We conducted a tank-based transmission experiment to compare the effects of antibiotic pre-treatment (yes or no) and the subsequent exposure to white band disease slurries (diseased vs. healthy doses) in a two-factor experiment (antibiotic x disease exposure) with five tank replicates per level and 20 replicate fragments from each of ten healthy A. cervicornis genotypes ( Figure 1 ). We examined the effects of antibiotic pretreatment and disease exposure on the infection rate of the coral fragments and changes in their associated microbiomes using 16S rRNA amplicon gene sequencing. Twenty replicate fragments from ten healthy coral genotypes were collected from Sebastian’s reef (9° 15’ 16.4” N, 82° 7’ 37.8” W), Bocas del Toro in July 2017 and one fragment of each genotype was randomly assigned to each of the 20 18-liter recirculating tanks filled with UV sterilized seawater, which were held at ambient seawater temperatures in the flow-through seawater system. After six hours of acclimation, ten tanks were treated twice with 100mg/l each of Kanamycin, Ampicillin, Chloramphenicol, and Tetracycline, with 24 hours between treatments. Both Ampicillin and Tetracycline have previously been shown to inhibit WBD transmission when used separately through the inhibition of cell wall synthesis and protein synthesis respectively (Kline and Vollmer, 2011). Due to the light sensitivity of Tetracycline it was supplemented with two additional antibiotics which act to inhibit protein synthesis, Kanamycin and Chloramphenicol. Two antibiotic doses were used to ensure the treatment was effective and not affected by the light sensitivity of Tetracycline. In the morning after the second antibiotic dose, the seawater was replaced with new UV sterilized seawater, a post-antibiotic, pre-exposure sample of each fragment was taken (day 0, see below for tissue sampling methods), and a Waterpik with 0.2 um filtered seawater (FSW) was used to create small (ca. 0.25 cm²) experimental lesions in the coral tissue to facilitate transmission (Gignoux-Wolfsohn et al., 2012). Five of the antibiotic treated tanks and five untreated tanks were designated as disease exposure with the remaining tanks designated as healthy exposure. The disease exposure tanks were dosed with 50ml of disease slurry produced from 22 WBD infected coral fragments while healthy exposed tanks were dosed with 50ml of healthy slurry created from 22 healthy fragments. Healthy and diseased coral fragments were collected from Sebastian’s reef 1 hour prior to dosing the tanks. Tissue slurries were produced by liberating diseased or healthy coral tissue from the skeleton of sampled corals using a Waterpik containing filtered seawater (FSW), normalizing the slurry doses to a standard ocular density of 0.4 at 600nm with FSW, and then dosing each tank with 50ml of slurry using sterile centrifuge tubes.

During the experiment, corals were monitored every 12 hours and new disease signs recorded for a total of eight days after slurry exposure. Kaplan-Meier survival curves (Kaplan and Meier, 1958) comparing the four treatment groups - antibiotic pretreated and untreated corals crossed with disease versus healthy exposure – were used to analyze the rate of infection over time across coral fragments and test for differences in infection rates using a log-rank test (Harrington and Fleming, 1982). To determine which combinations of treatment groups differed significantly from each other, we performed post-hoc pairwise log-rank tests, adjusting the p-values to account for the familywise error rate using sequential Bonferroni correction (Holm, 1979).

