Dichotomy between Regulation of Coral Bacterial Communities and Calcification Physiology under Ocean Acidification Conditions

Ocean acidification (OA) is a consequence of anthropogenic CO2 emissions that is negatively impacting marine ecosystems such as coral reefs. OA affects many aspects of coral physiology, including growth (i.e., calcification) and disrupting associated bacterial communities.

gradient of pH ranging from average ambient surface seawater to OA conditions projected to occur within the next 50 years (10). The multifaceted responses of the coral host together with its Symbiodiniaceae and microbiome (referred to as the "coral holobiont") to OA conditions requires multidisciplinary approaches. In this study, we used 16S rRNA gene amplicon sequencing and biogeochemistry approaches in three coral species (Pocillopora eydouxi, Porites lobata, and Porites rus) to examine (i) coral-associated bacterial community response (resistance, acclimation, or dysbiosis) and (ii) the ability of the coral host to upregulate internal pH with long-term exposure to low pH seawater.

RESULTS
Changes in bacterial communities across the pH gradient were coral speciesspecific. Bacterial communities from P. eydouxi, P. lobata, and P. rus samples are referred to here as "Poc-communities," "Plob-communities," and "Prus-communities," respectively. After quality filtering, 6,097,305 high-quality sequences were obtained from all coral samples analyzed (n = 82) ( Table 1). After the removal of mitochondrial, chloroplast, and unassigned reads, the total pool of sequences were assigned to 3,918 amplicon sequence variants (ASVs). All samples clustered into three distinct groups by coral species, but Plob-communities and Prus-communities were more similar to each other than either were to Poc-communities (see Fig. S1 in the supplemental material). Bacterial communities in each coral species were all significantly different from each other (analysis of similarities [ANOSIM]: P , 0.01; see Table S1 in the supplemental material). Thus, remaining analyses were conducted on each coral species separately.
Bacterial communities of all three coral species shifted significantly across the pH gradient, but in a species-specific manner (ANOSIM: P , 0.05, Fig. 2a to c, Table 2). Poc-communities from background-pH and mid-pH sites were highly similar with tight clustering of samples from these sites, but Poc-communities from the low-pH site dispersed, indicating changes in overall community structure near the vent ( Fig. 2a; Table  2). Plob-communities distinctly clustered at each site (Fig. 2b) and were significantly different at each site (Table 2). Prus-communities experienced changes across the pH gradient with much overlap between sites ( Figure 2c); however, a significant shift in Prus-community structure occurred at the low-pH site, but not between background-pH and mid-pH sites ( Table 2).
The species-specific responses to the pH gradient are more clearly visualized in the sample-to-sample variation, or community dispersion, in bacterial communities, represented by mean pairwise dissimilarity, within each site. Poc-communities destabilized across the pH gradient, with a significant increase in community dispersion occurring at the low-pH site (Fig. 2d). In contrast, Plob-communities stabilized across the pH gradient, with significant decreases in community dispersion occurring at both mid-and low-pH sites (Fig. 2e). Prus-communities also stabilized, but the decrease in community dispersion was at the low-pH site only (Fig. 2f). Both richness (observed ASVs) and Shannon diversity were consistent in Poc-communities and in Prus-communities across the pH gradient. In Plob-communities, both the Shannon diversity and the ASV richness were impacted across the pH gradient (Shannon Diversity: Kruskal-Wallis; H = 17.25, P = 0.0002) (ASV richness: Kruskal-Wallis; H = 19.07, P = 0.0001) ( Table 1). Richness and Shannon diversity were significantly higher at the mid-pH site, but at the low-pH site the Shannon diversity was significantly lower in Plob-communities. Plob-communities at the mid-pH site were distinct due to the presence of 1,940 ASVs (59.1% of the total Plob-community ASVs) that were unique to Plob-communities at the mid-pH site.
