Skip to main content

Differential learning by native versus invasive predators to avoid distasteful cleaning mutualists

Cite this dataset

Tuttle, Lillian; Lamb, Robert; Stringer, Allison (2021). Differential learning by native versus invasive predators to avoid distasteful cleaning mutualists [Dataset]. Dryad.


1. Cleaning symbioses on coral reefs are mutually beneficial interactions between two individuals, in which a ‘cleaner’ removes and eats parasites from the surface of a ‘client’ fish. A suite of behavioural and morphological traits of cleaners signal cooperation with co-evolved species, thus protecting the cleaner from being eaten by otherwise predatory clients. However, it is unclear whether cooperation between cleaners and predatory clients is innate or learned, and therefore whether an introduced predator might consume, cooperate with, or alter the behaviour of cleaners. 2. We explored the role of learning in cleaning symbioses by comparing the interactions of native cleaner fishes with both naïve and experienced, non-native and native fish predators. In so doing, we tested the vulnerability of the predominant cleaners on Atlantic coral reefs, cleaning gobies (Elacatinus spp.), to the recent introduction of a generalist predator, the Indo-Pacific red lionfish (Pterois volitans). 3. Naïve juveniles of both invasive (P. volitans) and native predators (Cephalopholis spp. groupers) initially attacked cleaning gobies and hyperventilated from a putative toxin on the gobies’ skin during laboratory experiments. After 1 to 5 such encounters, invasive lionfish often approached the cleaner closely, then turned away without striking. Consistent with learned avoidance, invasive lionfish rarely interacted with cleaning gobies in the wild, either antagonistically or cooperatively, and did not affect gobies’ abundance. Native predators showed little evidence of learning during early encounters; they repeatedly attacked the cleaner during laboratory experiments and hyperventilated less violently than did lionfish. However, consistent with learned cooperation, native predators rarely antagonised and were frequently cleaned by gobies in the wild. 4. We demonstrate that rapid, learned avoidance protects a distasteful cleaning mutualist from an invasive predator. The behavioural plasticity of this invader likely contributes to its success across its invaded range. Additionally, our results suggest that the cleaner’s chemical defence most likely evolved as a way to deter predation and reinforce cooperation with naïve individuals of native species.


Research Question 1: How do invasive and native predators interact with cleaners?

  1. Field

We explored how predators interact with two common cleaners on Atlantic coral reefs, the cleaner goby (Elacatinus genie, dominant cleaner on shallower reefs <12 m depth, but exists 0-30 m: Colin 1975) and the sharknose goby (E. evelynae, dominant cleaner on deeper reefs >12 m depth, but exists 0-50 m: Colin 1975).  In the Bahamas, we observed cleaning stations at natural patch reefs where both invasive lionfish and native predators were present.  A species was considered a potential predator of cleaning gobies if most of their diet is fish (Randall 1967).  From a distance of at least 3 m, in 10-minute intervals, and during daylight hours (07h00 to 20h00), SCUBA divers documented the nature (cleaning, predation, etc.) and duration of all interactions between cleaning gobies and other fishes that occurred at a focal cleaning station.  Cleaning behaviour was presumed when a cleaning goby made contact with another fish’s body for at least one second.  We conducted our observations on 4 coral patch reefs (depth: 6-20 m; surface area: 200-1300 m2) in the northern Exuma Sound in the Bahamas (24°47'10"N, 76°19'33"W) during the summer of 2013.  Observation time totaled 24 hr, 10 min, 13 sec, with observation-effort spread both across reefs (no less than 4.5 hr at a reef, with an average of 6.0 hr per reef) and within reefs (no fewer than 10 cleaning stations per reef).  This resulted in no fewer than 37 observations of cleaning at a reef (76.0±17.4, mean±SEM) at an average frequency of one cleaning interaction every 4.75 min.

  1. Laboratory

            We conducted a laboratory experiment to determine how invasive lionfish and two native groupers – graysby (Cephalopholis cruentata) and coney (C. fulva) – interact with the cleaner goby (E. genie), during the summer of 2011 at Lee Stocking Island, the Bahamas (23°46'00"N, 76°06'00"W) and Little Cayman, Cayman Islands (19°41'56"N, 80°3'38"W).  Graysby and coney are common native mesopredators, and are similar to lionfish in size and diet (Morris & Akins, 2009; Randall, 1967).  Divers used SCUBA and hand nets to capture juvenile (<15 cm total length TL) lionfish, graysby, and coney (hereafter “predators”) on nearby reefs at 3-15 m depth.  Predators were held in indoor aquaria at least 72 hours prior to their experimental trial and fed live mosquitofish (Gambusia sp.) once daily except for the 24 hours preceding each trial.

            Divers used SCUBA and handnets to capture two goby species: the cleaner goby and the bridled goby (Coryphopterus glaucofraenum).  The bridled goby is a known common prey of both lionfish (Albins & Hixon, 2008) and native groupers (Randall, 1967) and was therefore an indicator of predator hunger during lab trials. 

