Coral samples were collected from four sites along the northeastern coast of Taiwan from September to October 2020 ( Supplementary Figure S1 ; Supplementary Table S1 ). The selection of the study sites was based on their direct exposure to environmental and anthropogenic stressors. The shallow hydrothermal vent (HV) site is situated on Kueishan Island, approximately 10 km off the northeast coast of Taiwan. Kueishan Island is known for accommodating numerous gaseohydrothermal vents, ranging from depths of 10–300 m, and it is home to the world’s most acidic vents, with pH of as low as 1.52 (Chen et al., 2005a; Chan et al., 2016).

Carbon dioxide, along with hydrogen sulfide, constitutes the majority of the gases released by these vents (92%) (Tang et al., 2013; Chen et al., 2018). The coral samples from the HV site were collected from a depth of 20 m near white vents. The nuclear power plant (NPP) inlet is characterized by minimal or no direct ecological and man-made disturbances. T. aurea colonies were found to be attached to wave breakers in the intertidal region of the NPP inlet. Similarly, the site adjacent to the conservation zone (CZ) is less susceptible to direct natural and anthropogenic stressors; the coral colonies were collected from a rock at a depth of 2–3 m. In contrast to the other sampling sites, the YYS is characterized by a mixture of seawater and polluted water from the copper refinery of the Taiwan Metal Mining Corporation. This sampling site was referred to as the copper mining site (CM). Effluents with dissolved iron and aluminum contain chips of ferric and aluminous hydroxide upon contact with estuarine water, which in turn absorbs heavy metals suspended in clay and carries them to the YYS (Fang et al., 2003). Coral samples were collected from the CM site at a depth of 8–9 m.

Three T. aurea colonies were collected from each of the HV, CM, and CZ sites by scuba diving. At the NPP site, colonies were collected from concrete wave blockers during the lowest low tide. The size of the colonies had a width of 5–10 cm. In addition to coral colonies, ambient water and sediment samples were collected in triplicate. Water samples were collected in 1 L sterile glass bottles, and sediment samples were collected in sterile 50 mL falcon tubes in triplicate. The coral colonies and the sediments were immediately preserved in liquid nitrogen upon reaching the surface. The collected water samples were transported to the laboratory in an icebox and filtered using 0.22 µm mixed cellulose ester membrane filter paper (ADVANTEC, Tokyo, Japan). All samples were stored in a refrigerator at −80°C in the laboratory until further processing or DNA extraction.

The test samples contained in dry filter papers were digested for 3 h at 120°C in Teflon tubes with 1 mL nitric acid (Merck Suprapur®, 65%, Darmstadt, Germany) and 3 mL hydrochloric acid (Merck Suprapur®, 33%). The resulting digested solutions were diluted with ultrapure water to a final volume of 10 mL and stored at 4°C until analysis. Graphite atomic absorption spectrophotometry (GAAS) was performed to determine the contents of heavy metals (Cd, Zn, Pb, Cu, Ni, Cr, Fe, and Hg) in triplicate (Hitachi Z-8000, Japan; Pai, 1988). Standard reference material samples (NASS-3), including bovine liver (NBSSRM-1577) and orchard leaf (NASS-3), were utilized to assess data quality (NBS-SRM-1571).

Physical and chemical parameters were analyzed in accordance with the procedure described by Meng et al. (2008) and Huang et al. (2012). Nutrients such as ammonia, nitrate, nitrite, phosphate, and silicate were measured using a flow injection analyzer (FIA) and a spectrophotometer (Hitachi Model U-3000, Japan; Pai et al., 2001). Ammonia was determined using an indophenol–blue-based spectrophotometric method. Nitrate was converted to nitrite through cadmium reduction, and nitrite was measured by diazoting with sulfanilamide and coupling with N-(1-naphtha)-ethylenediamine-dihydrochloride. Phosphorus was determined using the molybdate–antimony technique. Silicate was detected using the silicomolybdate method and subsequently reduced by a metal–oxalic acid solution. Ammonia, nitrite, nitrate, phosphate, and silicate had the following precision and minimum detection limits (MDLs): 3.8% and 0.71 mM; 0.9% and 0.03 mM; 5.0% and 0.10 mM; and 4.5% and 0.06 mM, respectively. Duplicate, standard sample spiking, and quality assurance (QA) sample analyses were performed to standardize each nutrient measurement. For example, the duplicate, standard sample spiking, and QA sample phosphate examination were 0.16%, 4.1%, and 4.03%, respectively. The Winkler method, with an average of three readings at each sampling, was applied to measure dissolved oxygen (DO) by using a multiparameter monitoring sensor (YSI model 600XLM, Yellow springs, Ohio, USA). The precision and accuracy of the measurements of biological oxygen demand for 5 days (BOD₅) were ±2.38% and 98.3% ± 7.7%, respectively, which were verified using a control chart. pH was measured using a pH meter from Hanna Instruments (Smithfield, Rhode Island, USA). A spectrophotometric approach was applied to assess phytoplankton Chl-a. In this approach, water samples were filtered using a Whatman GF/C glass microfiber filter, and Chl-a was extracted in 90% aqueous acetone at 4°C in the dark for 24 h. The turbidity of the water samples was determined using a nephelometric approach.

