Pseudoalteromonas sp. strain McH1–7 cryostocks were maintained at -80°C and preserved in a 20% glycerol (final concentration) solution made with 0.22 µM filtered seawater (FSW) and sterilized by autoclaving before use. Cultures were revived by streaking onto sterile seawater agar (SWA) consisting of 2 g/L yeast extract, 15 g/L of agar, and 4 g/L tryptone mixed with 1 L of FSW (Ushijima et al., 2023), and then incubated at 28°C overnight. Two to three resulting colonies were inoculated into glycerol supplemented seawater broth (GSWB) consisting of 2 g/L yeast extract, 2 mL/L sterile glycerol, and 4 g/L tryptone mixed with 1 L of FSW (Ushijima et al., 2023) and incubated at 28°C while shaking at 150 rpm for approximately 15 h. These cultures were then diluted by a factor of 100 in fresh GSWB and incubated at 28°C while shaking at 150–200 rpm. The cultures were grown until reaching an optical density of 1.5–2.0, measured at 600 nm (OD_600nm), representing 1.55 x 10¹⁰ colony-forming units (CFUs) per mL according to plate counts. One-liter aliquots were centrifuged at 8,000 × g for 10 min and the resulting supernatant was decanted. For whole-colony bagging treatments, each pellet was resuspended with 250 mL of previously autoclaved FSW and then divided equally into five 60 mL Luer-Lock syringes, with 50 mL (~3.1 x 10¹² McH1–7 CFUs) in each syringe. A 10 mL sized air bubble was added to each syringe for oxygenation and mixing before capping for transportation.

A paste was created as a vehicle for the lesion-specific treatment of SCTLD consisting of 187.5 g polyvinylpyrrolidone (CAS number 9003-39-8), 11.25 g sodium alginate, 11.25 g sodium chloride, and 375 mL reverse osmosis (RO) water. The sodium alginate allows for the polymerization, or thickening, of the paste when in contact with divalent ions such as Ca²⁺ or Mg²⁺ in seawater. The sodium chloride was added to control salinity and avoid osmotic shock of the marine bacteria. The polyvinylpyrrolidone was used to further thicken the paste. All dry ingredients were mixed before adding the RO water. Once added, a hand mixer was immediately used to combine. A pellet of McH1–7 from one liter culture broth (above) was mixed with 15 mL of a sterile 3% NaCl solution made with RO water (used instead of FSW to prevent premature polymerization) before being folded into 600 g of paste with sterile tongue depressors until evenly mixed. The paste was loaded into the back of 60 mL catheter syringes using a silicone piping bag, typically used for icing cakes. Each of the resulting syringes contained ~2.6 x 10¹⁰ CFUs of McH1–7 per g of paste. To allow for a wider band of the paste to be released underwater along the coral disease lesion, the tips of the syringes were cut off leaving 1 cm of the tip remaining before capping for transportation to the reef site.

Assessing the survival of McH1–7 during transportation to in situ sites is crucial to ensure probiotic treatments are viable for application. Therefore, one liter of McH1–7 culture in GSWB with an OD_600nm between 1.5–2 was pelleted and resuspended in 250 mL of FSW. Conical tubes were filled with 50 mL of this bacterial suspension, leaving room for an air bubble, just as the syringes for the field were created. The tubes were capped and stored at 4 or 22°C and sampled at 2, 7, 14, 21, 28, 42, and 56 days (n=3 per temperature and timepoint). Samples were taken from three separate conical tubes at each time point, then serially diluted 10-fold (to 1:10⁸) in FSW before dilutions were spotted on SWA. For each dilution, 10 µL was spotted in 3 technical replicates onto the same SWA plate per conical tube replicate. Serial dilution plates for each time point were incubated for 24 hours at 28°C before colony counts were done and colony forming units (CFUs) per mL were calculated. An additional experiment was conducted using the same methodology to determine the viability of McH1–7 at five different temperatures to evaluate if this strain could survive being frozen (-20°C), refrigerated (4°C), on the countertop in an air-conditioned room (22 ± 4°C), and stable incubation (22°C and 28°C). CFU counts were taken initially, and after 1, 2, 3, and 5 days, the time it would typically take to prepare and distribute the probiotic along a coral reef.

To assess the viability of McH1–7 cultures, the CFUs per mL values were calculated from 10-fold serial dilutions that were spotted onto SWA using the equation:

where n c is the number of colonies per spot, d is the dilution factor, and v is the volume spotted in mL. For this experiment, 10 µL (0.01 mL) aliquots were spotted (“volume spotted”) from each dilution, and the colonies per spot were counted (“colonies per spot”). To determine which dilution would be used for the CFU/mL calculation, the standard practice of using only spots from a dilution with easily resolved colonies, containing >5 colonies per spot, and from the least diluted sample (i.e., the lowest dilution) was utilized. The average CFU/mL was determined from at least 3 technical replicates per biological replicate from plates incubated no longer than 24 h.

To analyze the viability of McH1–7 at different temperatures, the average of three technical replicates per sample was used to determine the number of bacteria present within each sample. A repeated measures analysis of variance (ANOVA) followed by Šídák’s multiple comparisons test were performed using GraphPad Prism (version 9.5.1 (733) for Windows, GraphPad Software, San Diego, CA, USA, www.graphpad.com).