16S rRNA amplicon gene sequencing was obtained by haphazardly sampling two polyps near the lesion site from six genotypes, so as to not sacrifice the whole fragment, in three disease exposed tanks for both antibiotic treated and untreated tanks sets across three time points: post-antibiotic treatment, pre-exposure (day 0, see above), two days post-exposure (day 2), and eight days post-exposure or when WBD symptoms developed, whichever occurred first (day 8). Fragments of the same six genotypes in the same three antibiotic treated and three untreated tanks were repeatedly sampled with repeated measurements statistically accounted for by including fragment nested within genotype and tank as random effects (see below). In all subsequent analyses the two post-exposure samples (day 2 and day 8) are analyzed as a single post-exposure treatment. Post-exposure samples at day 2 and day 8 were combined to improve statistical power while accounting for repeated sampling through the use of random effects to accommodate the sampling design (Hurlbert, 1984; Millar and Anderson, 2004). At each timepoint, polyps were sampled adjacent to the tissue lesion or disease interface using flame sterilized tweezers, the sampled polyps were placed into 150 ul of DNA/RNA shield (Zymo Research) and stored at -20C until extraction. Diseased corals were removed from the tank to prevent disease amplification. Genomic DNA was extracted from each sample using CHAOS extraction buffer (Fukami et al., 2004) and GenElute DNA extraction kits. 16S rRNA amplicon gene sequencing of the V3-V4 region was produced using Klindworth et al.’s (2013) protocol, V3-V4 (341F/785R) primer sets, and four lanes of Illumina MiSeq 2x300 bp sequencing. Reads were quality trimmed, overlapped and assembled into amplicon sequencing variants (ASVs) using the dada2 denoising algorithm and pipeline in R v4.2.1 (Callahan et al., 2016; R Core Team, 2022). Chimeras were removed and taxonomy was assigned to each ASV, first by using a Bayesian taxonomic classifier based on the NCBI 16S microbial database (downloaded on 3 Feb 2024) and classified to the lowest taxonomic level possible with greater than 80% classification confidence (Gao et al., 2017) and then using the Silva SSU r138 database modified for decipher for ASVs not classified by the Bayesian classifier (Quast et al., 2013). ASV sequences were aligned using decipher (Wright, 2016) and a neighbor-joining tree of the aligned ASV data was constructed using phangorn (Schliep, 2011). The resulting ASV table, taxa table and 16S rRNA tree was imported into phyloseq (McMurdie and Holmes, 2013) and merged with the sample metadata for downstream analyses. Samples were pruned to keep only samples with more than 1,000 16S rRNA gene reads and ASVs identified as cyanobacteria, mitochondria, and/or chloroplast sequences were removed as potential host or algal contaminants (Hanshew et al., 2013; Thomas et al., 2020). This removed 150 putative cyanobacteria ASVs, 133 of which would have been filtered due to low abundance with 17 passing the 10% prevalence filter. All 17 of these ASVs had significant BLAST hits against Symbiodiniaceae genomes suggesting they are host symbiont contamination ( Supplementary Table S1 ). Read counts were normalized for variable sequencing depth accounting for the compositional nature of 16S sequencing data using the robust centered log-ratio (rclr) of the number of reads (Gloor et al., 2017; Martino et al., 2019). This normalization method improves upon the additive log-ratio method used in the popular analysis software ANCOM in using the geometric mean of all taxa as the reference rather than one designated reference taxa and also improves the incorporation of 0 data to avoid the use of pseudo-counts which can bias results (Mandal et al., 2015; Kaul et al., 2017; Lin and Peddada, 2020b). Furthermore, the use of this normalization method, allows for planned post-hoc analyses (see below) using the same linear mixed-effects model framework which are not yet possible in the ANCOM or ANCOM-BC analysis software (Mandal et al., 2015; Lin and Peddada, 2020a, 2024).

To fully document the alpha diversity, we used a suite of common metrics assessing various aspects of the communities; these included observed richness, Camargo evenness (Camargo, 1992), Shannon Diversity (Shannon, 1948), Gini inequality (Gini, 1921) to assess community dominance, and Faith’s phylogenetic diversity (Faith, 1992). Prior to calculation of the alpha diversity metrics we rarified samples to an equal sequencing depth, with samples below that sequencing depth being removed, 1,000 times and calculated each alpha diversity metric. To determine the sampling depth which balances removing additional samples and having adequate sequencing depth to characterize the community we calculated rarefaction curves and Good’s coverage (Good, 1953; Sanders, 1968). The average of each diversity metric across all bootstrap rarefaction subsamples was modeled using linear mixed effects models using the lme4 package in R v4.2.1 (Bates et al., 2015; R Core Team, 2022).

