Bacterial inoculation elicits changes to the coral epigenome

Abstract

Environmental shifts can cause epigenetic modifications in corals associated with changes in gene expression and physiology, although it remains unclear if associated bacteria can also induce such changes. Here, we inoculate nubbins of the coral Pocillopora verrucosa with an opportunistic pathogen, Vibrio coralliilyticus, and/or a coral probiotic, Cobetia sp., and subject the nubbins to heat stress. We show that pathogen exposure leads to distinct DNA methylation changes compared to the control, probiotic, and co-inoculation groups. We also confirm that DNA methylation broadly correlates with coral gene expression; however, individual correlations are restricted to a subset of genes. We then highlight genes altered by pathogen inoculation that show similar responses in their expression and methylation. Notably, coral probiotic exposure mitigates specific epigenetic changes caused by pathogen inoculation, which correlates with increased stress resilience and higher coral survival rates. Although many aspects remain to be resolved, evidence that bacteria alter coral epigenomes may help clarify the role of microbiome interactions in coral resilience.

Pocillopora verrucosa corals were inoculated with a bacterial pathogen and/or a coral probiotic, then subjected to heat stress. Pathogen inoculation induced epigenetic and gene expression changes that were mitigated by probiotic co-inoculation.

Introduction

Recent work has shown that beneficial bacteria (probiotics) can improve coral health and resilience in response to stress caused by pathogens, increased temperature, and oil spills, including in in situ experiments^1–10. This is because bacteria are essential to the proper functioning of the coral holobiont^11–13, which consists of the coral animal and its associated symbiotic microbes. For example, inoculation of the coral Pocillopora damicornis with beneficial microorganisms for corals (BMCs) consisting of bacteria from the genera Yangia, Roseobacter, Phytobacter, and Salinicola, correlated with increases in energy reserves and calcification rates¹⁴. Furthermore, administration of BMCs during and after heat stress increased transcription of genes involved in recovery and stress attenuation and decreased those related to inflammatory and innate immune responses in the coral Mussismilia hispida⁵. These changes also correlated with improved coral survival and increased photosynthetic efficiency in the associated symbiotic microalgae (i.e., Symbiodiniaceae)⁵. High temperatures, though, can also lead to dysbiosis and increased establishment of opportunistic pathogens, causing coral bleaching and death^5,15. For example, inoculation with the temperature-dependent coral pathogen Vibrio coralliilyticus brought about a reduction in skeletal density and abnormal crystal morphology in P. damicornis³. This could be attenuated by inoculation with a BMC consortium composed of bacteria from the genera Pseudoalteromonas, Cobetia, and Halomonas³, showing the protective capacity of BMCs against pathogens as well as temperature stress.

Coral-associated microorganisms have recently been posited to have the ability to alter coral epigenomes, thus modifying phenotypic responses¹⁶. The most studied epigenome modification in corals is DNA methylation, which, depending on the location of the methylated bases, prevents transcriptional noise and increases fidelity^17–19. The interaction between coral epigenomes and their symbionts was recently demonstrated by exchanging the Symbiodiniaceae strain in the coral Montastraea cavernosa and applying thermal stress, which together produced changes in DNA methylation²⁰. Exchanging or manipulating bacterial symbionts may have similar impacts on the coral host; however, this has not been studied. These impacts could be either indirect, by changing the nutritional status or altering metabolic capacities¹³, or direct, by producing metabolites that act as cofactors for epigenome-modifying enzymes²¹ or encoding proteins that are homologs of bacteria-produced proteins capable of inducing epigenomic changes in mammals¹⁶.

Here, we seek to determine whether inoculation with bacteria leads to changes in the coral epigenome that align with improved resilience to thermal stress. To do this, we regularly inoculate Pocillopora verrucosa nubbins with either a BMC (Cobetia sp. strain pBMC4²²; hereafter “Cobetia”), a temperature-dependent pathogen (V. coralliilyticus strain BAA-450²³; hereafter “Vibrio”), or both bacteria (hereafter “Both”), and subject the corals to a temperature stress regime (from 27 °C to 33 °C), followed by a recovery period. Heat-stressed corals are hereafter denoted with “H_”, whereas corals kept at ambient temperature throughout the experiment are denoted with “A_”. The mesocosm experiment encompasses 28 days, with samples taken before any inoculation is applied (T0 = day 0), at the peak of the heat stress (T1 = day 19), and after a recovery period (T2 = day 28). DNA methylation differences at the gene level are analyzed and linked to increases in the abundance of the respective inocula, gene expression levels, and coral phenotypic responses, to show that bacterial inoculation, in combination with temperature stress, can modulate the coral epigenome. Through co-inoculation of a probiotic and pathogen, we further show that indicators of stress seen in the pathogen-inoculated corals are reduced when co-inoculated with the BMC. This pattern is observed in all components of the holobiont (host phenotype, host epigenome, host gene expression, symbiont photosynthetic efficiency, and microbiome composition) and underscores the potential for probiotic therapy as prevention of epigenomic disturbances caused by pathogenic infection.

Results

Probiotic inoculation increases coral resilience

To evaluate the effects of bacterial inoculation and temperature stress on P. verrucosa, we monitored nubbin health over a 28-day mesocosm experiment using photosynthetic efficiency measurements and nubbin condition/coloration. This showed that Vibrio inoculation negatively affected coral resilience to temperature stress during recovery, while the Cobetia and co-inoculation groups showed relatively higher resilience. The photosynthetic efficiency (Fv/Fm values measured by Diving-PAM II) was significantly lower in heat-stressed corals than in ambient temperature corals at T1 (Wilcoxon Rank Sum: W = 256, p = 1.54e−6) and T2 (Wilcoxon Rank Sum: W = 248, p = 6.64e−6) (Fig. 1A and Supplementary Data 1a). The Fv/Fm values were not significantly different between inoculation groups in either the ambient temperature corals (Kruskal-Wallis: X² = 4.87, df = 3, p = 0.182) nor the heat-stressed corals (Kruskal-Wallis: X² = 3.97, df = 3, p = 0.265) at T2, although the mean Fv/Fm for H_Vibrio was lower than the other heat-stressed groups (0.368 ± 0.245 for H_Vibrio, 0.523 ± 0.047 for H_Control, 0.514 ± 0.045 for H_Cobetia, 0.530 ± 0.012 for H_Both).