While Pseudoalteromonas sp. McH1–7 was effective at treating SCTLD infected M. cavernosa colonies in aquaria (Ushijima et al., 2023), the reaction of other Caribbean coral species to this strain remained unknown. Therefore, five other Caribbean species, including Colpophyllia natans (n=1), Stephanocoenia intersepta (n=2), Meandrina meandrites (n=2), Siderastrea siderea (n=2), and Orbicella faveolata (n=2), were treated and monitored for their response to ensure no adverse reactions occurred before using McH1–7 in situ. To do so, healthy coral colonies were collected from the field ( Supplementary Table S1 ), transported to the Smithsonian Marine Station in Fort Pierce, FL, held in buckets with 13 L of oxygenated FSW, and trimmed to approximately the same size (3–5 cm) with a masonry saw (Husqvarna MS 360) within 24 h of collection. The corals were held in large recirculating systems indoors until they were utilized for safety testing. For experimentation, each fragment was placed in individual 5 L tanks on top of egg-crate grating with aeration as previously described in Ushijima et al. (2023).

Experimental fragments were inoculated with McH1–7 according to Ushijima et al. (2023). Briefly, GSWB liquid cultures were grown to an OD_600nm of 1.0, which correlates to approximately 10⁹ CFUs per mL according to plate counts. Then, 50 mL aliquots of each strain were centrifuged and decanted, leaving the pelleted bacteria for treatment. The bacterial cells were resuspended with 1 mL of seawater from the tank receiving the treatment and then the solution was pipetted over the fragment with the aeration turned off for the following 2 h. The final concentration of the strain was approximately 10⁷ CFU per mL of tank water. Starting 24 h after inoculation, ~50% water changes were conducted tri-weekly with sterilized scoops. The visible health status of the corals was monitored daily over 21 days while photographing from a top-down view every other day to look for discoloration, bleaching, or tissue loss.

On September 1, 2020, a research site off the coast of Fort Lauderdale, Florida (26°9’2.358”N, 80°5’45.031”W) spanning 100 m of longitude and 40 m of latitude was established in an area of regional SCTLD prevalence (Toth et al., 2024) at a depth of ~9 m of shallow colonized pavement. Forty adult M. cavernosa colonies displaying signs of active SCTLD were tagged and mapped ( Figure 1 ). The establishment of such a small study area reduced the likelihood that variation in colony responses to treatments would be considerably confounded by environmental variation. Tagged colonies were haphazardly divided into five treatment groups: (1) probiotic bag (PB), (2) control bag (CB), (3) probiotic paste (PP), (4) control paste (CP), and (5) background control (BC).

The probiotic bag treatment was conducted by covering the whole colony with a 378.5 L volume plastic bag with a weighted line along the bottom to trap ocean water within. The bag was cinched at the top to allow approximately 8 cm of space between the inside of the bag and the surface of the colony. Aquarium tubing was then used to syringe 50 mL of a McH1–7 suspension (~3.1 x 10¹² CFUs) into the bag ( Figures 2A, B ). The aim was to achieve a minimum concentration of 10⁷ CFU/mL seawater, which is the concentration that was used in previous aquarium studies (Ushijima et al., 2023). Final concentrations varied depending on the size of the corals and amount of seawater in the bags, but dosages of 10⁷-10⁸ CFU/mL should have been achieved. A locking mechanism between the syringe and the tubing allowed the diver to lock and remove the tubing from the syringe after delivering the probiotic treatment to ensure McH1–7 was not released outside of the bag. The syringe was refilled with 50 mL of ocean water, reattached to the tubing, the locking mechanism was unlocked, and the ocean water was pushed through the tube to clear the bacterial culture from the tube into the bag ( Figure 2C ). The bag was left for 2 h before carefully removing it from the colony. The control bag treatment was the same as the probiotic bag treatment except the syringe was only filled with ocean water.

The probiotic paste treatment included applying paste filled with McH1–7 directly to the lesion with a 60 mL syringe ( Figures 2D, E ). A 2 cm wide band of paste was then flattened with a gloved hand along the whole lesion. Bleached or paling tissue around the lesion was also covered in probiotic paste ( Figure 2F ). The amount of paste used per coral varied with the size of the lesion. Small lesions required much less than one syringe of paste, whereas larger lesions may have used a full syringe or more. The control paste treatment consisted of flattening paste with no added McH1–7 over the disease lesion in the same manner. Finally, background control colonies, all infected with SCTLD, were not treated or handled and were only monitored over time to serve as a baseline for the disease progression at the site.