Each alpha diversity metric was modeled using the same model as the individual ASV models with a fixed treatment effect combining the sampling time (before/after disease exposure), antibiotic treatment, and disease state, resulting in five unique combinations ( Table 1 ). All metrics were also modeled with random effects for tank, genotype, and fragment nested within genotype to represent the experimental design. We used a priori contrasts to distinguish the effects of time, antibiotics, and disease. To test for an effect of time, we compared the average alpha diversity metric of healthy fragments, regardless of antibiotic pretreatment, before and after disease slurry exposure. To test for an effect of antibiotic treatment, we compared the average alpha diversity metric of the healthy fragments before and after disease exposure between antibiotic treated and untreated fragments. Finally, we examined the effect of disease exposure by comparing diseased fragments to the average alpha diversity metric of untreated healthy fragments before and after disease exposure.

ASVs rarely detected in coral microbiomes, less than 10% of samples, were removed prior to the analysis of beta diversity. Community differences in coral microbiomes were analyzed using distance-based redundancy analysis using the robust Aitchison distance metric, which is scale invariant, obviating the need for rarefaction (Aitchison, 1982; McArdle and Anderson, 2001; Martino et al., 2019). Specifically, we investigated differences in microbial community composition between coral fragments prior to disease homogenate exposure and after exposure, as well as among corals treated with antibiotics and those left untreated and those which develop the disease. Significance was assessed using permutational MANOVAs with 10,000 permutations (McArdle and Anderson, 2001). Homogeneity of dispersions was tested for to distinguish differences in microbial community composition from differences in microbiome variability within treatments (O’Neill and Mathews, 2000; Anderson, 2006). As post-hoc analyses for both the community composition change and homogeneity of dispersion analyses, we tested for differences between healthy corals before and after dosing with the disease homogenate and between antibiotic treated and untreated fragments, excluding in both cases corals which develop disease symptoms. Finally, we looked at community differences between diseased and healthy corals, excluding corals treated with antibiotics. All beta diversity analyses were performed using the R package vegan (Oksanen et al., 2013).

To analyze differences in individual ASVs that were retained after the low prevalence filter and which drove changes in alpha and beta diversity, abundances were modeled using linear mixed effects models using lme4 (Bates et al., 2015). We modeled ASV abundance as an independent effect, combining the effects of time, antibiotic treatment, and disease outcome for a total of five treatment combinations, as no antibiotic treated samples developed disease signs after disease exposure ( Table 1 ). To control for repeated measurements and potential tank effects, we included random effects of tank and fragment nested within genotype. Main effect p-values were calculated using Satterthwaite’s method of calculating denominator degrees of freedom and adjusted to account for multiple testing (Satterthwaite, 1946; Benjamini and Hochberg, 1995), with planned post-hoc contrasts applied only to those found to have a significant main effect. Given a significant treatment effect (i.e. at least one of the five treatment combinations had significantly different ASV abundance than the others), we used the following a priori contrasts to test for different effects. To test for an effect of time, we compared the average robust centered log ratio (rclr) of the healthy fragments, regardless of antibiotic pretreatment, before and after disease slurry exposure. To test for an effect of antibiotic treatment, we compared the average rclr of the healthy fragments before and after disease exposure between antibiotic treated and untreated fragments. Finally, we examined the effect of disease exposure by comparing diseased fragments to the average rclr of untreated healthy fragments before and after disease exposure. These p-values were adjusted to account for multiple comparisons with the significance and direction (positive vs. negative association) of these three effects being used to categorize ASVs based on how they responded to the experimental procedures.

Infection rates in our experimental transmission were impacted by the combination of antibiotic pretreatments and disease exposure (χ²⁽³⁾ = 236.7, p < 0.0001, Figure 2 ). Corals that were exposed to disease without first being treated with antibiotics had significantly higher infection rates than any other treatment combinations (all p < 0.0001), with 37.2% (± 3.3% SE) of coral fragments becoming infected with WBD by day seven. In contrast, corals that received antibiotics prior to disease exposure had significantly higher non-infection rates with 93.1% (± 2.3 SE %) of the antibiotic treated, disease exposed corals remaining uninfected until the end of the experiment. These high non-infection rates were comparable to untreated fragments exposed to the healthy slurry (92.4% ± 2.5% SE, χ²⁽¹⁾ = 0.61, p = 0.44). Antibiotic pretreatment even conferred an advantage to corals dosed with healthy slurries, as no corals in the antibiotic treated, healthy exposed treatment developed WBD infections after seven days.

