Dynamics of coral‐associated microbiomes during a thermal bleaching event

Abstract Coral‐associated microorganisms play an important role in their host fitness and survival. A number of studies have demonstrated connections between thermal tolerance in corals and the type/relative abundance of Symbiodinium they harbor. More recently, the shifts in coral‐associated bacterial profiles were also shown to be linked to the patterns of coral heat tolerance. Here, we investigated the dynamics of Porites lutea‐associated bacterial and algal communities throughout a natural bleaching event, using full‐length 16S rRNA and internal transcribed spacer sequences (ITS) obtained from PacBio circular consensus sequencing. We provided evidence of significant changes in the structure and diversity of coral‐associated microbiomes during thermal stress. The balance of the symbiosis shifted from a predominant association between corals and Gammaproteobacteria to a predominance of Alphaproteobacteria and to a lesser extent Betaproteobacteria following the bleaching event. On the contrary, the composition and diversity of Symbiodinium communities remained unaltered throughout the bleaching event. It appears that the switching and/or shuffling of Symbiodinium types may not be the primary mechanism used by P. lutea to cope with increasing seawater temperature. The shifts in the structure and diversity of associated bacterial communities may contribute more to the survival of the coral holobiont under heat stress.

Besides conferring the ability to fix carbon to the coral holobiont, Symbiodinium have been shown to influence the resilience of their hosts to thermal stress (Boulotte et al., 2016;Hume et al., 2016;Keshavmurthy et al., 2014). Attempts to understand the differences in the coral host responses to elevated water temperature have largely focused on genetic variation of Symbiodinium. Given the capacity of reef-building corals to host different types of algal symbionts, two potential adaptive mechanisms to ocean warming have been proposed. The first mechanism, symbiont "shuffling," is an approach by which corals resist or recover from heat stress through an adjustment in relative abundance of Symbiodinium types already in-hospite (Baker, Starger, McClanahan, & Glynn, 2004;Jones, Berkelmans, van Oppen, Mieog, & Sinclair, 2008;Rowan, 2004). The second mechanism, symbiont "switching," takes place when the coral hosts acquire new types of Symbiodinium through uptake from the environment such as water column and sediments (Boulotte et al., 2016;Fautin & Buddemeier, 2004). In addition to the proposed switching/shuffling mechanisms, Howells et al. (2012) demonstrated that adaptive differences in heat sensitivity/tolerance existed not only among different types of Symbiodinium, but also among locally adapted populations belonging to a single Symbiodinium type (Howells et al., 2012). Additionally, a study by Bellantuono, Granados-Cifuentes, Miller, Hoegh-Guldberg, and Rodriguez-Lanetty (2012),  revealed that short-term preconditioning of Acropora millepora to thermal stress allowed the corals to acclimatize to subsequent heat stress without any change in the make-up on their symbiont types .
