Study location and species
Field research and sample collection was conducted on the shallow reef habitat surrounding the Smithsonian’s Carrie Bow Cay Research Station, Belize (16°48’9.8316”N, 88°4’54.8148”W) between January-April 2018. Mysid swarms, identified primarily as Mysidium integrum50, are present on these reefs year-round, and this species was used in all experiments. Six damselfish species in the genus Stegastes were present at the study site. Three of these, the longfin damselfish (Stegastes diencaeus), the phylogenetically similar dusky damselfish (Stegastes adustus)31, and threespot damselfish (Stegastes planifrons), can be characterized as ‘intensive territorial grazers’ or ‘farmers’ that tend and aggressively defend the turf-algae communities on which they feed24. The others, the bicolor damselfish (Stegastes partitus), cocoa damselfish (Stegastes variabilis) and beaugregory (Stegastes leucostictus) tend turf-algae to a lesser degree and display limited territoriality. As the most common intensive-farming species, the longfin damselfish was used in all experiments.
Analytical software
Analyses were conducted using R51. Generalized linear mixed models (GLMMs) were fitted with the lme4 package52 and zero-inflated GLMMs were fitted with the glmmTMB package53. The multinomial logistic regression model used for algal analysis was performed with the packages nnet54 and effects55.
Associations between mysids and damselfish farms
To determine whether mysid swarms were associated with Stegastes farms, we conducted a series of transects. Thirty 30-m transects were laid haphazardly across the study site. For 1 m on each side of the transect we recorded: the total number of swarms, the total number of Stegastes and whether a swarm was associated with a Stegastes farm. Each Stegastes was recorded to species. We used a (chi)2 test to investigate whether the presence of intensive-farming Stegastes was associated with swarm presence.
Mysid swarm movement and site fidelity
Mysidium swarms will often leave the substrate at dusk to feed in the water column, and isotope tagging indicates that swarms regroup at the same location each morning where they remain during daylight hours29. To test whether swarms at our study site followed this pattern, and whether swarms remained associated with the same Stegastes farms, we tested for site fidelity across a 20-day period. Thirty locations within longfin damselfish farms that had mysid swarms were tagged with numbered flagging tape. Thirty longfin damselfish farms where swarms were absent were also tagged. All locations were tagged between 9 and 10 a.m. Each location was visited at day 1, 10, and 20 post-tagging, 1-h post-sunrise and 1-h post-sunset, with swarm presence or absence recorded. In addition to formal rechecking, sites were frequently reassessed throughout this period, both during the day and at night, to confirm the consistency of the patterns recorded. We used a Friedman test to investigate whether swarms exhibited site fidelity to particular farms during the day over the 20-day period. The full and final model included time as a predictor and farm identification as a blocking factor.
Mysid responses to habitat-related olfactory cues
Choice experiments were conducted to determine whether mysids used olfactory cues to actively seek out intensive-farming damselfish33. Experiments were conducted using a two-channel choice flume (13 cm length × 4 cm width)56. Header tanks contained two separate water sources, each gravity fed into separate sides of the flume at equivalent volumes (~100 mL min−1). The flume design ensures that once laminar flow is achieved each water mass remains separated on either side of the main chamber with no areas of turbulence or eddies, presenting an individual placed in the center with a choice between the two separate water sources/olfactory cues. Regular dye tests confirmed laminar flow and that the two water sources remained separated.
Five separate cue combinations were tested. Four cues were tested against a seawater control (seawater with no added cue): longfin damselfish (putative mutualism partner), farmed turf (putative mutualism partner’s environment), bicolor damselfish (non-intensive-farming damselfish that did not associate with mysids), and slippery dick wrasse (Halichoeres bivittatus, a diurnal predator of mysids). The fifth combination was a mysid-associated longfin damselfish versus a non-mysid-associated longfin damselfish. Cues were prepared by soaking an individual fish, or turf-covered rock from a longfin damselfish farm, in 10 L of seawater from the Carrie Bow Cay flow-through system for 1-h. All selected fishes and turf pieces had a similar biomass to minimize variation in cue concentration. For each trial set, both the cue and seawater control were produced concurrently, with both buckets sitting adjacent with constant aeration for the 1-h period. In this way, both cue and control had matching salinity, temperature, and O2 levels, minimizing any opportunity for behavioral bias due to the physical properties of the water and ensuring proper laminar flow. Each cue combination was split into three blocks: three separate fish or turf-covered rocks (one per block) were used to prepare treatments, with ten replicates obtained from each block (a total of n = 30 per combination). Individual mysids were only used in one trial.
All trials were conducted blind, with the tester having no knowledge of the cues tested or the side on which each cue was placed. A second observer was also present at all times. For each trial, a mysid was placed at the downstream center of the flume chamber. Following a 2-min habituation period, its position on either the left or right side of chamber was recorded at 5-s intervals for 2-min. Water sources were then switched to the opposite sides, and the chamber was allowed to flush for 1-min. The 2-min habituation period and 2-min test period were then repeated to exclude the possibility that mysids were exhibiting a side preference (i.e., spending 100% of time on one side of the chamber despite the water source being switched midway through the trial). Mysids that exhibited a side preference (n = 6 of 156) were excluded from analysis. Data were analyzed using paired t-tests, except for the longfin damselfish versus seawater comparison. Here, a Wilcoxon signed-rank test was used because these data did not meet the assumption of normality.
