Antibiotics reduce Pocillopora coral‐associated bacteria diversity, decrease holobiont oxygen consumption and activate immune gene expression

Corals are important models for understanding invertebrate host–microbe interactions; however, to fully discern mechanisms involved in these relationships, experimental approaches for manipulating coral–bacteria associations are needed. Coral‐associated bacteria affect holobiont health via nutrient cycling, metabolic exchanges and pathogen exclusion, yet it is not fully understood how bacterial community shifts affect holobiont health and physiology. In this study, a combination of antibiotics (ampicillin, streptomycin and ciprofloxacin) was used to disrupt the bacterial communities of 14 colonies of the reef framework‐building corals Pocillopora meandrina and P. verrucosa, originally collected from Panama and hosting diverse algal symbionts (family Symbiodiniaceae). Symbiodiniaceae photochemical efficiencies and holobiont oxygen consumption (as proxies for coral health) were measured throughout a 5‐day exposure. Antibiotics altered bacterial community composition and reduced alpha and beta diversity, however, several bacteria persisted, leading to the hypothesis that these bacteria are either antibiotics resistant or occupy internal niches that are shielded from antibiotics. While antibiotics did not affect Symbiodiniaceae photochemical efficiency, antibiotics‐treated corals had lower oxygen consumption rates. RNAseq revealed that antibiotics increased expression of Pocillopora immunity and stress response genes at the expense of cellular maintenance and metabolism functions. Together, these results reveal that antibiotic disruption of corals' native bacteria negatively impacts holobiont health by decreasing oxygen consumption and activating host immunity without directly impairing Symbiodiniaceae photosynthesis, underscoring the critical role of coral‐associated bacteria in holobiont health. They also provide a baseline for future experiments that manipulate Pocillopora corals' symbioses by first reducing the diversity and complexity of coral‐associated bacteria.

One approach to study host-microbe interactions in complex holobionts is to use controlled experiments that aim to manipulate the composition of the host-associated microbiota and reveal the specific microbial processes that contribute to holobiont functions (Chevrette et al., 2019). In idealized model organisms, this involves the elimination of the native microbiota with broad-spectrum antibiotics to produce axenic ("germ-free") or gnotobiotic ("known microbiota") animals that can be used in manipulative experiments to reveal how host-microbe interactions affect the development, physiology and health of the meta-organism (Bosch et al., 2019;Dobson et al., 2016). In axenic or gnotobiotic animals, researchers can investigate the effects of microbial colonization on host immune function, metabolism and gene expression to reveal how specific microbial metabolites interact with host cells and tissues to modulate host physiology and influence the development of various diseases (Kostic et al., 2013;Morgun et al., 2016). To date, no study in corals has achieved this ambitious goal, in part due to coral holobionts' highly complex and tightly associated microbial symbioses among the coral host, Symbiodiniaceae, bacterial communities and other viruses and micro-eukaryotes. Despite these limitations (or perhaps because of them), approaches that reduce the diversity and complexity of coral-associated microbiota can yield insights into how these communities affect overall holobiont health and will facilitate future experiments that target specific microbial functions.
In this study, we used high-throughput sequencing and physiological measurements to characterize the effects of antibiotics on 14 colonies of Pocillopora meandrina and P. verrucosa corals from Panama hosting diverse Symbiodiniaceae communities. We hypothesized that antibiotics treatment would (1) reduce coral-associated bacteria diversity to reveal tightly associated and/or antibioticsresistant bacteria, (2) cause changes in coral holobiont physiology and metabolism and (3) alter the expression of coral immune genes that mediate host-bacteria interactions. We anticipated that antibiotic suppression of coral bacterial communities would affect coral health and metabolism, symbiotic interactions and immunity and disease resistance and we aimed to measure experimental phenotypes to characterize these changes. Altogether, this study reveals how antibiotic disruption of the coral-associated bacterial community can negatively impact holobiont health and physiology by activating the coral immune response and suppressing net holobiont oxygen consumption and provide a methodological framework that will facilitate future efforts to manipulate corals' symbiotic interactions.
Multiple sequence alignment of an 828 bp region used in previous studies was completed using muscle (Edgar, 2004) and haplotype networks were visualized using the r package pegas (Paradis, 2010).