Corals have been shown to associate with a highly diverse group of bacteria that contribute important functions to the coral holobiont (Ainsworth et al., 2015;Bourne, Morrow, & Webster, 2016;Kim, 2006;Pootakham et al., 2017). Members of coral-associated bacterial communities appear to be involved in the provisioning and cycling of carbon, nitrogen and sulfur in coral reefs (Lema, Willis, & Bourne, 2012;Lesser et al., 2007;Rädecker et al. 2015;Raina, Tapiolas, Willis, & Bourne, 2009;Wegley, Edwards, Rodriguez-Brito, Liu, & Rohwer, 2007). These prokaryotes may also function in protecting their hosts against pathogenic microbes by preventing their colonization through physical occupation of otherwise available niches (Ritchie & Smith, 2004) or through the production of antibacterial compounds (Krediet, Ritchie, Paul, & Teplitski, 2013). A study by Gilbert, Hill, Doblin, and Ralph (2012) revealed that an intact bacterial consortium provided the coral holobiont resilience against thermal stress. The authors applied antibiotics to artificially manipulate the resident bacterial community and demonstrated that undisturbed bacterial consortium ameliorated the thermal stress response and promoted the recovery of their hosts from bleaching events (Gilbert et al., 2012). More recently, microbiome dynamics has been shown to be linked to patterns of coral heat tolerance (Ziegler, Seneca, Yum, Palumbi, & Voolstra, 2017). The authors demonstrated that the composition of bacterial community associated with Acropora hyacinthus was different across thermally variable habitats, and the community adapted to the new environment when corals were reciprocally transplated. Changes in coralassociated bacterial communities during a bleaching event have also been documented in A. millepora from the Great Barrier Reef (Bourne, Iida, Uthicke, & Smith-Keune, 2008) and Acropora muricata in Nanwan, Taiwan (Lee, Davy, Tang, & Kench, 2016). In one of the studies, the denaturing gradient gel electrophoresis (DGGE) patterns revealed a correlation between increasing seawater temperature and the appearance of Vibrio-affiliated sequences (Bourne et al., 2008).
The scleractinian coral Porites lutea is one of the dominant reef builders widely distributed in the Indo-West Pacific (Yeemin et al., 2009). Over the past two decades, a number of bleaching events have been reported in the Gulf of Thailand and Andaman Sea (Phongsuwan et al., 2013). The resilience of coral holobionts to environmental stressors, including elevated seawater temperature, appears to be strongly influenced by members of their associated microbial communities.
Even though the important contribution of coral-associated bacteria to the overall fitness and long-term survival of coral hosts have been demonstrated in several species (Ainsworth et al., 2015;Bourne et al., 2008;Gilbert et al., 2012;Tout, Siboni et al., 2015;Vega Thurber et al., 2009;Ziegler et al., 2017), there has not been a study that examines the shifts in bacterial and algal communities associated with P. lutea during a thermal stress.
Here, we investigate the dynamics of coral-associated microbiomes during a recent thermal bleaching event in the Andaman Sea. We employed a long-read PacBio SMRT sequencing technology to obtain full-length 16S rRNA and internal transcribed spacer (ITS) sequences and thoroughly examined the structure and dynamics of microbial communities associated with P. lutea prior to, during and following a bleaching event in 2016. While there were no significant changes in the structure and diversity of coral-associated Symbiodinium, we observed that the bacterial communities became more diverse as their coral hosts experienced elevated water temperature. The balance of the symbiosis shifted from a predominant association between P. lutea and Gammaproteobacteria to a predominance of Alphaproteobacteria following the bleaching event. The comparison of bacterial community compositions and potential shifts in microbiomes between pre-bleaching, bleaching, and post-bleaching corals may be useful in bridging the knowledge gap on how these associated bacteria contribute to the resilience of their coral host to thermal bleaching.