Effect of predation on the damselfish-mysid relationship
To first test whether mysids receive protection by residing within the boundaries of longfin damselfish farms we conducted a predation-risk experiment. Sixty trials were conducted (n = 30 inside, and n = 30 immediately outside of farms), which each consisted of three treatments: (1) live mysids, (2) an ‘imitation’ mysid control, and (3) a seawater control. Each treatment consisted of a weighted 3.5 L polyethylene bag filled with seawater. The live mysid treatment consisted of 150 mysids. The ‘imitation’ mysid control was included to account for the presence of objects within the bag and consisted of 150 1 mm long sections of 4 mm diameter silicon tubing. These lightweight sections were slightly negatively buoyant and moved within the bag due to external water movement. Finally, the seawater control consisted of an empty seawater-filled bag. For each trial, bags were sequentially placed on the substrate in random order.
Each trial was 1-min in duration, during which the focal bag was filmed using an HD video camera (GoPro). After 1-min, this bag was removed and a 1-min rest period was observed. A bag containing the next treatment was then placed in the same location, and the 1-min trial was repeated. This same procedure was then repeated for the third bag. Videos were analyzed to compare: the number of fishes that attacked each bag, the number of strikes taken, the species of the attacker(s), and the number and species of fishes that came within 1 m of the bag but did not attack.
We used a zero-inflated GLMM with a Poisson distribution to test whether the number of strikes directed by predators at bags differed according to treatment (empty bag, artificial mysids, and live mysids), and location (inside versus outside of farm). The full and final model included treatment, location, and the interaction between treatment and location as fixed effects and trial as a random effect. We used a GLMM with a Poisson distribution to test whether the number of species that directed strikes at bags differed according to treatment and location. The full model included treatment, location and the interaction between treatment and location as fixed effects and trial as a random effect; however, the interaction was removed from the final model as it was found to be non-significant. We used a GLMM with a negative-binomial distribution to test whether the number of individuals that directed strikes at bags differed according to treatment and location. The full model included treatment, location and the interaction between treatment and location as fixed effects and trial as a random effect; however, the interaction was removed from the final model as it was found to be non-significant. Finally, we used a zero-inflated GLMM with a Poisson distribution to test whether the number of chases by longfin damselfish directed towards mysid predators differed according to treatment, location and the interaction between treatment and location. The full model included treatment, location and the interaction between treatment and location as fixed effects and trial as a random effect; however, the interaction was removed from the final model after being found to be non-significant.
In addition, to test whether the persistence of naturally occurring swarms was dependent on the protection damselfish provide, we conducted a second field-based predation experiment. Thirty trials were conducted (n = 15 treatment, and n = 15 control) with replicates for both conducted in a random order. For treatment trials, a swarm within a damselfish farm was observed on SCUBA from a distance of 2 m for a 5-min period. During this time, all strikes on the swarm by predatory fishes were recorded with this number taken as the baseline predation rate. Immediately following this period, a second 5-min observation was conducted during which a second diver actively prevented damselfish from defending their territory, pressuring fish into reef structure by gesturing at them using a fiberglass pole. During this second period, all strikes on the swarm were again recorded with this number taken as the change in predation rate. Control trials accounted for the effect of the second diver’s actions on predator behavior. Control period 1 was as above; however, during the second period, the second diver made movements and noise using the pole but did not direct this at damselfish, allowing them to continue to defend their farm. All strikes on the swarm were recorded. During control trials, damselfish did not react to the second diver’s actions indicating that their territorial behavior was not affected. Differences in strikes between periods were determined using Wilcoxon signed-rank tests.
Effect of mysids on damselfish behavior
We conducted field observations to determine whether mysid-associated longfin damselfish behaved differently to those without mysids. Adult longfin damselfish that were (n = 30), or were not (n = 30), associated with swarms were observed for 30-min. For each observation, the focal fish was observed on SCUBA from a distance of at least 2-m. During each observation we recorded the number of bites on the farmed substrate, the number of strikes directed towards the swarms, the number of chases directed towards intruding fishes, the number of chases directed towards intruding fishes attempting to feed on farm-associated mysids and the number of non-aggressive interactions between the focal fish and the swarm. At the end of each observation, we recorded the total number of longfin damselfish associated with each farm with only one observation made per farm. An estimate of farm area was also made at this point by using a transect tape to measure the maximum length and width across the area that was actively defended and tended during the observation period.
We used a GLM with a Gaussian distribution to determine whether the number of chases by longfin damselfish was associated with the presence or absence of mysids. The full model included farm type (mysids present or absent), longfin damselfish group size, and the interaction between farm type and group size as fixed effects; however, the interaction between farm type and group size, and group size were removed from the final model as they were found to be non-significant. A GLM with a Gaussian distribution was used to test whether the number of bites on farmed substrate by focal longfin damselfish was associated with the presence or absence of mysids. The full model included farm type (mysids present or absent) longfin damselfish group size and the interaction between farm type and group size as fixed effects; however, the interaction between farm type and group size, and group size were removed from the final model because they were found to be non-significant. Whether farm area differed between farms with and without mysids was determined using a Wilcoxon rank sum test.
