Animal Cognition

, Volume 13, Issue 3, pp 507–513

A new method for studying problem solving and tool use in stingrays (Potamotrygon castexi)

Authors

    • Institute of Life Sciences, Department of Neurobiology and Interdisciplinary Center for Neural ComputationThe Hebrew University of Jerusalem
  • Ruth A. Byrne
    • Abteilung für Rheumatologie, Klinik für Innere Medizin IIIMedizinische Universität Wien
  • Gordon M. Burghardt
    • Department of PsychologyUniversity of Tennessee
Original Paper

DOI: 10.1007/s10071-009-0301-5

Cite this article as:
Kuba, M.J., Byrne, R.A. & Burghardt, G.M. Anim Cogn (2010) 13: 507. doi:10.1007/s10071-009-0301-5

Abstract

Testing the cognitive abilities of cartilaginous fishes is important in understanding the evolutionary origins of cognitive functions in higher vertebrates. We used five South American fresh water stingrays (Potamotrygon castexi) in a learning and problem-solving task. A tube test apparatus was developed to provide a simple but sophisticated procedure for testing cognitive abilities of aquatic animals. All five subjects quickly learned to use water as a tool to extract food from the testing apparatus. The experimental protocol, which gave the animals the opportunity of correcting a wrong visual cue decision, resulted in four out of five subjects correcting an error rather than making an initial right choice. One of five subjects reached 100% correct trials in the visual discrimination task. The ability to use water as an agent to extract food from the testing apparatus is a first indication of tool use in batoid fishes. Performance in the instrumental task of retrieving food from a novel testing apparatus and the rapid learning in the subsequent discrimination/error correction task shows that cartilaginous fish can be used to study the origins of cognitive functions in the vertebrate lineage.

Keywords

Tool useProblem solvingCognitionLearningFishStingrayPotamotrygon castexi

Introduction

The number of publications on learning and cognition in fish grew from about 70 published papers as cited in Thorpe (1963), Gleitman and Rozin (1971), and Keifer and Colgan (1992) to more than 500 in 2003 (Laland et al. 2003). A recent book was devoted exclusively to cognition and behavior in fish (Brown et al. 2006). Researchers have begun to systematically compare cognition in fish and terrestrial vertebrates (Bshary et al. 2002; Brown et al. 2006). Most of this work, however, is based on testing cognitive processes and learning in teleost fishes, a highly derived vertebrate group. Bony fish have become one of the more important taxa in the study of cognitive abilities such as problem solving, numerosity, Machiavellian intelligence, and social learning (Bshary et al. 2002; Brown and Laland 2003; Laland et al. 2003; Bergen et al. 2004; for reviews see Brown et al. 2006; Agrillo et al. 2007, 2008; Bshary et al. 2008).

Cognition and learning in elasmobranch fishes have received, in contrast, little attention, although play behavior has been documented in them as well as in teleost fishes (Burghardt 2005). Given the fact that elasmobrachiomorph fishes are among the oldest lineages of vertebrates, the lack of learning experiments with them is unfortunate, as they are important taxa for testing theories in comparative cognition. Apart from some recent research on large sharks (Klimley 2003; Carrier et al. 2004) and spatial cognition in fresh water stingrays (Schluessel and Bleckmann 2005), few studies on elasmobranch fish cognition exist. This is all the more surprising since there was some interest in the behavior of sharks in the 1960s and 1970s (Clark 1959; Aronson et al. 1967; Hodgson and Mathewson 1978; for a review see Hueter et al. 2004). For example, studies on visual discrimination showed that in simple light–dark discrimination tasks the performance of sharks was similar to that of teleost fishes and even mammals (Aronson et al. 1967; Graeber 1978).

