Fish predation hinders the success of coral restoration efforts using fragmented massive corals

Coral fragmentation

Colonies used in both experiments were collected from a seawall at Fisher Island, Miami, Florida (25.76°N, 80.14°W; depth = 1.8 m). Each parent colony was cut into small fragments (average size = 4.2 ± 1.9 cm² (mean ± SD) using a diamond band saw, and fragments were attached to ceramic plugs using super glue as described by Page, Muller & Vaughan (2018). After fragmentation, the height of the fragments ranged from 0.5–1.0 cm. The ceramic plugs with corals were placed on PVC frames and then fixed to coral trees (Nedimyer, Gaines & Roach, 2011) at the University of Miami’s in-water coral nursery (25.69°N, 80.09°W; depth = 9.4 m) where they were allowed to acclimate for 4–6 weeks before outplanting (Fig. 1A). After this recovery period, the fragments were strongly cemented (no corals were dislodged during the transport and outplanting steps) but had not fully skirted tissue onto the ceramic plugs. Due to funding constraints, the parent colonies used in this study were not formally genotyped. Nevertheless, fragments from every parent colony were represented in each reef and treatment and, thus, the results combine the survivorship of corals from three parent colonies per species in experiment one and five parent colonies in experiment two.

Outplanting experiment one: assessing outplant survivorship

The first outplanting experiment consisted of four coral species with massive or brain colony morphologies: O. faveolata (listed as threatened under the US Endangered Species Act), P. clivosa, P. strigosa, and M. cavernosa. Fragmented corals were glued onto ceramic plugs using polyurethane waterproof glue (Gorilla Glue) (Figs. 1A, 1B). The plugs with the corals were then mounted into cement pucks with 2-part epoxy putty (AllFix). Finally, these pucks were secured onto the reef using cement (1 part Portland cement, 0.1 part Silica fume), raising corals three cm above the substrate to limit sediment and algal interactions (Fig. 1C). Only corals with healthy (no discoloration or lesions) tissue were used. Corals were outplanted onto three reef sites in Miami, Florida, in June–July 2018: Reef 1 (25.70°N, 80.09°W; depth = 6.0 m), Reef 2 (25.68°N, 80.10°W; 7.5 m), and Reef 3 (25.83°N, 80.11°W; 6.4 m). These reefs have low topography and very low cover of stony corals (<1%, Supplemental Data). Corals were collected and outplanted under Florida’s Fish and Wildlife Commission Permit SAL-19-1794-SCRP. Corals were deployed within replicate grids (from 3 × 3 to 5 × 5 m) whose areas were determined based on substrate availability. Four such grids were deployed in Reef 1, 7 in Reef 2, and 6 in Reef 3. The corals were spaced 40–60 cm apart within each grid to maintain a consistent density of corals across replicate plots and grids were separated by at least 2 m. The coral outplants were placed at least 20 cm away from existing stony and soft corals, sponges, and the zoanthid Palythoa. In total, 53 M. cavernosa, 123 O. faveolata, 80 P. clivosa, and 41 P. strigosa were outplanted among the three reefs, with all 4 species represented in each plot.

For each outplant in this experiment, we documented presence/absence of the coral fragment (Fig. 1D) and prevalence (i.e., proportion of corals by species with signs of predation) of tissue mortality caused by predation on remaining fragments (e.g., missing polyps, feeding scars) (Figs. 1E, 1F) at one week, one month, and six months after deployment. The proportion of corals physically removed by fish predators from their outplanting platforms was compared among coral species, reefs, and time since outplanting using a Generalized Linear Model (GLM) following guidance by Warton & Hui (2011). Here, we used a GLM with a binomial distribution and a logit link function to model the probability of outplant removal. The model incorporated species, reef, and monitoring period (time) as fixed variables. Residuals diagnostics plots were used to test model assumptions, and D-squared values, which indicate the amount of deviance accounted for by the models (i.e., analogous to R²), were used to evaluate the goodness of fit of the selected models. Tukey post hoc tests were used to evaluate pairwise differences among the levels of the categorical variables in the models (species, reefs, time). All statistical analyses were performed in R v3.5.3 (R Core Team, 2017).