DESeq2 analysis identified 86 ASVs that were differentially abundant at either the low-pH or the mid-pH site, compared to the background-pH site (see Table S2). In Poccommunities, an ASV classified as "Candidatus Amoebophilus" was enriched at both mid-pH and low-pH sites, whereas five ASVs classified as unclassified Cyclobacteraceae or unclassified Spirochaetaceae were depleted at both mid-pH and low-pH sites. Also, 10 ASVs classified as Endozoicomonas were depleted in Poc-communities at either the mid-pH or the low-pH site. In Plob-communities, 53 and 36 ASVs were differentially abundant at mid-pH and low pH sites, respectively, with 22 ASVs being differentially abundant at both mid-and low-pH sites (see Table S2). Differentially abundant ASVs in Plob-communities were diverse, representing 26 bacterial families across 9 phyla. The ASV most enriched in Plob-communities at both mid-pH and low-pH sites was "Candidatus Amoebophilus," and the ASV most depleted at both mid-pH and low-pH sites was Nitrospira. Seven ASVs, classified as Endozoicomonas, were significantly enriched at the low-pH site. Prus-communities had only four ASVs that were differentially abundant compared to the background-pH site: two ASVs, classified as Endozoicomonas, were significantly enriched at the low-pH site, and two ASVs, classified as Algicola and unclassified Cellvibronaceae, were significantly depleted at the low-pH site (see Table S2). No ASVs were significantly differentially abundant in all three coral species. Endozoicomonas (order Gammaproteobacteria, family Endozoicomonadaceae) was the only bacterial taxon with differentially abundant ASVs present in all three coral species, but each coral species had different Endozoicomonas ASVs contributing to differences within that species (see Table S2). Therefore, the relative abundance of the genus Endozoicomonas (from all ASVs classified within this genus) was explored further. Poc-communities were dominated by Endozoicomonas, averaging 96.1 6 0.8% of the total community at the background-pH site. Endozoicomonas abundance in Poc-communities decreased across the pH gradient, averaging 93.0 6 1.7% and 67.4 6 4.6% at the mid-pH and low-pH sites, respectively (Fig. 3b), but was significantly lower only at the low-pH site (Kruskal-Wallis; H = 20.54, P = 0.0001). Plob-communities at the low-pH site were distinct resulting from a significant increase in Endozoicomonas abundance, averaging 63.4 6 3.6% of Plob-communities at the low-pH site, compared to 0.4 6 0.1% and 0.01 6 0.01% of Plob-communities at the background-pH and mid-pH sites, respectively (Kruskal-Wallis: H = 23.96, P = 0001; Fig. 3b). Prus-communities also had high abundance of Endozoicomonas, averaging 53.6 6 4.5% at the background-pH site and 63.7 6 4.8% at the mid-pH site, but Prus-communities also experienced a significant increase in Endozoicomonas abundance, averaging 87.2 6 2.7%, at the low-pH site (Kruskal-Wallis: H = 17.37, P = 0.002; Fig. 3b).
Chloroplast-derived reads were a large proportion of the total reads from all samples (79.6%). In P. eydouxi, chloroplast-derived reads increased significantly across the pH gradient (Kruskal-Wallis: H = 9.92, P = 0.007) (see Fig. S3a). Ostreobium abundance in P. eydouxi samples also increased across the pH gradient, comprising 2.2 6 1.0%, 96.2 6 1.9%, and 89.8 6 4.1% of the chloroplast-derived reads at the backgroundmid-, and low-pH sites (Kruskal-Wallis: H = 19.21, P = 0.0001) (see Fig. S3b). In P. lobata, chloroplast-derived read abundances were significantly less abundant at the mid-pH site (Kruskal-Wallis: H = 11.95, P = 0.002) (see Fig. S3c). Ostreobium comprised 68.0 6 2.5%, 39.1 6 2.2%, and 83.4 6 3.1% of the chloroplast-derived reads at the background-pH, mid-pH, and low-pH sites, respectively, but there was a significant difference in Ostreobium abundance at the mid-pH site only (Kruskal-Wallis: H = 23.11, P = 0.0001) (see Fig. S3d). Chloroplast-derived reads in P. rus had similar abundance across sites (Kruskal-Wallis: H = 0.71, P = 0.703) (see Fig. S3e), and Ostreobium abundance, which was 91.4 6 0.9% at the background-pH site, was also similar across the pH gradient (Kruskal-Wallis: H = 2.86, P = 0.239) (see Fig. S3f). Changes in calcification fluid pH across the pH gradient were coral species specific. Each coral species maintained its ECF pH higher than the external seawater pH at all sites, and P. rus maintained the highest ECF pH at each site (Fig. 4). In P. eydouxi, mean ECF pH decreased toward the vent (by 0.02 pH units), but there were no significant differences among sites (Kruskal-Wallis; H = 0.26, P = 0.878). In P. lobata, FIG 3 (a) Mean percent abundances of bacterial sequences for Pocillopora eydouxi, Porites lobata, and Porites rus coral samples at background-pH, mid-pH, and low-pH sites. Bacterial taxonomy was assigned to phyla or class using the Silva v138 database. All taxa of ,1% of the total community were grouped under the category "minor taxa." (b) Relative abundances of Endozoicomonas sequences in all three coral species across sites. Black, background-pH site; gold, mid-pH site; red, low-pH site. Data represent means with individual values. Kruskal-Wallis test with Dunn's multiple comparisons were conducted across sites within a species. *, P , 0.05; **, P , 0.01; ***, P , 0.001; ****, P , 0.0001. mean ECF pH increased modestly at the mid-pH site by 0.02 pH units and then dropped by 0.08 pH units at the low-pH site, compared to the background-pH site, but significant differences in mean ECF pH occurred only at the low-pH site (Kruskal-Wallis: H = 8.34, P = 0.015). In P. rus, mean ECF pH decreased by 0.08 pH units toward the vent, but like P. eydouxi, there were also no significant differences among sites chiefly as a result of increased variability in ECF pH values at the lower-pH sites (Kruskal-Wallis; H = 3.91, P = 0.142).