Effect of mysid swarms on longfin damselfish body condition
To determine the effect of mysid presence on longfin damselfish body condition, we compared the hepatosomatic index (HSI) of damselfish with and without mysids in their farms. This measure can reflect the amount of stored energy in the liver, and thus it can indicate of the relationship between diet and physical condition in damselfishes57,58,59,60. Thirty adult longfin damselfish (75–100 mm TL) were sampled from farms with or without associated mysids. Damselfish were collected on snorkel using hand nets and a 1:3:7 clove oil/ethanol/seawater mixture. Prior to euthanasia, fish were maintained in a 20 L flow-through aquaria for 24-h. Fish were not fed during this period. Fish were euthanized by immersion in a clove oil/ethanol/seawater solution to induce anesthesia followed by immersion in an ice slurry. Once euthanized, fish were measured (SL and TL) and weighed. The liver of each fish was removed and weighed, and the alimentary canal checked to confirm that all digested matter was evacuated. The HSI of each fish was calculated as the proportion of total weight contributed by the liver [(liver weight (g)/total weight (g)) × 100]. We used a GLM with a Gaussian distribution to test the effect of mysid presence on hepatosomatic index (HSI). The full model included damselfish length, farm type (mysids present or absent) and the interaction between damselfish length and farm type as fixed effects; however, the interaction and damselfish length were removed from the final model as they were found to be non-significant.
Effect of mysid swarms on algal composition within damselfish farms
Algal composition was assessed to determine the effect of mysid swarms on algae within longfin damselfish farms. Sixty farms were analyzed: 30 with and 30 without swarms. Three 20 cm × 20 cm quadrats were placed haphazardly within each farm, and a series of four photographs were taken, including one overhead shot encompassing the entire quadrat and three macro shots of the algae within the quadrat. Within each quadrat, algal composition and coverage was determined to phylum, including Chlorophyta, Rhodophyta, and Ochrophyta. Percent cover of these groups was assigned based on a categorical classification scheme: low (<10% coverage); medium (10–30% coverage); and high (>30% coverage). To obtain a single assessment of algal composition for each phylum within each farm, the categorical classification was averaged across the three quadrats photographed in each farm. We used a multinomial logistic regression model to assess how the percent cover of Ochrophyta within farms was affected by swarm presence. The response within each model was multinomial: low, medium, or high percent coverage of Ochrophyta. The model included the fixed effect of mysid presence or absence.
Estimates of mysid swarm density
Surveys were conducted to determine the average size and density of farm-associated swarms. The size and area of 30 focal swarms were determined by measuring the length, width and height across the widest points when first observed. Each swarm was then collected using hand nets and returned to the laboratory where the total number of mysids within each swarm was counted. Finally, swarm density was estimated by calculating the maximum ellipsoid volume based on the measured axes and dividing this volume by the total number of mysids, giving an estimate of mysids mL−1.
Mysid waste excretion and nutrient availability
We used an aquarium experiment to determine if mysid swarms produce key nutrients at concentrations that could enhance benthic algal growth. Artificial seawater was produced at sunrise by mixing deionized fresh water with a phosphate and nitrogen-free aquarium salt (Instant Ocean® Sea Salt) to 35ppt salinity. Once mixed, the absence of phosphate and ammonia was confirmed through color comparison using laboratory-grade test kits (Hach PO-19A test kit, Hach NI-SA test kit), and the temperature and pH of the water were also measured. Seawater was then distributed into a series of 1 L plastic containers, along with one air stone per container. Containers were covered to prevent water loss through evaporation.
Mysids were collected using hand nets and returned to the lab where they were allowed to habituate for 30-min in a bucket containing 3 L of artificial seawater. For sorting, mysids were removed from the habituation bucket using a hand net and placed into a petri dish containing artificial seawater. Mysids were individually selected using a sterile plastic pipette and moved into a second petri dish containing artificial seawater. The water in the second petri dish was then removed using the pipette, and mysids were placed into the appropriate container corresponding to one of three treatments: 0 mysids L−1 (n = 30), which served as a control, 100 mysids L−1 (n = 30) and 200 mysids L−1 (n = 30). These densities were selected as they are representative of the range found during the swarm density surveys. Each day, containers were haphazardly assigned a treatment prior to sorting, with an equal number of replicates for each treatment (n = 5) conducted each day. Trials were conducted from 08:00 to 16:00. This period was selected to represent the daylight hours when mysids are present, but was also short enough to prevent stress due to nutrient accumulation. Following this 8-h period, mysids were removed, and the concentration of phosphorous (P) and nitrogen-ammonia (NH3–N) in each container were recorded to the nearest 0.1 mgL−1 through color comparison. The temperature, salinity, and pH of each container were also recorded at this point.
Reporting summary
Further information on research design is available in the Nature Research Reporting Summary linked to this article.
Source: Ecology - nature.com