| EXPERIMENTAL PROCEDURES
2.1 | Field sample collection and temperature monitoring P. lutea samples were collected at the depth of 7 m from two locations in the Andaman Sea: Maiton (MT; 7°45′44.5″N 98°28′42.9″E) and Racha islands (RC; 7°36′8″N 98°21′58″E; Figure 1a). Three and four colonies were tagged at MT and RC, respectively, and coral samples were collected during the following stages: pre-bleaching (March), bleaching (May), and post-bleaching (August) in 2016. Coral colonies were visually healthy in March and August, but displayed signs of color loss in >80% of the colony surface area during the bleaching period in May (Figure 1c). One nubbin from each coral colony was collected underwater using scalpel blades, placed in sterile 2-ml screw-capped tubes without any air space and transported back to shore within  Palstra, Piquet, & Miller, 2001). In order to multiplex amplicons from several samples into a single SMRTbell library, we employed a previously described two-step PCR approach to add unique identifier (barcode) sequences to 16S rRNA/ITS amplicons from individual samples (Pootakham et al., 2017). First rounds of PCR were carried out using M13-tagged forward (27F or symITS1-FP) and reverse (1492R or ITS2REV2) primers in a final volume of 20 μl con-

| 16S rRNA and ITS sequence data analysis
RS-ReadsOfInsert protocol (SMRT Analysis software version 2.3) was used to demultiplex and process PacBio raw reads to obtain consensus sequences with a minimum of five full passes. ITS consensus reads shorter than 650 nt were removed (in order to exclude sequences derived from partial amplification) prior to downstream analyses. Filtered sequences were aligned to the ITS database (Tong et al., 2017) using BLASTN with an E-value cutoff of 10 −5 and the clade/subclade identification was assigned based on 97% sequence similarity. An OTU table was subsequently generated from the clade/subclade alignment data, and alpha diversity analyses were performed using the "vegan" package in R (Oksanen, Blanchet, Kindt, Legendre, & Minchin, 2013).
The analysis of similarities (ANOSIM) was used to compare community assemblages based on the Bray-Curtis distance matrix as implemented in the vegan function Adonis (permutation = 1000) (Oksanen et al., 2013).
For 16S rRNA CCS reads, we filtered out consensus sequences that were shorter than 1000 nt (in order to exclude sequences derived from partial amplification), and chimeric detection was performed on both strands, using the abundance-based algorithm implemented in UCHIME (Edgar, 2010) and a reference dataset from RDP (Cole et al., 2014). Filtered sequences were analyzed using QIIME software version 1.9.1 (Caporaso et al., 2010). Once the close-reference OTU picking was finished, the remaining sequences were clustered into OTUs based on an open-reference OTU picking method at 97% identity, using UCLUST (Edgar, 2010). A representative sequence for each OTU was selected, and taxonomy was assigned, using the RDP Classifier (Wang, Garrity, Tiedje, & Cole, 2007) retrained toward the Greengenes database (version 13.8) (DeSantis et al., 2006). Alpha and beta diversity analyses were performed using the QIIME pipeline. The OTU table was rarefied to an even depth of 8022 sequences per sample (corresponding to the number of sequences present in the smallest sample) in order to avoid biases from unequal sampling depth. ANOVA and Tukey's honestly significant difference (HSD) post hoc tests were used to test for differences in relative abundance of associated bacteria between samples collected in March, May, and August. A nonparametric permutational multivariate analysis of variance (PERMANOVA) was used to compare community assemblages based on the unweighted Unifrac distance matrix as implemented in the vegan function Adonis (permutation = 1000) (Oksanen et al., 2013). The PacBio 16S rRNA and ITS sequence data were deposited in the GenBank NCBI Sequence Read Archive (SRA) database under the accession number PRJNA392355.

| Performance comparison between fulllength and partial 16S rRNA sequences in species identification
To compare the ability of full-length 16S rRNA, V3-V4 and V5-V6 sequences in species identification, we followed the protocol from Pootakham et al. (2017). First, we extracted the V3-V4 and V5-V6 hypervariable regions from the full-length sequences using the same flanking sequences reported in Pootakham et al. (2017). The fulllength 16S rRNA sequences along with their respective V3-V4 and V5-V6 regions were subsequently aligned to the non-redundant RDP reference (bacterial type strain) sequences using BLAST with an Evalue cutoff of 10 −10 . Any query sequence that returned two or more best hits belonging to different species with identical E-value, bitscore and aligned region was considered ineffective in resolving taxonomic classification at the species level. For taxonomic classification at the genus level, any query sequence that had two or more best hits belonging to different genera was considered "unassigned."

| Identification of P. lutea core microbiome
Candidate members of the P. lutea core microbiome were identified at 75% sample coverage (present in at least 15 out of 19 coral microbiome samples) based on previous studies (Hadaidi et al., 2017;Lawler et al., 2016;van de Water et al., 2016). Sample coverage is defined as the minimal percentage of all samples in which an OTU must be present to be considered part of the core microbiome. Bacterial type strain sequences were downloaded from RDP and clustered at 100% identity using the CD-HIT software (Fu, Niu, Zhu, Wu, & Li, 2012) to obtain a set of nonredundant reference sequences. To identify candidate members of the core microbiome, full-length 16S rRNA sequences were aligned to the nonredundant RDP reference sequences, using BLAST with a sequence identity cutoff of >90%.

| Functional profiling of bacterial communities
We employed a computational approach, PICRUSt (phylogenetic investigation of communities by reconstruction of unobserved states), to predict the functional composition of bacterial communities using 16S rRNA marker data and a database of reference genomes (Langille et al., 2013). We first applied the command "normalize_by_copy_ number.py" to the OTU abundance table to account for differences in gene copy number. Functional predictions of KEGG Orthologous groups (KOs) were carried out using the command "predict_metagenomes.py," and the KOs were summarized to KEGG pathways (level 1, 2, 3) with "categorize_by_function.py." The LEfSe method (Segata et al., 2011) was used to identify significantly different metagenome functions of bacterial communities among samples collected before, during and following the bleaching event (LDA > 3.0).  (Tables S1 and S2). Following the removal of chimeras and 16S rRNA CCS reads shorter than 1000 nt, a total of 238,497 reads were obtained for all samples. The number of processed full-length 16S rRNA sequences ranged from 8022 to 26,438 per sample, with an average of 12,552 reads/sample (Table S1). After removing ITS sequences that were shorter than 650 nt, the number of processed reads ranged from 986 to 10,583 per sample, with an average of 5449 reads/sample (Table S2).

