until November 2009; therefore, this successional pattern was far from conclusive. Apart from abundant ones, minor bacterial genera in S. pistillata also changed in abundance throughout the study, which emphasized the need for longer surveys of coral-associated bacteria. Although Pantos et al. (2003) proposed some explanations for observed fluctuations in coral-associated bacterial communities, their hypotheses were only based on short-term observations. It has been suggested that coral-associated microbes can attenuate or intensify cumulative stressors that may disrupt coral reef ecosystems (Bourne et al., 2016). Therefore, we believe that these stressors, which varied in times and degrees from place to place, also contributed to spatial and temporal dynamics in coral-associated bacterial composition and abundance.

Different coral reproduction strategies (i.e., broadcast spawning and brooding) affect symbiotic bacterial composition differently, and S. pistillata, a brooder coral, harbored distinguishable bacterial communities strongly clustered by geographical location (Neave et al., 2017b), implying, in S. pistillata, vertical transfer of microbes from parents to offspring, thereby restricting the microbial structure and development and leading to the observed geographical grouping. As some bacteria have inhabiting specificity to coral species (Hong et al., 2009; Reis et al., 2009), discrepant microbial compositions among corals reported by different studies were plausible.

Among abundant genera, Acinetobacter, Propionibacterium, and Pseudomonas were more constantly detected in S. pistillata regardless of seasonal and geographical differences, whereas Endozoicomonas had geographical variations. Endozoicomonas spp. have diverse roles in different hosts as being symbiotic (La Rivière et al., 2013) and parasitic (Mendoza et al., 2013). The documented high proportion of repeats and insertion sequences in E. montiporae (Ding et al., 2016) and pathogenic strain Candidatus Endozoicomonas cretensis (Katharios et al., 2015) corresponded to their high genome plasticity for wide adaptation to various hosts (reviewed by Neave et al., 2017a). Since Endozoicomonas was abundant in Lyudao and Kenting but not in Yehliu, their adaptation appeared more favorable for tropical versus subtropical oceans.

Based on the presence/absence transformation of abundance, minor genera had significant differences among sampling years, reflecting their fluctuating nature in S. pistillata over time. Ralstonia and Propionibacterium were two acknowledged minor symbiotic genera in corals and intimately associated with dinoflagellate endosymbionts (Ainsworth et al., 2015). However, in our study, Propionibacterium was abundantly detected among three locations during sampling, whereas Ralstonia had constantly low abundance among sampling sites. Lynch and Neufeld (2015) suggested that prevalent, minor taxa were just conditionally minor in abundance, and would opportunistically grow with abundance under optimal conditions. This accounted for the discrepant abundance of Propionibacterium in the current versus previous studies. In contrast, consistent detection of Ralstonia’s low abundance described its potential of being a keystone taxon permanently holding rare abundance in coral (Giovannoni and Stingl, 2005; Lynch and Neufeld, 2015), although the functional role of Ralstonia in S. pistillata remains unclarified.

Many bacterial taxa were reported as abundant members in coral (Littman et al., 2009; Liu et al., 2012; McKew et al., 2012; Morrow et al., 2012; Carlos et al., 2013), and mostly on Spongiobacter and Endozoicomonas (Lee et al., 2012; Morrow et al., 2012; Bayer et al., 2013; Carlos et al., 2013; Lema et al., 2014). These abundantly, frequently detected bacteria seemed adapted to live in coral tissues and acting as coral nutrition, pathogens, probiont, or purely commensal bacteria in the coral holobiont (Klaus et al., 2005). In the present study, we categorized S. pistillata-associated bacteria into six resident types, based on their abundance and detection frequency, with Acinetobacter (in type I) being the prevalent genera. Acinetobacter had been detected in stony corals from various regions (Littman et al., 2009; McKew et al., 2012; Morrow et al., 2012; Carlos et al., 2013; Li et al., 2014), and potentially played either beneficial or detrimental roles in corals. For example, Shnit-Orland and Kushmaro (2009) considered Acinetobacter sp. as a first-line defender that assisted the coral holobiont against pathogens resistant to multiple antibiotics. However, in a study of Dark Spot Syndrome in coral Stephanocoenia intersepta, Acinetobacter was regarded as a potential pathogen (Sweet et al., 2013). Given the common presence in stony corals, the uncertain role of Acinetobacter in S. pistillata needs to be clarified.

Among the eight genera in resident type I, four (i.e., Propionibacterium, Corynebacterium, Arthrobacter, and Actinophytocola) belong to the phylum Actinobacteria, consistent with other reports regarding Actinobacteria as a dominant genus associated with corals. Propionibacterium was present in coral Cirrhipathes lutkeni (Santiago-Vázquez et al., 2007), Mussismilia hispida (de Castro et al., 2010) and Acropora digitifera (Nithyanand et al., 2011). Using Actinobacteria-specific primers, Yang et al. (2013) reported 19 actinobacterial genera in two corals, suggesting high diversity of Actinobacteria in both hard and soft corals. Coral-associated actinobacteria reportedly had antimicrobial activity (Sweet et al., 2013; Mahmoud and Kalendar, 2016) attributed to production of bioactive substances (Yang et al., 2013). Therefore, these four type-I actinobacterial genera warrant careful consideration in future studies, considering their broad distribution in corals and antimicrobial properties.

In the ocean, heterotrophic bacterial communities (in terms of species composition, spatiotemporal variations, and community structure) are largely affected by the availability, composition, and spatiotemporal distribution of organic substrates. Similarly, variations in coral-associated bacterial community are also controlled, to a great extent, by organic matter secreted by coral, similar to the situation of algae-associated microbial communities (Dang and Lovell, 2016). For example, the genome of Endozoicomonas has a wide spectrum of genes related to generic transport for uptake of extracellular organic compounds and carbohydrates (Ding et al., 2016; Neave et al., 2017a). Numbers of transport molecules also implied interactions between Endozoicomonas and compounds produced by corals (Neave et al., 2014; Ding et al., 2016). Similarly, Acinetobacter was reportedly able to metabolize dimethyl sulfide (Horinouchi et al., 1997) produced by symbiotic algae in corals; furthermore, its precursor (dimethylsulfoniopropionate) has served as a chemoattractant to heterotrophic bacteria in oceans (Seymour et al., 2010). Therefore, availability of extracellular compounds in coral holobiont is likely associated with fluctuations in the abundance of Endozoicomonas and Acinetobacter.