A recent study on the freshwater stingray, Potamotrygon motoro, showed that these animals could orient in a maze using a cognitive map of the environment built by visual clues and egocentric and/or other orientation strategies (Schluessel and Bleckmann 2005) and demonstrated that vision is highly developed and may play an important role in their ecology. Rodrriguez (2006) compared this study to their previous work on goldfish and concluded that Potamotrygon can use allocentric (navigation based on a complex map-like representation of the environment) as well as egocentric strategies to reach a goal through novel pathways.

The study by Schluessel and Bleckmann (2005) also demonstrated that the rays exhibited a lateralized turning bias inside the maze. This is important documentation of lateralized behavior in a phylogenetically ancient branch of the vertebrate linage (Rogers 2002; Rogers and Andrew 2002; Malashichev 2006), particularly as an early origin of behavioral lateralization in vertebrates has been postulated (Bisazza et al. 1998). Since countless accounts on lateral asymmetries have been published on a wide spectrum of vertebrate groups, lateral asymmetries have been claimed to be a homologous trait across all vertebrates (Rogers and Andrew 2002). Therefore, the existence of behavioral lateralization in elasmobranch fishes strengthens this claim, but needs further study. Andrew (2002) argued that lateralization originated with the evolution of visually controlled predation and claims that the key step in the origin of vertebrates was the appearance of predation in a free swimming filter feeder, an organism resembling larval Amphioxus.

Allometric analyses show great variation of brain/body mass ratios in different elasmobranch groups (Northcutt 1978, 1989; New 2001). In some families, the brain/body ratio is greater than in some avian and mammalian taxa. The batoids, members of the superfamily Dasyatoidea (stingrays) have the relatively largest brains. One of the key features of the elasmobranch brain architecture is a highly elaborated cerebellum that shows a level of differentiation and complexity comparable to birds and mammals (New 2001; Butler and Hodos 2005). The fresh water stingrays of the genus Potamotrygon have somewhat smaller brains than their marine relatives, but it is noteworthy that rays in this genus possess a passive electrical sense 10000 times less accurate their marine relatives (Collin and Whitehead 2004). Since computation of the signals of this sense requires large areas of the cerebellum in marine stingrays, one might speculate that the smaller brain in fresh water stingrays might be due to this lesser developed sense organ (Collin and Whitehead 2004) rather than the lack of other cognitive traits.

Although tool use in fish is a controversial topic (see Bshary et al. 2002), documented examples include the use of water-jets by teleost fishes (e.g.: Archerfish: Beck 1980; Schuster et al. 2004) and the water-jets rays use to uncover prey hidden in the sand by various sharks and rays (Hueter et al. 2004). These examples involve an action involving an agent to achieve a goal and follow the definition of tool use given by Beck (1980).

In this study, we introduced a novel pipe-like testing apparatus to fresh water stingrays (Potamotrygon castexi). First, we observed whether these animals are able to discriminate between a positive (rewarded) side and a negative (non-rewarded) side. Second, we investigated their ability to solve a problem by correcting a wrong choice by switching to the correct (rewarded) side. Third, we evaluated potential side preferences in approaching the testing apparatus in order to confirm the existence of lateralization in elasmobranches. Fourth, we asked whether fresh water stingrays could modify their use of water-jets to retrieve food rewards from the pipe-like apparatus.

Materials and Methods

Five experimentally naïve, subadult, captive bred, fresh water stingrays (Potamotrygon castexi), three males and two females, were used in this study. They were born at the Zoo Schönbrunn in Vienna in 2004. When the study was conducted in summer 2005, the subjects were still subadult with an average disc diameter of 30 cm for males and 18 cm for females.

Subjects were housed together in an all glass tank (170 cm× 110 cm× 65 cm) with a sand substrate and wood branches. Additionally, aquatic plants (water lenses, Lemnasp.) covered the water surface. Deionized fresh water was adjusted to mimic the water chemistry of the natural habitat (GH: 12, KH: 7, pH: 7.6 Nitrate: 0.2 mg/L). Two large sponge filters were employed, and water quality was monitored daily. Water temperature was maintained 26°C. Four 58-Watt daylight emission spectrum neon bars provided lighting.