The tissue samples were thawed at room temperature, and the mucus was collected by gently swabbing a sterile cotton ball on the coral surface. The coral colony was then washed with sterile seawater to remove any remaining mucus and loosely attached bacteria. Subsequently, the tissue was carefully scraped off the skeleton by using a sterile scalpel. The scraped tissue and some skeletal particles were ground to a fine paste with a sterile mortar and pestle. For DNA extraction, 0.25 g of the ground tissue paste was used. The cotton balls containing coral mucus were cut into small pieces with sterile scissors and forceps. From the cut cotton balls, 0.25 g was used for DNA extraction. Similarly, the filter papers (0.22 µm) were cut into pieces for the seawater samples, and 0.25 g of the materials were taken for DNA extraction. DNA was extracted from the tissue, mucus, sediment (0.25 g), and water samples from all sites by using a DNeasy PowerSoil kit (QIAGEN, Hilden, Germany) in accordance with the manufacturer’s protocol.

DNA was extracted from each sample and amplified with the V1–V9 full-length (~1.5 kbp) 16S rRNA gene by using 16S rRNA gene primer sets (Urban et al., 2021). The universal primers 27F and 1492R were utilized for bacterial 16S rRNA amplification. Both primers were combined with a unique barcode to identify different samples ( Supplementary Table S2 ). In the full-length PCR of bacterial 16S rRNA, the following were used: 20.5 µL of sterilized water, 25 µL of PCR master mix (2x Master mix red), 4.0 µL of primer set (2 µL forward and 2 µL reverse) with barcode sequences (10 µM), and 0.5 µL of sample DNA extract in a total volume of 50 µL. The following PCR conditions were used for amplification: initial denaturation at 95°C for 3 min; 30 cycles of 95°C for 30 s, 55°C for 30 s, and 72°C for 1.5 min; and a final extension at 72°C for 5 min.

The amplicons were purified and the full-length 16S rRNA gene products were concentrated using an AMPure XP purification kit (Beckman Coulter Life Sciences, Indianapolis, USA). After the amplicons were mixed with the AMPure XP purification kit, the reactions were left at 25°C for 20 min. Subsequently, the samples were washed with 70% fresh ethanol for one round, air-dried at room temperature for 1 min, and eluted in 30 µL of sterilized ddH₂O. Each purified sample was quantified using a Qubit 4 fluorometer dsDNA HS (Thermo Fisher Scientific, Waltham, United States). Forty-eight purified samples were pooled at equimolar ratios of approximately 2000 ng. Then, nanopore sequencing libraries were generated in accordance with the SQK-LSK109 protocol (Oxford Nanopore Technologies, Oxford, UK). The library (75 µL) was loaded into R9.4.1 MinION flow cells (Oxford Nanopore Technologies) and run for approximately 96 h. Afterward, the reads were base-called using Guppy (version 3.15) with a high-accuracy model (HAC), and the output DNA sequences with a quality score (Q) of >9 were saved as “.fastq” files for subsequent analyses.

The package “Porechop” (version 0.2.4, https://github.com/rrwick/porechop) was employed to demultiplex reads and trim adapters under the default parameters. Nanofilt (version 2.8.0, https://github.com/wdecoster/nanofilt) was used to remove any reads shorter than 1.4 kbp and longer than 1.6 kbp (De Coster et al., 2018). NanoStat (version 1.1.2, https://github.com/wdecoster/nanostat; De Coster et al., 2018) was utilized to examine read statistics, including quality scores and read lengths. Additionally, Pistis (https://github.com/mbhall88/pistis) was used to generate quality control plots. The k-mer-based read computation tool Centrifuge (version 1.0.4), which relies on a Burrows–Wheeler transform and the Ferragina–Manzini index, was utilized to obtain the taxonomic assignments of the reads (Kim et al., 2016). For this purpose, the Centrifuge database was constructed using 16s rRNA sequencing data of the type strain from the National Center for Biotechnology Information (NCBI; released on July 2022, https://ftp.ncbi.nlm.nih.gov). With the NCBI reference database, a more precise taxonomic classification than SILVA could be obtained, because the SILVA database lacks species-level information (Matsuo et al., 2021). The custom Centrifuge database has been uploaded to the Mendeley Data, an online repository (DOI: 10.17632/tt5ntb4zyv.1), along with the manufacturing process. The results were visualized using the Pavian package on the R language (Breitwieser and Salzberg, 2020). After the taxonomic assignments of the reads were obtained, the read data were rarefied by removing the following: (1) unclassified reads, (2) reads of cellular organisms, and (3) bacterial groups with reads comprising <0.05% of the total reads obtained. This procedure was accomplished using the Vegan package version 2.6.2. Rarefaction was set to 22,555 reads per sample ( Supplementary Table S3 ) and conducted to mitigate the potential false discovery rate caused by significant disparities in bacterial read counts between samples (Weiss et al., 2017).