Skeletal trace elements indicate exposure to vent emissions. Various trace elements present in the surrounding seawater are incorporated into the coral skeleton during calcification and can be used as proxies for local seawater conditions during calcification (36). Thus, the skeletal concentrations of trace metals that are emitted from the vent allow us to assess the exposure of individual colonies to vent emissions that are not spatially homogeneous within sites. Mn was the only trace element in coral skeletons that differed significantly across sites in all three coral species and serves as an indicator of the magnitude of vent exposure among and within sites (analysis of variance [ANOVA]: df = 2, P , 0.002) (see Table S3). At the background site, mean Mn/ Ca was similar across all coral species (;0.6 mmol/mol; Fig. 5). However, the mean Mn/ Ca level increased in P. eydouxi and P. lobata by an order of magnitude at the mid-pH site (6.6 6 0.7 mmol/mol and 5.1 6 0.9 mmol/mol, respectively) and almost doubled in P. eydouxi and tripled in P. lobata again toward the low-pH site (11.3 6 0.5 mmol/mol and 15.3 6 2.0 mmol/mol, respectively). The mean Mn/Ca level in P. rus also increased significantly toward the vent, 1.3 6 0.1 mmol/mol at the mid-pH site and 1.5 6 0.1 mmol/mol at the low-pH site (Fig. 5). However, the Mn/Ca level in P. rus at both mid-  Black circles, background-pH site; gold triangles, mid-pH site; red squares, low-pH site. Sites within a species were compared using ANOVA with Tukey's post hoc comparisons of log-transformed data. **, P , 0.01; ***, P , 0.001; ****, P , 0.0001. and low-pH sites was significantly lower than that in the other two coral species at the mid-pH site (Fig. 5). In P. lobata, Fe/Ca also increased significantly toward the vent (ANOVA: df = 2; P = 0.0001), but no other trace element displayed significant differences across sites in either P. eydouxi or P. rus (see Table S3).
Environmental drivers of bacterial community structure. Particulate and dissolved Mn, Al, and Fe are known to emit from the Maug vent (37); therefore, CO 2driven reduction in seawater pH may not be the only environmental factor influencing coral-associated bacterial community structure at the mid-or low-pH sites. Canonical correspondence analysis (CCA) included 8 bacterial families for P. eydouxi, 14 bacterial families for P. lobata, and 7 bacterial families for P. rus. A Monte Carlo permutation test found that the CCA was robust in each coral species (P , 0.05), indicating a strong correspondence between relative abundance of major coral-associated bacterial families and predictor environmental variables (see Table S4). In all three coral species, CCA triplots showed high correspondence of seawater pH to background-pH-site samples, whereas Mn showed high correspondence to low-pH-site samples (Fig. 6). For P. eydouxi, seawater pH and Mn were the strongest predictor variables, with CCA axis 1 correspondence coefficients of 20.67 and 10.74, respectively (see Table S4). For P. lobata, seawater pH, Mn, and Fe were strong predictor variables, with CCA 1 correspondence coefficients of 20.89, 10.92, and 10.81, respectively (see Table S4). For P. rus, seawater pH and Mn were the strongest predictor variables, with CCA 1 correspondence coefficients of 10.71, and 20.63, respectively (see Table S4). No bacterial taxa clustered with a particular trace element in all the three coral species. In the Porites spp., most taxa have positive associations with increased seawater pH (Fig. 6).

DISCUSSION
Microbiome restructuring in a new environment may lead to better stress tolerance in corals and/or may reflect host tolerance and maintenance of homeostasis under a new environmental regime (38)(39)(40), but microbiome restructuring has been shown to occur in a host-specific manner (41). Understanding host-specific microbial interactions under long-term OA is important to predict the future health and function of reefs, and we found that three coral species along a natural pH gradient exhibited changes in bacterial community structure and composition. In P. eydouxi, bacterial communities experienced increased community dispersion with lower seawater pH, following the Anna Karenina principle that dysbiotic individuals vary more in their microbial community composition than their healthy conspecifics (29). The potential breakdown of hostmicrobe interactions seen in P. eydouxi are consistent with other coral taxa, such as Acropora millepora and Porites cylindrica, at another naturally acidified reef in Papua New Guinea (26). However, Pocillopora spp. are proposed to be "microbial regulators" with relatively inflexible microbial associations, even under heat and nutrient stress (41,42). The disruption to microbial communities seen in P. eydouxi at the low-pH site at Maug suggests that OA may represent a chronic environmental stress capable of budging even the most intransigent of coral-microbe associations. In contrast, in Porites spp., bacterial communities across the same pH gradient converged onto more tightly clustered communities with similar community composition, potentially reflecting processes of host acclimation and tolerance to OA conditions. Our data contrast with other studies, also from Papua New Guinea, that showed that bacterial communities in massive Porites spp. are resistant to changes in seawater pH (43). At Maug, P. rus displays a massive growth morphology but also displayed significant changes in bacterial community structure with lowered seawater pH. These differences in bacterial community flexibility of massive Porites spp. under OA conditions at Papua New Guinea and Maug may simply reflect species-specific responses to OA but could be useful for testing hypotheses regarding the role microbial flexibility in adapting to OA.