| Assessment of P. lutea-associated bacterial diversity during a thermal bleaching event
To evaluate the diversity of bacterial community present within each sample and among samples, alpha and beta diversity indices were calculated. The highest number of OTUs was observed in RC May (4) sample (1,056 OTUs; Table S1). Rarefaction curves seemed to plateau off for most samples, suggesting that sufficient sampling has been performed to capture the total diversity of the microbiomes ( Figure S1). In MT, pre-bleaching corals (Mar) appeared to harbor bacterial communities with lower degree of diversity than those exposed to elevated seawater temperature (May and Aug; Figure 2a Similar to our observations in P. lutea corals, an increase in bacterial diversity during a bleaching period has previously been reported in A. millepora (Bourne et al., 2008) and A. muricata (Lee et al., 2016). Bourne et al. (2008) also noticed a return to pre-bleaching diversity levels of coral-associated bacterial communities as A. millepora colonies recovered from the thermal stress, similar to the phenomenon observed with RC corals in our study ( Figure 2a). Interestingly, the bacterial diversity of post-bleaching samples in MT remained relatively high, comparable to that of bleaching corals despite the fact that post-bleaching samples from both locations were from coral colonies that appeared visually healthy (Figure 2a). The difference suggests that local environmental conditions have a profound impact on the structure of coral-associated bacterial communities and how quickly these communities recover after the exposure of a coral holobiont to thermal stress.

| Distinct shifts in P. lutea-associated bacterial communities during a heat stress
A total of 229,982 out of 238,497 filtered reads from both MT and RC were assigned using the RDP classifier (Wang et al., 2007) with a confidence threshold of 80%. Sequences were classified into 18 phyla