16S rRNA gene sequencing allowed us to examine how antibiotic pretreatment and subsequent exposure to disease changed coral microbiomes over time. 16S rRNA gene sequencing data was obtained for 36 replicate coral fragments from six coral genotypes from which 18 samples were treated with antibiotics and 18 left untreated all of which were exposed to disease slurries. Over the course of the experiment six of the untreated fragments developed WBD. Two polyps from each fragment were sampled prior to disease exposure, on day two post-exposure, and either on day eight post-exposure or when WBD symptoms first developed, whichever came first. A total of 106 samples were taken, less than the planned 108 samples as two fragments developed WBD symptoms and were removed following the day 2 sampling. In an additional two samples the PCR failed to amplify leaving 104 samples which were sequenced. After quality control filtering to remove samples with fewer than 1,000 reads, we were left with 99 samples for the analysis. The 16S rRNA gene dataset contained 4,705 unique bacterial ASVs across 427 genera from 42 classes and 206 families ( Figure 3 ). Removal of low prevalence ASVs (< 10%) left 1,182 ASVs from 22 classes and 94 families. The average number of reads of the removed ASVs was 0.125 ± 0.009 SE), 54 times less than the average number of reads of the retained ASVs (6.8 ± 0.25 SE, t_(117,333) = 26.9, p < 0.001). Samples were collected from and analyzed as five unique treatment combinations of before/after disease dosage either with or without antibiotic pretreatment, with the fifth treatment combination being those corals after disease exposure that were not pretreated with antibiotics which became infected with WBD.

The lowest Good’s coverage value for all 99 samples was 95.9%, indicating that in all samples at most 4.1% of the reads are ASVs which appear only once in the sample ( Supplementary Figure S1 ). This along with the sample rarefaction curves shows that there are diminishing returns of increased sequencing depth leading to the observation of new ASVs ( Supplementary Figure S2 ). Given that these ASVs are definitionally rare and that filtering to improve the minimum Good’s coverage requires the removal of 20 additional samples with lower sequencing depth we decided to filter to the sequencing depth of the least sequenced sample (1,056 reads) to prioritize the breadth of the samples at the slight expense of sampling depth.

Antibiotic treatment led to changes in the coral microbiome ASV richness (F_{(4, 91.1)} = 2.74, p = 0.03), and dominance (F_{(4, 22.6)} = 2.53, p = 0.069), but not diversity (F_{(4, 12.2)} = 1.74, p = 0.21), evenness (F_{(4, 9.75)} = 1.45, p = 0.29) or phylogenetic diversity (F_{(4, 25.5)} = 1.60, p = 0.20, Figure 4 , Table 2 ). Microbiomes of untreated corals had 30.5 more ASVs (± 9.5 SE; z = 3.21, p = 0.023) than antibiotic treated corals. As a result, antibiotic treated samples were more dominated by a few ASVs (z = 2.91, p = 0.034).

Corals in different treatment combinations had significantly different microbial community compositions (r² = 0.10, F_{(4, 94)} = 2.67, p < 0.0001, Figure 5 ) with some treatment combinations having significant within treatment variation among coral fragments (F_{(4, 94)} = 2.54, p = 0.045). Antibiotic treatment resulted in coral fragments with different microbial communities than untreated fragments (r² = 0.039, F_{(1, 89)} = 3.65, p < 0.0001), likely attributable to significant within treatment variation in microbial communities among antibiotic treated coral fragments (F_{(1, 89)} = 5.47, p = 0.022), suggesting that the effect of antibiotic treatment on ASV richness is primarily through the removal of low abundance/rare ASVs. Diseased coral fragments were associated with different microbial communities than healthy fragments (r² = 0.037, F_{(1, 49)} = 1.91, p = 0.006) without significant within treatment variation in community composition (F_{(1, 49)} = 0.007, p = 0.93) indicating a shift from a healthy to diseased microbial composition ( Figure 3 ). Finally, there is evidence that microbial communities change in composition (r² = 0.026, F_{(1, 89)} = 2.37, p = 0.0005) without significant variation within treatments (F_{(1, 89)} = 2.54, p = 0.11) after exposure to coral slurries regardless of the slurry type.