Fig. 1. Coral phenotypic response and inoculated bacteria abundances throughout the experiment.

A Mean photosynthetic efficiency (Fv/Fm ratios; y-axis) from coral nubbins during the mesocosm experiment days (x-axis). The error bars represent standard deviations (n = 4 per group). Carets (^) below the schematic represent days of bacterial inoculation (every 3 days starting from day 1). B Average color intensity and C percent live tissue of the coral nubbins in each condition, as measured by ImageJ (T0: n = 12 per group; T1/T2: n = 4 per group). Lines represent the mean value for each condition, and the error bars represent standard deviations. Higher color intensities represent more paling, bleaching, or tissue loss. Color intensity was measured on the CoralWatch Coral Health Chart⁶⁶ for reference. Percent abundances are shown in boxplots of the bacterial inocula (Cobetia sp. (D) and V. coralliilyticus (E)) across inoculations and sampling times (n = 4 per group). Boxplots display the individual data points, medians, lower and upper quartiles, and whiskers displaying the 1.5x interquartile range. Abundances of each inoculum were compared between inoculation groups (i.e., Control, Cobetia, Vibrio, Both) within each temperature condition (i.e., ambient temperature or heat-stressed). B–E Statistical comparisons were performed using two-sided Kruskal-Wallis tests, followed by post-hoc Dunn’s multiple comparison tests with Benjamini-Hochberg adjustment. Asterisks denote significant differences between groups within a time point; Dunn test: B p = 0.043, C p = 0.014. D, E Letters denote significant differences between groups; Dunn test: (a) p = 0.040, (b) p = 0.031, (c) p = 0.047, (d) p = 0.039, (e) p = 0.020, (f) p = 0.009, (g) p = 0.014, (h) p = 0.023, (i) p = 0.002, (j) p = 0.023, (k) p = 0.045. Source data are provided as a Source Data file.

Color intensity of coral tissue was measured by converting each image to grayscale and calculating the average intensity (from 0 = black, to 256 = white) across the whole image of the coral nubbin. Thus, a lower average color intensity would represent darker, i.e., healthier, living tissue, and a higher color intensity would represent pale, bleached, or dead coral tissue. Similar to previous findings⁴, the combination of heat stress and Vibrio inoculation (H_Vibrio) resulted in significantly higher color intensity, i.e., ‘whiteness’ or paling, (Kruskal-Wallis: X² = 8.49, df = 3, p = 0.037; Dunn test: p = 0.043; Fig. 1B) and lower percent live tissue (Kruskal-Wallis: X² = 10.08, df = 3, p = 0.018; Dunn test: p = 0.014; Fig. 1C) when compared to the control group (H_Control) after the recovery period (T2) (Supplementary Data 1b). Notably, both the color intensity and percent live tissue in all other inoculation groups within the same temperature regime, including the co-inoculation group, were statistically indistinguishable from each other (Fig. 1B, C). Although the Vibrio-inoculated coral nubbins were paler than the nubbins from the other experimental groups at T2 in the heat-stressed corals, the color intensities of the live tissue remaining on the coral nubbins were not significantly different (Kruskal-Wallis: X² = 7.48, df = 3, p = 0.058; Supplementary Fig. 1).

Using 16S rRNA gene analysis, we found that the relative abundance of V. coralliilyticus was significantly higher in the corals treated with the pathogen than the control group at T1 (Kruskal-Wallis: X² = 13.059, df = 3, p = 0.005; Dunn test: p = 0.002) and T2 (Kruskal-Wallis: X² = 11.647, df = 3, p = 0.009; Dunn test: p = 0.045) (Fig. 1D, E and Supplementary Data 2). The abundance of Vibrio in the co-inoculation groups was lower than the Vibrio-inoculations at T1, but not at T2, although these differences were only significant between A_Vibrio and A_Both at T1 (Kruskal-Wallis: X² = 11.978, df = 3, p = 0.007; Dunn test: p = 0.023). This coincided with increased microbiome alpha diversity in all heat-stressed groups at T2 (Supplementary Fig. 2) and microbiome restructuring in the heat-stressed groups at T1 and T2, marked by a decrease in members of the family Endozoicomonadaceae and increases in Saprospiraceae and Rhodobacteraceae (Supplementary Fig. 3). Significant differences in beta diversity were also observed between temperature groups and between the bacterial inoculation groups within both temperature conditions at T1 and T2 (Supplementary Fig. 4, Supplementary Data 3a and 3b).

Coral epigenomes respond to temperature stress and bacterial inoculation

To explore whether the stress mitigation in co-inoculated coral nubbins was also attributed to interactions with the coral epigenome, genome-wide differences in DNA methylation were analyzed in response to temperature stress, bacterial inoculation, and their interaction. The profile of median methylation levels of genes across the genome of P. verrucosa was visualized with a non-metric multi dimensional scaling (nMDS) plot for each time point (Fig. 2, Supplementary Data 4a). Significant differences existed at the peak of stress (T1) between corals exposed to the heat stress and those that were not (PERMANOVA: R² = 0.107, F = 3.245, p = 0.001; PERMDISP: p = 0.981), showing that methylation marks can change after twelve days of constant exposure to temperature stress (and 19 days of experimentation).

Fig. 2. Non-metric Multi Dimensional Scaling (nMDS) plots based on Bray-Curtis dissimilarities of the median methylation levels of genes across the genome of P. verrucosa.

