All experimental procedures were carried out in strict accordance with the European Directive 2010/63/EU on the protection of animals used for scientific purposes. The study design was approved by the Ethics Committee in Animal Experiment of Finistère and have received authorization from the French Ministry of Education, Research and Innovation under the agreement number 2020-C2EA74-VC-01-V2. This study was carried out in compliance with the ARRIVE 2.0 guidelines.
Animals and maintenance conditions
Animals were female triploid rainbow trout (Oncorhynchus mykiss) fertilized and reared at INRAE-PEIMA fish farming experimental facility (DOI: 10.15454/1.5572329612068406E12) from a delayed autumnal strain. At 100 days post fertilization (dpf), 1872 fish (2,71 g mean weight) were randomly chosen and weighed under anaesthesia (50mg/L tricaine and 50mg/L sodium bicarbonate) before being equally distributed into 12 rearing tanks (156 fish/tank) supplied with spring water. Tanks were uniform in colour (grey opaque walls), in size (70x70 cm), in volume of water (196 L) and in water flow (one renewal per hour). All tanks were equipped at both sides with a yellow light bulb (Leds 4000 K, 9.6 W, ELVADIS) controlled by a programmer panel to provide a 12L:12D photoperiod, a 35 % light intensity, and a progressive lighting over 15 minutes. For bubble diffusion, each tank was provided with an air diffuser (4x6 mm PVC transparent air tubes attached to the end of the water inlet pipe) and two air pumps supplied six air diffusers at the same time throw a 6-valve low pressure air distributor. The valves of the dispensers were adjusted so that the bubble size was equal in each tank. Water temperature remained constant at 12±0.2°C and density was maintained below 20 kg/m3 throughout the experiment. Ammonia, nitrite and nitrate concentrations were checked at the beginning and at the end of the experiment and were always lower than 0.27, 0.00 and 35.90 mg L-1, respectively. From this stage (100 dpf), fish were fed five times a day, in equal amounts, with extruded and commercial flowing pellets (60% protein, 33% lipid, 7% starch, 1.1 mm pellets, BioMar, France) in accordance with fish growth rate. The feed was delivered by ARVOTEC feeders at precise schedules using the computer-controlled Imetronic software (version 2008). Rearing tanks were all equipped with a video-surveillance camera (Full HD: 1920 x 1080 px, 105°, VIZEO - Adrien Alarme) installed above each tank allowing continuous recordings.
Conditions of fish raised with different possibility to predict feeding
After an acclimation period of 18 days, four feeding predictability treatments were applied (three tanks per treatment containing 156 fish each) during twelve days (Fig. 7A). In each treatment, fish received each day five diffusions of bubbles (during 15 seconds each) and five feedings (total of 60 sequences of bubbles and feedings) (Fig.7B).
(i) Treatment BUBBLE: a signal of 15 seconds of bubble diffusion was systematically followed 5 seconds later by feeding (20-40 seconds). For this treatment, daily feedings were delivered each day between 9.45am and 7.20pm according to a manually randomized schedule with a minimum of 15 minutes and a maximum of three hours between two feedings to avoid any starvation or digestion issues.
(ii) Treatment TIME: the occurrence of daily feedings could be predicted by the time of the day. Each day, fish received systematically their feed at the same time schedules: 10am, 12am, 2pm, 4pm and 6pm. Fish from this condition also received bubbles five times per day, at the same time schedule as the one used for treatment BUBBLE but systematically separated from at least 15 minutes from feedings.
(iii) Treatment BUBBLE + TIME: both bubbles and time were predictive of feed delivery. Fish received daily feedings at the same time schedules as treatments TIME. Each feeding was also systematically preceded by 15 seconds of bubbles (as treatment BUBBLE).
(iv) Treatment RANDOM: neither bubbles nor time were predictive of feed delivery. Daily feedings were delivered according to the same randomized schedule as the one used for treatment BUBBLE. The bubbles were systematically delivered separated from at least 15 minutes from feedings.