In captivity P. castexi eat a broad variety of fish and invertebrates. Except for testing days, our subjects were fed daily ad libitum with insect larvae, Tubifex worms, and frozen krill. During the experimental testing, either frozen shrimp (Palaemon sp.) or pieces of squid (Loligo sp. 1 × 0.5 cm in size) were used as reinforcement; both types of food were not used in regular feeding.

During the experiments, we divided the home tank into two compartments, the testing arena and the holding compartment (Fig. 1a), by putting opaque dividers into the tank 15 min before the beginning of the experiment. The dividers were designed in a way that they could easily be used as a door to exchange the rays when they were individually tested. The experiment consisted of two stages, the training phase and the problem-solving task. The testing apparatus was always positioned in the center of the testing area 2, parallel to the front screen of the tank.
https://static-content.springer.com/image/art%3A10.1007%2Fs10071-009-0301-5/MediaObjects/10071_2009_301_Fig1_HTML.gif
Fig. 1

a Experimental tank, divided into a residential area, where animals were kept before and after their trials and an experimental arena where the tests were conducted. The testing apparatus was placed in area 2 while the subject had to remain in area 1 before the onset of each trial. b Female 1 during a correct trial. The enlarged area shows the inside of the tube-like testing apparatus. The mesh size of the non-magnetic divider was 1 mm

Training phase

In the training phase, we used a straight light grey plastic pipe (PP-R GF™, 20 cm long and 3.2 cm ø). On the first two days of training, pieces of squid and Tubifex worms were used to reward the rays for learning that they could find food inside the tube. The food was placed in the center of the tube. During the training phase, we did not close the partition, so that all animals had access to the testing tube at the same time, located in the center of area 1. The session was continued until all animals had eaten their daily quantity of food. Chi-square and binomial tests were used to document significant differences in times and frequencies during the training sessions. We are well aware that in this situation the animals were not acting completely independently, and such tests are merely descriptive of individual differences in the social training context.

Problem-solving task

For the problem-solving task, we used the same 20-cm plastic pipe with two connection pieces attached to each side. One of these connection pieces held a metal mesh divider in place inside the tube (Fig. 1b). The mesh allowed unhindered passage of water and odors, but the rays could only retrieve food from one side of the tube. In addition, a black adhesive tape marked the ‘closed’ end of the pipe and a white adhesive tape marked the ‘open’ end. At the start of each experimental session, the individual animals were lured to the experimental arena and isolated from their conspecifics in the keeping compartment (Fig. 1a). During the experiments, the experimental animal had no visual or physical contact with the other animals. Animals were given up to 3 min to obtain the food reward from inside the pipe. A trial was labeled successful if the animals managed to retrieve the food reward either by an initial correct choice, or by switching from the wrong side of the apparatus to the right side (correction method). In all trials, the testing apparatus was placed in the center of the testing area 1 of the testing arena (Fig. 1a). Orientation of the apparatus was altered using a restricted randomized series (Fellows 1967). Timing began as soon as a subject moved into area 2 of the testing arena. Inter-trial intervals were at least 60 s. During a test session, animals were tested until they had reached 5 successful trials per session. All trials were videotaped for later analysis. After a minimum of 30 min following completion of a testing session, the animals were allowed to eat Tubifex worms or insect larvae ad libidum. Frame by frame data analysis was conducted using a Panasonic AG-DV 2500 digital video recorder. We analyzed the direction of approach, the duration of the trials, initial and final choices, and the strategy used to retrieve the food from the pipe. One person recorded the initial experiments and later another observer analyzed the film material, agreement exceeded 90% for all recorded data. Data were analyzed using SPSS 11 for Mac OSX, for all statistical calculations. Binomial tests were used throughout the study to look for significant variations. A linear regression was used to document the decrease in latency across the first 5 sessions of the testing phase.