| Stability of P. lutea-associated Symbiodinium communities throughout a bleaching event
In contrast to the shifts in bacterial community observed when the corals were under heat stress, minimal changes were observed in Symbiodinium community structures ( Figure S3). The profile of associated Symbiodinium remained essentially the same across samples from different locations and temperature conditions (ANOSIM on Bray-Curtis distance matrix, p = .626). Consistent with previous investigations (Hume et al., 2015;Tanzil et al., 2016;Yamashita, Suzuki, Hayashibara, & Koike, 2011), only members of Symbiodinium clade C were detected in our P. lutea samples collected from both MT and RC islands. Most of the ITS sequence reads (75%) were predominantly affiliated with subclade C116, and the remaining sequences belonged to subclades C15f, C15g, C15h, C15.2, C15.6, and C15.7 ( Figure S3). Even though ITS sequences are the most widely used genetic marker for assessing the diversity of Symbiodinium community, it is worthwhile to mention the caveat of this marker. The multicopy nature and known variability of ITS sequences within individual Symbiodinium cells render it nearly impossible to accurately identify the number of biological entities the sequence data represent (Stat et al., 2011). The presence of intragenomic variations in ITS can potentially confound estimates of microbial diversity in a community (Thornhill, Lajeunesse, & Santos, 2007). Nevertheless, the consequences of under-or overestimating the proportion of F I G U R E 2 Alpha and beta diversity estimates of bacterial communities associated with P. lutea in MT and RC prior to (Mar), during (May), and following (Aug) the bleaching event. (a) A bar chart displaying Shannon diversity indices of coral-associated microbiomes. Bars with different superscript letters (a,b) denote significant differences among RC samples (Tukey's HSD; p < .05). Numbers of independent data points for each sample are as follows: MT Mar n = 2; MT May, MT Aug and RC Mar n = 3; RC May and RC Aug n = 4. Shannon diversity indices were not significantly different among MT samples. (b) Principal coordinate analysis (PCoA) was used to plot the beta diversity of bacterial communities using the unweighted Unifrac Matrix. Red symbols indicate samples collected from MT whereas blue symbols indicate samples collected from RC. The ellipses indicate 95% confidence intervals of each sample group (note that that confidence interval could not be drawn for MT-Mar group since n < 3) each Symbiodinium clade/subclade and the diversity indices in our study are unlikely to affect our conclusion that there was no significant change in the coral-associated Symbiondinium community structure during a thermal bleaching event.
P. lutea is one of the coral species in which eggs or larvae acquires Symbiodinium from their female parents through a vertical transmission (Baird, Guest, & Willis, 2009). Vertically transmitted symbiont communities are often found in brooding corals with internal fertilization and are hypothesized to be of lower diversity and higher fidelity (Baird et al., 2009). A low degree of diversity observed in P. lutea-associated Symbiodinium may be due to this direct transfer of symbionts from parents to offspring (Byler, Carmi-Veal, Fine, & Goulet, 2013). This type of symbiont inheritance may reduce the likelihood of progeny obtaining new type/clade of Symbiodinium from the benthos.
Several studies have demonstrated shifts in Symbiodinium clades, by either switching or shuffling less heat tolerant for more heat tolerant types, enabling acclimation of coral holobiont to increasing seawater temperatures (Boulotte et al., 2016;Jones et al., 2008;Rowan, 2004 (Jones et al., 2008). Contrary to what was reported for other coral species (Boulotte et al., 2016;Jones et al., 2008;Rowan, 2004), the effect of increasing seawater temperature on the distribution of each subclade within the P. lutea-associated algal community F I G U R E 3 Composition of bacterial communities associated with P. lutea prior to (Mar), during (May) and following (Aug) a thermal bleaching event in MT and RC. Depicted is taxonomic classification of OTUs present in each sample group at the class/order levels based on Greengenes database, using QIIME software. Fourteen most abundant orders from nine classes are plotted, and the remaining taxa are grouped under "others" F I G U R E 4 Distribution of 18 most prevalent bacterial species associated with P. lutea samples from MT and RC before (Mar), during (May) and after (Aug) a thermal bleaching event in the Andaman Sea. An asterisk (*) and a dagger ( †) denote significant differences (Tukey's HSD; p < .05) among samples from MT and RC, respectively was negligible (PERMANOVA on Bray-Curtis distance matrix, F = 1, p = .412). The structure of Symbiodinium communities harbored by P. lutea from both MT and RC remained fairly stable throughout the bleaching event ( Figure S3). Similar to earlier observations, we noticed that P. lutea collected in the Andaman Sea exclusively harbored Symbiodinium clade C (Hume et al., 2015;Tanzil et al., 2016;Yamashita et al., 2011). A plausible explanation for the absence of a symbiont partner shift in our study is that these hosts have already established an optimal symbiotic relationship with a specific consortium of Symbiodinium that allows the coral holobionts to survive and thrive in their local environments. After multiple occurrences of thermal bleaching in the past few decades (1991, 1995, 2003, and 2010), the specific association of P. lutea with Symbiodinium clade C may be an optimal one that allows both partners to survive an increase in water temperature (Chavanich, Viyakarn, Adams, Klammer, & Cook, 2012;Wall et al., 2015). Switching or shuffling Symbiodinium types may not be the primary mechanism used by this coral species to acclimate to increasing seawater temperature, and shifts in the structure and diversity of and our study suggest that physiological acclimatization of the host and symbionts is a significant part of the response to heat stress. Host physiology and ability to induce stress response through the upregulation of heat-shock proteins has been suggested to play a role in heat stress tolerance in addition to harboring heat-resistant symbiont (Fitt et al., 2009;Rodriguez-Lanetty, Harii, & Hoegh-Guldberg, 2009). To further investigate changes at the cellular level in both corals and their symbionts, future work on gene expression analysis in P. lutea during heat stress will be required.