We used linear mixed effect models to identify individual ASVs that differ significantly due to antibiotic treatment, disease outcome, and/or across time. These differential abundance analyses identified 14 out of 1,182 ASVs that differ significantly in our main effect model combining antibiotic pretreatment, disease outcome, and time (i.e. before/after exposure to the disease dose, Table 3 , Supplementary Table S2 ). Out of the 14 ASVs, 9 ASVs differed due to antibiotic pretreatment, 6 ASVs differed due to coral disease outcome, and 11 ASVs differed depending on if the coral fragment was sampled before or after disease exposure ( Table 3 ).

The ASVs which were affected by antibiotic pretreatment included one Flavobacteriaceae (ASV 544, Ascidiaceibacter salegens), which was more abundant in antibiotic treated corals, and eight ASVs which were negatively affected by antibiotic treatment ( Table 3 ), including three strains of Rubritaleaceae (ASV 12, 185, 217) and one strain each of Erthrobacteraceae (ASV 40), Paracoccaceae (ASV 304), Roseobacteraceae (ASV 93), Verrucomicrobiaceae (ASV 20), and Vibrionaceae (ASV8, Figure 6 , Table 3 ). The Vibrionaceae (ASV 8, Vibrio sp.) and Roseobacteraceae (ASV 93, Thalassovita mediterranea) were both also more abundant in diseased corals, indicating that they were impacted by antibiotic treatment and were also strongly associated with WBD outcomes. Specifically, ASV 8 and 93 had low initial abundances before the antibiotic pretreatment (0.66 ± 0.26, 1.08 ± 0.17, respectively), and were knocked-down by the antibiotics (0.01 ± 0.26, -0.16 ± 0.17, respectively). In untreated corals, these ASVs increased 5-fold in abundance on diseased corals after disease exposure (2.56 ± 0.47, 2.49 ± 0.38, respectively). Our differential abundance analyses detected one ASV that was only associated with disease outcome, a Fastidiosibacteraceae (ASV 25, Cysteiniphilum litorale), and another which was associated with disease but increased in healthy corals after disease exposure, an unclassified Roseobacteraceae (ASV 9, Figure 6 , Table 3 ). In contrast to ASV 8 and 93 these ASVs were not detected initially (ASV 25: 0.14 ± 0.2, ASV 9: 0.15 ± 0.24) on healthy corals and thus were unaffected by the antibiotic treatment. Both ASVs were highly abundant on diseased corals (2.58 ± 0.37, 2.04 ± 0.36, respectively, Figure 6 ) and strongly associated with disease outcome, with ASV 9 also inhabiting healthy corals after receiving the disease dose (0.57 ± 0.26), unlike ASV 25 (0.35 ± 0.24).

The two top pathogen ASVs identified by the multi-year analysis in Selwyn et al. (2024) – Cysteiniphilum litorale (ASV 25) and Vibrio sp. (ASV 8) – both were strongly associated with disease outcomes in our tank-based experiment. Furthermore, Vibrio sp. was significantly negatively affected by antibiotic pretreatment while C. litorale was not detected on coral fragments prior to being dosed with the disease slurry. While both ASVs are clearly important to the WBD pathogenicity, it is unclear what roles they play in the WBD etiology.

Cysteiniphilum litorale is a recently described Fastidiosibacteraceae that has been linked to shrimp farm derived skin infections in humans (Liu et al., 2017; Xu et al., 2021). Prior to its description, it was described as Francisella-like (Liu et al., 2017; Qian et al., 2023), a genus which has been associated with WBD (Gignoux-Wolfsohn et al., 2017; Walton, 2017). In Florida a recent transmission experiment found Cysteiniphilum on A. cervicornis which developed disease symptoms following the experimental grafting of disease tissue (ASV 5b79cf6d5a5a9bf0bb866aed449eff44; Rosales et al., 2019) with additional studies in Florida finding Cysteiniphilum was present on both WBD resistant (Klinges et al., 2023) and WBD susceptible (Klinges et al., 2022) nursery reared Acropora cervicornis genotypes. The Cysteiniphilum genome, isolated in Wenzhou, China, contains a partial copy of the Francisella pathogenicity island (Qian et al., 2023) which facilitates Francisella being pathogens across a broad taxonomic range, including multiple marine species (Nano and Schmerk, 2007; Birkbeck et al., 2011; Colquhoun and Duodu, 2011).