Bacterial read data were statistically analyzed using R version 4.2.0 and Past version 4.05. A rarefied taxon rank data table was used for statistical analysis. The Vegan package version 2.6.2 (https://CRAN.R-project.org/package=vegan) was used to calculate the alpha-diversity indices; for visualization, box plots were created using the ggplot2 package (Wickham, 2009). The Bray–Curtis dissimilarity index was calculated and visualized through principal coordinate analysis (PCoA) to assess the difference in bacterial community composition between sites. The significance of the dissimilarity was determined through two-way PERMANOVA with 999 permutations using the Adonis function of the Vegan R package (https://CRAN.R-project.org/package=vegan). PERMDIS was conducted using the Vegan R package to verify whether the observed variations in multivariate datasets are influenced by differences in the spread or variability of data points. ANOVA was performed to assess the significance of PERMDIS analysis. Similarity percentage analysis (SIMPER) based on the Bray–Curtis dissimilarity matrices was carried out in PAST 4.05 to identify the bacterial genera with a higher contribution to the dissimilarity between sites. A Wald chi-square test was conducted to identify the significantly expressed dominant bacterial groups between sites by using the DESeq2 package (Love et al., 2014). The Functional Annotation of the Prokaryotic Taxa (FAPROTAX) database was employed to predict the relevant function of the bacteria (Louca et al., 2016). FAPROTAX can transform taxonomic bacterial community profiles into possible functional profiles by considering the species present in a sample. Spearman’s rank correlation was performed using the microbiomeSeq package (Torondel et al., 2016) to understand the correlation of dominant bacterial taxa with hydrographic parameters and the environmental microbiome. A heatmap was generated using abundance data by ClustVis (Metsalu and Vilo, 2015) to examine the difference in the distribution of Endozoicomonas species between sites. All analyses for testing the significance and correlation were corrected for multiple comparisons by using the Benjamini–Hochberg false discovery rate (FDR) method.

The bacterial communities in the tissues and mucus of the coral collected from four different sites were subjected to nanopore sequencing to investigate the variation in the bacterial community structure associated with the tissue and mucus of T. aurea in various habitats; ambient seawater and sediment were also analyzed to examine any influence of the ambient environment on changes in the bacterial community composition. A total of 11.2 million reads were obtained from 48 collected samples. Among them, 79.1% (n = 8,889,171) of raw reads were taxonomically assigned to the genus level ( Supplementary Figure S2 ). For statistical analysis, the overall rarefied taxonomy count data were used across different sites and among different sample types ( Supplementary Table S3 ).

Alpha-diversity was computed using the rarefied taxonomy count data to estimate the diversity of bacterial communities associated with not only the tissue and mucus samples of T. aurea but also the water and sediment samples within each habitat. The bacterial richness (Chao1), diversity (Shannon), and evenness (Pielou) of the sediment were significantly lower at the HV site than at the three other sites (p = 0.023, 0.024, and 0.054, respectively; Tukey’s post hoc test; Figure 1 ). The richness of the tissue samples at the HV site was also lower than that at the three other sites. The average alpha-diversity of the water samples from the HV site was lower than that from the other sites ( Figure 1 ). However, the alpha-diversity of the mucus and water samples varied across the sites, and no significant differences were observed ( Figure 1 ; Supplementary Table S4 ).

The variation in bacterial community composition among different samples was assessed using Bray–Curtis dissimilarity with PCoA based on the rarefied reads. The bacterial community structure of all samples, including the separated tissue, mucus, water, and sediment samples from the HV site, differed from that of the samples from the three other sites ( Figure 2 ). PERMANOVA revealed that the sites significantly affected the bacterial community structure of the coral tissue, mucus, and environmental samples ( Supplementary Table S5 ). The tissue and mucus samples respectively accounted for 57.4% and 50.3% of the variance in bacterial communities between sites. Conversely, the sediment sample contributed to 85.6% of the variance in the bacterial community ( Figure 2 ; Supplementary Table S5 ). The CM site samples showed a significantly high variation of 76.1% in the bacterial community within the site ( Supplementary Table S5 ; Supplementary Figure S3 ). These results indicated that different environmental sites could shape the distinct microbial communities of the coral tissues and mucus. At the CM site, the bacterial community structure significantly varied among different sample types, accounting for 76.1% of the variance (p < 0.001; Supplementary Table S5 ). According to the null hypothesis, dispersion did not differ among the groups. Considering that p > 0.05 ( Supplementary Table S5 ), we accepted the null hypothesis, indicating that no significant differences in dispersion were observed among the groups. Therefore, our results confirmed the presence of position effects (observed by PERMANOVA results, Supplementary Table S5 ) in comparison with dispersion (indicated by PERMDISP results, Supplementary Table S5 ). This finding further supported the validity of our PERMANOVA analysis. However, no significant differences were found using the PERMDIS (beta-dispersion) test ( Supplementary Table S5 ). Furthermore, no significant differences were observed when the samples were grouped by sites (HV, CM, CZ, and NPP) or sample types (tissue, mucus, water, and sediment). Nevertheless, the trend of the pseudo-F value in PERMDISP resembled that in PERMANOVA ( Supplementary Table S5 ).