| Changes in P. lutea-associated microbiomes during a heat stress
In contrast to an insignificant change in coral-associated algal communities during heat stress, shifts in bacterial communities were evident when P. lutea were exposed to elevated seawater temperature.
There was a marked decline in the apparent relative abundance of Gammaproteobacteria, particularly members of Oceanospirillales ( Figure 3 and Figure S2). Concomitantly, the microbiome structure of bleaching corals shifted toward a predominance of Alphaproteobacteria (Rhizobiales, Rhodobacterales, Caulobacterales, Rhodospirillales) and to a lesser extent Betaproteobacteria (Burkholderiales). Similar to our observation, a shift from Oceanospirillales-to Rhodobacteralesdominating community was observed over a 7-day heat stress experiment (McDevitt-Irwin, Baum, Garren, & Vega Thurber, 2017;Tout, Siboni et al., 2015). Tout, Jeffries et al. (2015); Tout, Siboni et al. (2015) also reported that P. damicornis-associated bacterial community experiencing heat stress had significantly higher diversity. Furthermore, studies performed in A. millepora (Bourne et al., 2008) and A. muricata (Lee et al., 2016) demonstrated that bacterial community became more diverse as the coral hosts experienced thermal stress and that the community shifted from being dominated by Gammaproteobacteria to Alphaproteobacteria at higher temperatures. The thermal-induced shift of bacterial population away from a stable microbiota appears to be a common response among scleractinian corals. Such changes in the bacterial communities are likely to affect the physiological function of these communities. Lee et al. (2016) (Lee et al., 2016). A number of studies observed a prominent increase in relative abundance of Vibrio-affiliated sequences during bleaching (Bourne et al., 2008;Lee et al., 2016;Tout, Siboni et al., 2015). Vibrio species are often acknowledged for their roles as opportunistic or pathogenic bacteria associated with coral diseases (Roder et al., 2014;Sweet & Bythell, 2012). Interestingly, significant shifts in the relative sequence abundance of Proteobacteria classes between healthy and white plaque disease affected tissues observed in Diploria strigosa and Siderastrea siderea were similar to the changes observed in coral tissues experiencing heat stress (Cardenas, Rodriguez, Pizarro, Cadavid, & Arevalo-Ferro, 2012). Diseased tissues showed a consistent increase in Alphaproteobacteria (Rhodobacterales) associated with a decrease in Gammaproteobacteria (Oceanospiralles) compared to samples from healthy tissues. Despite the similarity in the bacterial profile shifts associated with disease and thermal bleaching, we did not observe an increase in relative abundance of Vibrio or other known coral pathogen species under elevated seawater temperature.
As a matter of fact, Vibrio-related sequences were barely detected in our samples across locations and time periods, suggesting that the shifts in coral-associated microbiomes observed in this study were unlikely to be related to the effects of a pathogenic agent. The increase in relative abundance of Alphaproteobacteria may simply reflect an opportunistic colonization of this group of bacterial communities on coral tissues damaged from the heat stress. The reciprocal decrease in relative abundance of Gammaproteobacteria may favor the increase in Alphaproteobacteria by a loss of antagonism or other growth control mechanisms (e.g., antibiotic production) in nonstressed microbiomes.
Even though bacterial communities associated with P. lutea samples from MT and RC exhibited a similar shift from being Gammaproteobacteria-dominated to Alphaproteobacteria-dominated during heat stress, only the microbiomes associated with samples T A B L E 1 A list of bacteria present in P. lutea core microbiome (species that were found in at least 75% of all samples), their relative abundances (average abundance across all 19 samples) and their average ubiquity (defined as a percentage of P. lutea samples in which the species was detected)

ND
Asterisks denote species that were not identified as parts of the core microbiome themselves in the previous study (Pootakham et al., 2017) but their closely related species in the same genus were. ND indicates that the species was not included as members of the core microbiome in the previous study. Species in bold fonts were identified as part of the P. lutea core microbiome in both studies.
from RC appeared to recover and return to the structure and diversity observed prior to the bleaching event within 3 months (Figure 3 and Figure S2). In contrast, the profiles of MT-Aug microbiomes remained similar to those of MT-May samples, suggesting that the dynamics and the recovery of coral-associated bacterial communities were affected by local environmental factors. Daily water temperatures between January to May appeared to fluctuate more strongly in RC compared to MT (Figure 1b). These high-frequency step changes in temperature were likely due to large-amplitude internal waves, which are strongest during the dry northeast monsoon season (January through March) (Wall et al., 2015). RC island is located on the continental shelf in the Andaman Sea and is exposed to large-amplitude internal wave impact while MT island is sheltered from those waves (hence, less daily temperature fluctuation). A study by Wall et al. (2015) demonstrated that the cooling effect of large-amplitude internal waves was beneficial to coral reefs during thermal stress as coral groups from the area exposed to large-amplitude internal waves showed milder bleaching than those from the area sheltered from those waves. Even though MT and RC colonies chosen for this study exhibited similar bleaching profiles (signs of color loss in >80% of the colony surface area in May and completely recovered in August), RC coral-associated microbiomes might benefit from the cooling effect of large-amplitude internal waves in such a way that allowed them to return to the community structure observed prior to the bleaching event within a few months. It is also plausible that other environmental factors in RC (e.g., salinity and pH) influence the rate at which the structure and diversity of P. lutea microbiomes returned to the pre-bleaching configuration. The composition of post-bleaching microbiomes in MT may take longer to return to their original state. If we were to collect another set of samples toward the end of 2016, we might observe the microbial community structure that was similar to that of the pre-bleaching samples. These results highlight the importance of local environmental factors on the dynamics of coral-associated bacterial communities in response to thermal stress.