Nine diseased M. cavernosa colonies were treated with PP, six with PB, four with CP, five with CB, and four were untreated BC on September 1, 2020 (day 0). On October 14, 2020 (day 43), six additional corals were found with SCTLD: two of these colonies were treated with CB and four were treated with CP. All other previously treated colonies were also re-treated with their respective treatments at this time. All colonies were re-treated with their respective treatments again on January 15, 2021 (day 136). Six newly infected colonies were added to the site and treated for the first time that day: one with CP, one with PP, two with PB, and two were BC. The final treatment of 40 corals included eight PB, ten PP, seven CB, and nine CP treated corals, and six BC untreated corals ( Table 1 ). M. cavernosa newly showing signs of SCTLD were added during the first few months of the experiment to increase replication and keep sample sizes similar among treatments. Relative to the duration of the experiment (2.5 years), all colonies were added in the first 4.5 months of the experiment allowing ample time to observe the outcome of the treatment effects.

Each colony was imaged using a Nikon COOLPIX W300 by setting the camera on slow continuous mode taking a photo every second and then slowly circling each colony from approximately 1 meter distance. Once the coral was circled, the photographer swam and photographed over the top of the coral to ensure all sides had been captured. Lastly, while still in slow continuous mode, the photographer swam towards each lesion to approximately 30 cm distance to capture fine detail. Colonies were monitored and photographed 14 times within 2.5 years. A few colonies were not photographed on some monitoring dates due to difficulties relocating them ( Table 1 ).

Three-dimensional (3D) models of each M. cavernosa colony ( Figure 3 ) were generated using the proprietary software Agisoft Metashape Professional version 1.8.4 (Agisoft LLC) on desktop (Dell Precision 5820, Intel^® Xeon^® W-2225 central-processing unit @ 4.10 GHz, 32 GB of random-access memory, and an Advanced Micro Devices Radeon Pro W5500 graphics-processing unit) largely following the methods outlined in Meiling et al. (2020). Photos of each colony from each imaging date (e.g., Figure 3A ) were imported into Agisoft Metashape Professional and aligned to create a sparse-point cloud of each colony  ( Figure 3B ). Briefly, camera alignment was achieved by using a suite of computer vision, outlier detection, and minimization algorithms to detect identical features in overlapping images (Harris and Stephens, 1988; Lowe, 1999), which are triangulated into 3D coordinates and then used to estimate (Fischler and Bolles, 1981; Haralick et al., 1994; Nistér, 2004) and refine (Levenberg, 1944; Marquardt, 1963) the position and orientation of each camera (Mouragnon et al., 2006). Then, image-depth mapping (Geiger et al., 2011) of the sparse-point cloud created a mesh ( Figure 3C ) that was texturized from the red-, green-, and blue-channel values from the source imagery ( Figure 3D ). The mesh face count (i.e., resolution) was set to zero to disable decimation, thereby maximizing the accuracy and precision of the downstream surface area measurements. The mesh vertices color is redundant with the image texture and was therefore disabled to reduce processing time and file size.

The texturized models for each colony were scaled using a scale bar and manually aligned to each other in 3D space to easily visualize the extent of disease progression through time ( Supplementary Figure S1 ). Separate meshes representing the total tissue area (i.e., any living tissue present on the colonies, including healthy, pale, or bleached tissue, Figure 3E ) and the total skeletal area (i.e., healthy, bleached, and pale tissue, as well as exposed skeleton) were prepared for each colony by duplicating the original mesh and removing the appropriate background mesh depending on the measurement type. Next, each mesh was smoothed to eliminate fine tissue surface details while preserving the underlying skeletal surface topography ( Figure 3F ). High detailed models required a higher smoothing factor (e.g., ~5000, where allowed values range from 0 to 9999) to eliminate fine scale detail, whereas low detailed models required a lower smoothing factor (e.g., ~500). This step was important to maximize the downstream measurement precision because varying physical conditions (e.g., light, turbidity, wave action), imaging behavior (e.g., swimming speed, camera steadiness), and polyp extension between imaging dates influence the degree of fine surface detail that can be captured by the source imagery and subsequently generated in the mesh. Lastly, 3D surface area measurements of each smoothed mesh were obtained by using the automated mesh measurement tool.

The initial total tissue area of in situ M. cavernosa colonies greatly varied from 5.0 to 6045.7 cm² so differences in initial sizes among treatments were analyzed with a one-way ANOVA using GraphPad Prism. To account for the difference in sizes among colonies, treatment success was expressed as the proportional change in total tissue area (the total tissue surface area (cm²) of each individual at each time point t > 0 divided by its initial total tissue surface area at time t = 0). A generalized linear mixed effects model to this response was fit to a normal distribution using glmmTMB (Brooks et al., 2017). Maximum likelihood estimation is robust to uneven replication among groups. The following fixed effects were included: the experimental treatment (PB, PP, CB, CP, and BC); the log₁₀-transformed cumulative days since treatment; the log₁₀-transformed initial (t = 0) total tissue area of the colony; and an index of initial condition (proportion of initial total tissue area over the total skeletal area). Additionally, a treatment-by-cumulative days interaction term was modeled to test whether the effect of the treatments varied over the course of the experiment. For the random effect, an observation-level effect of colony was fit, and both the intercept and the slope were allowed to vary over time, as individual colonies varied in their disease-progression trajectories. This random effect should further account for any differences in trajectories of colonies over time depending on how many days had passed since they were treated. Several post-hoc contrasts were conducted: between the paste and bag treatments and their respective controls, and between the two treatment controls and the background control, as implemented in the emmeans R package (Lenth, 2023). An updated model was fit with a fifth covariate, the number of treatments, to test whether the results were affected by how many applications of the treatment each colony received. This covariate was not statistically significant (p = 0.356) and its inclusion did not appreciably increase the likelihood of the model (ΔAIC = 1.159). Further, it did not alter our original inferences, and so we have presented the model lacking this term in the main text. All models were fit in R version 3.4.0 (R_Core_Team, 2024) and evaluated using the DHARMa package (Hartig, 2022). Pseudo-R² values were computed using the piecewiseSEM package (Lefcheck, 2016). All code and data that are necessary to reproduce analyses can be accessed from online platform Figshare (see Data Availability).