Results

Training phase

All subjects learned to retrieve food rewards from the pipe within the first two days of training. All three male rays showed a significant decrease in the number of non-successful trials between the first and the second sessions (male 1: N = 33, χ2 = 7.64, P = 0.006; male 2: N = 65, χ2 = 21.79, P = 0.001; male 3: N = 46, χ2 = 16.52, P = 0.001). A female subject (female 1) had only successful contacts; we also noticed that this subject appeared to watch the other animals before her first, and correct, attempt. The other female (female 2) had some unsuccessful trials, especially when chased away by the larger males.

All animals showed a side preference when approaching the test apparatus. Females (females 1 and 2) approached the apparatus more often from the right (R), whereas the three male rays approached the apparatus more often from the left (L). Based on the binomial test, these biases were significant in three animals: female 1, R: 4, L: 1, N = 5, P = 0.37; female 2, R: 18, L: 3, P = 0.001; male 1, R: 6, L: 26, P = 0.001; male 2, R: 29, L: 36, P = 0.46; male 3, R: 8, L: 35, P = 0.0001.

Different strategies were used to retrieve the food from the pipe. The two smaller females used undulating fin movements to create a current inside the pipe transporting the food toward them. The three males used their bodies to produce a current by undulating fin movements (27% of observed cases) or employing their disc-like body like a suction cap (57% of observed cases) or combining suction and undulation (16% of observed cases). In the initial training, no animal used water-jets to get the reward out of the pipe. During the training trials, the animals did not change their strategy to retrieve food from the pipe.

Problem-solving phase

In this more demanding task, all animals learned to retrieve the food reward within 8 experimental sessions and the overall rate of unsuccessful trials dropped from 14 trials in the first session to 1 in the last session (Fig. 2). Performances differed between the subjects. Male 3 was the only animal to achieve 5 out of 5 correct trials based on the visual black and white cue. All other animals solved the problem by trial and error learning and switched from the incorrect to the correct side, avoiding unsuccessful trials. Female 1 needed 11 unsuccessful trials overall in 6 experimental sessions (30 trials), female 2 had an initial success rate of 5 correct to 1 incorrect trials on the first two days and then started to correct an initially wrong decision, although one unsuccessful trial occurred during the last experimental session. Males 1 and 3 were the fastest to learn the task; they needed two experimental sessions and 7 and 5 incorrect trials, respectively. Male 2 managed to perform successful trials only after 14 incorrect trials after session 5 (Table 1).
https://static-content.springer.com/image/art%3A10.1007%2Fs10071-009-0301-5/MediaObjects/10071_2009_301_Fig2_HTML.gif
Fig. 2

Pooled performance of all subjects during the experimental sessions (N = 256). The total number of unsuccessful trials is used to show the improvement of the subjects over time

Table 1

Number of errors (no retrieval of the food reward within 3 min) per session during the problem-solving phase

 

Session 1

Session 2

Session 3

Session 4

Session 5

Session 6

Session 7

Session 8

Female 1

3 (8)

3 (8)

0 (5)

3 (8)

1 (6)

1 (6)

0 (5)

0 (5)

Female 2

1 (6)

1 (6)

2 (7)

2 (7)

2 (7)

4 (9)

2 (7)

1 (6)

Male 1

4 (9)

3 (8)

0 (5)

0 (5)

0 (5)

0 (5)

0 (5)

0 (5)

Male 2

3 (8)

2 (7)

5 (10)

2 (7)

2 (7)

0 (5)

0 (5)

0 (5)

Male 3

2 (7)

3 (8)

0 (5)

0 (5)

0 (5)

0 (5)

0 (5)

0 (5)

The total number of trials per session is given in parentheses. All subjects had to conduct five successful trials per session. Note the fast decline of errors in male 1 & 3

The females were especially prone to initially try to retrieve the reward by exhibiting additional behaviors after making an initial incorrect choice. These behaviors ranged from biting the apparatus to grabbing it with the mouth to even tossing it around.

