Study system
The cuckoo catfish (Synodontis multipunctatus) belongs to the African catfish family Mochokidae. The genus Synodontis, with 131 species distributed across African freshwaters57, gave rise to a small radiation in Lake Tanganyika, with 10 described endemic species58. The taxonomy of the group is not well established59 and we use the name S. multipunctatus as this species is confirmed as a brood parasite30 and the name was used in previous studies4,30,32,37,42. Cuckoo catfish primarily parasitise mouthbrooding cichlids from the tribe Tropheini30, but species from other lineages can also be parasitised59.
Experimental design
All experiments took place between January and August 2020 at the Institute of Vertebrate Biology, Czech Republic. Prior to experimental use, fish were housed in mixed-sex groups in tanks equipped with shelter and internal filtration. Cuckoo catfish were F1 generation of commercially imported wild-caught parents (10 pairs). Host cichlids were descendant of wild fish imported from Kalambo, Zambia. Experimental tanks (420 L; length 150 cm, depth 70 cm, height 40 cm) were equipped with internal filtration, fine gravel (2–4 mm diameter), half a clay pot as a shelter on each side of the tank, and one artificial plant in the centre of each tank. Water temperature was maintained at 27 °C (±1 °C) and the dark – light regime was set to 11 h:13 h. In total, we stocked 18 tanks with 4 males and 12 females of the mouthbrooding cichlid Astatotilapia burtoni and introduced 3 cuckoo catfish pairs of one of three different experience levels. Naïve catfish (n = 36 individuals) had no prior experience with cichlids. Experienced catfish (n = 36) were housed together with reproductive cichlids for 12 months prior to the experiment and were age-matched to naïve catfish (5 years old). Highly experienced catfish (n = 36) were raised, coexisted and reproduced with cichlids for 7 years (and were on average 7–8% larger than both naïve and experienced catfish; mean ± SE, naïve: 116.2 ± 1.9 mm, experienced: 117.1 ± 1.5 mm, highly experienced: 125.6 ± 1.4 mm; Linear Model (LM): experienced vs. highly experienced, estimate ± S.E = 8.44 ± 2.29, t = 3.68, P = 0.0004, experienced vs. naïve, estimate ± S.E = −0.94 ± 2.29, t = −0.41, P = 0.681, n = 108). Additionally, both naïve and experienced cuckoo catfish were bred using in-vitro fertilisation32 to avoid cichlid imprinting (i.e., priming with cichlid cues), while highly experienced catfish were bred under natural conditions within the buccal cavities of their hosts. Each experimental tank contained catfish with the same experience level. Due to space limitations, we split the experiment into two consecutive phases with 3 replicate tanks for each treatment within both phases (in total 9 experimental tanks per phase). Between the two experimental phases, host cichlids were placed together and haphazardly assigned to new experimental tanks. During the second phase, we removed some cichlids from the tanks because of injuries caused by their intraspecific aggression (3 males and 3 females in total), and those hosts were not replaced. Over an experimental phase, cuckoo catfish and cichlids freely interacted for 15–16 weeks. During this period, each tank was checked for mouthbrooding cichlids twice each week (Tuesday and Friday). We caught the mouthbrooding females, gently washed the eggs out of their mouths using a jet of water from a Pasteur pipette, measured their body size to the nearest mm, and released them back to their experimental tank. For each female, we counted the number of cichlid eggs and cuckoo catfish eggs (if present). At the end of each experimental phase, we measured body size of all cuckoo catfish to the nearest mm. There was no significant difference between the number of cichlid spawnings between naïve and experienced catfish treatments (Generalised Linear Models with negative binomial error distribution, estimate ± S.E.: −0.093 ± 0.145, z = −0.644, P = 0.519), nor between naïve and highly experienced catfish (estimate ± S.E.: −0.269 ± 0.148, z = −1.810, P = 0.070).