Slurries of surface mucus and tissue of tagged corals were collected with sterile needle-less syringes for microbiome characterization at three time points ( Table 1 ): 1) on August 19, 2020, which was 13 days before treatment; 2) on October 30, 2020 or day 59, which was 16 days after the second probiotic treatment; and 3) on January 15, 2021 or day 136 which was 93 days after the second probiotic treatment and immediately before the third treatment was applied. Diseased colonies were sampled at apparently healthy tissue (HD) and diseased lesion tissue (DD). Nearby healthy colonies (HH) were completely covered in apparently healthy tissue suggesting they had not been previously infected with SCTLD. Mucus+tissue slurries were frozen at -20°C and shipped on dry ice to the University of Florida for analysis. Samples were thawed and excess seawater was decanted from mucus+tissue slurries after microcentrifugation. Deoxyribonucleic acid (DNA) was extracted from mucus+tissue slurries according to the manufacturer’s instructions with a Qiagen DNeasy PowerSoil extraction kit. Two process control blanks were prepared with the same kits and reagents as field samples, from DNA extraction to amplicon sequencing, without the addition of biomass. The V4 hypervariable region of bacterial and archaeal 16S ribosomal ribonucleic acid (RNA) genes was amplified with the Earth Microbiome Protocol (Caporaso et al., 2012; Parada et al., 2016) using the 515F (Parada et al., 2016) and 806R (Apprill et al., 2015) primers, as previously described (Ushijima et al., 2023). Barcoded amplicon libraries were sequenced at the University of Florida Interdisciplinary Center for Biotechnology Research NextGen DNA Sequencing Core Facility (RRID: SCR_019152) on an Illumina MiSeq with the 2 x 150 base pair (bp) v. 2 cycle format. Raw sequencing reads were deposited in the National Center for Biotechnology Information’s Sequence Read Archive under BioProject PRJNA1143377.

For the microbiome analysis, primers and Illumina adaptors were removed with cutadapt v 3.4 (Martin, 2011) and all downstream analyses were conducted with R v. 4.4.0 in RStudio v. 2024.04.1 + 748 (R_Core_Team, 2024). Briefly, sequencing reads with primers and Illumina adaptors removed were used for quality-filtering, merging of reads, chimera removal, and selection of amplicon sequence variants (ASVs) within each sequencing run using dada2 v. 1.32.0 (Callahan et al., 2016). ASV tables from multiple sequencing runs were merged and taxonomy was assigned with the SILVA small subunit ribosomal RNA database v. 138.1 (Yilmaz et al., 2013). Low abundance ASVs with a mean read count across all samples of less than five were removed from the analysis. Zero counts of sequencing reads were replaced using the count zero multiplicative method with the zCompositions package in R (Palarea-Albaladejo and Martín-Fernández, 2015). Zero-replaced read count data was transformed by centered-log-ratio (CLR) and the Aitchison distance of CLR-transformed data was determined with CoDaSeq v.0.99.7 (Gloor et al., 2017). Analysis of alpha and beta diversity were conducted with functions in phyloseq v. 1.48.0 (McMurdie and Holmes, 2013) and vegan v. 2.6-6 (Dixon, 2003). A Shapiro test was used to determine the normality of alpha diversity measures, and statistical differences in alpha diversity measures were determined by 2-way ANOVA on ranked data. A permutational analysis of variation (PERMANOVA) was conducted with the adonis function in vegan on the Aitchison distance of CLR-transformed data with 999 permutations. A pairwise PERMANOVA was conducted with the R package pairwiseAdonis v. 0.4 (Martinez Arbizu, 2020). Dispersion of the Aitchison distance was calculated with the betadisper function in vegan and fitted to a linear model to test the significance. The differential abundance of taxa by treatment was determined with corncob v. 0.4.1 (Martin et al., 2020). Plots were generated with ggplot2 (Wickham, 2016) and ggbreak (Xu et al., 2021). The full reproducible R script can be accessed from online platform GitHub (see Data Availability).