All animals showed a linear decrease in latency (the time from the beginning of the trial to the first contact with the apparatus) from session 1 to 5 (Fig. 3, N = 169, F = 11625, P = 0.01, R2 = 0.997, df = 1, −4.98 × day + 33.96). During later days, latency increased again.
https://static-content.springer.com/image/art%3A10.1007%2Fs10071-009-0301-5/MediaObjects/10071_2009_301_Fig3_HTML.gif
Fig. 3

Decrease in latency to approach the testing apparatus for all subjects in seconds (N = 256). After a rapid initial decline in latency, all subjects showed a slight increase in latency in the last three sessions. Error bars show the standard error

Four animals retained the side preferences from the testing phase to approach the pipe, and these were statistically significant (binomial test): female 1 (R: 35, L: 16, P = 0.01); female 2 (R: 35, L: 14, P = 0.01); male 2 (R: 2, L: 49, P = 0.0001); male 3: R: 0, L: 44, P = 0.0001). Male 1 showed a significant side preference different to the training phase (R: 10, L: 35, P = 0.0001).

During the first problem-solving sessions, the rays continued to use the same strategy as during the training phase; however, subsequently all animals showed a shift in their strategies for retrieving the food reward. This shift was generally from using undulating fin movements or suction to a combination of both. Male 3 was the only subject to use water-jets blown into the pipe to retrieve food rewards (Fig. 4).
https://static-content.springer.com/image/art%3A10.1007%2Fs10071-009-0301-5/MediaObjects/10071_2009_301_Fig4_HTML.gif
Fig. 4

Change of strategies by male 3 to retrieve food from the testing apparatus over the experimental sessions (N = 45). Toward the end of the experiments, all animals shifted to use a combination of suction and fin movements (s & u) as did male 3 here. S = suction, u = undulating fin movements, s & u = a combination of suction and fin movements, and b = blowing a water-jet into the apparatus

Discussion

The general aim of this study was to construct a testing apparatus and establish a simple but effective procedure enabling researchers to test cognitive abilities of aquatic vertebrates. Specifically, we wanted to investigate learning performances and problem-solving capacities in batoid fish. All 5 subjects rapidly learned to use water as a tool to extract food from the testing apparatus. Beck (1980) defined tool use as an action that involves an agent (here water) to achieve a goal (here retrieval of a food reward from a pipe). Our study shows that fish can be tested using a tube-trap apparatus. While recent studies in mammals focus on more complex trap-tube tasks (e.g.: Visalberghi and Limongelli 1994; Visalberghi et al. 1995; Reaux and Povinelli 2000; Horner and Whiten 2007), the initial starting point of tube testing paradigms was similar to ours (Yerkes 1943).

Only, one of the five subjects reached 100% correct responses based on the visual observation cues. The experimental protocol, which gave the animals the opportunity to trial and error learning, resulted in 4 out of 5 subjects making an initial error that had to be corrected. Correcting an incorrect choice can be very demanding for stingrays, involving inhibition of a strong chemical food recognition cue. Olfactory stimuli from the food reward inside the testing apparatus must have been intense, especially when the animals created a flow of water toward them. Learning to ignore the stimulation of one of the prime senses in batoid fishes (Hueter et al. 2004) and to venture away from the food source and approach the other side of the apparatus is a rather remarkable detail. However, compared to some studies we did not continue testing for long periods, and the animals may have done better with more trials.