| Antibiotic experimental treatments
Pocillopora microfragments were prepared by using stainless steel bone cutters to remove ~5 mm branch tips, which were attached to labelled tags and allowed to recover for 19 days (Methods S1, Figure S1). Microfragments of each colony were assigned to the "baseline," "control" or "antibiotics" treatments, with four replicates of each colony in each treatment (n = 168 total, Figure S1).
Microfragments in the baseline treatment were sampled at the start of the experiment and microfragments in the control and antibiotics treatments were kept in 24-well plates, in 3.4 mL wells with 2.5 mL of either filter-sterile seawater (FSW) or antibiotic seawater medium (ASM) consisting of 100 μg/mL ampicillin, 50 μg/mL streptomycin and 10 μg/mL ciprofloxacin in FSW for the 120 h duration of the experiment (Methods S1).

| Coral microfragment physiological measurements
Symbiodiniaceae photochemical efficiency was measured as the maximum dark-adapted quantum yield of photosystem II (F v /F m ) using an imaging pulse-amplitude-modulated (I-PAM) fluorometer (Walz GmbH). After 30 min of dark acclimation, images were taken to record microfragment F v /F m values. After a baseline F 0 image, F v /F m values were measured from 24 areas of interest (AOIs) in the images of each of the six plates on each of the 5 days of the experiment.
Microfragment dark oxygen consumption rates were measured using a 24-well microplate respirometer (#SY25040, Loligo Systems) placed in a dark, environmentally controlled room kept at 26°C.
Oxygen sensors in 1700 μL wells measured real-time oxygen concentration (pO 2 in % air saturation) every 15 s over 45-min incubations ( Figure S2). Microfragment oxygen consumption rates were measured for six plates per day over 4 consecutive days, with 3-6 blank wells to correct for background changes in oxygen concentration (Methods S1; Figure S3).

| Statistical analysis of physiology data
Tests for normality and homogeneity of variances were performed on the Symbiodiniaceae F v /F m values. Linear mixed-effect models in the r packages lme4 and lmerTest were used to test for differences F I G U R E 1 (a) Map of Panama depicting the collection locations of the 14 Pocillopora colonies used in the experiment and the proportions of different mtORF types from each location. (b) Haplotype networks of the mtORF region for the Pocillopora colonies used in the antibiotics experiment. Two unique haplotypes were detected in an 828-bp alignment, corresponding to Pocillopora type 1 (orange) and type 3 (cyan) sensu Pinzón et al. (2013). (c) Representative images of Pocillopora microfragments from each colony on the first day of the experiment, outlines are coloured according to mtORF type and dominant symbiont type (C/D) is indicated in parentheses. All colonies in the first row of images were used in transcriptome analyses. in the photosynthetic efficiencies between different experimental treatments, Pocillopora colonies, source locations and dominant symbiont types.
Respirometry data were analysed using the r package respR (Harianto et al., 2019). The analysis of microfragment oxygen consumption rates (MO 2 ) used oxygen concentration data collected after the initial 15 min of the incubation since microfragments often required this amount of time to equilibrate to conditions on the plate and begin consuming dissolved oxygen at consistent rates. Microfragment To test for significant differences in microfragment MO 2 while accounting for the effect of microfragment mass, a linear regression of log(MO 2 ) against log(mass) was performed and the model residuals were used as the mass-corrected MO 2 ( Figure S4; Drown et al., 2020). Linear mixed-effects models were used to assess the effects of experimental treatment, measurement day and their interaction on microfragment F v /F m values and mass-corrected MO 2 , with Pocillopora mtORF type and dominant symbiont type included in the model as random effects.

| Coral fragment sampling
The experiment ended after 5 days, when microscope observations revealed that a single microfragment in the antibiotics treatment had begun to exhibit signs of coenosarc dissociation and polyp bailout  For sequencing quality control, triplicate extractions were also completed using the ZymoBIOMICS microbial community standard as a positive control and triplicate no-template negative controls were processed using only PCR-grade water.