            To account for order of exposure to prey, we randomly assigned predators to one of two groups – (1) bridled goby then cleaner goby, or (2) cleaner goby then bridled goby – such that each predator was offered exactly two gobies less than one-third their total length (lionfish consume prey up to one-half their own length in the wild: Morris & Akins 2009).  On the day of the trial, individual predators were placed in a transparent, 208-L indoor aquarium (122 x 33 x 51 cm) and allowed to acclimate for at least 10 minutes.  A trial began when we released the first goby into the aquarium with the predator.  We observed all subsequent behaviour of the predator for 20 minutes, deemed during preliminary trials as sufficient time for a predator to detect a goby of either species.  We made observations from a distance of 2 m for lionfish, and from behind a viewing blind for grouper (observer could see the grouper, but the grouper could not see the observer) because grouper seldom hunted when people were visible.  If the first goby was eaten, then we waited 10 minutes for digestion before placing the second goby into the aquarium with the same predator.  If the first goby was uneaten at the end of 20 minutes, then we removed it from the aquarium, and replaced it with the second goby.  The trial then followed the same protocol as described above. 

            There were no instances of cleaning, so all analyses focused on predation behaviour, for which we excluded all trials in which the predator did not strike at either goby (n=11 of 42 lionfish, 9 of 32 graysby, and 18 of 30 coney).  We then calculated the proportion of trials in which a predator ate a cleaner goby and compared this proportion among predator species using Fisher’s exact tests.  We also used binary logistic regression to determine whether predation on the cleaner goby was affected by eating the bridled goby, the order of exposure to the prey species, the total lengths of the predator and the cleaner goby, and the region (Bahamas or Cayman Islands) where the trial was conducted.  A unique regression model was created for each predator species in R v3.2.1 (R Core Team, 2020).


Research Question 2: Does the interaction of invasive and native predators with cleaners change over time with repeated exposures?

            To determine whether novel and native interactions change over short time scales, we repeatedly exposed individual lionfish and graysby to a cleaner goby (a different goby each trial) over a two-week period and monitored the predators’ behaviours for any changes in response to cleaner versus bridled gobies.  This work was done at laboratory facilities in southern Eleuthera in the Bahamas (24°49'53"N, 76°19'43"W) during the summers of 2013, 2014, and 2015, and followed approximately the same protocol as described above (differences described below).  To minimise the likelihood of individual predators having prior experience with the cleaner goby and therefore maximise our ability to detect changes in the predators’ behaviours, we captured predators from small, isolated patch reefs (³100 m from nearest reef) without cleaner gobies.

            To determine if the cleaner goby is distasteful to predators, as has been previously suggested (Colin, 1975; Lettieri & Streelman, 2010), we quantified predators’ gill ventilation rates before and after goby consumption.  On the day of each trial, we allowed each predator to acclimate to the observation aquarium for at least 10 minutes, then determined their baseline gill ventilation rate by counting the number of times their gill opercula beat during a 10-second interval.  We then introduced the first goby to the tank and proceeded with the trial as described for the previous lab experiment.  Immediately after a predator ate a goby, we again determined the predator’s gill ventilation rate, and continued doing so every minute thereafter for no less than 3 minutes, and until the predator’s ventilation rate returned to its baseline level.  After completing a trial, we returned each predator to its holding aquarium where it was fasted until its next trial, approximately 48 hours later.  To allow for learning, we exposed individual predators to a cleaner goby no fewer than four times, collecting predators’ gill ventilation rates before and after goby consumption for each trial.

We conducted a separate experiment at Lee Stocking Island in 2007, during which we repeatedly exposed juvenile lionfish (n=9) to a cleaner.  We used a similar method as previously described except that lionfish were offered a cleaner goby a total of 8 times at an interval of once every 2, 4, or 6 days, with trials lasting 1 minute.

            To assess predator learning, we first excluded from analyses all trials in which the predator did not strike at either goby species, and all individuals that ate during no more than one trial (n=0 of 29 lionfish and 3 of 15 graysby).  We then divided the predators into two mutually exclusive groups: those that struck at a cleaner goby at least once, and those that did not (i.e., the predator struck at the bridled goby but never at a cleaner goby).  Of those predators that struck at a cleaner at least once, we calculated the proportion of individuals that developed an aversion to the cleaner (i.e., learned not to eat the goby).  A predator was considered to have developed an aversion if after striking at or eating a cleaner goby in an initial trial (1) it did not strike at a cleaner for 3 trials in a row, or (2) it approached a cleaner in hunting posture then turned away without striking, even if the lionfish was hungry (as demonstrated by eating a bridled goby during the same trial).  To test for learning, we used Cochran’s Q test (and when significant, p ≤ 0.05, pairwise McNemar’s tests with Bonferroni adjustments for multiple comparisons) to compare the proportions of each predator species that struck at a cleaner over time, beginning with the first trial in which a predator struck at a cleaner.