SIMPER was performed to identify the bacterial genus that contributed the most to the difference in the total bacterial communities between sites in each sample type. Endozoicomonas contributed the most to the Bray–Curtis dissimilarity matrix of the tissue samples from different sites, followed by Synechococcus, Kistomonas, and Sulfurovum ( Supplementary Table S6 ; Supplementary Figure S4 ). The bacterial phenotype that accounted for the dissimilarity in the bacterial communities of the mucus samples from different sites was Methylobacterium, followed by Sulfurovum and Brucella. Additionally, Thiomicrorhabdus, followed by Lebetimonas, Synechococcus, Marinifilum, and Candidatus Pelagibacter, played a crucial role in driving the differences in bacterial communities in the water samples from different sites. Sulfurovum and Woeseia were also responsible for community differences in sediments among different sampling sites ( Figure S4 ).

The abundance data showed that the coral tissue and mucus samples from different habitats had similar bacterial genera, while the relative abundance of the dominant genera varied across habitats ( Figure 3A ). For example, the dominant bacterial genera such as Synechococcus and Ruegeria were found in tissue samples from all sites. However, their relative abundance was comparatively higher in the tissue samples from CZ (5.8%) and NPP (8.3%) sites ( Figure 3A ) than in the other sites. The tissue samples from the HV and CM sites were dominated by the bacterial group Endozoicomonas (18.72% and 7.5%, respectively). Methylobacterium, Ralstonia, and Brucella were present in the mucus samples from all sites; however, their abundance was higher in the mucus samples from the HV (5%, 5.4%, and 3.3%) and CM (14.2%, 4.9%, and 7.7%) sites ( Figure 3A ). Similarly, Candidatus Pelagibacter and Synechococcus were observed in water samples from all sites. Although Candidatus Pelagibacter (7.3%) had higher dominance in the water samples from the CM (12.7%) site, Synechococcus was comparatively abundant in the water samples from CZ (13.3%) and NPP (6.5%) sites ( Figure 3A ). Likewise, Woeseia had higher dominance in the CZ (9%) sediment sample even though this genus was common at all sites ( Figure 3A ).

At the HV site, the bacterial community in the coral tissue sample was dominated by Endozoicomonas (18.73%), followed by Sulfurovum (6.12%), Synechococcus (5.19%), Vibrio (5.15%), and Ruegeria (3.38%). In the mucus sample at the HV site, Sulfurovum (13.64%), along with Ralstonia (5.38%), Methylobacterium (5%), Brucella (3.34%), and Sulfurimonas (2.72%; Figure 3B ), showed a comparatively higher abundance. Among all bacterial genera obtained from the HV site, the most dominant was Sulfurovum, which was found in significantly higher numbers in the sediment samples (64%). At the metal-ion-enriched CM site, Endozoicomonas was dominant (14.30%) in tissue samples.

To understand the significantly different bacterial groups in terms of relative abundance, we performed the Wald chi-square test by using DEseq2. Our results represented only the bacterial groups that were significantly different among the top 15 genera ( Figures 3C–F ). In comparison with the two other sites (CZ and NPP), HV site tissue samples were significantly enriched with Sulfurovum, Sulfurimonas, and Vibrio (p < 0.001, Wald chi-square test), and the mucus sample was significantly enriched with Sulfurovum, Sulfurimonas, and Ruegeria (p < 0.05, Wald chi-square test; Figures 3C, D ). Endozoicomonas and Methylobacterium were significantly enriched in CM site tissues and mucus samples (p < 0.001, Wald chi-square test; Figures 3E, F ).

The metabolic and ecological functions of the bacterial groups in the coral host and the environmental samples were predicted by FAPROTAX. FAPROTAX results suggested that chemoheterotrophs and aerobic chemoheterotrophs were the predominant functional groups observed in all the collected samples except in the HV site sediment samples, which were dominated by nitrate reducers (13.22%), followed by bacteria involved in nitrogen respiration (13.08%) and nitrate respiration (13.03%; Supplementary Figure S5A ). Furthermore, we analyzed the difference in the mean proportion of the relative abundance of the hypothesized functional groups from all samples between different sites and compared the bacterial functional groups between the samples from the HV sites without vent sites (CM, CZ, and NPP). We observed that the percentage mean proportion of sulfur-related functional groups (such as dark oxidation of sulfur compounds, dark sulfur oxidation, dark sulfide oxidation and sulfur respiration) were significantly higher in the HV site samples than outside the HV sites ( Figure 4 ). However, the difference in the mean proportion of photoheterotrophy, nitrogen fixation, and predatory or ectoparasitic taxa was significantly higher in the CM site samples (p = 0.017, p < 0.001, and p < 0.001, respectively), and phototrophy was significantly higher in the CZ site samples (p = 0.045) than in the HV site samples ( Figure 4B ). Phototrophic (p < 0.001), photoautotrophic (p < 0.001), oxygenic photoautotrophic (p < 0.001), and photosynthetic cyanobacteria (p < 0.001) were significantly lower in the HV site samples than in the NPP site samples ( Figure 4C ).