| Bacterial 16S rRNA gene amplicon sequencing
Bacteria 16S rRNA gene amplicon libraries were created following the Earth Microbiome Project (Gilbert et al., 2014), without pooling triplicate PCR technical replicates (Marotz et al., 2019). The 16S rRNA gene V4 region was amplified using the 515F and 806R primer sets (Apprill et al., 2015). PCR reactions were prepared in a sterile hood and consisted of 20 μL of Platinum Hot Start PCR Master Mix (Thermo Fisher), 1 μL each of the 10 μM forward and reverse primers, 2 μL of template DNA and 26 μL of PCR-grade water for a total reaction volume of 50 μL. Thermocycler conditions consisted of 3 min at 94°C initial denaturation, then 35 PCR cycles of 45 s at 94°C, 60 s at 50°C and 90 s at 72°C followed by a final extension for 10 min at 72°C before holding at 4°C. PCR amplifications were also completed using the ZymoBIOMICS microbial community standard and PCR-grade water as a positive and negative control.
Amplicons were checked for quality and size distribution on a 1.5% TE-agarose gel and DNA concentration was determined using the Qubit dsDNA BR kit (Thermo Fisher). 16S rRNA gene amplicons

| Bacterial community data analysis
Raw 16S rRNA gene sequence reads were processed using qiime2 version 2020.2 (Bokulich et al., 2018). The plugin q2-demux was used to visualize read quality and DADA2 (Callahan et al., 2016) was used to remove primer sequences, trim poor-quality bases, dereplicate reads, identify chimeric sequences and merge pairedend reads. q2-feature-  (Katoh & Standley, 2013) and q2-phylogeny (Price et al., 2010) were used to create a phylogenetic tree for further downstream analyses and q2-taxa was used to remove all mitochondria and chloroplast sequences from the dataset. Finally, output tables containing ASV counts, phylogenetic trees and sample metadata were exported from qiime2 and imported into r for statistical analyses.

| Transcriptome profiling
Transcriptome libraries were prepared using the Lexogen Quantseq 3′ mRNA-Seq Library Prep Kit FWD (Lexogen) following manufacturer's instructions. Libraries were submitted for two rounds of sequencing at the University of Miami CGT in December 2020 and June 2021.
Libraries were sequenced on the Illumina NovaSeq with 100-bp single-end reads in the first round and 100-bp paired-end reads in the second round, with only forward reads kept for downstream analysis.

| Transcriptome data analysis
Transcriptome data were analysed according to Lexogen recommendations. Raw sequences were checked for quality and adapter dimers using fastqc (http://www.bioin forma tics.babra ham.ac.uk/ proje cts/fastq c/), poor quality bases and adapters were removed using bbduk (sourc eforge.net/proje cts/bbmap/) and reads were aligned against the P. damicornis reference genome using star v2.7.9a (Dobin et al., 2013). Reads were quantified at the gene level using featureCounts (Liao et al., 2014). Gene counts were filtered to include only genes with mean counts >3 in at least 90% of samples and counts were normalized using the variance-stabilizing transformation (vst) in the r package DESeq2 (Love et al., 2014). PCA was used to visualize the overall transcriptome patterns and the r package VariancePartition was used to partition total transcriptome variance among explanatory factors including sequencing round, colony, dominant symbiont genus and treatment (Hoffman & Schadt, 2016). All scripts used in physiology, microbiome and transcriptome data analysis are available at: https://github.com/micha eltco nnell y/ coral_antib iotics_physi ology.

| Pocillopora colony mtORF type and Symbiodiniaceae diversity
Fourteen Pocillopora colonies were used for microfragment preparation and colony traits are summarized in Table 1. Six colonies were from the Gulf of Chiriquí and eight colonies were from the Gulf of Panama while nine colonies belonged to mtORF type 1 (P. meandrina) and five colonies belonged to mtORF type 3 (P. verrucosa, Figure 1b,c). According to RNAseq read alignments, four colonies were dominated by Cladocopium and 10 colonies were dominated by Durusdinium (Table 1, Figure S6). Microfragment masses ranged from 0.08 to 3.60 g (0.148 ± 0.058 g, mean ± SD).

| 16S rRNA gene amplicon sequencing results
16S rRNA gene V4 amplicon sequencing yielded 19.5 million reads, with 4831-428,372 reads per sample (106,245 ± 58,090, mean ± SD) across all samples in the baseline (n = 54), control (n = 56) and antibiotics (n = 56) treatments ( Figure S7). Two baseline samples were removed for low read depth (<1000 reads). The r package decontam identified nine contaminant ASVs with higher abundances in the negative controls than all other samples. After the removal of contaminant sequences and 309 unassigned, chloroplast and mitochondrial sequences from the dataset, a total of 1835 ASVs remained for analyses of bacterial community diversity and composition. Furthermore, Kruskal-Wallis tests revealed significant differences in bacterial community beta diversity between treatments as measured by the distances to centroids (Kruskal-Wallis chi-squared = 34.85, df = 2, p < .001; Figure 3c), indicating reduced community dissimilarity among antibiotics-treated samples relative to baseline and control samples.

