The relevant methods as described in Vompe et al. (2023):

Site and in situ experiment

Our experimental study site is located on the northern forereef in Mo'orea, French Polynesia (S17° 28.386′ W149° 49.059′). Mo'orea is a tropical, volcanic high island in the Central South Pacific Ocean. A shallow, ~1 km wide lagoon and barrier reef surround the island. The forereef gradually slopes downwards toward the open ocean and is composed of coral spur and sand groove formations. At the inception of our experiment in August 2018, this reef was dominated by scleractinian corals with low abundance of fleshy macroalgae. Coral cover was 56.0 ± 1.0% (mean ± SE) and macroalgae cover was 0.8 ± 0.2% (mean ± SE).

At this site, we have an ongoing in situ experiment investigating tipping points of coral benthic and microbial ecology in response to nutrient enrichment and herbivore reduction, as in Adam et al. (2022). Briefly, our experimental platform is a factorial design at 10 m depth on the forereef, consisting of four herbivore exclosures (~1 m2 each) placed over eight natural 30-m2 reef plots. The plots are exposed to two levels of nutrient enrichment (four plots ambient/four plots enriched) and four levels of herbivory (exclosures with different size holes of 2.5 cm × 2.5 cm, 5.0 cm × 5.0 cm, 7.5 cm × 7.5 cm, or open top, with one exclosure of each herbivory condition at each plot). Nutrient enrichment was achieved in the plots via PVC tubes with Osmocote® (19-6-12, N-P-K) slow-release garden fertilizer. These tubes were wrapped in plastic mesh to contain the fertilizer. The nutrient enrichment tubes were replaced every 12–16 weeks, except for two periods during the COVID-19 pandemic when travel to Mo'orea was not possible. See Supplementary Methods (Vompe et al. 2023) for a full description of the experimental setup.

Coral sampling for microbiome analysis

To investigate how the microbiomes of different coral species respond to environmental stress, samples of Acropora retusa, Porites lobata species complex, and Pocillopora spp. were collected over 2 years (July 2018–August 2020), 3× a year, in March, July or August, and November. Corals in the P. lobata species complex will be referred to as P. lobata below for brevity. However, we acknowledge there may be cryptic diversity in our samples (Brown et al., 2021). A nonmetric multidimensional scaling (NMDS) ordination of Bray–Curtis distances between P. lobata sample microbiomes from July 2018 suggests that the possible presence of cryptic members of the P. lobata species complex in our dataset was unlikely to affect P. lobata microbiome variation, as there are no obvious sample microbiome composition clusters. The taxonomic name Pocillopora spp. is used for this study because Pocillopora species have high cryptic diversity (Johnston et al., 2022), which makes it difficult to visually delineate among species. We selected Pocillopora spp. specimens that had consistent phenotypes similar to those now defined as Pocillopora meandrina or Haplotype 8a as described in Figure 1 of Johnston et al. (2022). Different coral species, even genotypes, tend to have distinct microbiomes (Bourne et al., 2016; Dunphy et al., 2019; Rosales et al., 2019). A NMDS ordination of Bray–Curtis distances between Pocillopora spp. sample microbiomes from July 2018 suggests that the possible presence of cryptic Pocillopora species in our dataset was unlikely to affect Pocillopora spp. microbiome variation, as there are no obvious sample microbiome composition clusters.

All colonies of each coral species appeared healthy when initially selected for microbiome sampling. Live tissue on these focal colonies was repeatedly sampled throughout the study regardless of subsequent visual phenotype, as long as live tissue remained. Live tissue was sampled at haphazardly chosen locations on the colonies at each time point. For A. retusa and Pocillopora spp., haphazardly chosen live branch tips were sampled. For P. lobata, live tissue was sampled from haphazardly chosen locations around the center of the colony. Coral samples were collected in July 2018, November 2018, March 2019, August 2019, November 2019, March 2020, and August 2020, covering a 28-month period. Additional coral colonies were sampled in November 2018, March 2019, and August 2019 to increase sample sizes and to account for initial focal colony mortality. Colonies of each species were also added to the dataset in March 2020 and August 2020 to restore sample size due to colony mortality. Bleaching and mortality data for coral colonies added to the microbiome sampling effort after the start of the experiment were collected retroactively. This was possible because these corals were already present in the exclosures and data could be collected from our photomosaic time series from before they were added to the microbiome sampling effort.

During each sampling event, coral fragments <1 cm3 were snipped from each of the focal colonies using bone cutters that were flame-sterilized with 95% ethanol at the surface. Corals were sampled between 08:00 and 14:00 h to help minimize diel microbiome variation. Fragments were immediately placed in sterile 207 mL Whirl-Paks. This volume of sample is sufficient to produce accurate microbiome data without significantly damaging the focal colony (Zaneveld et al., 2016). Upon surfacing, Whirl-paks were placed on ice and transported to shore (~15 min) then transferred to Qiagen DNeasy PowerSoil lysis matrix tubes, containing a guanidinium thiocyanate preservative, using 95% ethanol flame-sterilized forceps. Tubes were stored at −40°C prior to transport on Techni Ice to Oregon State University where they were stored at −80°C until further processing.

Coral bleaching and mortality data collection and analyses

At each microbiome sample collection point, except during March 2020, ~64 high-resolution digital photographs were taken of each plot using an Olympus TG camera. The digital photographs were then used to generate orthorectified photomosaics using Agisoft Metashape Professional Version 1.7.0 (AgiSoft PhotoScan Professional, 2024) software. These high-resolution photomosaics allowed identification of benthic organisms to a low taxonomic resolution. Percent cover of benthic organisms was quantified using 225 random points per m2 plot using CoralNet software (Lozada-Misa at al., 2017) in manual annotation mode. Photomosaics were also used to create digital maps of the plots in which each focal coral was identified and circled. For each focal colony, we recorded whether coral bleaching or mortality was present (prevalence) and, if so, estimated the percentage of the colony surface area (to the nearest 5%) that was bleached or dead (severity). Because corals undergo natural, seasonal variation in Symbiodiniaceae density that can affect their coloration, we defined bleached tissue only as tissue that had lost all pigmentation.

Mortality and bleaching were tracked for focal corals at all time points, except March 2020. Comparisons between average per colony bleaching and mortality levels were assessed across coral species and time points using one-way analysis of variance (ANOVA) with TukeyHSD (honestly significant difference) post hoc tests. In March 2020, samples for microbiome analysis were collected, but the photomosaics were not conducted due to the interrupted field season resulting from the COVID-19 shutdown. However, qualitative notes on bleaching and health status during coral microbiome sample collection were recorded for focal colonies. Additionally, coral bleaching and mortality data were gathered for A. retusa and Pocillopora spp. during the peak of the 2019 heatwave in May, as described above; however, these data were collected from individual photos of coral plots taken from the top, not photomosaics. In this study, we only analyzed correlations between microbiome dynamics and host bleaching and mortality for months during microbiome sample collection so that phenotypic and microbiome data for each coral colony could be paired.