The abundance of the hypothesized functional groups in the tissue samples from the HV site was compared with the mean abundance of the functional groups in the tissue samples from the CZ and NPP sites. Phototrophic, photoautotrophic, oxygenic photoautotrophic, and photosynthetic cyanobacteria significantly decreased in the HV site tissue samples ( Figure 5A ). Similarly, in the HV site mucus samples, phototrophy, photoautotrophy, and aerobic chemoheterotrophy significantly decreased ( Figure 5B ). The functional groups in the tissue samples from the CM site showed no significant changes compared with those in the tissue samples from the CZ and NPP sites ( Figure 5C ). Conversely, sulfur respiration, sulfur compound respiration, phototrophy, photoautotrophy, fermentation, dark thiosulfate oxidation, dark sulfur oxidation, dark sulfur compound oxidation, anoxygenic phototrophic sulfur oxidizers, and anoxygenic phototrophs in the mucus samples from the CM site were significantly reduced compared with those in the mucus samples from the CZ and NPP sites ( Figure 5D ).

Seawater quality parameter analysis revealed that the NPP and CZ sites shared similar conditions ( Table 1 ), while the HV and CM sites had different water quality parameters from those in the other sites. The HV site was comparatively more acidic, with the lowest pH (5.74), followed by CM (pH 7.57). CM was characterized by lower salinity (25.31) and higher concentration of heavy metals such as copper (104.05 µgL^-1) and iron (183.11 µgL^-1). The BOD was greater at the HV site, while turbidity was higher at the CM and HV sites than at the NPP and CZ sites ( Table 1 ). The concentrations of heavy metals such as Cr, Ni, and Cd were extremely low at the four sites, while Zn had a higher concentration at the NPP and CZ sites with 6.65 and 3.80 µgL^-1, respectively ( Table 2 ). The concentrations of nutrients, such as NO₃, NO₂, PO₄, and NH₃, were relatively similar across all four sites, whereas the concentration of SiO₂ was higher at the HV (1.52 mg/L) and CM (1.24 mg/L) sites than at the NPP (0.098 mg/L) and CZ (0.9 mg/L) sites.

Spearman’s rank correlation analysis was performed to analyze the relation between the hydrographic parameters and bacterial communities of the tissue and mucus samples. Several bacterial genera were positively and negatively correlated with hydrographic parameters ( Figure 6A ; Supplementary Table S7 ). In tissue samples, Vallitalea showed the highest positive correlation with salinity (r = 0.96, p < 0.001; Figure 6A ). Thermostilla, Roseivirga, and Denitrobaculum showed a higher negative correlation with BOD₅, SiO₂, and NH₃ (r < -0.8, p < 0.001); they also had a higher positive correlation with pH (r > 0.8, p < 0.001; Figure 6A ; Supplementary Table S7 ). Endozoicomonas was significantly and negatively correlated with pH and salinity, but it was significantly and positively correlated with BOD₅, SiO₂, NH₃, and turbidity ( Figure 6A ) in the tissue samples. For environmental metal concentrations, Endozoicomonas had a highly negative correlation with Zn, Cd and Pb, whereas it was positively correlated with iron (Fe; Figure 6B ). These results demonstrated that certain bacterial groups were highly positively correlated with some of the heavy metals. Therefore, these bacterial groups might be considered indicators of those elements. Vallitalea, Thermostilla, Methyloceanibacter, Kryptousia, Hoeflea, Denitrobaculum, Bythopirellula, Aliterella, and Actinomarinicola were significantly and positively correlated with heavy metals, such as Cd, Zn, and Pb (p < 0.05, Figure 6B ; Supplementary Table S8 ).

We observed that hydrographic parameters such as pH, turbidity, salinity, and BOD₅ at the HV and CM sites differed from those at the two other sites. Changing environmental parameters, such as temperature, salinity, turbidity, reduced pH, and nutrient level loads, cause stressful conditions that can alter coral-associated bacterial communities (Thurber et al., 2009). Hydrographic parameters at the NPP inlet and CM sites did not show any extreme variations ( Table 1 ). Hence, they are considered regular coastal habitats. Studies have not yet reported the environmental conditions prevailing at the CZ site. Lee et al. (2018) suggested that the nuclear power plant inlet has an inflow of natural seawater that cools power plant chambers. Since natural seawater with zooplankton continuously flows, the NPP inlet provides better conditions for the growth of T. aurea. This phenomenon could explain the observed variation in the dominance of bacterial communities and functional groups between sites ( Figure 7 ).