Outplanting experiment two: reducing predation impacts

Based on the high level of predation observed during Experiment One, a second study was designed to determine if predation impacts could be minimized through modifications to the outplanting method. This experiment tested the role of the skeletal profile (i.e., the height of the coral fragment) and attachment medium (glue vs cement) on coral removal and predation rates. This experiment used coral fragments (average size = 2.8 ± 6.5 cm² (mean ± SD)) from five P. clivosa colonies. A high level of predation and abundance of fish predators were recorded at Reef 1 in the first experiment, so this location was chosen as the study site for the second experiment.

Compared to fragments with healed, skirting edges, corals with exposed skeletal walls may provide easier access to the coral tissue and encourage growth of endolithic or turf algae that could attract grazing by fish. Hence, we hypothesized that the exposed skeletal profile (height) and the presence/absence of bare skeleton on the sides of a fragment would influence predation patterns (Figs. 2A–2B). We further hypothesized that the rate of the physical removal of outplanted fragments would be related to the attachment method. To test this, we developed a triangular cement platform (cement “pizza”; Figs. 2C–2D) that used cement (in lieu of glue) to secure corals and allowed the height of the fragments to be adjusted by varying the amount of cement used. Corals were secured to the cement pizzas as four treatments:

1. “raised exposed”, with fragments placed on top of the cement so that vertical walls (devoid of tissue) protruded from the cement treatment (Fig. 2E);

2. “raised covered”, with fragments placed on top of the cement so that vertical walls (covered with tissue) protruded from the cement treatment (Fig. 2F);

3. “flushed”, with fragments embedded into the cement so that the fragment was level with the cement platform and only the surface of the coral was visible (Fig. 2G);

4. “embedded”, with fragments embedded into the cement so that the coral was positioned one cm below the surface of the cement to prevent access by fish (Fig. 2H).

Individual coral fragments were attached in groups of three (triads) onto each cement pizza to foster coral fusion and faster colony development as described in Paget et al. (2018). The pizzas were cemented individually onto the reef pavement within plots (n = 120 corals placed onto 40 pizzas). Each plot consisted of 10 pizzas (n = 3–4 pizzas per treatment), with each pizza spaced 30–50 cm apart. Plots were separated by 1 m. In addition to using the cement pizzas, coral fragments were mounted onto ceramic plugs using glue and outplanted directly onto the reef (as used in Page, Muller & Vaughan, 2018) to serve as controls (Figs. 1B, 1D). All fragments used as controls had healed skeletal walls covered in tissue. Control corals mounted on plugs were grouped as triads with spacing between corals similar to the pizzas, and deployed directly onto the substrate within the same plots as the experimental corals using cement. Each plot received 5 control triads (n = 20 triads, 60 corals). All corals were outplanted in August 2019 and coral condition surveys were conducted immediately after deployment and again after one and three weeks to document the presence/absence of coral fragments and evidence of tissue mortality caused by predation. The average percent tissue removal was estimated visually for each coral using 10%-classification bins. Lastly, the proportion of the tissue area covered by sediments for each coral outplant was visually evaluated at one and three weeks using the methods just described. Values for these two metrics were averaged within pizzas/triads and compared among treatments using ANOVA.

Coral cover, fish abundance, and predation surveys

The percent cover of stony corals at the three reefs selected was calculated using the point-count method as described by Lirman et al. (2007). At each reef, three plots (10 m in diameter) were haphazardly selected in the vicinity of where the coral fragments were deployed. Within each plot, 25 images were haphazardly collected at a distance of 50 cm from the bottom. The cover of stony coral was calculated using 25 random points overlaid onto each image using the Coral Point Count with Excel extension (CPCe) software (Kohler & Gill, 2006). The proportion of random points placed over stony corals was divided by the total number of points (n = 25 per image) to calculate the proportional cover of corals. Mean percent coral cover was calculated for each plot (n = 3 plots per reef) and averaged for each reef.