Droplet digital polymerase chain reaction (ddPCR) was used to quantify gene copies of two biomarkers: one for McH1–7 and one for the putative coral pathogen Vibrio coralliilyticus. The McH1–7 biomarker was a 398-bp section of the Kor23 gene in the korormicin biosynthetic gene cluster (see Data Availability), amplified with primers designated KOR F (5’- ACGTTACCCGCTATCTGTGG-3’) and KOR R (5’- CGCTTTCCTAAAGCACTTGG-3’) (Ushijima et al., 2023). The V. coralliilyticus biomarker, a 197-bp portion of the vibriolysin-like metalloprotease gene (vcpA) (see Data Availability) was amplified with the vibriolysin F primer (5′- GGCGAACCAACTTTACTGGA-3′) and vibriolysin R primer (5′- GGTCAGTCACTGGCGTACCT-3′) (Ushijima et al., 2020). Extracted DNA and primers were sent to the University of Florida Interdisciplinary Center for Biotechnology Research Gene Expression and Genotyping Core (RRID: SCR_019145) for droplet formation, amplification with EvaGreen master mix, and analysis with the BioRad QX200 Droplet Digital PCR system. Both reactions were performed in triplicate with an annealing temperature of 60°C and 2 ng of input DNA per reaction. Negative controls with no added template were also amplified with each assay. Gene copies per ng of DNA were calculated as an average from the triplicate reactions. One-way ANOVA analyses with TukeyHSD post-hoc tests were used to detect significant differences in biomarker gene copies by treatment type, sample month, or health state using R v. 4.4.0 in RStudio v. 2024.04.1 + 748 (R_Core_Team, 2024). The R scripts are also available on GitHub (see Data Availability).

The survival of McH1–7 under two different temperature conditions, 4°C and 22°C, was investigated over a 56-day period to simulate transportation ( Figure 4A ). Time (repeated measures ANOVA, p< 0.0001), but not temperature (p = 0.165), had a significant impact on McH1–7 viability (as CFU/mL). The viability at day 7 (Šídák’s Multiple Comparisons test, all comparisons p< 0.0001) and day 14 (all comparisons p< 0.0001) was significantly higher than all later days. Although not significant, a general trend showed treatments incubated at 4°C had the greatest potential for optimal viability conditions, only dropping by one log unit over the first 14 days compared to a 3-log reduction at 22°C. Both time (repeated measures ANOVA, p = 0.061) and temperature (p = 0.115) did not significantly impact McH1–7 viability over five days at five different temperatures ( Figure 4B ).

Before bringing McH1–7 into the ocean to treat M. cavernosa, it was important to investigate if this strain may cause adverse reactions in other Caribbean coral species. Therefore, five coral species ( Supplementary Table S1 ) that are known to be susceptible to SCTLD (Sharp et al., 2020) were inoculated with McH1–7 and their reaction was monitored over 21 days. All ten fragments showed 100% healthy appearing tissue remaining ( Supplementary Figure S2 ), suggesting that McH1–7 does not elicit a visible negative reaction on these fragments of these species.

Although highly variable, the initial total tissue area of in situ M. cavernosa colonies was not significantly different among treatments ( Figure 5 ; one-way ANOVA, p = 0.428). All colonies experienced a non-linear loss of the proportion of total tissue area remaining over time ( Figure 6A ). The severity of these losses varied depending on the experimental treatment (analysis of deviance from generalized linear mixed effects model, treatment-x-time interaction, p = 0.006; Table 2 ). However, neither initial condition (p = 0.680, Table 2 ) nor initial total tissue area (p = 0.255, Table 2 ) significantly influenced the response. Further investigation of the proportion of tissue area remaining, derived from the generalized linear mixed effects model controlling for covariates, revealed that the probiotic bag treated colonies exhibited a lesser decline in this proportion than the other treatments ( Figure 7 ). In total, the model explained 38.2% of the deviance in the proportion of total tissue area remaining based on the fixed effects alone, increasing to 90.1% when including the random effect of colony.

Considering just the final proportion of total tissue area remaining at the end of the experiment—an indicator of the total cumulative treatment effect over the course of the experiment—the probiotic bag treated corals had the highest average proportion remaining at 0.790 ± 0.124 (mean ± standard error) ( Figure 6B ). Post-hoc contrasts between the probiotic paste and bag treatments and their respective controls, as well as between the two treatment controls and the background control [as implemented in the emmeans R package (Lenth, 2023)], showed that the average final proportion of total tissue remaining was significantly higher than for those colonies receiving the control bag [based on post-hoc contrasts (p < 0.001)]. The control bag treated colonies only had 0.533 ± 0.160 proportion of final total tissue area remaining at the end of the experiment. Oppositely, the colonies treated with the probiotic paste had a lower proportion of tissue remaining (0.278 ± 0.110) than the control paste treated colonies (0.592 ± 0.117; p < 0.001). Both treatment controls did, however, perform significantly better than the background control, which had only 0.336 ± 0.151 proportion total tissue area remaining by the end of the experiment (p< 0.001 relative to both the bag and paste controls).