The relative abundance of the dominant bacterial communities in the tissue, mucus, sediment, and water samples varied among the different sites ( Figure 3A ). The dominant genera from the hydrothermal vent samples were Sulfurovum and Sulfurimonas of Epsilonproteobacteria; the dominant genus from the water sample was Thiomicrorhabdus of Gammaproteobacteria ( Figure 3B ). Previous reports showed that bacterial communities from the hydrothermal vent region are dominated by Epsilonproteobacteria and Gammaproteobacteria (Wang et al., 2015; Wang et al., 2017). Sequences related to Sulfurovum have also been found in shallow-water hydrothermal vents off the coast of Milos Island, Greece (Giovannelli et al., 2013). Recent research indicated that Epsilonproteobacteria are chemolithoautotrophs responsible for primary production (Hügler et al., 2005); among them, Sulfurovum can fix CO2 by using elemental sulfur or thiosulfate as the sole electron donor and oxygen or nitrate as the electron acceptor (Inagaki et al., 2004). Similarly, Thiomicrorhabdus can fix CO2 and oxidize sulfide to elemental sulfur (Magnuson et al., 2020). Therefore, the higher abundance of these bacteria in the samples from the HV site could be linked to vent fluids with a surplus of CO₂ and H₂S.

In the present study, Endozoicomonas was the predominant group in the coral tissue samples from the HV and CM sites ( Figure 3A ). The species-level classification of the bacterial groups revealed that nine Endozoicomonas species were present in the tissue samples from the HV and CM sites ( Supplementary Figure S6 ). Endozoicomonas is a core microbiont of corals (Ding et al., 2016) since it commonly provides the dominant bacterial taxon in healthy coral colonies (Vezzulli et al., 2013; Meyer et al., 2014). Previous studies confirmed the occurrence of Endozoicomonas in the associated microbiota of T. coccinea (Engelen et al., 2018). Endozoicomonas has less specificity toward the sun coral because they are less tightly associated with the coral host. The present study showed a contrastingly higher abundance of Endozoicomonas in the tissue of T. aurea collected from the HV (18.73%) and CM (14.3%) sites with ambient water pH of 5.74 and 7.57, respectively. This finding was different from the NPP and CZ sites with high pH of 8.18 and 8.28, respectively ( Table 1 ). Similar to our observation, previous findings showed that Endozoicomonas is enriched in two Porites species at low pH (Shore et al., 2021). Both results suggested that Endozoicomonas might help specific coral hosts survive under common pH conditions. Studies have shown that environmental factors can play a crucial role in shaping microbial communities in coral reefs, leading to variations in the composition of these communities (Zaneveld et al., 2016; Ziegler et al., 2019). T. aurea may use coral-mediated tolerance mechanisms, such as gene expression, genetic adaptation, or other metabolic/genetic processes, to withstand stressful conditions and consequently sustain symbiotic relationships with important microbes such as Endozoicomonas (Reigel et al., 2021).

The differential expression of Endozoicomonas species in extreme environments compared with that in more common habitats shows that they are more acclimated to extreme environmental conditions. The presence of Endozoicomonas may be a good indicator of healthy T. aurea in stressful environments, such as HV and CM sites. The different microbial profiles of the relative abundance of Endozoicomonas species and different major Endozoicomonas species varied between the HV and CM sites ( Supplementary Figure S6 ). E. acroporae dominated the HV site tissue samples, and it is considered the dominant coral-associated Endozoicomonas species, which participates in sulfur cycling via DSMP metabolism (Tandon et al., 2020). E. elysicola and E. montiporae, which were dominant in the HV tissue samples, are involved in nitrogen cycling (Neave et al., 2017). E. euniceicola, E. atrinae, and E. numazuensis, which were dominant in the CM site, are closely related to E. elysicola and E. montiporae. E. eunceicola isolated from octocorals, E. numazuensis isolated from purple marine sponges, and E. atrinae isolated from the gut of a comb pen shell grow in seawater with an optimum NaCl concentration of 2%–3%, which could explain the higher dominance of these species in CM site tissue samples. The comparative analysis of the genomes of E. elysicola, E. atrinae, E. montiporae, E. eunceicola, and E. numazuensis revealed a high proportion of the substantial presence of transposable elements within the genomes of Endozoicomonas, suggesting that these elements likely facilitate rapid evolutionary adaptations inhosts or environmental niches (Neave et al., 2016a). The differential expression of Endozoicomonas species between the tissue samples from different sites could be attributed to the coral-mediated tolerance mechanisms and the presence of transposable elements within the bacterial genomes.

The functional role of Endozoicomonas in hosts is unknown (Neave et al., 2016b; Engelen et al., 2018). Various Endozoicomonas strains can provide various functions, such as producing vitamins and co-factors, cycling carbohydrates, and synthesizing amino acids (Neave et al., 2017). In addition, Endozoicomonas contributes to the health of their host by supplying macromolecular nutrients that are otherwise unavailable to the host, altering the associated microbial community through signaling molecules and antimicrobial action, and contributing to nitrogen and or sulfur cycling (Morrow et al., 2015).