| WGCNA identifies co-expressed gene modules correlated with antibiotics treatment and physiology metrics
Weighted Gene Co-Expression Analysis identified 17 gene coexpression modules that contained 42-4096 genes and were labelled according to module size along a standard sequence of colours within the WGCNA r package ( Figure S11). Module eigengenes of 12 modules were significantly correlated (p < .05) with one or more treatments or physiology metrics and six modules were significantly enriched for one or more KOG terms ( Figure 5, Table S10).
The "brown" module (n = 2869 genes) was the most negatively correlated with the antibiotics treatment and most positively correlated with the control treatment and final MO 2 values ( Figure 5).
Conversely, the "lightyellow" module (n = 4096 genes) was the most positively correlated with the antibiotics treatment and most negatively correlated with the control treatment. The "lightyellow" module was enriched for the KOG term "signal transduction mechanisms" and the BP GO terms (n = 132, p adj < .05) "regulation of cell activation," "calcium ion transmembrane transport," "G protein-coupled receptor signaling pathway," "activation of adenylate cyclase activity," "positive regulation of interleukin-6 production" and "regulation of ERK1 and ERK2 cascade." The module hub gene (pdam_00021164) and 11 other top-connected genes (kME > 0.9) had no matches to any proteins in the UniProt database (Table S9), however, other highly connected genes were similar to pentraxin-4 (PTX4, pdam_00021627), protein decapentaplegic F I G U R E 3 Antibiotics effect on Pocillopora coral-associated bacteria community composition, diversity and core microbiome taxa. (a) Principal components analysis (PCA) plot of Pocillopora bacterial communities in the baseline, control and antibiotics treatments reveals that antibiotics samples cluster separately from baseline and control samples and have a narrower distribution along the first two PC axes. Baseline samples are coloured grey, control samples are coloured blue and antibiotics samples are coloured red. Convex hull polygons for each treatment show a clear separation of antibiotics-treated samples, whereas baseline and control samples overlap in their distribution. Violin plots of bacterial community (b) alpha diversity and (c) beta diversity show that bacterial community diversity is significantly reduced in the antibiotics treatment. Alpha diversity is represented as the Chao1 metric and beta diversity is represented as the distance to treatment centroid using the Aitchinson distance PCA. (d) Area proportional Euler diagram of treatment "core" (>75% prevalence across samples) identifies unique and shared bacterial amplicon sequence variants (ASVs) across treatments. (e) Bubble plot of the relative abundances of all treatment core ASVs (n = 54 total) reveals that bacterial community shifts in the antibiotics treatment are driven by depletion of 27 ASVs and persistence of 10 ASVs.  gamma-glutamyl peptidase 3 (GGP3, pdam_00013248) and a CuZnsuperoxide dismutase (SODCP, pdam_00019857).

F I G U R E 4
The "orange" module (n = 42 genes) was also positively correlated with the antibiotics treatment and negatively correlated with the control treatment and negatively correlated with final MO 2 values, but had no significantly enriched KOG or GO terms. While module hub gene (pdam_00017535) had no matches to any proteins in the UniProt database (Table S9), the other top-connected genes were similar to X-box-binding protein 1 (pdam_00016926) and prothrombin, or coagulation factor II (F2, pdam_00022018).
The "black" module (n = 1317 genes) was positively correlated with final F v /F m values and was not enriched for any KOG terms but was enriched for the BP GO terms "positive regulation of lipid metabolic process," "negative regulation of growth" and "regulation of phosphatidylinositol 3-kinase signaling" with the module hub gene, the transcription factor protein c-FOS (pdam_00011993, Table S9).
The "royalblue" module (n = 62 genes) was also positively correlated with final F v /F m values and had no significantly enriched KOG or GO terms ( Figure 5).
The "lightcyan" module (n = 98 genes) was negatively correlated with final F v /F m and MO 2 values and was enriched for the KOG term "extracellular structures" despite having no enriched GO terms and the module hub gene was similar to an Acropora millepora skeletal F I G U R E 5 Weighted Gene Co-Expression Analysis module correlation heatmap reveals Pocillopora co-expressed gene modules are correlated with treatment responses and physiological metrics and enriched for KOG terms representing key biological processes. Significant treatment and physiology-correlated modules are listed in bold and module correlation coefficients and p-values are shown in the heatmap cells. Significantly enriched KOG terms for each module are marked with asterisks (***p < .001, **p < .01, *p < .05).  WGCNA module correlation heatmap and KOG enrichment scores matrix protein (pdam_00023883, Table S9). The "darkgrey" module (n = 131 genes) was also negatively correlated with final F v /F m and MO 2 values but had no enriched KOG or GO terms and an unknown hub gene ( Figure 5, Table S9).