Samples from each time point were separated using Bray-Curtis dissimilarity and plotted along the first two dimensions. Circles = 95% confidence intervals around temperature groups. Both temperature and inoculation were tested as factors driving the separation between groups using PERMANOVA (1000 permutations, two-sided) at A T0, B T1, and T2. R² and p-values are displayed, representing the magnitude and significance attributed to each of the factors. Homogeneity of multivariate dispersions was assessed using ANOVA on Bray-Curtis distance-based dispersions (betadisper). Pairwise PERMANOVA comparisons were conducted with Benjamini-Hochberg adjustment for multiple comparisons (n = 4 per group, except A_Control at T1, H_Control at T0/T1/ T2, and H_Both at T1, where n = 3). nMDS plots for T2 are shown highlighting differences based on C temperature and D bacteria inoculation. Source data are provided as a Source Data file.

The difference between samples subjected to different temperature conditions was also observed after 9 days of recovery (28 days total; T2) (PERMANOVA: R² = 0.061, F = 1.955, p = 0.002; PERMDISP: p = 0.399). Additionally, ‘timepoint’ was a significant factor in separating the heat-stressed corals using genome-wide DNA methylation levels, and pairwise comparisons showed significant differences between each of the time point pairs T0-T1, T1-T2, and T0-T2 (Supplementary Fig. 5). Bacterial inoculation groups also significantly contributed to the separation of samples at T2 (PERMANOVA: R² = 0.112, F = 1.197, p = 0.028; PERMDISP: p = 0.412). However, there was no statistical support for the interaction between temperature and bacteria inoculations (PERMANOVA: R² = 0.106, F = 1.133, p = 0.106). Therefore, all further methylation analyses focused on temperature and inoculation as separate factors causing differences in DNA methylation. Pairwise comparisons showed that pathogen-inoculated corals were significantly different from the control corals (PERMANOVA: R² = 0.101, F = 1.456, p = 0.029), but there was no statistical support for pairwise differences between any of the other inoculation groups at T2 (Supplementary Data 4b).

Pathogen inoculation leads to changes in gene methylation

The significant genome-wide differences in median methylation levels observed between the temperature conditions at T1 and the temperature conditions and bacteria inoculation groups at T2 (Fig. 2) were further investigated using pairwise generalized linear models (GLMs). A total of 932 genes were significantly associated with temperature differences at T1, and 343 genes were significantly associated with the different temperature groups at T2 (Supplementary Fig. 6, Supplementary Data 5a and 5b). At T2, 17 genes were found to be significantly influenced by Vibrio inoculation in comparison to the control group, 15 of which had depleted median methylation levels correlated with Vibrio inoculation (Supplementary Data 5c). Extending the comparison of these 17 genes to the other inoculation groups using hierarchical clustering based on Euclidean distances shows that coral nubbins inoculated with Vibrio separated from the other three conditions at T2 (Fig. 3). These results corroborate the stressed physiological response of the coral nubbins caused by Vibrio inoculation observed at T2 (Fig. 1B, C) because it shows that the other inoculation groups (Cobetia and Both) cluster more closely to the Control group than the Vibrio inoculation in these genes. This demonstrates that, although pathogen inoculation significantly alters coral DNA methylation, co-inoculation of the pathogen and BMC leads to methylation patterns similar to the control group, supporting the notion that BMCs negate impacts caused by pathogen stress on coral health.

Fig. 3. Genes with median methylation levels significantly associated with Vibrio inoculation according to GLMs compared to samples from other inoculation groups at T2 (n = 4; n = 3 for H_Control).

There is a separation between the corals inoculated with V. coralliilyticus and the other inoculation groups (x-axis). Hierarchical clustering was performed using Euclidean distances and the “ward.D2” agglomeration method. Two clusters were formed at a height of 14.1, which separated Vibrio-inoculated corals from the other inoculation groups. Two clusters of methylated genes were defined by cutting the dendrogram at a height of 18.3. Colors represent the z-score for any given gene. The enriched GO terms using TopGO display the significant functions with depleted methylation levels in pathogen-stressed corals (cluster β). All enriched GO terms are in Supplementary Data 6a. All GO terms associated with these genes are in Supplementary Data 7. Source data are provided as a Source Data file.

Increases in gene body methylation reduce spurious transcription in highly expressed genes, thereby reinforcing the epigenetic control of genes as expression increases^17–19. Our findings here revealed that methylation within P. verrucosa occurred primarily within the gene body (Supplementary Fig. 7, Supplementary Data 8). Genes with depleted methylation in the corals treated with Vibrio only (cluster β in Fig. 3) highlight the link between pathogen stress and diminished epigenetic control of genes involved in cell maintenance and functioning and genes related to the immune system, which potentially correlates with decreased gene expression.

The enriched GO terms in cluster β included “adenylate cyclase-inhibiting G protein-coupled glutamate receptor signaling pathway” and the “regulation of synaptic transmission (glutamatergic)”, which both rely on the neurotransmitter glutamate. Glutamate production is necessary for algal symbiosis in the cnidarian model Aiptasia²⁴. Moreover, in the coral Stylophora pistillata, heat stress was found to disrupt glutamate biosynthesis and trigger glutamate catabolism, which contributed to coral bleaching²⁵. Therefore, pathogen inoculation may exacerbate losses in glutamate caused by heat stress by reducing the epigenetic control of glutamatergic signal transduction. This may be further supported by significant differences at T2 in the GO term “glutamatergic synapse” (GO:0098978) due to temperature conditions (Supplementary Data 6c), although additional research in this area is necessary, as heat stress did not uniformly alter methylation patterns in all of the affected glutamatergic synapse genes (Supplementary Fig. 8). Further coral immune functions with depleted median methylation levels due to Vibrio inoculation include “lamellipodium morphogenesis”, “early endosome to Golgi transport”, and the stress response mechanisms “interstrand cross-link repair” and “regulation of DNA repair”.

An analysis to identify differentially methylated regions (DMRs) revealed 96 DMRs attributable to temperature (ambient vs. heat-stressed), and 8 DMRs attributable to Vibrio inoculation; no DMRs were detected when comparing Cobetia or the co-inoculation to the control group (Supplementary Data 9). All eight pathogen-associated DMRs were located within gene bodies (4 in exons, 4 in introns; Supplementary Fig. 9).