Vibrio was found at the four sites, but its abundance was significantly high in the tissue and mucus samples of the HV site ( Figure 3A ). Several studies have reported an increased abundance of Vibrio in stressed corals (Ben-Haim et al., 2003; Rosenberg and Falkovitz, 2004; Geffen et al., 2009; Rypien et al., 2010; Tait et al., 2010). Meron et al. (2011) observed that the dominance of Vibrio sp. increases when the coral Acropora eurystoma is exposed to reduced pH conditions. This finding suggests that Vibrio can be a natural and common member of the coral-associated microbiome and act as an opportunistic pathogen (Bourne and Munn, 2005). Culture-independent studies have also reported that Vibrio is a vital component of healthy coral tissues (Harris et al., 2001; Cooney et al., 2002; Frias-Lopez et al., 2002; Rohwer et al., 2002). When the balance of the bacterial consortium becomes disrupted under stressed conditions, Vibrio species may become dominant; in some cases, they activate pathogenic traits (Mao-Jones et al., 2010; Littman et al., 2011; Kimes et al., 2012). Therefore, the higher abundance of Vibrio in the tissue samples of the HV site could be attributed to stressors such as reduced pH. However, none of the coral samples collected from these sites showed any symptoms of infectious diseases possibly because some Vibrio strains are permanent members of diazotrophic assemblages found in tissues of some corals (Olson et al., 2009; Lema et al., 2014; Li et al., 2014; Zhang et al., 2016).

The bacterial composition results revealed that the tissue and mucus samples of T. aurea from the NPP and CZ sites and the tissue samples from the HV site had a higher abundance of Synechococcus ( Figure 3A ). We assumed that such a higher abundance might have been influenced by high seawater salinity, and the lower abundance of this genus at the CM site might be due to low salinity, which is 25 psu ( Table 1 ). As a support to our assumption, Xia et al. (2017) reported that the abundance and diversity of Synechococcus in freshwater-dominated low-saline waters are low. The marine clade of Synechococcus likely grows abundantly in high-saline seawater with low turbidity rather than in low-saline regions with high turbidity (Li et al., 2021). This condition could possibly explain the higher abundance of Synechococcus in the tissue and mucus samples at the NPP and CZ sites and the tissue samples at the HV sites, where salinity is higher and turbidity is lower than those at the CM site ( Table 1 ). Cyanobacteria are responsible for N fixation in the tissues of hermatypic corals (Shashar et al., 1994). Jeong et al. (2012) reported that Synechococcus is an important source of nitrogen for the associated Symbiodinium in zooxanthellate corals. However, studies have yet to describe the putative functional role of Synechococcus in ahermatypic corals. We hypothesized that T. aurea may trap Synechococcus from the surrounding water column (Naumann et al., 2009; Hoadley et al., 2021), which then contributes to nitrogen cycling in the coral.

Ruegeria contributed to more than 2% of the bacterial communities in HV, NPP, and CZ ( Figure 3A ). It is one of the abundant bacterial genera associated with several coral hosts, including healthy and diseased ones (Ziegler et al., 2016; Keller-Costa et al., 2017; Huggett and Apprill, 2019). This genus has been recognized as robust among most bacterial lineages and found in a wide range of marine environments; it also has various metabolic capabilities and regulatory mechanisms (Luo and Moran, 2014). Ruegeria species show horizontal gene transfer that may help hosts and their associated microbes quickly adapt to changing environmental conditions (McDaniel et al., 2010; McDaniel et al., 2012). These species also require NaCl or sea salt as an important growth factor (Lee et al., 2007). In our study, the analyzed hydrographic parameters indicated that the CM site had the lowest salinity, which was approximately 30% less than the standard seawater salinity. We assumed that salinity could be the driving factor of the reduced abundance of Ruegeria at the CM site.

Coral mucus samples from the CM and HV sites had a higher abundance of Methylobacterium ( Figures 3A, F ). Its participation in the nitrogen cycle is well established in deep-sea corals, several other animal and plant hosts, and marine environments (Kellogg, 2019). Additionally, some studies have reported the antimicrobial activities of Methylobacterium isolates against pathogenic microorganisms (Balachandran et al., 2012; Photolo et al., 2020). Methylobacterium species also possess a Type VI secretion system (T6SS), which helps them compete with other competitive bacterial groups (Kempnich and Sison-Mangus, 2020). However, the role of Methylobacterium in the coral holobiont is poorly understood. Previous studies observed that the abundance of Methylobacterium is positively correlated with the presence of Si in seawater (Kempnich and Sison-Mangus, 2020). Its siderophore has high concentrations of Fe (III) ions (Simionato et al., 2006). Consistent with these findings, our results revealed a higher abundance of Methylobacterium in coral mucus from the CM and HV sites ( Figure 3 ), which are characterized by high concentrations of Si and Fe in ambient water ( Table 1 ).

