DataSet for "Path integration guides homing in a coral reef fish"

We provide all raw data and code used in "Path integration guides homing in a coral reef fish" . Please read the README.txt file for a full description of the codes and datasets, including column information. See below for the methods used to collect the raw data and the statistical analyses description. Methods Overview We tested whether domino damsel fish can navigate back to a shelter using path integration and whether idiothetic path integration or landmark cues were favoured when the two cues gave conflicting information. We first allowed individuals to acclimatise overnight to a large circular experimental apparatus (i.e. pool) with two landmarks and a shelter. The next morning, we placed a trap with food rewards in the centre of the pool and a spotlight outside the curtain surrounding the pool. Fish were forced to leave their shelter in order to reach the food reward in the centre of the apparatus. Once trapped, we either (1) removed the landmarks and shelter from the pool; (2) removed the shelter and kept the landmarks; (3) removed the shelter and shifted the landmarks position by 90 degrees; (4 and 5) repeated tests 1 and 2 with an additional shift to the spotlight. Fish were then released from the trap, and free navigation movements were recorded and analysed. Ethics This study was reviewed and approved by the AWERB ethical approval (Ref. No. APA/1/5/ZOO/NASPA/Burt/Turbidity). Animal Husbandry We used 19 domino damselfish (Dascyllus trimaculatus), sourced from national suppliers (Tropical Marine Centre, Hertfordshire, WD3 5SX and The Goldfish Bowl, Oxford OX4 1RQ) in experiments. Individuals were wild-caught and kept by the supplier for a minimum of one month before being housed in the laboratory. We housed these territorial individuals in separate 0.35 m x 0. 32 m x 0.60 m tanks within a single flow-through marine system. We enriched each tank with 1 cm depth of coral gravel substrate, a large rock, a cave shelter (13 cm x 12 cm x 8 cm Aqua One Marble Cave), and a 30 cm tall plastic plant. The fish were kept on a 12-hour light/12-hour dark cycle with fluorescent light, and fed with crisp (TetraPRO Energy Multi-Crisps) in the morning and mysis shrimp in the afternoon to provide supplementary nutrients. We maintained water parameters at healthy levels for this species (Temperature: 26.0 °C, Salinity-specific gravity: 1.024, pH: 8.2, KH: 7-8 dKH, Nitrite: 0 ppm, Ammonia: 0 ppm, Nitrate: <20 ppm) by carrying out water tests, water changes, and tank cleans once per week. Room temperature, water temperature, and individual health were checked at each feed. We quarantined all individuals in this system for 4 weeks before starting experiments. The laboratory space allowed us to keep and maintain 10 fish at a time. The first 10 fish underwent the first test (Path Integration test), and only six participated in the experiment by entering the trap. Additional tests (e.g. tests 2 to 5) were undergone by the subsequent nine fish. In this second group of 10 fish, the 10th fish was kept as a sentinel in its home tank and was not tested in the experiment (Ethical requirement by the University of Oxford). Experimental Apparatus The experiments were run in a 2 m-diameter blue plastic circular pool, which was filled with 10 cm depth coral gravel substrate and 20 cm depth water (from water surface to top of substrate). We kept pool water at identical parameters to the home tank system and carried out weekly water changes and tests. To help maintain these parameters, a water heater and a filter pump were placed in the pool outside of the experiments. During the acclimation period and in experiments, we placed two 14.0 cm x 14.6 cm x 18.1 cm polyurethane resin replica green brain coral landmarks and a shelter (14 cm x 11 cm x 8 cm Aqua One Marble Cave) in the pool. To reduce the shelter’s saliency, we covered it with coral gravel substrate using fish-safe silicone. Indeed, it was crucial for the shelter to be inconspicuous to prevent fish from remembering it as a salient landmark or beacon. In addition to covering the shelter with substrate, the shelter was slightly buried (i.e. positioned below the substrate level), forcing the fish to search for it when exploring the experimental pool. To reduce the likelihood of individuals using global visual cues from the laboratory room, we surrounded the pool with white curtains (side and top). To record footage during acclimation and experiments, we positioned a GoPro Hero 8 Black 2 m above the centre of the pool (recording at 1440p, 30 fps, Linear FOV). To provide a low-latency live video feed of individual fish location, we positioned a USB webcam next to the GoPro and connected it to a laptop outside of the curtains. This second camera allowed for monitoring fish movement in real time for