Group behaviours during the 12 days of conditioning
Previous studies showed that fish activity was a relevant behaviour as an indicator of anticipation of feed delivery 10. Thus, group activity in the rearing tank was automatically analysed by EthovisionXT® software (v. 15.0.1418) (Noldus, Wageningen, The Netherlands). Activity was given as a proportion of pixels that changed between two successive images (25 images per second). Data are given as percent of activity (%). Behavioural recordings were conducted at three different feedings or bubble sequences (1st, 3rd, and 5th sequences of the day) on the 1st and the 12th days of the conditioning to examine its progression over the experiment. Depending on the period targeted, analyses were performed into the bubble area or into the whole tank (Fig. 7C).
Fish ability to use bubble diffusion as a predictor of feed delivery was assessed by analysing their activity during the 15-seconds sequence of bubble diffusion added to the 5 seconds of time gap that followed. This analysis was performed in the bubble area. Fish activity was also analysed during neutral periods (i.e., 6-minute period with no feed delivery or bubble diffusion) in the bubble area.
To assess whether fish could use time as a predictor of feed delivery, fish activity was analysed for the 6 minutes that preceded feedings (for treatments TIME and RANDOM) or bubble diffusions (for treatments BUBBLE and BUBBLE + TIME). Previous studies showed that agonistic and stereotyped behaviours are observed when fish anticipate feedings relying on time as the main predictor 27. In this aim, frequency of agonistic (bites, chases) and stereotypic behaviours (jump, high accelerations with fish that crossed more than the half of the tank with high speed) was also counted in the whole tank during the 6-minute period. An occurrence was considered to end when the fish interrupted the behaviour for at least 5 seconds. Agonistic and stereotypic behaviours were scored in the whole tank.
Feed omission tests
To assess whether fish could associate bubbles to feed delivery after 12 days of conditioning, three feed omissions were subsequently performed (no feed was delivered at the expected feeding moment): one on day 13 and two on day 14 (Fig. 7A).
During each omission, group activity was analysed during a period of time that included the 15 seconds of bubble diffusion, the 5 seconds of time gap before the expected feed and 25 seconds after the omission (totalling 45 seconds of observation). For treatments BUBBLE and BUBBLE + TIME, the last period of 25 seconds corresponds to the moment when feed was supposed to be delivered. For treatments TIME and RANDOM, the period of 25 seconds started 5 seconds after the end of bubbles. Similar to the conditioning phase, this analysis was performed in the bubble area.
As the omission of an expected reward is often characterized by an increased activity and aggressions in salmonids 22,23,41, the total number of agonistic and stereotypic behaviours (total/min) as well as fish activity (in %) in the whole tank were also scored during a 6-minute period that followed each omission. Total number of agonistic and stereotypic behaviours (total/min) was also analysed during neutral periods (i.e., 6-minute period with no feed delivery or bubble diffusion) in the whole tank on the day before the feed omissions (day 12) to detect any specificity due to feed omissions.
Individual fish emotional reactivity
The emotional reactivity of an individual animal refers to its propensity to react to emotion-provoking stimuli such as suddenness and novelty 31. Considering the emotional reactivity of an animal is therefore crucial for assessing how it perceives its environment (in a threatening way or not), its ability to cope with environmental changes, and thereby its welfare state. Fish emotional reactivity was evaluated after 12 days of conditioning on days 15 and 16 (Fig.7A), in a novel-tank test following Colson et al., (2018) 43. Ninety-six fish per treatment (24 fish/treatment and 8 fish/tank) were randomly chosen, netted, and introduced individually into a novel tank (68×33×32 cm). Fish were all fed the day before the test with 3/5 of the feed ration in one feeding so that the tested fish could experience the same fasting interval. Bubble diffusions were also stopped the day before testing. Water was changed every four tested fish. Behavioural responses were video recorded for 28 min. We analysed the first 20 min, by 5 min time-bins with the Ethovision ® XT software. At 23 min, individuals automatically received 80 feed pellets (approximately 100 mg) and the remaining pellets were counted after 5 min to measure a possible inhibition of feed intake (anorexia) following a stressful situation (novel tank exposure). Indeed, feed consumption recovery after an acute stressor is a classical stress indicator in fish 16. The following behavioural parameters were scored for each individual: total distance moved (in cm), maximum swimming velocity (in cm/s), angular velocity (in °/s) (i.e., erratic swim), and time spent (in %) in the border zone (i.e., thigmotaxis).