Pathogen inoculation affects gene expression

To determine whether the methylation changes correlated with changes in gene expression levels, the transcriptomes of the T2 nubbins were analyzed using RNA-seq. Heat stress and pathogen inoculation caused distinct transcriptome changes, with H_Vibrio samples separating from the other groups in a principal component analysis (PCA) plot along PC1 (which accounted for 74.93% of the variance) (Fig. 4A). Hierarchical clustering also showed that three of the four H_Vibrio samples separated distinctly from the other groups (Supplementary Fig. 10). Likelihood ratio tests identified 12,165 differentially expressed genes (DEGs) affected by temperature, 5255 DEGs affected by inoculation, and 3690 DEGs affected by their interaction (p < 0.05). Interaction effects were largest in Vibrio-inoculated corals compared to control (1838 genes), Cobetia (1387 genes), and co-inoculated corals (1259 genes) (Fig. 4B; Supplementary Data 10). The median methylation levels across the genome positively correlated with the relative expression level of genes (R² = 0.084, p < 2e−16; Fig. 4C). Of the 17 genes with altered methylation due to pathogen inoculation, four showed a significant correlation between changes in methylation and gene expression across experimental groups (Fig. 4D and Supplementary Data 11). Three of the genes, g13505, g16333, and g8986, were positively correlated with gene methylation, and these were defined by GO terms related to transmembrane transporter activity, negative regulation of transcription by RNA polymerase II, and cilium movement, respectively (Supplementary Data 7). One gene, g1533, was negatively correlated with gene methylation and was not associated with a GO term (Supplementary Data 7).

Fig. 4. Gene expression levels are affected by heat stress and bacterial inoculation and correlated with methylation levels at T2.

A PCA generated from the transcripts per million (TPM) expression estimates of P. verrucosa genes (n = 4 per group). B The number of DEGs across each pairwise comparison using Wald tests (n = 16 for temperature groups, n = 8 for inoculation groups, and n = 4 for interaction groups). C A linear model shows that median methylation levels for all the genes across all 31 samples were significantly and positively correlated with gene expression (using Log₂ (TPM + 1)). The correlation between median methylation and gene expression was assessed using a two-sided linear regression model (R² = 0.084, p < 2e−16). D The 17 genes with median methylation percents affected by pathogen inoculation are plotted against their respective gene expression. Four of these genes had median methylation levels that significantly correlated with gene expression (denoted with asterisks). Each dot represents the gene expression (using Log₂ (TPM + 1)) of a gene in a sample (32 samples total). Shaded regions represent 95% confidence intervals of the linear regression trendlines for each gene. Source data are provided as a Source Data file.

Gene methylation differences are correlated with phenotypic response

To determine which genes affected by Vibrio inoculation (Fig. 3) correlated most with the paling and tissue loss observed in the corals (Fig. 1), linear models were implemented using the median methylation levels and color intensity as factors (Supplementary Data 12). The genes showing the highest correlation between methylation percent and phenotypic response (R² > 0.100 and p < 0.05) were g16333, g1533, g17345, g14350, and g8986 (Fig. 5). Example GO terms describing these genes were “negative regulation of transcription by RNA polymerase II” (GO:0000122), N/A, “ceramide transport” (GO:0035627), “adenylate cyclase-inhibiting G protein-coupled glutamate receptor signaling pathway” (GO:0007196), and “cilium movement” (GO:0003341), respectively (Supplementary Data 7). Of the genes with defined functions, three are related to cellular processes, while g17345 (“ceramide transport”) functions in macromolecule localization within the cell. In all of the defined genes, Vibrio-inoculation led to a significant depletion in DNA methylation when compared to the control group in either the ambient temperature or heat-stressed conditions. In all but one gene (g16333), co-inoculation with Cobetia prevented similar decreases in DNA methylation when compared with the control group. The genes g14350, defined by the GO term “adenylate cyclase-inhibiting G protein-coupled glutamate receptor signaling pathway”, and g1533, had significantly different median methylation levels between Vibrio-treated corals and the co-inoculation in the heat-stressed condition (ANOVA: F = 4.46, df = 3, p = 0.028, TukeyHSD: p = 0.033 for g14350; ANOVA: F = 11.62, df = 3, p = 0.001, TukeyHSD: p = 0.014 for g1533).

Fig. 5. Methylation profiles of genes with significant correlations (R² > 0.100, p < 0.05) between methylation levels and phenotypic response (coral nubbin color intensity).

Linear regression plots show the correlation lines of coral nubbin color intensity to the median methylation percent in each sample (n = 91 total coral nubbins) along with the 95% confidence intervals (shaded regions). Boxplots showing the median methylation percent of the samples at T2 (after recovery), were used to compare between groups with one-way ANOVA, followed by TukeyHSD (n = 4; n = 3 for H_Control). Boxplots display the individual data points, medians, lower and upper quartiles, and whiskers displaying the 1.5x interquartile range. Letters denote significance between inoculation groups in the boxplots; TukeyHSD: (a) p = 0.027, (b) p = 0.003, (c) p = 0.014, (d) p = 0.017, (e) p = 0.001, (f) p = 0.014, (g) p = 0.031, (h) p = 0.014, (i) p = 0.002, (j) p = 0.026, (k) p = 0.033, (l) p = 0.015, (m) p = 0.007. Source data are provided as a Source data file.

Discussion

Although hypothesized, the interaction between changes in the coral-associated microbiome and the coral epigenome was yet to be demonstrated^16,21. The findings of this study are four-fold, as we show that, in corals, (1) the addition of a single probiotic (Cobetia sp.) was able to reduce the indicators of stress caused by inoculation with the pathogen V. coralliilyticus; (2) bacterial inoculation, as well as temperature stress, contribute to differences in DNA methylation; (3) gene expression is, on the whole, positively correlated with methylation levels and affected by pathogen and heat stress; (4) pathogen inoculation causes changes in DNA methylation and gene expression to specific cellular processes and immune responses that are mitigated by probiotic co-inoculation.