trapping. Acclimation To allow fish to acclimatise to the pool overnight, we moved individuals from their home tank to the experimental pool in the afternoon the day before an experimental test. We ensured that water temperature, salinity, and pH conditions were identical just before moving an individual into the pool to acclimatise. Once moved, we fed the fish and recorded the following hour of the acclimation period, allowing us to calculate an exploratory score for each individual. Experimental Setup Before beginning an experimental trial, we recorded the pool water temperature. To eliminate confounding navigational cues within the pool, we removed the water heater and filter pump for the duration of the experimental trial. To provide a directional cue analogous to the sun, we positioned a spotlight outside the experimental enclosure at a height of 128 cm and a horizontal distance 48 cm away from the pool, facing the curtains with a 20° negative inclination. This created a localised bright spot that the fish could use as a directional cue or heading indicator for path integration (Goodyear & Ferguson, 1969; Guilford & Taylor, 2014). We then placed the trap at the centre of the pool. The trap consisted of a 12.5 cm diameter by 25.3 cm height white opaque plastic cylinder with a 6.4 cm diameter entrance hole cut into one side 1.6 cm from the bottom. We used the trap to contain the fish during the trial and prevent it from directly observing any experimental manipulations. To get the trap to appear more like a natural shelter, we covered the exterior and interior with the coral gravel substrate. A 17.5 cm by 7.5 cm rectangular white opaque plastic door could be opened or closed vertically by sliding along plastic rails fixed behind the trap’s entrance hole. We were able to operate the door and remove the trap remotely and outside the experimental apparatus by using fishing lines attached via a pulley system. To provide additional enrichment and encourage the fish to explore the central area of the pool, we placed a plastic plant by the trap door. We then started each trial by setting the GoPro to record, adding mysis shrimp inside the trap, and remotely raising the door. Experimental Procedure We recorded the fish movements in each trial for up to two hours, and ended experimental trials if individuals did not enter the trap within this period. As soon as we observed fish entering the trap on the live low-latency USB camera feed, we remotely closed the trap door. We then removed the shelter and the plastic plant. Additionally, we removed or shifted the landmarks and the spotlight depending on the experimental test run : 1) The landmarks were removed and spotlight stayed at a fixed position (test 1 “PI”, n=15 fish). 2) The landmarks and the spotlight stayed at a fixed position (test 2“LF”, n=9 fish). 3) The landmarks were shifted 90 degrees clockwise and the spotlight stayed at a fixed position (test 3 “LS”, n=9 fish). 4) The landmarks were removed and the spotlight was shifted 130 degrees clockwise (test 4 “LightM+PI”, n=8 fish). 5) The landmarks stayed at a fixed position and the spotlight was shifted 130 degrees counter clockwise (test 5 “LightM+LS”, n=8 fish). To prevent the fish from using chemical cues from the water or visual patterns in the substrate to identify the shelter location, we also mixed and smoothed out any disturbed substrate and water in the experimental pool. These manipulations took less than one minute. Finally, we released the fish by lifting the trap vertically out of the water using the pulley system. We recorded five minutes of fish free movement before ending the trial. The first six fish only performed test 1 “PI” and were returned directly to their home tank afterwards. The other nine fish underwent five tests. In this case, the shelter, landmarks, pump and heater were placed back in the experimental tank, and the fish was left alone until another test started either in the afternoon (minimum 3 hours between tests) or the following day. The first three tests (i.e. “PI”-“LF”-“LS”) were run in a Latin square order (1-2-3; 2-3-1; 3-1-2) and two additional tests, where the spotlight was shifted, were run after the first three tests. Fish Tracking and Trajectory Analyses Experiments were run between November 2023 and May 2024. Data were collected from 55 successful trials (i.e. the fish entered the trap), out of 94 experimental tests performed. Fish movements during the acclimation and pre-trapping part of the experimental trial were tracked using AnimalTA (Chiara & Kim, 2023). Fish movements after release from the trap were recorded using a custom MATLAB code (MathWorks, R2023b), where the fish coordinates were extracted every second for one minute. This latter method prevented interferences due to the trap and water movement after release. The shelter coordinate was extracted