| DISCUSS ION
In this study, we show that antibiotics treatments alter Pocillopora coral-associated bacterial communities by selectively reducing bacteria diversity, which is linked to reduced holobiont oxygen consumption, increased expression of coral immune response genes and decreased expression of cellular housekeeping and symbiosis maintenance genes. There was no significant difference in Symbiodiniaceae photochemical efficiency between control and antibiotics-treated samples (Figure 2a), indicating that the antibiotics used here do not impair Symbiodiniaceae photosynthetic machinery, contrary to previous studies (Gilbert et al., 2012;Soffer et al., 2008). In contrast, we hypothesize that the significant reduction in microfragment MO 2 values in the antibiotics treatment was due to the direct killing, elimination and/or metabolic suppression of susceptible coral-associated bacteria ( Figure 2b). Antibiotics treatment may also have directly affected coral host and Symbiodiniaceae respiration to diminish overall holobiont MO 2 rates. Overall, our results suggest that antibiotics can cause a decrease in coral holobiont health by depleting beneficial microbes and/or suppressing holobiont metabolism, which allows a few persistent microbes to flourish in a potentially dysbiotic state and trigger costly host immune responses.

| Antibiotics treatments selectively alter coralassociated bacterial communities by reducing overall community diversity and dissimilarity
The Pocillopora corals that were examined in this study belong to two species across divergent mitochondrial lineages, host different dominant Symbiodiniaceae genera and were collected from environments with different thermal regimes in Panama. However, there was no significant effect of coral colony, mitochondrial lineage or source location on bacterial community composition and communities were similar across baseline and control samples (Figure 3a). This may be because all colonies were cultured together in a commongarden aquarium at the University of Miami for 5-16 years after collection. As expected, a 5-day treatment with the broad-spectrum antibiotics ampicillin, streptomycin and ciprofloxacin in filter-sterile seawater significantly altered bacterial community composition and reduced community diversity (Figure 3), similar to previous studies that have used antibiotics to alter coral microbiomes (Connelly et al., 2022;Glasl et al., 2016;Sweet et al., 2011Sweet et al., , 2014. These changes were driven by the near elimination of ASVs in the families Nitrincolaceae, Rhodobacteraceae and Alteromonadaceae, among others, that were highly prevalent in baseline and control samples but reduced in antibiotics-treated samples (Figure 3d, Figure S8).
These "antibiotics-susceptible" core bacteria included representatives from the family Rhodobacteraceae and genera Neptuniibacter and Kordiimonas and each may have important functions within the coral holobiont.
Genomic evidence suggests that Pocillopora corals cannot synthesize eight essential amino acids, including the branched-chain amino acids isoleucine, leucine and valine and the sulphur-containing amino acid methionine (Cunning et al., 2018;Li et al., 2022), so the downregulation of branched-chain amino acid transport and sulphur compound catabolic process genes in antibiotics-treated corals suggests that Rhodobacteraceae may have important roles in supporting these metabolic exchanges (Figure 4c, Table S8). Similarly, bacteria in the genus Neptuniibacter (family Nitrincolaceae) associate with diverse marine organisms including corals, medusozoans and ascidians, where they have been hypothesized to provide various health benefits such as stimulation of growth and development (Fragoso Ados Santos et al., 2015;Keller et al., 2021;Schreiber et al., 2016).
Bacteria in the genus Kordiimonas have also been detected in general association with Pocillopora corals, anemones and ascidians (Blasiak et al., 2014;Cahill et al., 2016;Zhang et al., 2021), however, specific functional roles have yet to be identified.