            Gill ventilation rates were quantified and compared in two ways: the number of gill opercular beats per minute upon consuming a goby, and the number of minutes after consuming a goby that it took for the predator’s gill ventilation rate to return to “normal,” defined as within 6 gill opercular beats/minute of the baseline level.  We used rank-sum tests to compare gill ventilation rates among predator species.


Research Question 3: Does the invasive predator affect wild populations of cleaners?

            To determine whether invasive lionfish affect densities of Elacatinus spp. cleaning gobies in the wild, we conducted a manipulative experiment on 8 coral patch reefs in the northern Exuma Sound in the Bahamas (24°47’10”N, 76°19’33”W) during the summer of 2013.  Reefs were surrounded by sand and seagrass and the nearest hard substrate was at least 80 m away.  We paired reefs by similarity in size (surface areas 200-1300 m2), depth (6-20 m), vertical relief, and benthic community (coral percent cover) to create 4 experimental reef pairs.  We randomly assigned one reef in each pair to have periodic lionfish removals (“low-lionfish”; approaching 0 lionfish/m2), and the other reef to have periodic lionfish additions (“high-lionfish”; about 0.04 lionfish/m2, similar to unmanipulated densities in the Bahamas, mean±SD: 0.039±0.014 lionfish/m2: Green & Côté 2009).

            Before manipulating lionfish densities, we conducted full reef censuses of the two cleaning gobies present at these reefs: the cleaner goby (E. genie) and the sharknose goby (E. evelynae).  After baseline goby censuses were complete, we manipulated lionfish densities (low vs. high) for the next 10 weeks.  We conducted full-reef censuses of cleaning gobies, lionfish, graysby, and coney 5-6 times over 10 weeks, at approximately 2-week intervals. At the end of the experiment, lionfish were removed from all experimental reefs.

            We used a linear mixed effects model (LME) to assess the effect of lionfish on changes in cleaning goby density, with lionfish treatment (low- vs. high-lionfish densities) as a categorical fixed effect, time in days (day 0 was the time of baseline census) as a continuous fixed effect, the interaction between treatment and time as a fixed effect (treatment*time), and reef (8 reefs) as a random effect (with weighted terms to allow for variance among reefs) (Bolker et al., 2009).  We first fitted models with and without (1) random effects (reef and reef pair, 4 reef pairs), (2) weighted terms to allow variance to differ among reefs, and (3) AR1 structures to allow for temporal autocorrelation within reefs, using restricted maximum likelihood estimation (REML).  We chose the best-performing model per Akaike’s Information Criterion (AIC) and p-values from likelihood ratio tests (LRTs) (see Table S1 in Supporting Information).  Residuals from the final model (full fixed effects + random effect of reef + weighted variance among reefs) indicated that all assumptions were met.  We conducted our analyses using the statistical software R v3.2.1 (R Core Team, 2020) with the associated packages nlme v3.1-118 (Pinheiro et al., 2016) and MASS v7.3-35 (Venables & Ripley, 2002).


Ethics and Permits

Oregon State University's Institutional Animal Care and Use Committee (IACUC; ACUP 3886), the Department of Marine Resources of the Bahamas, and the Marine Conservation Board of the Cayman Islands Department of the Environment all approved our work.


Methods References

Albins, M. A., & Hixon, M. A. (2008). Invasive Indo-Pacific lionfish Pterois volitans reduce recruitment of Atlantic coral-reef fishes. Marine Ecology Progress Series, 367, 233–238.

Bolker, B. M., Brooks, M. E., Clark, C. J., Geange, S. W., Poulsen, J. R., Stevens, M. H. H., & White, J.-S. S. (2009). Generalized linear mixed models: a practical guide for ecology and evolution. Trends in Ecology & Evolution, 24(3), 127–135.

Colin, P. (1975). The Neon Gobies. Neptune City, NJ: T.F.H. Publications, Inc.

Green, S. J., & Côté, I. M. (2009). Record densities of Indo-Pacific lionfish on Bahamian coral reefs. Coral Reefs, 28(1), 107–107.

Lettieri, L. B., & Streelman, J. T. (2010). Colourful stripes send mixed messages to safe and risky partners in a diffuse cleaning mutualism. Journal of Evolutionary Biology, 23(11), 2289–2299.

Morris, J. A., & Akins, J. L. (2009). Feeding ecology of invasive lionfish (Pterois volitans) in the Bahamian archipelago. Environmental Biology of Fishes, 86(3), 389–398.

Pinheiro, J., Bates, D., DebRoy, S., & Sarkar, D. (2016). nlme: Linear and Nonlinear Mixed Effects Models. R package version 3.1-128. Retrieved from

R Core Team. (2020). R: A language and environment for statistical computing. Retrieved from

Randall, J. E. (1967). Food Habits of Reef Fishes of the West Indies. Studies in Tropical Oceanography, 5, 665–847.

Venables, W. N., & Ripley, B. D. (2002). Modern Applied Statistics with S. (4th ed.). New York, New York: Springer.

Usage notes

Please see for a description of the files associated with these data and analyses.


National Science Foundation, Award: OCE 08-51162

National Science Foundation, Award: OCE 12-33027