for each trial. We focused on and analysed three navigational periods in detail: Acclimation: To characterise fish exploratory behaviour during the one-hour acclimation period, we calculated average speed (cm/s), total distance travelled (cm), and area of the pool explored (cm2) in AnimalTA. We also recorded exploration latency (the first time at which each fish exited the shelter and became entirely visible on camera) and measured the greatest distance it swam from its shelter (cm). Each fish was given an exploratory index calculated by multiplying normalised versions of the above metrics. Last outward trajectory: To characterise fish behaviour during the experiment before the fish was trapped, we analysed fish movement during the last outward trajectory. This is defined as the period between the fish leaving its shelter for the last time during the trial and it being trapped in the centre. We calculated the length (cm) and time taken (s) for the last outward trajectories in AnimalTA. To obtain measures of the tortuosity of the last outward trajectories, we calculated their straightness indices and sinuosities in R using the ‘trajr’ package (version 1.5.1; McLean & Skowron Volponi, 2018). The straightness index is defined as D/L, where D is the beeline distance between the first and last points in the trajectory and L is the path length travelled (Batschelet, 1981). Sinuosity, defined by Benhanou (2004), uses the average step length, the average cosine of turning angles, and the coefficient of variation of step length (standard deviation/average) to calculate a measure of tortuosity. Homeward path: To characterise fish behaviour after being released from the trap, we analysed the extracted fish coordinates using the ‘trajr’ package in R (McLean & Skowron Volponi, 2018). We used the above metrics of straightness index and sinuosity to determine the transition point between the homeward path ending and the initiation of search behaviour. We identified this transition through visual inspection of the straightness index and sinuosity over time. We defined the end of the homeward path based on when two sequential coordinates showed a breakdown in straightness index or increase in sinuosity after a plateau. Once the homeward paths were identified, we calculated the homeward path direction (angle between first and last homeward path coordinates, degrees), beeline distance (straight line distance between first and last homeward path coordinates, cm), length (cm), time (s), average speed (cm/s), straightness index, and sinuosity. We then used the homeward path direction to calculate homeward direction error (absolute angular difference between homeward path direction and shelter direction, degrees). We also calculated the homeward beeline distance error (absolute difference between shelter distance and homeward beeline distance, cm) and the homeward path length error (absolute difference between shelter distance and homeward path length, cm). Statistical analysis All statistics were carried out in R (version 4.2.2; R Core Team, 2022, R studio 2022.12.0). We used circular statistics to analyse direction data. The significance threshold α of tests was set at 0.05. For each experimental test (1 to 5), we performed the following analyses (unless mentioned otherwise): 1 – Homeward path direction analysis We conducted a “Rayleigh test of uniformity” to determine if the distribution of homeward path directions differed significantly from a uniform distribution (i.e. homogeneously spread over 360 degrees). If so, we used a model-based approach with maximum likelihood with the R package ‘CircMLE’ (version 0.3.0; Fitak & Johnsen, 2017) to investigate whether fish homeward path directions followed a unimodal or multimodal distribution. The fish direction data were converted to circular class data using the package ‘circular’ (version 0.5.0; Lund et al., 2017). These directions were then compared to ten direction models, and the best-fitted model was estimated using AIC values (for more details, see Fitak & Johnsen, 2017). 2 – Homeward path direction relative to shelter We analysed whether homeward path directions differed significantly from the direction to the shelter, for tests with unimodal homeward paths (tests 1 to 3). First, we conducted a Watson’s two-sample test of homogeneity between the distribution of homeward path directions and a simulated distribution centred on the average shelter location with the same concentration parameter as the homeward path directions. The concentration parameter used (e.g., 2.582 for test 1 “PI”) was estimated during the maximum likelihood analysis with ‘CircMLE’. Second, we calculated a circular 95% confidence interval around the average homeward path direction to see if this contained the shelter direction. The average direction and confidence intervals were calculated using the R package ‘circular’ (Lund et al., 2017). The confidence intervals were calculated by bootstrapping with replacement over 10000 iterations for the parameters of a von Mises distribution (considering average direction and concentration parameter). 