Conditions at Maug had a pronounced effect on the abundance of the bacterial genus, Endozoicomonas. This tissue-residing bacterium is found in various coral species across the globe (44) and is thought to be a symbiont (as opposed to a commensal) and skeletal trace element concentrations) and the mean abundances of major bacterial families (.1% of total bacterial community within a coral species). All panels: black circle, background-pH site; gold triangle, mid-pH site; red square, low-pH site. (45). We detect significant losses of Endozoicomonas toward the vent system in P. eydouxi. In contrast, both Porites species display significant increases in Endozoicomonas abundance as ambient seawater pH decreased. P. lobata in particular, increased the relative abundance of this taxon from 0% to almost 50% when in proximity to the vent. Endozoicomonas did not strongly correspond to other vent emissions (such as Mn, Fe, or Al), suggesting that lowered seawater pH, or the coral host's response to lowered seawater pH, has a stronger influence on its abundance. Interestingly, different Endozoicomonas ASVs were enriched or depleted in each coral species, suggesting species-specific associations between different Endozoicomonas spp. and their coral hosts. Loss of Endozoicomonas in response to OA has been described in other coral species, such as A. millepora and even in massive Porites spp. (26,43,46), but our study documents gains of Endozoicomonas under low-pH conditions. Even though P. eydouxi experienced losses in Endozoicomonas, this genus was still the dominant bacterial taxa at the low-pH site. Gains in Endozoicomonas in both Porites species could suggest that Endozoicomonas may be beneficial to these coral under OA conditions. Alternatively, low-pH conditions may be lowering the Porites corals ability to control the growth of these intercellular bacteria. Either way, it raises the question as to why some coral species lose abundance of this bacterial taxon with OA and highlights the potential of this bacterial genus as an indicator of tolerance to OA. OA is likely directly and indirectly influencing the structure and composition of coral-associated bacterial communities. The changes seen in Endozoicomonas abundance, for example, are likely due to indirect effects because Endozoicomonas, as a tissue-residing bacterium, would not be exposed to seawater. Rather, changes to host physiology in response to OA likely are influencing Endozoicomonas abundance. In contrast, mucus-associated and skeleton-associated bacteria do interact with seawater; thus, the direct impacts of changing seawater pH would be most evident in these communities. Because whole fragments were processed for amplicon sequencing, we cannot tease apart these potential direct versus indirect effects or differentiate patterns that may be driven by bacterial localization. Future studies should consider these partitions in coral-associated microbial communities to reveal more nuanced insights into the impact of OA on coral microbiome.
OA is also known to alter coral interactions with skeleton-associated eukaryotic endophytes. In particular, OA leads to higher abundance of the green algae Ostreobium in Porites skeletons, leading some to describe it as a harmful bioeroder (47,48). However, Ostreobium is commonly found in the skeletons of living corals and provides photoassimilates to coral during thermal bleaching, leading others to suggest it is a beneficial coral symbiont (49). Assessment of eukaryotic microalgae communities in corals is best addressed by using 23S or ITS2, rather than 16S, rRNA sequencing, but we were able to classify Ostreobium sequences with high confidence using the Protist Ribosomal Reference database (50). Similar to previous studies, P. eydouxi and P. lobata at Maug displayed increased abundances of Ostreobium with decreasing seawater pH, and P. eydouxi had similar (but high) abundances of Ostreobium across the pH gradient. The high Ostreobium abundance, especially at the low-pH site, in all corals suggests that it may be acting as a harmful bioeroder.
We also document genus-specific responses in calcification physiology in response to long-term low-pH conditions. In P. eydouxi, calcification physiology was highly regulated across all seawater conditions, whereas both Porites experienced a mean decrease in ECF pH near the vent. P. rus has been suggested as an OA-resistant coral (13,14), and this species appears the most OA tolerant in this study in terms of calcification physiology, maintaining the highest ECF pH of all three species across all sites and experiencing only a slight decrease in ECF pH across the pH gradient. The variable response in pH upregulation in P. rus (and other coral species) may reflect localized differences in exposure to vent emissions, as discussed below, or a variable response to low-pH conditions by different individuals. Our data show that different coral species have varying capability to raise ECF pH and maintain calcification rates under OA.
Expanding this analysis to other coral species and understanding the genetic/molecular mechanisms will help identify OA resilient corals species and/or populations.