Regarding the phenotypic response, differences in color intensity and tissue loss observed in the coral nubbins at T2 between the control group and the Vibrio inoculation in the heat-stressed corals may be a delayed effect of the establishment of the pathogen in the heat-stressed samples at T1. Interestingly, the relative abundance of V. coralliilyticus was lower at T2 than T1, coinciding with the observed tissue loss. The heat-stressed groups at T1 and T2 were also defined by a decrease in the family Endozoicomonadaceae, which is thought to be a stable²⁶ and possibly beneficial²⁷ group of bacteria within Pocillopora microbiomes. These results, and the increases to bacterial community diversity indices in heat-stressed nubbins, indicate that the holobiont is becoming increasingly dysbiotic²⁸. Furthermore, the co-inoculation with the BMC mitigated Vibrio establishment in the microbiome during the peak of stress at T1, and controlled the relative abundances of Vibrio to a similar degree at T2. The co-inoculation also had a positive impact on the phenotype in heat-stressed corals at T2, contributing to the increased stress resilience and survival observed in this group versus the Vibrio-inoculated group. Together, this supports the notion that, although heat stress leads to overall changes in the microbial community, attenuation of the pathogen by the addition of probiotics boosts coral resilience.

DNA methylation has been shown to be responsive to environmental changes^{18,20,29–32}, correlate with phenotypic acclimatization^18,32, and be heritable³³. Plasticity of DNA methylation in corals thus far has been tested on the scale of weeks^32,34–36 and months^20,30,31,37 to a year or more^17,18,29,37. The findings here further prove the plasticity of coral epigenomes during stress, as 12 days of sustained heat stress induced DNA methylation changes across the genome. Furthermore, the difference continued during a nine-day recovery period, which exemplifies the persistence of epigenome modifications brought about by environmental change.

Recently, the concept of “epigenetic memory”, where intragenerational differences between epigenomes of different individuals lead to vertical transmission of distinct epigenomes, was confirmed in the brain coral Platygyra daedalea³³. This concept has been recognized in plants for many years (reviewed in ref. ³⁸). Here, the changes in epigenomes driven by environmental differences, as well as bacterial inoculations, signify the potential for long-term persistence after removal of the environmental stress and show promise for microbiome manipulation experiments with implications for long-term acclimation and coral resilience.

When analyzing the methylation levels of the genes and putative functions, we can conclude that Vibrio inoculation caused changes to methylation levels in metabolic and cellular processes relevant to the observed phenotypic responses (Fig. 1). The difference between the Vibrio-inoculated corals and the other three inoculation groups (control, inoculation with Cobetia sp., and the co-inoculation) in these genes at T2, independent of the temperature regime, demonstrates the coral probiotic’s ability to prevent changes to the coral epigenome in key functions altered by the pathogen (V. coralliilyticus). This effect was associated with increased resilience to stress and higher survival rates.

Moreover, methylation of functions related to algal symbiosis (e.g., “adenylate cyclase-inhibiting G protein-coupled glutamate receptor signaling pathway” and “lamellipodium morphogenesis”) and immune responses (e.g., “early endosome to Golgi transport”) was depleted in Vibrio-inoculated corals. As mentioned above, pathogen stress may reduce epigenetic control over glutamate receptor pathways, disrupting algal symbiosis and compromising coral health. Glutamate mediates interactions between corals and their algal symbionts²⁴, and inhibiting glutamate receptors disrupts this relationship³⁹. Lamellipodia, and by association “lamellipodium morphogenesis”, also contribute to coral-algal interactions⁴⁰, as well as wound healing in the cnidarian model Clytia hemisphaerica⁴¹. Reduced methylation of “early endosome to Golgi transport” supports previous findings that Vibrio targets endosomal pathways during infection⁴². Specifically, toxins produced by V. coralliilyticus disrupt endosome development and downregulate immune genes in the host^42,43. These methylation changes indicate the Vibrio’s ability to elicit a response to coral host defense genes, and the lack of difference between co-inoculated corals and the other inoculations shows the probiotic’s ability to mitigate this effect.

The DMR analysis further showed the effect of Vibrio inoculation on gene body methylation, as all eight pathogen-linked DMRs were located within exons and introns. Notably, the absence of DMRs in Cobetia-inoculated or co-inoculated corals suggests that methylation remodeling at the regional level is selectively triggered by Vibrio. None of the DMRs overlapped with the differentially methylated genes, which may be due to the stringent filtering that was intended to retain only high-confidence methylation calls. While these thresholds helped to minimize false positives at individual CpG sites, they may reduce the resolution of site-specific DMR analyses but not the more conservative median methylation level analyses.

Interestingly, the effect of Vibrio inoculation on gene expression was significantly altered by its interaction with heat stress, as evidenced by the separation of inoculation groups in the PCA (Fig. 4A) and the large number of interaction-driven DEGs identified by pairwise Wald tests (Fig. 4B). While temperature and Vibrio inoculation independently shaped both expression and methylation, their interaction also drove changes in gene expression. Although there is an association between DNA methylation and gene expression, questions remain about the direction and magnitude of this relationship and its implications for acclimation and adaptation^35,36. This has led to the hypothesis that coral gene body methylation responds to transcription levels, rather than functioning as a gene regulatory mechanism, as it does in mammals^17,31,36. However, the decoupling of methylation and gene expression responses to the combined stressors exemplifies the complex and not yet fully elucidated relationship between these two molecular components in corals.