| The P. lutea core microbiome
Despite the differences in microbial communities associated with P. lutea from MT and RC throughout the bleaching event, we examined the conserved bacterial species that were consistently present in at least 75% of the samples regardless of the relative abundance of each species. Following this criterion, we identified 17 species belonging to a diverse group of bacteria from 9 families, 6 orders and 4 classes as members of the core microbiome (Table 1).
Gammaproteobacteria and Alphaproteobacteria constituted the greatest proportions of the core microbiomes, representing 63.51% and 27.33%, respectively (Table 1). Full-length 16S sequences have an advantage of covering all hypervariable regions of the 16S rRNA genes and enable high-resolution taxonomic classification of the core microbiome members at the species level. The majority of the conserved bacterial community belonged to the genus Endozoicomonas, with five species identified as members of the core microbiome (Table 1). Three highly conserved core members consistently present at 100% sample coverage across spatial and temporal variations were members of Hahellaceae (E. elysicola, E. montiporae) and Bradyrhizobiaceae (Bradyrhizobium pachyrhizi) families. Eleven out of seventeen species have also been identified as P. lutea core microbiome members in a previous study, and it is noteworthy to mention that E. elysicola have consistently been detected in all P. lutea samples collected from seven sampling sites in the Gulf of Thailand and Andaman Sea (Pootakham et al., 2017). Compared to other Oceanospirillales symbionts, Endozoicomonas genomes appear to be enriched for genes involved in transport activities, especially carbon sugar transport and the secretion of proteins (Neave, Michell, Apprill, & Voolstra, 2017). This enrichment in transport and secretion may allow efficient transfer of organic molecules between the symbionts and corals, suggesting that Endozoicomonas species play important roles in the upcycling or carbohydrates and the provision of proteins to the host (Neave et al., 2017). Furthermore, the ability of E. montiporae and E. elysicola to utilize a wide variety of sugars found in coral mucus (Lee et al., 2016) probably allows these two Endozoicomonas species to remain associated with P. lutea throughout the bleaching event, regardless of the change in coral mucus composition resulting from elevated seawater temperature.
The majority of species identified as members of the P. lutea core microbiome were present in low abundance (Table 1). Of 17 species found within at least 75% of all samples, 11 had a relative abundance lower than 5% within the whole community, including the three highly conserved core members (present at 100% sample coverage). This is consistent with previous findings that bacteria that form stable and species-specific associations may be present at low relative abundance in the coral microbiome (Ainsworth et al., 2015;Hernandez-Agreda, Gates, & Ainsworth, 2017;Littman, Willis, Pfeffer, & Bourne, 2009;Pootakham et al., 2017). In addition to the low abundant members of the core microbiome, we observed the presence of two highly abun-