While antibiotics treatment was successful in reducing bacterial community diversity and eliminating many transiently associated and presumably antibiotics-susceptible bacteria, not all core bacteria were eliminated and several ASVs were observed to increase in relative abundance (Figure 3d, Table S6), despite the caveat that 16S rRNA gene amplicon sequencing data cannot distinguish whether these bacteria are still alive and metabolically active. These persistent ASVs included a single Vibrio ASV with 100% sequence similarity to the known coral pathogen V. harveyi, which could be driving the observed activation of coral immune responses through niche expansion and opportunistic infection following initial resistance to antibiotics (Luna et al., 2010;Meyer et al., 2015;Smith et al., 2015).
Indeed, cultured Vibrio bacteria from Pocillopora corals have been shown to possess genomes encoding antibiotic resistance genes, virulence factors and quorum-sensing inhibitors (Li et al., 2022;Ma et al., 2018), indicating the potential for pathogenesis following microbiome disruption. Three Micavibrionaceae ASVs also increased in relative abundance in antibiotics treatment samples and because Micavibrionaceae have been documented to increase during heat stress in Pocillopora corals and sea sponges, this suggests their role as another group of opportunistic microbes (Li et al., 2021;Posadas et al., 2022).
Additionally, out of a total of seven unique Endozoicomonas ASVs detected in the dataset, a single Endozoicomonas ASV was identified in the core microbiome across all colonies and persisted and increased in relative abundance in antibiotics treatment samples.
Bacteria in the genus Endozoicomonas are among the most conspicuous endosymbionts of Pocillopora corals, having been identified as core taxa across numerous studies throughout the genus' range, living in multicellular aggregates within coral gastrodermal tissues and being vertically transmitted between generations in maternal oocytes (Damjanovic et al., 2020;Hernández-Zulueta et al., 2016;Neave et al., 2017). Endozoicomonas bacteria have been shown to resist the effects of nutrient enrichment and temperature stress in P. verrucosa from the Red Sea (Pogoreutz et al., 2018) and ampicillin and streptomycin antibiotics treatment in P. acuta and P. damicornis from Taiwan (Connelly et al., 2022), indicating an ability to persist in coral tissues under various disturbances. In other corals, Endozoicomonas bacteria have been shown to metabolize dimethylsulfoniopropionate (DMSP) and upregulate host cell attachment proteins and vitamin B biosynthesis processes following exposure to holobiont cues (Pogoreutz et al., 2022;Tandon et al., 2020).
Future research should seek to characterize the genomic diversity, metabolic capacity and cellular niches of Pocillopora-associated Endozoicomonas bacteria to better understand the nature of this important symbiotic relationship.
Another persistent core ASVs belonging to the genus Thalassospira, which have been documented in symbiosis with corals, medusozoans and ascidians, are detected in seawater after P. damicornis spawning and produce a wide variety of secondary metabolites including the immunosuppressive thalassospiramides A, D and G (Blasiak et al., 2014;Um et al., 2013).
A single ASV in the family Arcobacteraceae also persisted after antibiotics treatment and these bacteria are found in association with corals, cuttlefish and sea urchins and are capable of thiosulfate oxidation and production of fibronectin-binding proteins for host cell attachment (Li et al., 2022). Together, these bacteria should be investigated for their ability to persist within coral host tissues during targeted disturbances and assessed for their metabolic ability to contribute to holobiont nutrient exchanges and mediate-immune system crosstalk.
Other notable bacteria that were observed to persist in antibiotics-treated samples have been recently documented in intimate associations with cultured and in hospite Symbiodiniaceae.