RNA-seq analysis confirmed that overall gene expression was positively correlated with gene methylation, which adds to the growing evidence that basal coral epigenomes and gene expression levels are correlated^17,18,35,44. The concordant down-methylation and down-regulation of genes related to transmembrane transporter activity (g13505), cilium movement (g8986), and negative regulation of transcription by RNA polymerase II (g16333) due to pathogen exposure may indicate direct mechanisms or host responses to the stress. For example, transmembrane transporters are essential for transporting nutrients between corals and their symbionts; reduced expression of related genes may signal a breakdown in symbiosis⁴⁵. The pathogen’s impact on cilium movement expression and methylation also shows that the stressed coral nubbins may have been unable to remove excess reactive oxygen species⁴⁶, which could have contributed to tissue bleaching and death observed in our samples. RNA polymerase II, on the other hand, helps control spurious transcription by recruiting the histone-methylating enzyme SetD2 to transcribed DNA^17,19. Therefore, pathogen-induced repression of this gene demonstrates a disruption in both the transcriptional and epigenetic control of gene fidelity. The significant positive correlations between DNA methylation in the latter two genes (g8986 and g16333), as well as in the glutamate receptor pathway gene (g14350), and the coral’s phenotypic response to Vibrio inoculation further demonstrate the importance of integrating coral epigenomes into coral immune research.

The direct mechanisms by which bacteria are able to elicit changes to the coral epigenome are still to be explored, although metabolites and proteins produced by coral-associated bacteria are suggested to play a role^16,21. Additionally, the crosstalk between coral epigenomes and coral-associated bacteria may not be unidirectional, as pathogen susceptibility has been hypothesized to be at least partly controlled by epigenetic factors in aquaculture fish⁴⁷. By connecting rapid microbiome changes to more long-term and heritable epigenetic changes (Fig. 6), bacterial inoculation has the potential to induce lasting differences, even without direct establishment in the bacterial community. Thus, as future studies explore this interaction further, the current paradigm that microbiome interventions only lead to short-term and volatile changes⁴⁸ might shift. Therefore, as research in this field progresses, there is considerable potential to shape the use of customized microbial therapies for increasing resistance and resilience in vulnerable corals^49,50.

Fig. 6. Rethinking the flexibility and longevity of coral microbiomes via crosstalk with coral epigenomes.

Microbiome composition can change rapidly and may be transient⁶⁷. Our results linking microbiome manipulation to changes in coral DNA methylation suggest, however, that microbiome perturbations can leave a more persistent molecular imprint on the host through modification of the coral epigenome. This interaction provides a potential mechanism by which short-lived changes in the bacterial community could generate longer-lasting and potentially heritable effects on coral phenotypes.   The figure therefore extends  the linear framework proposed  in ref. ⁶⁸ by introducing crosstalk between the bacterial community and host epigenetic regulation, linking microbiome flexibility to longer-term host responses  (left purple arrow). Solid lines indicate study-supported responses due to bacterial inoculation, whereas dashed lines represent direct, indirect, or potentially cascading effects of this crosstalk.

Methods

Coral collection and experimental setup

This study was conducted under the remit of the King Abdullah University of Science and Technology Institutional Biosafety and Bioethics Committee (IBEC) and approved under protocol number 22IBEC003, in compliance with all research compliance guidelines of the Kingdom of Saudi Arabia National Committee of Bioethics (KSA NCBE). In order to prevent intergenomic variability that would obfuscate identification of methylation marks⁵¹, clonal fragments from Pocillopora verrucosa were collected from the Al Fahal reef in the central Red Sea (22°18.3071’ N, 38°57.8741’ E) on 14 March 2022 (in situ seawater temperature = 25 °C) and brought to the aquariums at the Coastal and Marine Resources Core Lab (CMR) in King Abdullah University of Science and Technology (KAUST). The coral colony was acclimated to CMR conditions (using a 300 L open-system aquarium tank, with a maximum intensity of 180 μmol of photons m ^− 2 s⁻¹ between 10:00 and 14:00, and a 12-hour day/night cycle changing at 7:00 and 19:00) and brought to 27 °C. Corals were further fragmented into finger-sized nubbins on 4 April 2022. After a five-week recovery period, nubbins were divided into the eight experimental tanks, which housed the replicates for each inoculation group (10 L aquarium tanks, using the same light parameters as the acclimation tank above) containing 0.22 µm filtered seawater. The coral nubbins were acclimated for seven days (until the average Fv/Fm ratio, a measure of photosynthetic efficiency and proxy for coral health, from the coral nubbins in each experimental tank was >0.600, using a Diving-PAM II; Walz, Germany).

The bacteria used in the experiment were the BMC Cobetia sp. strain pBMC4 (isolated from Pocillopora verrucosa²²) and the opportunistic pathogen Vibrio coralliilyticus strain BAA-450 (isolated from Pocillopora damicornis²³). The bacterial cultures used in this experiment were grown using Marine Broth 2216 (Becton Dickinson, USA) for 24 hours at 27 °C before each inoculation. To minimize additional carbon input from the culture media, the cultures were spun at 6000 x g for 5 min and the bacterial pellets were resuspended in sterilized 3.5% NaCl solution. This process was repeated for a total of two washes, then the bacteria were resuspended in sterilized 3.5% NaCl solution for inoculation. Abundance and purity of each of the cultures were checked by the plate counting method using Marine Agar 2216 (Becton Dickinson, USA).

Starting on day 1, bacteria were inoculated into each of the tanks every three days at a final concentration of 10⁵ cells/mL. Thirty percent of the water was replaced with new 0.22 µm filtered seawater (kept at the same temperature as the experimental tanks) every other day. The experiment was performed over 28 days. Of the eight experimental tanks, four were maintained at 27 °C throughout the entirety of the experiment (denoted with an “A” for ambient temperature). In the other four experimental tanks, starting from day 1, temperature was increased from 27 °C to 33 °C (by increasing the temperature 1 °C per day for 6 days), then maintained at 33 °C for 12 days (until the average Fv/Fm across all conditions was <0.450), then returned to 27 °C (by decreasing the temperature 1 °C per day) where they were kept until the end of the experiment (4 days). These tanks are denoted with an “H” for heat stress. The four experimental tanks in each of A or H, contained one of the following experimental conditions: corals with no inoculation (“Control”), corals inoculated with Cobetia sp. at a final concentration of 10⁵ cells/mL (“Cobetia”), corals inoculated with V. coralliilyticus at a final concentration of 10⁵ cells/mL (“Vibrio”), and corals inoculated with both Cobetia sp. and V. coralliilyticus, each at a final concentration of 10⁵ cells/mL (“Both”). In total the eight conditions were: A_Control, A_Cobetia, A_Vibrio, A_Both, H_Control, H_Cobetia, H_Vibrio, and H_Both.