Species-specific upregulation of ECF pH in response to OA has been documented but during short-term experiments (19,51). Similarly, species-specific changes in coral bacterial communities in response to OA conditions have been documented but under far more extreme pH reductions (pH 7.5 or less) (26,33,52). Here, we document changes in ECF pH and the composition of coral bacterial communities with long-term in situ exposure to conditions (20.1 pH units) projected to occur at the end of this century under midrange emissions scenarios (10). Seawater pH on coastal coral reefs may decline faster than open ocean predictions, especially for coral reefs in lagoons or enclosed bay with less mixing, because community metabolism and local watershed influences can drive large declines and/or high variability in local seawater pH (53,54). Therefore, coral reefs around the globe may soon experience, or may already experience, the levels of OA stress needed to impact calcification and microbial symbiosis.
The different responses to OA stress described at Maug may have important longterm consequences to coral population dynamics. ECF pH upregulation may allow P. eydouxi to sustain normal calcification in spite of further OA, but bacterial community destabilization below a pH threshold may increase susceptibility to other environmental disturbances or cause mortality by allowing opportunistic pathogens to proliferate or reduce energy supply. In contrast, P. lobata and P. rus may experience similar reductions in calcification or skeletal density as other corals under OA conditions but may benefit from a more stable bacterial community. Given increasing seawater temperatures and recurrent global bleaching events, a stable, mutualistic bacterial community will be an important factor in reef persistence even at relatively pristine reefs like those found at Maug.
Investment trade-offs can occur among different physiological functions in corals under stress, including OA (55). For example, spawning female colonies of Astrangia pocluata, which require more energy for the production of gametes, experienced decreased calcification under OA conditions compared to spawning male colonies (56). However, it is unclear whether other physiological functions, such as regulation of bacterial communities, also impact calcification sensitivity to ocean acidification, or vice versa. Short-term OA experiments reveal complex coral calcification and bacterial symbiosis responses. Experimental OA reduced both calcification rates and microbial nitrogen fixation rates in Seriatopora hystrix (34), and dual OA and temperature stress destabilized the bacterial community and decreased calcification in a thermally sensitive coral (Acropora millepora) (31), suggesting that both calcification and bacterial symbiosis are impacted negatively with OA. In a more thermally tolerant coral (Turbinaria reniformis), dual OA and temperature stress neither reduced calcification nor destabilized bacterial communities, suggesting that both physiological processes can be maintained in some corals during short-term stress (31).
Given the physical separation of mucus-and tissue-associated bacteria from the host tissues undergoing calcification, a direct link between bacterial community structure/composition and calcification physiology is unlikely. However, the dichotomy in ECF pH upregulation and bacterial community structure seen in this study reveals a potential investment trade-off by the coral host under long-term OA that warrants further investigation. Corals have cellular mechanisms to optimize calcification conditions in the ECF, which consume ATP (17,57,58), and the energy requirements to elevate ECF pH relative to seawater pH increase exponentially under increasing OA (59). It remains poorly understood how corals select for and regulate their bacterial communities, but it is thought to involve the composition and shedding of mucus (60)(61)(62) and/ or interactions with immune defenses (63)(64)(65), both energetically costly functions (66)(67)(68)(69). Thus, by maintaining high ECF pH at the Low pH site, P. eydouxi is likely investing more energy into calcification, compared to P. lobata and P. rus, potentially at the cost of resources needed to maintain stable bacterial communities. We did not measure skeletal density or linear extension rates in the corals collected for this study; however, Enochs et al. report depressed calcification and linear extensions rates for Porites spp. at Maug (70). Controlled, manipulative experiments will be needed to mechanistically link changes in coral bacterial communities and calcification physiology under long-term OA conditions to determine how different coral species may invest in OA tolerance.
Mn/Ca in our coral skeletons provides a useful covariate to approximate long-term vent exposure in lieu of discrete seawater pH measurements taken at the time and place each coral colony was sampled within the sites. P. eydouxi and P. lobata display strong increases in skeletal Mn/Ca across sites, suggesting a robust gradient of vent exposure across the three collection sites. P. rus displayed a much weaker signal. Interspecific differences in Mn incorporation could explain lower Mn/Ca values in P. rus at mid-and low-pH sites (71,72). However, the natural distribution of corals within each site and heterogeneous distribution of vent emissions across the reef could also explain the weaker Mn signal in P. rus. At the low-pH site, P. rus colonies form a large, monospecific wall with high relief that grows to the edge of the vent zone. In contrast, P. eydouxi and P. lobata colonies are sparsely distributed throughout the vent zone (R. Day, unpublished data). Thus, the P. rus colonies may not be experiencing as great an OA gradient compared to P. eydouxi and P. lobata.