Amplicon libraries were successfully sequenced for 194 coral microbiome samples and two process control blanks ( Supplementary Table S2 ). Microbiomes were successfully sequenced from 51 samples before treatment, which included three healthy colonies and 24 paired samples of healthy tissue and disease lesion from diseased colonies. Microbiomes were also successfully sequenced from 80 samples collected sixteen days after the second probiotic treatment (experiment day 59), which included 16 healthy colonies and mostly paired samples of disease lesions (n=33) and healthy tissue (n=31) from diseased colonies. This translated to microbiomes of six to eight coral colonies per treatment, with samples of both healthy and diseased tissue. Likewise, microbiomes were successfully sequenced from 63 samples 93 days after the second probiotic treatment (experiment day 136), which included five healthy colonies and mostly paired samples of disease lesions (n=22) and healthy tissue (n=36) from diseased colonies. This translated to microbiomes of six or seven coral colonies per treatment, with samples of both healthy and diseased tissue. In total, microbiomes from healthy and diseased tissue were successfully characterized from 24 colonies at all three timepoints as well as 16 additional colonies without the full complement of samples spanning three timepoints, five treatments, and two tissue types ( Table 1 ). After quality-filtering, joining reads, and chimera removal, an average of 43,139 sequences per sample were retained for analysis and a total of 20,164 amplicon sequence variants (ASVs) were detected. A total of 76 ASVs were detected in the two process control blanks ( Supplementary Table S3 ). After removal of low-abundance ASVs, 756 ASVs were detected in coral samples and used for further analysis.

Alpha diversity, measured as observed ASVs, Shannon diversity index, and Simpson diversity index, varied by sample collection date (2-way ANOVA, p< 0.001), but not with treatment. All three measures indicated that alpha diversity was lowest two weeks after the second probiotic treatment on experimental day 59 ( Supplementary Figure S3 ) and Tukey multiple comparisons of means showed that alpha diversity was lower on experimental day 59 compared to before treatment (p-adj< 0.001) and compared to three months after treatment on experimental day 136 (p-adj< 0.01).

Analyses of beta diversity revealed that overall, the sample collection date explained more of the variation in microbial community composition (PERMANOVA, R² = 0.085, p< 0.001) than treatment (PERMANOVA, R² = 0.042, p< 0.001) or health condition (PERMANOVA, R² = 0.033, p< 0.001) ( Figure 8 ). However, it is important to note that 84% of the variation between microbial communities was not explained by three factors examined here. After controlling for sample collection date, treatment did not explain microbiome variation. In August, before the start of probiotic treatments, microbiome variation was explained by health condition (PERMANOVA, R² = 0.076, p = 0.01), where microbiome composition in disease lesions was statistically different from both healthy colonies (pairwise PERMANOVA, R² = 0.111, p< 0.001) and healthy tissue on diseased colonies (pairwise PERMANOVA, R² = 0.0648, p = 0.001). On the two sample collection dates after probiotic treatment (16 days and 93 days after the second treatment), health condition continued to explain microbiome variation, but treatment did not. To further explore what drove variation among the M. cavernosa colonies over the early treatment phases, we examined the differential abundance of taxa among treatment types, controlling for sample collection date. A total of five ASVs were differentially abundant by treatment and the applied probiotic strain Pseudoalteromonas sp. McH1–7 was not one of the differentially abundant taxa ( Supplementary Figure S4 ). One ASV classified as Halodesulfovibrio and one classified as Endozoicomonas had lower relative abundances in all treatment types compared to no treatment, while one Photobacterium ASV had higher relative abundance in all treatment types compared to no treatment. One ASV classified as Ilumatobacter was generally low (< 1% relative abundance) and lower in treated corals except for one sample treated with the control bag in which this ASV was at 4% relative abundance. One ASV classified only to family Rhodobacteraceae was detected at 1% relative abundance or less in all samples, but some corals treated with control paste or probiotic paste had higher relative abundances compared to the bag treatments.

We also examined the twenty most abundant ASVs over time, by treatment and by tissue type ( Figure 9 ). The most abundant taxa included one Actinobacteria (Renibacterium), one Desulfovibrionia (Halodesulfovibrio), three Alphaproteobacteria (Nautella, Ruegeria, Thalassobius), and one Verrucomicrobiae (Roseibacillus). The remaining 14 abundant taxa were Gammaproteobacteria, including multiple ASVs of Pseudoalteromonas, Vibrio, Endozoicomonas, and Halomonas ( Figure 9 ). These twenty predominant ASVs were reflective of the most common bacterial orders. For example, the most predominant bacterial orders across all timepoints and treatments were the gammaproteobacterial orders Enterobacterales and Pseudomonadales as well as the alphaproteobacterial order Rhodobacterales ( Supplementary Figure S5 ). The relative abundance of these predominant taxa changed by sampling period, irrespective of treatment type, likely reflecting the response of the microbial community to seasonal environmental changes. For example, among the Vibrio taxa, Vibrio ASV2, ASV3, and ASV4 were most abundant before treatment in August, Vibrio ASV1 was more abundant 16 days after the second treatment in October, and all four Vibrio taxa were at lower abundance 93 days after the second treatment (January 2021). Similar seasonal fluctuations were also seen in the genus Endozoicomonas, where Endozoicomonas ASV1 was most abundant in August, while Endozoicomonas ASV2 and ASV3 were more abundant in January. Endozoicomonas ASV1 was also detected as enriched in untreated corals in the differential abundance analysis. In addition, both Pseudoalteromonas and Halomonas taxa varied seasonally, with highest abundances in October, lower abundances in January, and lowest abundances in August. Pseudoalteromonas ASV1 is an exact sequence match over 253 nucleotides of the 16S ribosomal RNA gene to McH1–7 that was used in the probiotic treatments. Pseudoalteromonas ASV2 was one nucleotide different from Pseudoalteromonas ASV1. Overall, Pseudoalteromonas taxa were most abundant 16 days after the second probiotic treatment was applied, even in untreated corals ( Figure 9 ).