Samples of four coral nubbin replicates (n = 4) per tank were taken at the start of the experiment before any bacterial inoculation (T0), on day 19 (peak of stress, T1), and on day 28 (after recovery, T2). At each sampling time, the apical tip of the sampled nubbin was removed and flash-frozen in liquid nitrogen for nucleic acid analysis. Frozen nubbins were then transferred to -80 °C until processing.

Coral health parameters

Coral health was monitored during the experiment using a Diving-PAM II (Walz, Germany) to measure the Fv/Fm ratios. Fv/Fm measurements were taken after dark, acclimating the coral nubbins for at least 1 hour the night before each inoculation day (every three days starting on day 0) throughout the entirety of the experiment (n = 4 for each experimental tank). Fv/Fm measurements were compared between temperature groups using Wilcoxon Rank Sum tests⁵², and between inoculation groups Kruskal-Wallis H tests⁵³. Images were taken of the coral nubbins at T0 and at their respective sampling times (T1 or T2).

ImageJ was used to analyze the percent live tissue, color intensity, and color intensity of live tissue in each of the images of the nubbins from sampling times T0 (n = 12 for each group), T1 (n = 4 for each group), and T2 (n = 4 for each group). For the percent live tissue, the area of live tissue on the coral nubbin was divided by the total area of the coral nubbin in each respective image. The color intensity, as described in the main text, was obtained using the unweighted average color intensity for each coral nubbin, after converting the image to 8-bit grayscale (offering 2⁸ = 256 levels of differentiation). The average color intensity incorporates the color intensities (from 0 = black to 256 = white) of all pixels within a coral nubbin. This, then, incorporates the color of the live tissue and any exposed coral skeleton that may be present in coral nubbins with tissue loss. For the color intensity of live tissue, only the unweighted average in color intensity observed in the live tissue portion of the image was considered. Percent live tissue, color intensity, and color intensity of live tissue measurements were compared between the samples with Kruskal-Wallis H tests⁵³, followed by Dunn’s multiple comparison test⁵⁴ using Benjamini-Hochberg adjustment.

16S rRNA gene amplicon sequencing

DNA and RNA were extracted from airbrushed coral tissue using AllPrep DNA/RNA Mini Kit (Qiagen, Germany), following manufacturer’s instructions. Extracted DNA was used for analysis of the bacterial community as well as identification of the methylation marks within the coral genome. For bacterial community analysis, the DNA composing the V3-V4 region of the 16S rRNA gene was sent to Novogene (Novogene, China) for sequencing using a NovaSeq 6000 (Illumina, USA) with the primers 341F (5’-CCTAYGGGRBGCASCAG-3’) and 806R (5’-GGACTACNNGGGTATCTAAT-3’).

Library preparation and sequencing of methylated DNA

Libraries of converted methylated DNA were prepared using the NEB Enzymatic Methyl-seq Kit (New England Biolabs Inc., USA) with the EM-seq Index Primers (New England Biolabs Inc., USA). Prepared libraries were sent to Novogene (Novogene, China) for sequencing using a NovaSeq 6000 (Illumina, USA). Due to insufficient DNA quality in 5 samples, 91 of the 96 samples were used for downstream analysis (n = 4 for each experimental group; except A_Control at T1; H_Control at T0, T1, and T2; and H_Both at T1, which all had n = 3).

Coral microbiome analysis

For microbiome analysis, primer sequences were removed using cutadapt⁵⁵. Sequences were further filtered and trimmed using the DADA2 pipeline⁵⁶. Taxonomy was assigned using the SILVA_nr99_v138.1 database⁵⁷. Sequences identified as Bacteria that did not match Chloroplast or Mitochondria were further analyzed using phyloseq⁵⁸. ASVs with a 100% match to the 16S rDNA sequence of the inocula (Cobetia sp. or Vibrio coralliilyticus) via BLASTN⁵⁹ were determined to be representative ASVs of the inoculated bacteria. Statistical differences between relative abundance data in different groups were determined using the vegan package in R (version 4.3.1). Bacterial abundances were not normally distributed (Shapiro-Wilk: p < 0.05), so significant differences between groups were determined by non-parametric tests. Kruskal-Wallis H tests⁵³, followed by Dunn’s multiple comparison tests⁵⁴, were used to find significant differences between inoculation groups (n = 4 for each experimental group).

Microbiome composition differences were calculated between inoculation groups in each of the temperature conditions and time points using principal coordinate analyses (PCoA). Significant factors driving the separation between samples and sample dispersion were determined using the adonis2 and betadisper commands in the vegan R package (version 4.3.1). Pairwise differences between inoculation groups were tested using the pairwise.adonis command in the pairwiseAdonis R package.

Coral DNA methylation

DNA methylation was analyzed bioinformatically following the steps described in Liew et al. (2020)³³. Reads were first trimmed using TrimGalore v0.6.10⁵⁵. Then, the filtered reads were mapped to a reference genome for Pocillopora verrucosa⁶⁰ and deduplicated, before methylation marks were scored per-position using Bismark v0.23.1⁶¹. Methylated DNA positions were only considered bona fide if they met the criteria defined in Liew et al. (2020)³³. Briefly, the probability of methylated positions resulting from sequencing error needed to be below 0.05 after correcting for false discovery rate, and the methylated reads were only kept if the methylation occurred in all replicates of at least one of the experimental conditions. Also, to minimize type I errors, the median coverage was set to ≥ 10, with minimum coverage set to ≥ 5 across all of the samples, in order to be considered for further analyses^17,18,33 (Supplementary Data 13).