Vibrio are well known opportunistic pathogens (Munn, 2015), which have been implicated in numerous coral diseases (Bourne et al., 2009) and are frequently commensally associated with corals under homoeostatic environmental conditions (Munn, 2015) where they can exist on ~20% of healthy corals (Ben-Haim et al., 2003; Gibbin et al., 2019; Selwyn et al., 2024). However, these vibrios become pathogenic under various environmental conditions, including increased temperature (e.g. V. coralliilytics causing tissue lysis in Pocillopora damicornis; Ben-Haim et al., 2003). One key trigger for the conversion of commensalist vibrios into pathogens is by quorum sensing initiated by the introduction of autoinducers (Liu et al., 2013). When autoinducers are introduced to healthy coral microbiomes, WBD symptoms develop (Certner and Vollmer, 2015) and the spread of WBD can be arrested by introducing quorum sensing inhibitors (Certner and Vollmer, 2018). Our results suggest that pretreating corals with antibiotics may prevent the initiation of quorum sensing, as quorum sensing is a density dependent behavior and pretreatment significantly reduced the relative abundance of Vibrio in the microbiome, which we assume to result from an absolute reduction in Vibrio abundance rather than an increase in other taxa following antibiotic treatment; this prevents the Vibrio sp. from becoming pathogenic and arrests any further cascade of opportunistic bacterial growth, preventing the shift to a diseased microbial community.

Quorum sensing is the process of intercellular communication among bacteria via the release and detection of small signaling molecules called autoinducers. This allows bacteria to alter behaviors based on bacterial density, such as activating virulence factors in situations of high cell density (Abisado et al., 2018). Bacterial quorum sensing is important in the transmission of WBD (Certner and Vollmer, 2015, 2018). Generally, autoinducers fall into two classes, species-specific (AI-1) and universal (AI-2), with AI-1 type autoinducers being variations of the class of molecules known as acylated homoserine lactones (AHLs) (Fuqua et al., 2001; Miller and Bassler, 2001; Schauder and Bassler, 2001; Xavier and Bassler, 2003). The enzymatic pathways required for both AI-1 (AHL) and AI-2 production and detection are well characterized across many Vibrio spp (Schauder et al., 2001; Winzer et al., 2002; Henke and Bassler, 2004), including those known to cause other coral diseases (Tait et al., 2010), and have been shown to regulate pathogenicity in many marine Vibrio spp (Henke and Bassler, 2004; Natrah et al., 2011).

In WBD, it is unclear what causes the initial production of autoinducers which initiates quorum sensing and triggers pathogenesis. We propose two potential initiators: either an overabundance of Vibrio sp., or the introduction of C. litorale. If vibrios are the initiators, then the densities of Vibrio spp. likely pass a threshold, initiating pathogenesis; this is analogous to infection by Clostridium difficile in humans, which are commensal bacteria in the digestive system and become pathogenic only after reaching high densities, often induced through the removal of competitors with antibiotics (Ng et al., 2010; Kamada et al., 2013). In our study, the disease dose likely introduced sufficient Vibrio sp. to pass this threshold in the untreated corals, which still contained indigenous Vibro sp., but not in the antibiotic treated corals where Vibrio ASV 8 was knocked down to be either absent or in very low abundance. In nature, changes in Vibrio sp. density could be caused by environmental changes such as increased temperature, which has been associated with increased WBD prevalences (Randall and van Woesik, 2015; Selwyn et al., 2024). The second mechanism is that C. litorale is a keystone pathogen which promotes pathogenicity in the indigenous bacteria (Hajishengallis and Lamont, 2016; Vega Thurber et al., 2020) through the production of autoinducers and initiation of quorum sensing, analogous to the role of enterotoxigenic Bacteroides fragilis which produces a biofilm that causes inflammation and changes in the gut microbiome resulting in colon cancer (Sears and Pardoll, 2011; Cheng et al., 2020).