Weight and size of fish
Fish body weight (W) and length (L) were measured after the novel-tank test under anaesthesia (50mg/L tricaine and 50mg/L sodium bicarbonate). For each fish, the condition-factor was calculated as followed: K-factor = 100 (W/L33).
All tests and graphs were performed using the 1.4.1717 version of RStudio and were plotted using the packages ggplot2, ggthemes, and effects.
Group activity analysis
Group activity was analysed by performing linear mixed-effects models (LMER)s, with the lme4 package. Group activity (given in % by Ethovision) during the 6 minutes that preceded feed was analysed with treatments (BUBBLE+TIME, BUBBLE, TIME, RANDOM) and feedings number (feeding 1, 3 or 5), and their interaction, as fixed factors. Group activity during bubble diffusion (15s bubbles + 5s gap) was analysed with treatments and bubble diffusion number (diffusion 1, 3 or 5), and their interaction, as fixed factors. Rearing tanks were considered as random factors. The following LMER was used: lm(Mean.activity~Treatment*Trial.number + (1|Rearing tank). During feed omissions tests, group activity was analysed with treatment and omission number (1, 2 or 3), and their interaction, as fixed factors. The following LMER was used: lm(Mean.activity~Treatment*Omission.number + (1|Rearing tank). Similar to the previous model, rearing tanks were added as random factors.
The day (day 1 or day 12) was not included as a fixed factor into the models since data given by Ethovision did not account for fish growth between days 1 and 12 of the experiment, resulting in more pixel changes on day 12 compared to day 1.
All data were checked for normality before running a model and the dispersion of residuals was subsequently checked. If data did not meet the assumptions for parametric statistics, even after data transformation (inverse, log- or log10-transformation), we used a generalized linear mixed-effects models (GLMER)s instead of (LMER)s, assuming a Gamma distribution with inverse function. For each model, the effects of fixed factors on each variable were evaluated using the analysis of deviance table with ANOVA of type III. Significant effects of fixed factors and their interaction are presented in the results section.
Agonistic and stereotypic behaviours analysis
Agonistic and stereotypic behaviours occurrences over the 12 days of conditioning were analysed by using generalized linear mixed-effects models (GLMER)s considering a Poisson family with log function. Group behaviours during the 6 minutes that preceded feed and the 6 minutes during neutral periods were analysed with treatments (BUBBLE+TIME, BUBBLE, TIME, RANDOM), days (day 1, day 12), and their interaction, as fixed factors. Rearing tanks were considered as random factors. The following GLMER was used: glmer(Mean.activity~Treatment*Day + (1|Rearing tank), family=poisson(link=“log”)).
For feed omissions tests, agonistic and stereotypic behaviours occurrences were analysed for the 6 minutes following feed omissions and the 6 minutes during neutral periods with treatments (BUBBLE+TIME, BUBBLE, TIME, RANDOM), omission number (1, 2 or 3) or period number for neutral periods (1, 2 or 3), and their interaction, as fixed factors. Rearing tanks were considered as random factors.by using this GLMER: glmer(Mean.activity~Group*Omission.number + (1|Rearing tank), family=poisson(link=“log”)).
For each model, the effects of fixed factors on each variable were evaluated using the analysis of deviance table with ANOVA of type III. Significant effects of fixed factors and their interaction are presented in the results section. When significant, only the effect of the interaction of fixed factors is presented in the results section.
The analysis of the emotional reactivity consisted in testing the effect of the treatment on each variable (total distance moved, swimming velocity, angular velocity, time spent in thigmotaxis). LMERs or GLMERs considering a gamma family in case of non-normally distributed data, were used for all variables. The rearing tank was defined as a random factor. We used a chi-square test to analyse the effect of the treatment on the number of pellets eaten at the end of the novel-tank test.
Fish weight, length, and K-factor were compared between treatments using One-way ANOVA.
When significant effects were found, post-hoc analyses were performed using HSD-Tukey tests. P-values < 0.05 were considered statistically significant for all statistical analyses.