The median methylation levels across genes were compared between groups. Each methylated position within a given gene was given a score between 0 (no methylation in any of the cells) and 1 (fully methylated across all cells), and the median across each gene was used for further analyses. Only genes with five or more methylated positions were used for downstream analyses. Nonmetric multidimensional scaling (nMDS) plots were used to compare the median methylation across the genome of the coral using Bray-Curtis dissimilarity to separate samples. Significant factors driving the separation between samples and sample dispersion were determined by PERMANOVA using the adonis2 and PERMDISP using betadisper commands in the vegan R package (version 4.3.1). Pairwise differences between groups were tested using the pairwise.adonis command in the pairwiseAdonis R package.

To extrapolate the genes related to the difference observed between the Control inoculation and the Vibrio-treated corals at T2, generalized linear models (GLMs) were implemented in R using the following formula:

$$\mathrm{glm}\left(\mathrm{median}\_\mathrm{methylation}\_\mathrm{levels}~\mathrm{temperature}+\mathrm{inoculation},\mathrm{family}=\mathrm{gaussian}\left(\right)\right)$$ 1

for the data from T2, where median methylation levels for each gene were compared with the predictor variables temperature (ambient vs. heat-stressed) and inoculation (by using the Control inoculation as a baseline to compare with each of Cobetia-treated, Vibrio-treated, and the co-inoculation inoculation samples). Genes with fewer than 5 methylated positions were removed from the analysis. Genes with Vibrio inoculation as a significant predictor variable (after correcting for false discovery rate) were plotted on a heatmap with all of the T2 samples using hierarchical clustering with Euclidean distances and the “ward.D2” agglomeration method.

Differentially methylated regions (DMRs) were then discovered to determine whether there were correlations with genes that were differentially methylated due to pathogen inoculation at T2. DMRs were analyzed using the Dispersion Shrinkage for Sequencing data (DSS) method following the same additive model as for the GLMs above⁶². DMRs were determined using the suggested parameters (minimum length = 50 bp, minimum of three CpG sites, p < 0.001).

Functional enrichment of differentially methylated genes

GO term annotations were collected from the published P. verrucosa genome⁶⁰. GO term enrichment analysis was performed using TopGO⁶³ on the genes with significantly different methylation levels from the previous step. Enriched GO terms with p  <  0.05 were considered significant.

RNA-seq analysis

T2 RNA samples (n = 32) extracted using the AllPrep DNA/RNA Mini Kit (Qiagen, Germany) were lyophilized onto a GenTegraRNA plate (Gentegra, USA) and sent to the MR DNA (Molecular Research LP, USA) for RNA-seq analysis. There, the lyophilized RNA was resuspended in 30 µL RNase-free water, and 500–750 ng of total RNA was used for library preparation using the poly-A selection method of KAPA mRNA HyperPrep Kits (Roche, Switzerland) following the manufacturer’s instructions. The libraries were then pooled in equimolar ratios of 0.6 nM and sequenced using a NovaSeq 6000 (Illumina, USA). RNA-seq reads were first trimmed using TrimGalore v0.6.10⁵⁷. The adapter-trimmed reads were pseudo-aligned to P. verrucosa gene models⁶⁰ using kallisto v0.50.1⁶⁴ to quantify gene expression levels. Transcripts per million (TPM) obtained from sleuth v0.30.1⁶⁵ were then used to compare across inoculation and temperature groups from T2 (n = 4 for each condition) using multiplicative models. Likelihood ratio tests (LRTs) were conducted to determine the overall significance of each factor (temperature, inoculation, and the interaction term) in extracting differentially expressed genes (DEGs) (Benjamini-Hochberg adjusted p-values < 0.05). Pairwise Wald tests were performed between each comparison, and significantly upregulated or downregulated genes were determined from Benjamini-Hochberg adjusted p-values (p < 0.05) and absolute log₂ fold changes >1 (n = 16 per temperature group, n = 8 per inoculation group, and n = 4 per interaction group). Correlations between TPM and median methylation levels across genes were determined using the linear model:

$$\mathrm{lm}\left({\log }^2\left(\mathrm{TPM}+1\right)~\mathrm{median}\_\mathrm{methylation}\_\mathrm{levels},\mathrm{data}=\mathrm{gene}\right)$$ 2

Significant correlations in individual genes were determined using Benjamini-Hochberg adjusted n-values (p < 0.05).

Correlating median methylation with phenotypic parameters

The genes predicted to be associated with Vibrio inoculation in the previous step were then tested against a phenotypic response variable using linear models with the formula:

$$\mathrm{lm}\left(\mathrm{median}\_\mathrm{methylation}\_\mathrm{levels}~\mathrm{color}\_\mathrm{intensity},\mathrm{data}=\mathrm{gene}\right)$$ 3

to determine which of these genes correlated with phenotypic differences observed in the corals affected by temperature and inoculation. The genes with R² > 0.100 and p < 0.05 (after Benjamini-Hochberg correction) were plotted using linear regressions with 95% confidence intervals. Boxplots showing the median methylation levels of these genes in each sample at T2 were then used to compare differences between groups using one-way ANOVA followed by TukeyHSD.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Author contributions

A.R.B., H.D.M.V., G.C., T.T., M.A., C.R.V., and R.S.P. designed the experiments; A.R.B., H.D.M.V., P.M.C. conducted the mesocosm experiment and sample collection; A.R.B. and N.D. performed DNA extractions; A.R.B. and F.C.G. performed analyses; A.R.B. wrote the manuscript with input from all authors; A.S.R. and R.S.P. provided guidance and funding during the experiment.

Peer review

Peer review information

Nature Communications thanks Javier Rodriguez-Casariego and the other anonymous reviewer(s) for their contribution to the peer review of this work. A peer review file is available.

Funding

This work was supported by KAUST grant number BAS/1/1095-01-01.

Data availability

EM-seq, RNA-seq, and raw 16S rRNA gene amplicon sequencing data generated in this study have been deposited in the NCBI Sequence Read Archive under the BioProject accession number PRJNA1123499 (https://www.ncbi.nlm.nih.gov/sra/PRJNA1123499). Source data are provided with this paper.

Code availability

The code used for data analysis can be found at https://github.com/arbarno/pver_epibac.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at https://doi.org/10.1038/s41467-026-77646-2.