| Full-length 16S rRNA sequence data allow taxonomical classification at the species level
In this study, we utilized the long-read PacBio sequencing technology to capture full-length 16S rRNA sequences to investigate the structure and diversity of coral-associated bacterial communities during a bleaching event. To demonstrate a key advantage of using fulllength sequences in 16S rRNA gene-based community surveys, we compared the resolution of microbial community analyses performed using short hypervariable regions and full-length 16S rRNA sequences. We extracted the commonly used V3-V4 and V5-V6 regions from the full-length reads and aligned both the in silico amplicons (420-bp V3-V4 and 255-bp V5-V6 fragments) and their respective full-length sequences against non-redundant reference sequences in the RDP (Cole et al., 2014). We observed marked differences in the proportions of sequences that could be classified at the species level using V3-V4, V5-V6 regions or full-length 16S amplicons ( Figure 5).
Over 99.4% of the full-length 16S rRNA sequences were taxonomically classified at the species level while 59%-98% and 59%-80% of the V3-V4 and V5-V6 sequences, respectively, could be assigned to specific species. These results showed that partial 16S rRNA sequence information from the hypervariable V3-V4 or V5-V6 regions is often insufficient for high-resolution taxonomic assignments at the species level.

| Microbial functional profiles change during thermal stress
To explore the relationship between functional genes of coralassociated microbial communities and seawater temperatures, we employed a computational approach PICRUSt (Langille et al., 2013) to assess putative functional changes underlying the distinct microbial communities associated with corals during a heat stress ( Table 2) damicornis during heat stress (Garren et al., 2014). However, since we did not detect pathogen-related sequences in any of our microbiome samples, the enrichment of functional profiles related to chemotaxis observed here is likely associated with successful establishment and maintenance of specific coral-bacteria associations, which ultimately influence the health and stability of the coral holobionts under optimal conditions. Functional profiles of microbial communities associated with P. lutea during heat stress were characterized by the enrichment of functions related to amino acid metabolism (Table 2). Several proteins involved in amino acid transport system (permeases and ATP-binding proteins) were overrepresented in the microbiomes associated with corals experiencing thermal stress. We observed F I G U R E 5 A bar chart illustrating the percentages of 16S rRNA sequence reads that are classifiable at the species level, using fulllength (blue) or partial amplicons (V3-V4, red; V5-V6, green) T A B L E 2 Enrichment of functional proteins in P. lutea-associated microbial communities prior to (Mar) and during a heat stress (May) an enrichment in guanosine-5′-triphosphate,3′-diphosphate pyrophosphatase, an enzyme that catalyzes the conversion of pppGpp (guanosine-5′-triphosphate,3′-diphosphate) to ppGpp (guanosine 3′,5′-bis(diphosphate)), a cytoplasmic signaling molecule that controls the "stringent response," an adaptive process that allows bacteria to respond to amino acid starvation (Hara & Sy, 1983;Hauryliuk, Atkinson, Murakami, Tenson, & Gerdes, 2015). Under optimal conditions, coral hosts may be sharing nutrients with their associated microorganisms. However, an increase in seawater temperature may trigger the breakdown of this symbiotic relationship and as a consequence, the associated microbes may be experiencing an amino acid starvation. Recently, Zhou et al. (2017) demonstrated that five gene ontology (GO) terms overrepresented in heat stressed P. damicornis were related to TNF signaling pathway, apoptosis and cell death . Similar results were also reported in Acropora aspera (Rosic et al., 2014), A. hyacinthus (Seneca & Palumbi, 2015) and Orbicella faveolata (Pinzón et al., 2015). Amino acids released from bleached coral tissues undergoing apoptosis and cell death are likely scavenged by their associated microbes. This amino acid scavenging activity appears to be congruent with the observed enrichment in amino acid transport proteins in the functional profiles of microbial communities associated with corals during thermal stress.
A number of studies have highlighted the importance of coralassociated microorganisms and their roles in fitness and survival of the host animals (Ainsworth et al., 2015;Bourne et al., 2008;Gilbert et al., 2012;Tout, Siboni et al., 2015;Vega Thurber et al., 2009;Ziegler et al., 2017). Nonetheless, prior to this work, very little was known about the effects of thermal stress on the dynamics of P. lutea-associated Symbiodinium and bacterial communities. Our study thoroughly examined the shifts in both bacterial and algal populations associated with Porites corals during a natural bleaching event. While there were minimal changes observed in Symbiodinium community diversity and composition, we provided the evidence of dramatic shifts in structure and diversity of associated bacterial communities when corals were exposed to elevated seawater temperature. The information on how coral-associated microbiomes shifts their compositions between pre-bleaching, bleaching, and post-bleaching stages may help us further our understanding on how these associated bacterial contribute to the resilience of their coral host to thermal bleaching.