Fish surveys to compare fish abundance at coral outplant sites were conducted as part of experiment one at each reef site using the Reef Visual Census (RVC) method (Bohnsack & Bannerot, 1986). Using this method, the surveyor recorded the abundance of fish taxa from a stationary point at the center of the study site within a cylindrical 15-m diameter survey area, extending from the substrate to the surface of the water column. Each survey was completed in 15 min and all fishes observed were identified to species level. Between May 2018 and February 2019, we completed 13 RVC surveys at Reef 1, 10 surveys at Reef 2, and 14 surveys at Reef 3. During the 10-month monitoring period, all three reefs were surveyed within one month, with Reefs 1 and 3 surveyed opportunistically additional times. All surveys were completed by a single, expert observer. The mean abundance of all corallivorous or predatory fish (Robertson, 1970; Randall, 1974) was compared among reefs using ANOVA.

In addition to the visual fish surveys conducted during the coral deployment for experiment one, we deployed a video camera immediately after each coral deployment focused on the newly outplanted corals to document the fish species observed interacting with the corals after the divers had left the plot for both experiments. Each video was viewed and a list of species interacting with the outplants was compiled. The duration of these deployments was variable (1–7 hrs) based on the time spent by the divers at the site during deployment and was only intended to compile a list of fish species approaching and/or biting the corals.

Outplanting experiment one: assessing predation impacts

Two types of fish-predation impacts were documented: (1) physical removal of outplanted fragments, and (2) tissue removal from corals that remained attached to outplanting platforms. The probability of outplant removal was explained by coral species, reef, and time as fixed effects (GLM χ²-test, p < 0.05) and the model explained 67% of the deviance (Fig. 3, Tables S1, S2). One week after deployment, 8% of M. cavernosa (n = 53 fragments outplanted), 12% of O. faveolata (n = 123), 23% of P. clivosa (n = 80), and 27% of P. strigosa (n = 41) fragments were physically removed from the outplant platforms by fish (all sites combined) (Fig. 4A). The ranking of the probability of removal for the four species was consistent across reefs and time (Fig. 3). There was a minor, but significant increase in the probability of removal over time (Fig. 3, Table S2). The majority of removal occurred during the first week, but corals continued to be removed over time, with an additional 1.9% of M. cavernosa, 6.6% of O. faveolata, 7.1% of P. clivosa, and 7.0% of P. strigosa removed between one and six months after deployment (Fig. 4A).

The species with the highest prevalence of fish bites one week after deployment were the two Pseudodiplora species, followed by O. faveolata. M. cavernosa was the only species that did not show any signs of predation on remaining corals after one week (Fig. 4B). Similar to the rate of removal, predation prevalence slowed over time, as only an average of 0.3% of surviving corals of all four species combined showed fish bites at the one-month survey compared to 9.2% after the first week. After six months, no signs of predation were observed for surviving M. cavernosa and P. strigosa, and <1% of colonies of the remaining two species showed evidence of fish bites (Fig. 4B).