| Antibiotic bacterial community suppression activates coral immunity and stress response gene expression
Pocillopora transcriptome patterns showed strong colony variation. However, co-expression network analysis and functional enrichment tests revealed that antibiotic bacterial community suppression is correlated with enhanced expression of coral immunity, inflammation and stress response genes at the expense of cellular housekeeping functions such as DNA replication, energy production, translation and cytoskeleton maintenance (Figures 4 and 5).
Specifically, the positive correlation of the "lightyellow" and "orange" modules and negative correlation of the "brown" module with the antibiotics treatment indicate the existence of trade-offs between the activation of cellular immune responses, such as interleukin signalling pathways and coagulation cascades (Poole & Weis, 2014) and cellular housekeeping functions, such as ATP production and cytoskeleton maintenance (Figures 4c,d and 5). These changes may be related to the increased relative abundance of opportunistic microbes such as Vibrio bacteria in the coral holobiont following antibiotics treatment (Figure 3d) because these bacteria are known coral pathogens (Ben-Haim et al., 2003;Luna et al., 2010) that may possess antibiotics resistance genes and could shift to a pathogenic lifestyle following antibiotic disruption of the healthy community (Gao et al., 2021;Meyer et al., 2015;Smith et al., 2015). Prior studies implicated Alteromonas and Vibrio bacteria in driving dysbiosis following antibiotics treatment and heat stress in P. acuta and P. damicornis corals from Taiwan (Connelly et al., 2022) and while Alteromonas bacteria were reduced following the addition of ciprofloxacin in this study, several of the same key immune genes, including endoribonuclease ZC3H12B (pdam_00012426), ETS-related transcription factor Elf-4 (pdam_00016137) and ETS domain-containing protein Elk-1 (pdam_00003296), were upregulated following antibiotics treatment (Table S7). Future studies should aim to characterize the prevalence and diversity of coral-associated bacteria that possess antibiotic-resistance genes and virulence factors and determine the immune mechanisms that allow healthy corals to resist infection to better understand the potential for breakout infections following environmental disturbances.

| Antibiotics treatments as a practical tool for coral microbiome manipulation studies
Our experiment reveals the utility of antibiotics as an experimental tool for coral microbiome manipulation studies, as we succeeded in reducing the diversity of Pocillopora coral-associated bacteria by approximately 60% in only 5 days using a combination of broadspectrum antibiotics in filter-sterile seawater. The persistence of potentially antibiotics-resistant and internally associated bacteria highlights the difficulty in obtaining truly germ-free corals for future experiments, as many of these bacteria are opportunistic pathogens, key mutualists with beneficial functions that support the health of the coral holobiont or have other context-dependent symbiotic roles. Additionally, our experiment cannot attribute the observed responses to antibiotics treatment to depletion of the native microbiota, persistence of antibiotic-resistant microbes or possible direct effects on host tissues (Morgun et al., 2016) and future research is required to determine the relative impact of these different factors.
Nonetheless, antibiotics treatments are promising tools for manipulating coral bacterial communities and identifying which of these holobiont functions (e.g. nutrient cycling, amino acid and vitamin synthesis, pathogen resistance, heat tolerance, etc.) are negatively impacted by a diminished microbiota (Domin et al., 2018;Fraune et al., 2015;Klimovich et al., 2020;Murillo-Rincon et al., 2017;Weiland-Bräuer et al., 2020). Experiments that use antibiotics followed by subsequent "recovery" treatments will yield insights into whether bacterial communities can be returned to healthy states by re-introduction of beneficial microbes (Assis et al., 2020;Damjanovic et al., 2019;Rosado et al., 2018;Santoro et al., 2021) and measuring the resulting holobiont phenotypes with integrative omics methods will reveal the genetic and molecular basis of coral-holobiont interactions in unprecedented detail.

ACK N O WLE D G E M ENTS
We would like to thank Dr. Stephanie Rosales for bioinformatics assistance and NOAA's Ocean Chemistry and Ecosystems Division for the provision of the optical respirometer. This research was supported by the 2019 RSMAS David Rowland Fellowship awarded to MTC and by National Science Foundation Award #1951826 awarded to NTK.

CO N FLI C T O F I NTER E S T S TATEM ENT
The authors declare no conflict of interest.

DATA AVA I L A B I L I T Y S TAT E M E N T
Raw 16S rDNA gene amplicon reads and RNAseq reads were deposited into the NCBI Sequence Read Archive (SRA) under BioProject accession PRJNA818888. All the scripts used in the 16S and RNAseq data analysis are available on GitHub: https://github.com/ micha eltco nnell y/coral_antib iotics_physi ology.

B EN EFIT-S H A R I N G S TATEM ENT
Benefits from this research accrue from sharing data, analysis scripts and results on public databases (described above) and this research addresses priority concerns in advancing coral conservation. More broadly, our group is committed to equitable international scientific partnerships and institutional capacity building.