To complement the conserved 16S ribosomal RNA taxonomic marker survey, specific biomarker assays using high-sensitivity ddPCR were conducted to detect functional genes associated with the probiotic strain Pseudoalteromonas sp. McH1–7 or the putative pathogen Vibrio coralliilyticus. These functional gene assays showed low abundances (less than 25 gene copies per ng of DNA) of the biomarkers for both the probiotic and putatively pathogenic strains over the three time points, regardless of treatment type ( Figure 10 ). The McH1–7 biomarker copies varied by treatment, with higher gene copies in the probiotic paste versus untreated corals (p-adj = 0.005). In addition, the McH1–7 biomarker copies were higher in January than in August (p-adj< 0.001) or in October (p-adj< 0.0001), while the 16S rRNA survey showed highest relative abundances of Pseudoalteromonas taxa in October. This indicates that while members of the Pseudoalteromonas genus were most abundant two weeks after the second probiotic treatment, these taxa may not be the strain that was applied since the strain-specific functional gene was generally not detected in the same DNA samples. Gene copies of the vcpA biomarker were also low (less than 10 gene copies per ng of DNA) across the sample months and were not significantly different among treatments, sample month, or tissue type. The predominant Vibrio taxa detected by the conserved 16S ribosomal RNA taxonomic marker survey may not contain the putative virulence factor vcpA from Vibrio coralliilyticus since it was present at very low gene copy numbers.

The whole-colony, and not the lesion-specific technique, successfully slowed or stopped the progression of SCTLD. In aquaria, McH1–7 slowed or stopped SCTLD progression and transmission among M. cavernosa without dominating the host microbiome (Ushijima et al., 2023), suggesting its ability to assimilate with the microbiome and enhance the coral’s functioning without causing a disruptive imbalance. The colony-wide colonization of McH1–7 in situ may be necessary to produce any significant therapeutic effects. The whole-colony bagging technique was effective at trapping McH1–7 against whole M. cavernosa colonies for a 2-hour duration, which our results suggest is a sufficient time for microbial colonization. Alternatively, the lesion-specific paste treatment may have provided only a localized or otherwise inadequate inoculum, contributing to its ineffectiveness. Interestingly, the probiotic paste treatment contributed to more tissue loss compared to the control paste treatment. The reason for this result remains unclear; however, it is possible that high concentrations of McH1–7 negatively impacted the already stressed tissue at the lesion site. This could be due to localized hypoxia, changes in the microbiomes at the lesion site (McDevitt-Irwin et al., 2017; Zaneveld et al., 2017), or production of antibiotics (Ushijima et al., 2023) or other metabolites by McH1–7 that were harmful at high concentrations. In contrast, the application of the paste control led to less tissue loss, on average, than the untreated colonies. One potential explanation is that the paste is changing the relationship between the coral host and mucus, as mucus is central to the host’s defense mechanism (Brown and Bythell, 2005).

Assessment of the coral microbiome response during early phases of the probiotic treatments in situ demonstrated that the applied probiotic strain McH1–7 did not bloom or drastically alter the resident microbiome, in agreement with previous aquaria trials (Ushijima et al., 2023). Instead, microbiome variation in both treated and untreated corals in this study showed patterns of seasonal changes in composition or differences between SCTLD-impacted corals and visually healthy corals. This suggests that McH1–7 can have a beneficial impact on the host without establishing high relative abundances for extended periods.

From our data and a previous publication (Ushijima et al., 2023), McH1–7 appears to persist on the corals but at low concentrations. In previous tank trials, ddPCR was used to track the genes for production of the antibiotic korormycin. This genetic signal was not detected after ~24 h, but we were able to detect McH1–7 again weeks later. In the current study, we detected McH1–7 at all three time points tested, albeit at low levels. Further, we found that McH1–7 does not dominate and significantly shift the host microbiome. We hypothesize that McH1–7 persists on the coral at levels at or below the detectable limit of ddPCR, and therefore, it may persist on the corals longer than it can be accurately detected. Also, because we have not verified the mechanism through which McH1–7 protects corals, we are currently unable to conclude anything about the long-term protective effects. For example, if it is based on the production of an antimicrobial or immune system-stimulating molecule, we could attempt to determine the concentration needed for a protective effect. Further, if the effect is based upon McH1–7 producing a trans-acting factor (e.g., an antibiotic), then the cellular concentration can be low while the effective concentration of the factor is reached.