Previous studies have stressed the need to account for other hydrothermal emissions (such as metals) as potential confounding variables when utilizing volcanic vents as natural laboratories of future OA conditions (73,74). Sulfur-rich compounds have not been detected in proximity to the vent, but Maug caldera does emit dissolved and particulate Mn, Al, and Fe (37) that, in addition to increased pCO 2 , may affect host physiology and bacterial community structure. To address this concern, we utilized CCA on a subsample of our bacterial community data and found that seawater pH was the strongest predictor variable to bacterial community structure in all three coral species, followed by Mn and Fe. Both Mn and Fe may impact bacterial community structure because these trace elements are used in various biological processes, including photosynthesis, redox reactions, nutrient acquisition, cell adhesion and biofilm formation, and these metals can be toxic to microbes (75,76). Marine bacterial taxa that are known to be Mn or Fe reducers or oxidizers (77)(78)(79) or taxa known to associate with Mn-enriched marine environments (80,81) were not abundant (,1% of total bacterial reads) in any coral sample at any site. For example, Chlorobiaceae require Fe for anoxygenic photosynthesis, but this abundant family was not positively correlated with Fe concentrations. These data suggest that trace element emissions from the Maug vent do not directly alter coral-associated bacterial communities. However, the direct impacts of these other vent emissions on mucus-associated bacteria may be difficult to detect. In addition, we cannot currently assess whether other vent emissions could be affecting bacterial community structure and/or composition via impacts on host physiology.
Utilizing both 16S rRNA gene amplicon sequencing and biogeochemistry analysis provides deeper insight into the long-term impacts of OA on coral physiology. The species-specific responses to long-term OA described here and their potential ecological implications highlight the need to understand the mechanisms behind differential susceptibility and resilience of reef-building corals to OA. This study did not investigate the response of an important member of the coral holobiont (i.e., Symbiodiniaceae) or seek to answer how the responses of the coral holobiont influence each other. With the ability to control for volcanic influence using skeletal biogeochemistry approaches, it is imperative to leverage the high research potential of CO 2 vents, such as Maug caldera, to better understand how changes in coral holobiont physiology will impact health and function of reef ecosystems as global CO 2 emissions continue to increase.

MATERIALS AND METHODS
Study site and sample collection. Maug is an uninhabited group of three islands located in the Northern Mariana Islands (20°1N, 145°13E) (Fig. 1a) that make up a sunken volcanic caldera. The reefs studied within the caldera are shallow (;9-m depth), with localized submarine volcanic vents that bubble CO 2 creating a localized gradient of reduced pH and aragonite saturation state (37,47). In addition to CO 2 , the volcanic vent at Maug emits particulate and dissolved iron, manganese, and silica (37). No significant concentrations of sulfur-rich compounds such as sulfate, sulfide, or hydrogen sulfide have been detected in proximity to the vent (37). Three study sites were previously established along this 1,500-m gradient and are labeled as background pH, mid pH, and low pH (Fig. 1b). Previous work conducted by Enochs et al. (70) characterized the mean seawater pH, mean pCO 2 , and benthic composition at these three discrete sites in order to capture the temporal variability in carbonate chemistry, described briefly below. The mean seawater pH levels (6 the standard deviations) measured over a 3month interval (n = 3,984 measurements) at these sites were 8.04 6 0.016, 7.98 6 0.027, and 7.94 6 0.051, respectively, with the low-pH site reaching a minimum of pH of 7.72. The mean pCO 2 levels (6 the standard deviations) measured from discrete water samples over a 2-day period were 401.3 6 4.61, 441.2 6 21.23, and 502.0 6 29.67 matm, at the background-pH, mid-pH, and low-pH sites, respectively (70). Percent coral cover also decreases from .50 to ;20% to ,1% at the background-pH, mid-pH, and low-pH sites, respectively (70). Temperature, measured over the same 3-month period, did not vary significantly across sites (70). Light, measured as daily dose of photosynthetically active radiation, was reported to decrease significantly across the pH gradient: 10.9 6 2.46, 9.5 6 2.46, and 6.5 6 4.16 mol photons per m 2 at the background-pH, mid-pH, and low-pH sites, respectively (70). However, light levels were not found to be a major driver of benthic community structure at Maug (70), and the high variability in light levels, especially at the low-pH site, suggest that more measurements over a longer period of time (light measurements data were collected over the course of just 2 days) may be necessary to better understand differences in light regime across sites.
One fragment was collected from 10 colonies of three coral species (Pocillopora eydouxi, Porites lobata, and Porites rus) via hammer and chisel at each of the three sites via SCUBA in 2014. At Maug, P. eydouxi forms discrete branching colonies, P. lobata forms discrete mounding colonies, and P. rus forms massive or discrete mounding colonies. Fragments were collected from an accessible terminal branch or lobe at the top of each colony. At each site, coral fragments were collected from colonies close to the location of seawater chemistry measurements, within an area of ;50 m 2 . At the low-pH site, the massive colonies of P. rus are on the immediate south perimeter of the vent compared to the samples collected from P. eydouxi and P. lobata, which were distributed throughout the vent. Coral fragments were placed in individual bags and snap-frozen in liquid nitrogen at the surface. Frozen coral fragments were kept in liquid nitrogen at the National Institute of Standards and Technology's Marine Environmental Specimen Bank until further processing. Corals were collected under NOAA Pacific Islands Fisheries Science Center permits, approved by the Commonwealth of the Northern Mariana Islands Department of Fish and Wildlife, and the collection was conducted in accordance with applicable rules and regulations governing fieldwork and sample collection at the study site.