Behavioural recording
Over the experimental period, we successfully recorded 18 videos of spawning events (Lamax x3.1 ATLAS cameras; naïve catfish treatment, n = 9; experienced catfish treatment, n = 6; highly experienced catfish treatment, n = 3). One camera was placed near the spawning site approximately 20 cm away from spawning activity and a second camera was placed outside the experimental tank to obtain an overall view. Nine spawnings were recorded from the start (n = 7 naïve catfish experiments and 2 experienced catfish experiments) and nine spawnings were recorded from the timepoint when we recognised ongoing spawning activity (n = 2 naïve, 4 experienced, and 3 highly experienced catfish experiments). From the video footage taken for each spawning, we scored all overt aggression that host cichlids directed towards cuckoo catfish, counted the number of intruding catfish during each distinct cichlid spawning behaviour (i.e., male and female cichlid interact in a repeated succession of quivering and T-positions), measured the delay of intruding catfish to each distinct spawning behaviour (i.e., the time from the start of spawning behaviour until the first catfish directly approaches the spawning cichlids), and recorded the presence or absence of catfish during each spawning behaviour. Additionally, we recorded whether cichlids used the available shelters for spawning as this might have impeded catfish recognition of the spawning activity. When spawning was recorded from the start, scoring started 100 s before we detected the first egg laid (cichlid or cuckoo catfish). When spawning was already ongoing, the scoring started immediately after the cameras were in place. Mounting of the cameras did not interrupt the normal behaviour of cichlids or catfish. For all video footage, scoring ended 100 s after the last male-female interaction within the spawning site. To estimate the duration of male T-positions during spawnings, we measured the time period from the start of male nuzzling near female genital papilla until the female turned around either to collect eggs or start nuzzling near the male´s genital papilla (n = 115 male T-positions from 12 cichlid spawnings).
Statistical analysis
We used R v. 3.5.1 (R Development Core Team, 2018) for all statistical analyses. All statistical tests were two-sided. First, we compared body size among the three cuckoo catfish experience levels using a Linear Model with catfish size (mm) as response variable and ‘treatment’ (naïve, experienced, and highly experienced catfish) as predictor variable. Second, we formally tested whether the number of host spawnings varied between the treatment groups (total numbers: naïve = 191 spawnings, experienced = 174 spawnings, highly experienced = 146 spawnings). To obtain an insight into temporal dynamics of cichlid spawning, we calculated the number of cichlid spawnings for each treatment in each quarter of the duration of the experimental period. We fitted a GLM with a negative binomial error distribution (to account for slightly overdispersed data) with the number of cichlid spawnings as the response variable and our treatment groups as predictors.
To test how experience with host spawning (treatment) affected cuckoo catfish ability to place their eggs in the care of the host, we compared (1) the number of parasitised cichlid clutches among the three catfish experience groups (prevalence of parasitism), (2) the mean number of catfish eggs introduced into cichlid clutches among the three treatment levels (mean parasite egg abundance, the mean number of catfish eggs calculated across all cichlid broods, (3) mean parasite clutch size (the number of catfish eggs calculated only across parasitised cichlid broods), and examined (4) temporal dynamics of all three measures of parasite success within each treatment group throughout the duration of the experiment.
To test for differences in prevalence of parasitism among different cuckoo catfish experience treatments, we applied a Generalised Linear Mixed-effects Model (GLMM, R package glmmTMB)60 with a binomial error distribution. We fitted the occurrence of ‘catfish parasitism’ (1 = yes, 0 = no) as the binary response variable and ‘treatment effect’ (i.e., ‘catfish experience’), ‘time progress of experiment’ (1–113 days) and ‘host female body size’ (in mm) as predictor variables. We additionally fitted an interaction between treatment (‘catfish experience’) and ‘time progress of experiment’ to the model to test whether parasitism rate changed over time at treatment-specific rates. We included tank identity (‘tank ID’) as a random intercept to account for nonindependence of data obtained from the same tank.
Next, we tested whether the mean number of parasite eggs that were accepted by host females during one spawning bout differed between catfish experience treatments. We applied two GLMMs (R package glmmTMB)60 with a negative binomial error distribution (i.e., nbinom1) to account for over-dispersed count data. We applied GLMMs on the mean abundance of catfish eggs (across all host clutches) and on mean clutch size of cuckoo catfish using a subset of clutches that were parasitised. For both GLMMs, we included the ‘number of cuckoo catfish eggs per clutch’ as the response variable and treatment (‘catfish experience’), ‘time progress of experiment’, and their interaction as predictor variables. We additionally fitted ‘host female body size’ as a predictor variable because larger female cichlids are capable of laying more eggs and may appear more attractive hosts to cuckoo catfish. Further, a higher number of host eggs may increase the number of opportunities for cuckoo catfish to deposit their own eggs in the host clutch. ‘Tank ID’ was included as random intercept to account for nonindependence of data.