Bacterial functional group analysis demonstrated the heterogeneity in the functional groups of the bacterial genera associated with the coral tissue and mucus samples and with the environmental samples ( Supplementary Figure S5A ). Bacterial functions in coral host microbiota have been explored. Although some of the bacteria appear to have commensal or anti-infectious functions (Nissimov et al., 2009; Olson et al., 2009), other bacteria are dynamic in nature and change with varying environmental conditions (Bruno et al., 2007). Our samples contained abundant chemoheterotrophic bacteria, which derive their energy from various processes, including the reduction and respiration of nitrate and nitrogen; fermentation; oxidation of sulfur compounds, thiosulfate, and hydrogen; and degradation of aromatic compounds. The cycling of nutrients such as nitrogen, sulfur, phosphorous, and carbon in a holobiont is crucial for the health and development of host organisms in oligotrophic environments (Zhang et al., 2015; Babbin et al., 2021).

Our bacterial function results revealed a higher abundance of nitrate-reducing and nitrogen/nitrate-respiring bacteria in the tissue and mucus samples from the HV site ( Supplementary Figure S5B ). Yang et al. (2013) reported that the bacterial symbionts of the coral T. coccinea have potential denitrifiers and ammonia-oxidizing bacteria, which consist of the nitrite reductase gene nirK and ammonia monooxygenase subunit A (amoA) genes, respectively. Bacterial groups that contribute to nitrogen cycling are important in maintaining homeostasis (Rädecker et al., 2015), and they may be critical for the resilience of hermatypic corals during ocean acidification to keep up with the higher photosynthetic rate expected under elevated CO₂ conditions (Santos et al., 2014; Rädecker et al., 2015; Marcelino et al., 2017). However, the role of nitrogen-cycling bacteria in azooxanthellate Scleractinia corals remains understudied. Biagi et al. (2020) first reported an increase in the abundance of nitrogen-fixing bacteria in the scleractinian coral Asteroides calycularis, which has no symbionts and lives under acidified conditions near CO₂ vents on Ischia Island, Gulf of Naples, Italy. Our study revealed a higher abundance of bacteria involved in the reduction and respiration of nitrate and nitrogen from the HV site, where the ambient seawater had pH 5.7 ( Table 1 ; Supplementary Figure S5B ).

Chemoautotrophic bacteria with sulfur-oxidizing potential are the common primary producers in vent ecosystems, while phototrophic prokaryotes and eukaryotes are far less abundant (Wang et al., 2015). Wang et al. (2015) reported that the mesophilic chemoautotrophs Sulfurovum and Sulfurimonas are dominant in the vent sediment on Kueishan Island. These Epsilonproteobacteria have the potential to oxidize sulfur compounds (Yamamoto et al., 2010; Han and Perner, 2016). This ability might explain the significant dominance of dark sulfur- and sulfur-compound-oxidizing prokaryotes in our HV site samples ( Figure 4 ). The HV site had a higher abundance of Sulfurovum and Sulfurimonas, while the water sample was dominated by sulfur-oxidizing Gammaproteobacteria, namely, Thiomicrorhabdus (Kojima and Fukui, 2019; Liu et al., 2020; Figure 3 ).

Photoautotrophic prokaryotic functional groups were significantly abundant in the coral tissue samples of the NPP and CZ sites ( Figures 4A, C ). Endolithic algal and cyanobacterial photosynthates serve as a nutrient source for the coral host during extreme events (Fine and Loya, 2002; Yang et al., 2019). Yang et al. (2013) reported the presence of cyanobacteria in the tissue samples of T. coccinea. However, knowledge about the interaction between phototrophic prokaryotes and azooxanthellate Scleractinia is limited. These functional groups may fulfil the nutritional requirements and support the skeletal growth of the host T. aurea. However, a comprehensive study should be performed to enhance our understanding of the complex relationship between the coral host and its symbiotic phototrophic functional groups.

Some bacterial species can be used to predict certain environmental conditions (Burger, 2006). In the present study, we observed the dominance of certain coral-related bacteria, which might serve as bioindicators of high concentrations of Cd, Zn, or Pb in their respective aquatic environments ( Figure 6B ). These genera include Vallitalea, Thermostilla, Methyloceanibacter, Kryptousia, Hoeflea, Denitrobaculum, Bythopirellula, Aliterella, and Actinomarinicola. Additionally, Marispirochaeta and Kistimonas are highly correlated with Fe pollution in specific water bodies ( Figure 6B ). However, we could not find any published reports supporting the use of these bacteria as indicators for monitoring heavy metal pollution. Given the adaptability of T. aurea to thrive in reef and non-reef habitats, as well as in extreme environments such as HV and CM sites, we propose that the microbial communities associated with this coral, as revealed by nanopore sequencing technology, could be used to monitor environments with high sulfur and copper contents in the future.