Skeletal biogeochemistry analysis. Preprocessing for skeletal biogeochemistry included subsampling half of the coral samples (n = 5 from each coral species at each site) under liquid nitrogen, followed by lyophilization for 24 h. Dry coral samples were then scraped using a scalpel to collect the exterior layer of skeleton most closely associated with living tissue. An aliquot of the coral powder (;100 mg) was treated twice with a 5% sodium hypochlorite solution at room temperature for 24-h periods to remove the soft tissue. Skeletal samples were then washed with boron-free Milli-Q water (.18.2 MX·cm) and then further lyophilized to remove moisture. Carbonate powders were further subsampled (;5 mg) and subjected to additional oxidative cleaning in warm 1% H 2 O 2 (buffered in ammonium hydroxide) to chemically remove remaining organic matter. These samples were then given a weak acid leach (0.0005 M HNO 3 ) to remove any readsorbed ions. Once cleaned, samples were dissolved in a minimal volume of Optima 0.5 M HNO 3 (Fisher Scientific).
Trace elements in each dried coral skeleton were measured using Thermo Element II inductively coupled plasma mass spectrometry (ICP-MS) at the National Institute of Standards and Technology (NIST). Dissolved samples were diluted in 0.5 M HNO 3 to 80 mg/g [Ca] for analysis. Samples were run using multimode detection and low and medium mass resolution using a method modified from Marchitto (82). Multielement external calibration using gravimetrically prepared matrix-matched standards were used to quantify Li, B, Al, Na, Mg, V, Mn, Fe, Co, Ni, Cu, Zn, Rb, Sr, Mo, Cd Sb, Ba, Nd, Pb, and U analytes relative to Ca. Blanks were run between each sample/standard, which were used to blank-correct counts of each element before ratios were determined. The limit of detection for each element for each run was determined as three standard deviations above the mean of the blanks. The percent relative standard deviation was calculated for each elemental ratio from replicate measurements (n = 37) of a matrix-matched control material (NIST RM 8301 Coral) to assess analytical precision (83). The majority of the measured trace elements had an uncertainty of #62% (Li/Ca, Al/Ca, Na/Ca, Mg/Ca, Co/Ca, Cu/Ca, Rb/Ca, Sr/Ca, Cd/Ca, Sb/Ca, Ba/Ca, Nd/Ca, Pb/Ca, and U/Ca), with B/Ca, Fe/Ca, Ni/Ca, and Zn/Ca # 65% and V/Ca, Mn/Ca, and Mo/Ca # 610%.
Boron isotopes analysis was conducted according to established methods (84,85). Boron in the remaining solutions (;200 ng of B) was separated from the carbonate matrix using 20-ml microcolumns (Amberlite IRA 743 boron-specific anionic exchange resin). After elution of the boron fraction, additional elutions were checked to ensure . 99% of sample boron was recovered in the sample. The purified boron samples were diluted to 100 ppb [B] for analysis. The d 11 B of samples were measured on a multicollector Nu Plasma II MC-ICP-MS against NIST SRM 951a. The accuracy and precision of d 11 B results were assessed using carbonate standards NIST RM 8301 Coral and JCp-1. Measured values for these reference materials during sample analysis were, respectively, 24.35% 6 0.26% (2 standard deviations; n = 34) and 24.06% (n = 2), which were within uncertainty of interlaboratory consensus values (83,86). Seven total procedural blank measurements were made alongside samples in this study (average of 104 pg of boron). These blanks were found to be small (,0.06% of sample boron) resulting in minimal impact on d 11 B sample results (i.e., less than analytical uncertainty), hence a total procedural blank correction was not applied.
Extracellular calcifying fluid pH calculations. The ECF pH of each sample was estimated from measured skeletal d 11 B using the following equation (87,88): ! where a B (1.027) is the fractionation factor between boric acid and borate (89), d 11 B sw (39.61) is the boron isotopic composition of seawater (90) and pK B * (8.54) is the dissociation constant of the two boron species calculated using the Seacarb package in R with site-specific temperature of 30°C and salinity of 35 ppt (70). Analytical uncertainty on d 11 B measurements contributes to a ,0.02 pH unit shift in calculated internal pH.