To test whether cuckoo catfish presence affected cichlid spawning activity, we applied a GLMM (R package glmmTMB)60 with Gaussian error distribution (which provided superior model fit compared to Poisson and negative binomial distributions by ‘simulateResiduals’ and ‘testDispersion’ functions in the R package DHARMa)61. We fitted the ‘number of host eggs’ per clutch as the response variable and treatment (‘catfish experience’), ‘host female body size’, ‘time progress of experiment’, and ‘experimental phase’ (1st or 2nd phase) as predictor variables. We also included ‘tank ID’ as random intercept to account for nonindependence of data. The full model further included an interaction between treatment and ‘time progress of experiment’ to accommodate the possibility that host egg numbers may be affected differently across catfish experience treatments over time. As this full model predicted no difference in temporal aspect of host clutch size among treatments (‘catfish experience’: ‘time progress’, experienced: z = 0.92, P = 0.360, highly experienced: z = 1.46, P = 0.143), we subsequently dropped the interaction term from the final model.
We used data collected from video footage to investigate whether naïve, experienced and highly experienced cuckoo catfish differed in their response to host spawnings and, additionally, if catfish from the three treatments elicited different host reactions towards them by applying Linear Mixed-effect Models using the R packages lme462 and glmmTMB60. To account for different starting times of recordings, we calculated either the rate of behaviour per minute of observation (i.e., for aggression) or their relative values (i.e., for the number of host courtships that cuckoo catfish missed).
First, we tested whether host spawning pairs increased their aggressions towards cuckoo catfish over the experimental period to rule out the presence of host adaptation to cuckoo catfish intrusions, which would interfere with our aim of understanding parasite learning. We fitted a Generalised Linear Mixed-effects Model (GLMM, R package glmmTMB) with a negative binomial error distribution. The number of overt aggressive behaviours that the spawning pair performed towards cuckoo catfish per minute of catfish presence at the spawning site (summed over male and female cichlid) was fitted as the response variable and treatment (‘catfish experience’) as the predictor variable. We further included ‘time progress of experiment’ and ‘experimental phase’ as predictors to account for their possible effect on host aggression. We additionally included ‘tank ID’ as random intercept in the model to account for individual variation in host aggression levels among experimental tanks.
To investigate if naïve cuckoo catfish missed more opportunities to parasitise cichlids than experienced and highly experienced catfish, we fitted a GLMM (R package lme4) with a binomial error distribution. We included the proportion of missed spawning behaviours (coded as ‘missed spawnings behaviours’ versus ‘intruded spawning behaviours’, based on count data for each spawning) as the response variable (‘spawnings missed’) and treatment (‘catfish experience’) as a predictor variable. We fitted ‘tank ID’ as a random intercept to the model to account for nonindependence of data within tanks, and we additionally fitted a random intercept based on whether the spawning was covered by a shelter or not (‘sheltered spawn’, yes / no) since spawning in a shelter may have been less apparent to catfish.
We tested whether cuckoo catfish experience played a role in the timing of their intrusion to specific spawning behaviours by fitting a GLMM (R package lme4) with a Gamma error distribution to account for a positive skew in the data distribution. We included the mean delay of catfish to the first appearance of cichlid T-position in seconds (‘catfish delay’, see main text and Supplementary Movie 1 for a detailed description of cichlid spawning sequence) as the response variable and ‘catfish experience’ as the predictor variable. We included ‘tank ID’ and ‘sheltered spawn’ as random intercepts.
Finally, we fitted a GLMM with a Poisson error distribution to test whether cuckoo catfish learn to synchronise their intrusion behaviour as they gain experience through interactions with their hosts. We included the maximum number of catfish during a specific cichlid spawning behaviour (‘intruder number’, count data) as the response variable and ‘catfish experience’ as the predictor variable. To account for nonindependence of data within experimental tanks and spawnings, we included a random intercept where each spawning was nested within ‘tank ID’ in the model.
Ethical compliance
Research adhered to all national and institutional animal care and use guidelines, was administered under permit No. CZ62760203 and was approved by ethical boards of the Institute of Vertebrate Biology and the Czech Academy of Sciences (approval No. 32-2019).
Reporting summary
Further information on research design is available in the Nature Research Reporting Summary linked to this article.
Source: Ecology - nature.com