While examining the role of seasonality in M. cavernosa microbiomes was not the goal of this study, we observed shifts over time in two predominant genera, Pseudoalteromonas and Vibrio, that may play a role in SCTLD dynamics. First, we saw higher relative abundances of two predominant Pseudoalteromonas ASVs in October and January across all corals, regardless of treatment. This enrichment coincides with the cool and dry season in southeast Florida and the concurrent decline in SCTLD presentation in the region (Walker et al., 2021; Toth et al., 2024). In contrast, multiple Vibrio ASVs were in lower relative abundance in both October and January during the cool/dry season compared to August, the warm/wet season. The decline in the relative abundance of Vibrio ASVs during the cool/dry season and the increase in the relative abundance of Pseudoalteromonas likely reflects seasonal environmental changes that favor one group over the other. However, these patterns may also be impacted by the microbe-microbe interactions that inspired the development of the probiotic treatments described here. Pseudoalteromonas are well known for their genetic capacity to generate diverse antimicrobial products (Paulsen et al., 2019; Eze et al., 2023; Ushijima et al., 2023; Wang et al., 2024) and the probiotic strain McH1–7 has demonstrated inhibition of multiple strains of Vibrio in laboratory studies (Ushijima et al., 2023). Similarly, the low relative abundance of V. coralliilyticus at all sampling time points coincides with the consistent detection of Ruegeria, members of which have demonstrated inhibition of V. coralliilyticus (Miura et al., 2019; Xu et al., 2024). The seasonal variation in the resident microbiome suggests that timing of probiotic treatments during the year has the potential to influence host outcomes, an area that warrants further investigation.

After repeated whole-colony bagging treatments occurring over 4.5 months, the probiotic strain McH1–7 may provide protection from SCTLD for the following 2+ years, as the amount of total tissue area remaining only dropped 7.1% compared to 35.1% on the control bag treated corals over this time. Probiotics used as prophylactic treatments have been found to reduce coral bleaching by preventing the reduction in bacterial diversity, fostering beneficial strains while simultaneously alleviating pathogenic-induced dysbiosis (Xu et al., 2024). Beneficial microorganisms can also mitigate bleaching prophylactically by causing changes to the coral host including lipid maintenance and transcriptional reprogramming of cellular repair and stress protection (Santoro et al., 2021). However, not much is understood about using prophylactic treatments for SCTLD. Since McH1–7 potentially produces at least four types of antibiotic compounds (korormicins, tetrabromopyrrole, marinocine, and pseudoalterins) (Ushijima et al., 2023), it is predicted that this strain reduces pathogenic bacteria on the coral host. The potential to reduce multiple pathogenic strains of bacteria may be instrumental as SCTLD has been found to disrupt the coral microbiome (Meyer et al., 2019; Rosales et al., 2020; Ushijima et al., 2020; Work et al., 2021; Becker et al., 2022; Huntley et al., 2022; Rosales et al., 2022, Rosales et al., 2023). That said, the pathway for the use of McH1–7 prophylactically remains unknown and needs further investigation to maximize the effectiveness of the treatment.

The viability of McH1–7 did not significantly differ over the first 120 hours of storage at five different temperatures. However, when investigating viability over 56 days at both 4°C and 22°C, the survival of McH1–7 significantly declined after the first two weeks of storage, regardless of temperature. These results suggest a two-week shelf life of this strain. Although not significant, there was a general trend that colony counts were higher for McH1–7 in the 4°C treatment during this time, suggesting that refrigeration should be utilized if available but is not necessary for McH1–7 survival. These methods allow for the probiotic strain to be grown in a sterile laboratory setting and then dispersed to even remote reefs within 14 days.

All five alternative species of Caribbean corals conveyed no visual negative effects over 21 days after being inoculated with McH1–7 in aquaria, which provides a promising outlook for the safety of McH1–7 treatments in the reef ecosystem. Since McH1–7 was isolated from a local healthy M. cavernosa, did not dominate the host coral microbiome, and did not negatively impact healthy controls when applied in aquaria (Ushijima et al., 2023), it is generally understood that this probiotic does not disadvantage M. cavernosa or, based on our laboratory trials, other common stony corals in the Caribbean reefscape. Further, Delgadillo-Ordoñez et al. (2024) found probiotics were able to successfully alter the host microbiome without changes to the surrounding seawater and sediment environment. Although, higher replication among a greater diversity of reef species is necessary to better understand any wider reaching impacts of McH1–7 on the reef.

We do not believe that limited application of this probiotic would impact the surrounding reef environment for several reasons. First, this probiotic was isolated from Florida corals, so nothing foreign is being introduced (Ushijima et al., 2023). Most bacterial populations are controlled by natural phage populations, so a native bacterium being reintroduced into the same environment would likely be controlled (Batinovic et al., 2019; Chevallereau et al., 2022). Coral microbiomes have been restructured by probiotic applications without microbial shifts in surrounding seawater or sediments (Delgadillo-Ordoñez et al., 2024). It is possible that the probiotic is pathogenic to some other reef organism; however, SCTLD has already significantly shifted the reef ecosystem and McH1–7 is already present on Florida’s coral reef. Nonetheless, the impacts of widespread application of McH1–7 remain unknown. Therefore, the impact of McH1–7 or other probiotics on the surrounding reef setting, especially in larger quantities, should be further characterized to ensure the sustainable application on wild corals. Further investigation is necessary to determine if a single application is sufficient to replicate the benefits we demonstrate in this study.