Bacterial community analysis. Preprocessing included subsampling from each coral fragment (n = 90) and subsequent homogenization (minimum of 2 min at 25 reps per second) all under liquid nitrogen using a Retsch Cryomill (Retsch GmbH, Haan, Germany). An aliquot of each coral homogenate (;300 mg) was preserved in 1 ml of TRIzol reagent and stored at 280°C until nucleic acid extraction. Preserved DNA was separated from RNA using organic solvent phase separation with chloroform, and subsequent DNA extraction was conducted using back-extraction buffer (BEB; 4 M guanidine thiocyanate; 50 mM sodium citrate; 1 M Tris [pH 8.0]) (91). After removal of the aqueous phase containing RNA, BEB was added to the organic phase and interphase and then incubated at room temperature for 10 min. Samples were then centrifuged at 13,200 rpm for 15 min at 4°C. The upper, aqueous phase (now containing DNA) was removed, an equal volume of 100% isopropanol was added, and the samples were incubated overnight at 220°C. The samples were then centrifuged at 13,200 rpm for 30 min at 4°C to pellet DNA. The supernatant was removed, and the DNA pellets were washed twice with 70% ethanol and then resuspended in Nanopure water. Extracted DNA was quantified using a NanoDrop spectrophotometer and submitted to the BioAnalytical Services Laboratory at the Institute of Marine and Environmental Technology for high-throughput sequencing of V1-V3 hypervariable regions of the 16S rRNA gene using 27F (59-AGAGTTTGATCCTGGCTCAG-39) and 534R (59-ATTACCGCGGCTGCTGG-39) on an Illumina Mi-Seq (paired-end 2 Â 300 read) platform.
Sequence analysis was conducted using QIIME2 v. 2019.10 pipeline (92). Paired-end, demultiplexed reads were quality filtered, trimmed of poor-quality bases, dereplicated, chimera filtered, pair merged, and identified as amplicon sequence variants (ASVs) using the DADA2 plug-in (93). Taxonomy was assigned by training a naive-Bayes classifier on the V1-V3 region of the 16S rRNA gene in the SILVA version 138 database (94) using the feature-classifier plugin (95) to match the primers used. Nonprokaryotic ASVs (i.e., mitochondria, eukaryote, and unassigned) were then removed. Sequences classified as "chloroplast" were also removed from the analysis of bacterial sequences but were saved as a separate data set. Sequences identified as chloroplast via the SILVA database were further classified using the Protist Ribosomal Reference (PR 2 ) database (50). Rarefied bacterial ASV tables (rarefied to 1,100 reads per sample) were used to calculate alpha-diversity metrics and to conduct beta-diversity analyses using weighted UniFrac distance matrices. For each coral species, alpha rarefaction curves of bacterial ASVs did reach a plateau, indicating sufficient sampling depth at to 1,100 bacterial reads per sample (see Fig. S3).
Statistics. Skeletal trace element ratios were log transformed to meet assumptions of normality and then compared among sites, separated by species, using one-way ANOVA with Tukey's post hoc comparisons and Bonferroni correction. Data for Mn were further compared among site and species groups as a proxy of vent exposure, using one-way ANOVA with Tukey's post hoc comparisons. Due to unequal variances in other data sets, even after transformation, a nonparametric Kruskal-Wallis (with Dunn's post hoc comparisons and Bonferroni correction) was used to compare ECF pH, alpha-diversity metrics, betadiversity metrics, and relative abundances of major microbial taxa among sites (within each coral species). All data are represented as means 6 the standard errors, unless stated otherwise.
The weighted UniFrac distance matrix was used to calculate the beta diversity to assess dispersion in bacterial communities between samples at each site (within each species). The weighted UniFrac distance matrix was also used to construct principal-component analysis (PCoA) plots to visualize differences in bacterial community structure between sites. PCoA was conducted for all samples and then for each species individually. Pairwise analysis of similarities (ANOSIM) was used to test for significant differences in bacterial communities among sites. To assess differences in ASV abundance across the pH gradient, the R package DESeq2 (v1.26.0) was used to fit a negative binomial model using the unrarefied ASV table for each species, and then Wald tests were used to test for differences in taxon abundance between mid-or low-pH sites versus the "control" (background-pH site). Benjamini-Hochberg false discovery rate tests were used to account for multiple comparisons, and ASVs with P values of ,0.05 were identified as significantly differentially abundant.
To test which environmental variables (vent-associated trace elements and external seawater pH) have significant relationships to bacterial community structure, canonical correspondence analysis (CCA) was performed using the program PAST v3 (96). Inputs for CCA analysis include concentrations of trace elements with significant differences among sites or known to be emitted from the vent, the mean seawater pH of each site as measured previously (47), and the relative abundances of major bacterial families (.1% of total bacterial community). A Monte Carlo test with 999 permutations was carried out to ensure the significance of the canonical axes.
Data availability. The raw sequence data files have been submitted to the NCBI Sequence Read Archive under accession number SRP174887. Data files, including concentrations of trace elements, an ASV table, an ASV taxonomy assignment, and R script for DESeq analysis, are available on Figshare at https://figshare.com/account/home#/projects/88106.

SUPPLEMENTAL MATERIAL
Supplemental material is available online only. SUPPLEMENTAL FILE 1, PDF file, 0.4 MB.