3 – Effect of last outward trajectory behaviour on homeward path behaviour Next, we tested how homeward path behaviour was influenced by last outward trajectory behaviour by fitting a series of linear mixed models using the R package 'lme4' (Bates et al., 2015). We tested whether the last outward trajectory length and straightness index (fixed effects) influenced homeward path direction error (response variable), as expected due to error accumulation during path integration. As fish undertook multiple tests, the order of the test was added as a fixed but also as a random intercept (multiple fish were tested in the same order) in initial models. We tested models with different random intercepts and without non-significant fixed effects. Model selection was based on AIC values and we used the R package DHARMa for model validation. 4 - Homeward path distance relative to shelter We then analysed whether the homeward path beeline distance or homeward path length differed significantly from the distance to the shelter. In test 1 (sample size =15) the three metrics were normally distributed (Shapiro-Wilk normality test for the three metrics: homeward beeline_ homeward length _ distance to shelter: Test 1= W=0.89, P=0.09 _ W=0.93, P=0.25 _ W=0.97, P=0.85) and the homoscedasticity of variance was verified (Brown_Forsythe Test: F = 2.213, DF = 2, 23, P=0.132), we conducted a Welch Two Sample t-test between homeward path beeline distances and shelter distance and between homeward path lengths and shelter distance. We conducted a Wilcoxon rank sum exact test between the variables for the fish tested in the conditions 2, 3, 4 and 5 as the sample sizes were small (<10), and some of the data did not meet homoscedasticity. 5 - Effect of acclimation on homeward path behaviour We then tested how homeward path behaviour was influenced by acclimation behaviour by fitting a series of linear models _ in test 1 “PI” only . We tested whether a selection of biologically relevant acclimation metrics set as explanatory variables (acclimation distance travelled, area explored, and exploratory index) had any influence on homeward path behaviour (response variables included: homeward direction error, homeward beeline distance error, homeward path length error, total time, average speed, straightness index, and sinuosity). Model selection of random and fixed effects was based on AIC, and the results of the best models were reported. 6 – The use of Path Integration versus landmarks in Domino Damsel navigation We explored whether the shift in landmark cues had a significant effect on the fish orientation behaviour using a Watson’s two-sample test of homogeneity between test 1 “PI” and test 3 “LS” homeward path directions. 7 – Path integration and the use of a heading indicator The use of the spotlight as a compass during the Path integration test was verified by measuring the shift in fish orientation after the spotlight was shifted (Test 1 “PI” versus Test 4 “LightM+PI”). We measured a circular 95% confidence interval around the average homeward path direction using the package ‘circular’ (Lund et al., 2017). References Bates, D., Mächler, M., Bolker, B., & Walker, S. (2015). Fitting linear mixed-effects models using lme4. Journal of statistical software, 67, 1-48. Batschelet, E. (1981). Circular statistics in biology. Academic Press. Benhamou, S. (2004). How to reliably estimate the tortuosity of an animal’s path: Journal of Theoretical Biology, 229(2), 209–220. https://doi.org/10.1016/j.jtbi.2004.03.016 Chiara, V., & Kim, S. Y. (2023). AnimalTA: A highly flexible and easy-to-use program for tracking and analysing animal movement in different environments. Methods in Ecology and Evolution, 14(7), 1699–1707. https://doi.org/10.1111/2041-210X.14115 Fitak, R. R., & Johnsen, S. (2017). Bringing the analysis of animal orientation data full circle: model-based approaches with maximum likelihood. Journal of experimental biology, 220(21), 3878-3882. https://doi.org/10.1242/jeb.167056 Goodyear, C. P., & Ferguson, D. E. (1969). Sun-compass orientation in the mosquitofish, Gambusia affinis. Animal Behaviour, 17(4), 636–640. https://doi.org/10.1016/S0003-3472(69)80005-9 Guilford, T., & Taylor, G. K. (2014). The sun compass revisited. Animal Behaviour, 97, 135–143. https://doi.org/10.1016/j.anbehav.2014.09.005 Lund, U., Agostinelli, C., & Agostinelli, M. C. (2017). Package ‘circular’. Repository CRAN, 775(5), 20-135. McLean, D. J., & Skowron Volponi, M. A. (2018). trajr: An R package for characterisation of animal trajectories. Ethology, 124(6), 440–448. https://doi.org/10.1111/eth.12739