Our findings along with other studies (Aronson et al. 1967; Schluessel and Bleckmann 2005) show that the cognitive abilities of elasmobrachiomorph fishes rival not only those of teleost fishes, but also those of reptiles, birds, and mammals. Aronson et al. (1967) were the first to undertake a comparative learning study testing mice and nurse sharks (Ginglymostoma cirratum). They showed that the learning of a black and white discrimination task by a juvenile nurse shark was similar to that of mice. Both groups of subjects reached and average performance of 80% correct trials on the fifth experimental day (with an average of 15–30 trials per day). After 14 days of testing, the sharks did not make any mistakes discriminating the targets. In addition, they compared their results to an earlier study on Tilapia (Tilapia heudeloti macrocephala) by Wodinsky and Bitterman (1957) showing that teleost fish perform similar to both shark and mice. Now that we have established our new method, we plan to repeat classic experiments by Bitterman (1965), who tested fish, pigeons, and rats in serial reversal and probability learning experiments. He concludes that there are qualitative differences between fish and rat, and pigeons performed fish-like in one (probability learning) and rat-like in the other experiments (serial reversal). While his claims have been widely criticized (Mackintosh 1969; Bitterman and Mackintosh 1969; see Burghardt 1977) his studies remain an interesting starting point for new comparative studies on basic learning processes in vertebrates.

Northcutt (1977, 1989) showed that elasmobrachiomorph fishes possess large and highly complex brains. He compared his findings to those of Jerison (1970, 1973) and proved that elasmobrachiomorph brain size and organization is closer to those of birds and mammals than to the brains of the more ‘modern’ teleost fishes. Recently, New (2001) reviewed the architecture and function of the cerebellum in elasmobrachiomorph fishes. The highly developed cerebellar structures in both sharks and rays might not only be important for locomotion and electrical signaling, but also have other significant functions. For example, Yarom and Cohen (2002) speculated about the function of the cerebellum as a universal timer for the execution of behavioral programs. This would include an important function in the timing and execution of responses during cognitive processes.

All subjects showed a strong side preference when approaching the testing apparatus. In three subjects, these preferences remained constant in both the training and the testing phase. One animal (male 1) showed different side preferences in the two phases of the experiment. The side preferences to approach the testing apparatus in our study agree with the results on turning preferences shown in the maze experiments by Schluessel and Bleckmann (2005). This is important, because previously teleost fishes were the most basal vertebrates tested and shown to exhibit lateralized behavior (Rogers 2002; Rogers and Andrew 2002; Malashichev 2006).

The use of water as a tool in batoid fishes can be compared to tool use in teleost fishes. In both groups of fish, water is used as an agent to retrieve food (Beck 1980; Bshary et al. 2002; Schuster et al. 2004). In addition, our subjects also showed the ability to modify and potentially optimize their strategy to extract food from the testing apparatus. Although the animals exhibited different strategies (both between and within subjects) at the beginning of the experiments, they ended up using a similar technique toward the end of the experiments. Tool use in elasmobrachiomorph fish could have evolved either early on or at least twice in the vertebrate linage as elasmobrachiomorph fishes are thought to have split from the general vertebrate linage in the late Silurian or early Devonian.

The present study shows that cartilaginous fishes can be used to study the origin and evolution of cognitive traits. Their phylogenetic relationship to the other vertebrates (Grogan and Lund 2004; Hamlett 2005) also makes them ideal candidates for examining the evolution of the central nervous system and complex behaviors. These rather tractable and relatively small batoid fishes might prove useful subjects for combining neuroscience and behavioral experiments.

Acknowledgments

The authors of this study want to thank the zoo of Vienna for providing the animals. We would especially like to thank E Wolf, A Weissenbacher and R Halbbauer at the zoo of Vienna for their help and cooperation during this study. We want to express our special gratitude to E Sonntag of the Detroit zoo who used food containing plastic pipes as a behavioral enrichment for the stingrays housed at the Detroit zoo. This example gave rise to our idea to create a pipe-like testing apparatus.

Supplementary material

10071_2009_301_MOESM1_ESM.divx.
Supplementary material 1 (DIVX 2136 kb). Animal 2 choosing the correct side of the tasting apparatus. The correct side to extract the food is marked by white tape

Copyright information

© Springer-Verlag 2009