Two family groups of otters—five giant otters (Pteronura brasiliensis) and four Asian small-clawed otters (Aonyx cinerea)—participated in this study. None of the subjects had previous experience with experimental studies. The animals were housed at Leipzig zoo. The giant otter group consisted of an adult female (9.5 years) with her four subadult offspring (2 females, all 1.5 years of age). The small-clawed otter group consisted of four males (siblings) aged between 5.0 and 6.6 years (see Supplementary Material 1 for details concerning husbandry and enclosures). One juvenile giant (Erna) otter stopped participating during the training and was therefore not included in any analysis.
The individual training apparatus consisted of small, square PVC platforms (giant otters: 30 × 30 cm, small-clawed otters: 15 × 15 cm) with a rope attached to it (see Fig. 2a). The cooperation apparatus was modeled on the original design by Hirata and Fuwa (2007). Our “Hirotter” board consisted of a long, flat, U-shaped platform (giant otters: 200 × 60 cm, small-clawed otter: 100 × 30 cm, see Fig. 2b). The training and test boards were located on the floor outside the enclosure. During the test, we baited both ends of the board with food rewards with preferred food types as indicated by the caretakers (pieces of fresh fish for the giant otters and cat food for the small-clawed otters, after trying grapes in the first sessions). A rope ran around three vertical screws (for the giant otters) or through two eyebolts (for the small-clawed otters) that were protruding from the platform at both ends (and in the middle for the giant otters). At both sides of the platform, the ends of the rope extended into the otter enclosure underneath the mesh. The otters could access the food on the platform if two individuals were cooperating either by pulling at each end of the rope simultaneously or by holding one end of the rope, while the partner was pulling the other end of the rope. One individual pulling the rope alone resulted in removal of the rope from the apparatus without moving the baited platform. Thus, pulling only one end of the rope resulted in loss of access to the food as the second individual could no longer reach the rope.
The entire study was conducted in a group setting per species, i.e., no individuals were separated from the group at any point. Subjects were first trained in an individual string pulling task before they entered the cooperation test phase (see Supplementary Material 1 for details). The cooperation test phase encompassed five conditions that were administered in this order: Simultaneous I (6 sessions/103 trials), Simultaneous II (6 sessions/93 trials), Delay I (3 sessions/28 trials), Long-rope-delay (3 sessions/29 trials) and Delay II (1 session/14 trials) (see Supplementary Material 2 for examples of simultaneous and delay trials in both species). For the giant otters, the number of trials per session varied (between 5 and 30 trials) depending on the food availability as the amount and size of fish provided to us by the zoo varied. For the small-clawed otters, we matched the number of trials to the giant otters. In all conditions, both sides of the platform were baited at the same time. In the simultaneous conditions, the experimenters slid both ends of the rope underneath the mesh of the enclosure at the same time when at least one subject was present on each side of the apparatus. Subjects could therefore access the two ends of the rope simultaneously; no waiting was necessary. When one individual pulled harder than the other one, the platform sometimes tilted so that one side of the platform became accessible before the other one. When this happened, the former individual typically released the rope to eat the food. For this reason, the other individual could not retrieve its food reward. In the Simultaneous I condition, this resulted in an uneven food distribution in some trials (proportion of trials with uneven food distribution in Simultaneous 1: giant otters: 0.40; small-clawed otters: 0.20). In Simultaneous II, the experimenters pushed the other side of the platform forward when the platform tilted to maintain a consistent reward contingency.
In the delay conditions, all individuals in the group were lured to another compartment of the enclosure as far from the apparatus as possible where every individual would receive a piece of food. While the test compartment was empty, the platform was baited and the two ends of the rope were pushed into the test compartment (see Fig. 3). The delayed access to the rope ends was achieved by the delayed entry of the otters because one rope was closer to the door to the adjacent compartment, so that when the otters returned to the testing compartment, the first individual could access this end first and they would have to wait before another individual could move around to the other end of the rope. In Delay I and II, the rope was the same length as in the simultaneous conditions (giant otters: 4.0 m total length, approx. 0.3 m inside the cage at either end; small-clawed otters: 2.0 m, approx. 0.15 m inside). In the Long-rope-delay condition, we extended the length of the rope, thereby relaxing the need for temporal synchronization of pulling (giant otters: 5.4 m, approx. 1 m inside; small-clawed otters: 2.7 m, approx. 0.5 m inside) and providing the otters with further opportunity to learn the affordances of the delay conditions.
Coding and analysis
For each trial, we coded whether or not the participating otters were successful. Trials in which the board was pulled in only on one side were also coded as success (proportion of all trials: giant otters: 0.45; small-clawed otters: 0.27). A second coder, blind to the purpose of the study, coded a random selection of 20% of test trials from video. There was a very high agreement of 96.36% between the two coders (Cohen’s Kappa; Κ = 0.92). Furthermore, we coded live which subject pulled on which end of the rope (left or right). In most unsuccessful trials, one subject started pulling on the rope, while the other end was unoccupied. For these trials, we coded the subject who pulled on the rope. For the delay conditions, a third coder coded the time between the arrival of the first otter and the arrival of the second otter at the board from video.
The dependent variable was the binary success code. To analyze the data, we used a generalized linear mixed model (GLMM) with a binomial error structure. All models were fitted in R (R Core Team 2012) using the function glmer of the R-package lme4 (Bates et al. 2015). We used likelihood ratio tests (LRT) to assess whether the inclusion of predictors and their interactions improved the general fit of a model to the data by comparing models with and without the respective effects (Dobson and Barnett 2008).
The full model comprised of species, condition and their interaction as fixed effects and trial and session number as covariates. We compared this model to a reduced model comprising of only the covariates (trial number and session number). To test the significance of the interaction, we compared the full model to a reduced model without the interaction. Given the interaction turned out to be nonsignificant, we tested the significance of each fixed effect (species and condition) by comparing a model comprising them to a model lacking them. We accounted for the identity of the first individual pulling one end of the rope (left and right) and the specific dyad by including them as random intercept terms (see Supplementary Material 1 for details).