Cover of stony corals recorded at the three outplant sites was very low. Mean coral cover was 0.85 (±1.0) for Reef 1, 0.8 (±0.4) for Reef 2, and 0.04 (±0.07) for Reef 3. The prevalence of fish predation, including complete fragment removal and fish bites, was highest at Reefs 1 and 2, which coincided with the significantly greater abundance of corallivorous fish taxa recorded at these two sites compared to Reef 3 (ANOVA; Tukey–Kramer HSD test; p ≤ 0.05) (Figs. 4C, 4D). The average number of fish observed interacting with the coral outplants (i.e., parrotfishes, damselfishes, butterflyfishes, surgeonfishes, triggerfishes) was 2.7 individuals survey⁻¹ ± 6.0 (mean ± SD) at Reef 1, 2.0 ±4.7 at Reef 2, and only 0.8 ± 2.3 at Reef 3 (Fig. 4B). Complete coral removal was 17% at Reef 1, 26% at Reef 2, and only 7% at Reef 3 after one week (Fig. 4C). Similarly, signs of fish predation were higher among the remaining corals at Reef 1 (13.7% corals with evidence of predation) and Reef 2 (13.1%), while no evidence of fish bites was observed at Reef 3 after one week (Fig. 4C). The fish taxa observed biting coral fragments included butterflyfishes, parrotfishes, and damselfishes (Table S3). Wrasses and surgeonfishes were also observed approaching the coral outplants but not necessarily biting the coral tissue. While no direct evidence of predation by triggerfishes (a known coral predator) was captured, this taxon was seen in the vicinity of outplants in the video collected in experiment two. Grunts, surgeonfish, and wrasses were the most consistently sighted fish across all 3 sites and were recorded during all 37 surveys. Parrotfishes and damselfishes were also regularly observed at all three outplant locations, having been recorded as being present within 34 and 35, respectively, out of the 37 surveys conducted. Chaetodontidae were recorded within 27 of the 37 surveys, and were present during all 13 surveys at Reef 1, 8 out of 10 surveys conducted at Reef 2, but only 6 out of 14 surveys completed at Reef 3. It is important to note that no evidence of fish removing the corals from their outplanting platforms was captured in our video surveys. Nevertheless, the removal by fish was considered the only driver of the missing corals as corals did not detach during transport nor during their time at the nursery where no fish predators are observed (pers. obs.)

In addition to fragment removal and partial tissue mortality caused by predation, complete coral mortality was observed. After one week, 4% of P. clivosa, 5% of O. faveolata, 9% of M. cavernosa, and 17% of P. strigosa fragments that remained attached to the outplant platforms showed 100% tissue mortality (all sites combined). After six months, the cumulative prevalence of complete mortality was 4% for P. clivosa, 16% for M. cavernosa, 27% for P. strigosa, and 41% for O. faveolata fragments. When removal and complete tissue mortality were combined for all corals and sites combined, 26% of corals died after one week, 30% of corals died after one month, and 51% of corals died within six months of outplanting. Overall, M. cavernosa suffered 26% losses (removal + 100% tissue mortality), followed by P. clivosa (40%), O. faveolata (62%), and, finally, by P. strigosa (73%). While it was not possible to ascertain the cause of mortality (besides that visibly caused by predation) among the coral outplants, no evidence of active stony coral tissue loss disease (SCTLD), which affected the reefs of South Florida in recent years, was observed on outplanted or wild corals at any of the sites during either experiment.

Outplanting experiment two: reducing predation impacts

The mode of attachment of outplanted corals influenced removal patterns. After one week, 14% of the corals fixed to ceramic plugs using glue were removed, while none of the corals outplanted using cement within the pizzas were missing. After three weeks, still none of the corals deployed on pizzas were removed, whereas 54% of the corals outplanted using plugs were missing. While none of the corals in any of the four cement pizza treatments were removed, fish predation impacts were significantly affected by coral treatment within the cement bases (ANOVA, p < 0.05) (Fig. 5). No significant influence of plot was documented and data wer thus grouped for all plots. After three weeks, the average percentage of tissue removed by predation was significantly lowest for corals within the “embedded” treatment and highest for corals placed within the “raised exposed” treatment and corals outplanted using plugs (Tukey-Kramer HSD test; p < 0.05) (Fig. 5). No significant differences were found between corals in the “raised covered” and “flushed” treatments (Fig. 5). Predation impacts were lowest among embedded corals, but only corals within this treatment experienced sediment accumulation on the surface of the colony. For corals within the embedded treatment, the average of the total surface area of the coral outplants covered by sediments was 3.1% ± 2.4 (mean ± SD) after one week and 3.8% ± 3.5 (mean ± SD) after three weeks. Neither controls outplanted using plugs nor corals within the other three pizza treatments accumulated sediments on the coral surfaces.