J Comp Physiol A DOI 10.1007/s00359-014-0887-1

Original Paper

Visual generalization in honeybees: evidence of peak shift in color discrimination J. Martínez‑Harms · N. Márquez · R. Menzel · M. Vorobyev 

Received: 24 January 2013 / Revised: 16 January 2014 / Accepted: 23 January 2014 © Springer-Verlag Berlin Heidelberg 2014

Abstract  In the present study, we investigated color generalization in the honeybee Apis mellifera after differential conditioning. In particular, we evaluated the effect of varying the position of a novel color along a perceptual continuum relative to familiar colors on response biases. Honeybee foragers were differentially trained to discriminate between rewarded (S+) and unrewarded (S−) colors and tested on responses toward the former S+ when presented against a novel color. A color space based on the receptor noise-limited model was used to evaluate the relationship between colors and to characterize a perceptual continuum. When S+ was tested against a novel color occupying a locus in the color space located in the same direction from S− as S+, but further away, the bees shifted their stronger response away from S− toward the novel color. These results reveal the occurrence of peak shift in the color vision of honeybees and indicate that honeybees can learn color stimuli in relational terms based on chromatic perceptual differences. Keywords  Honeybees · Color vision · Visual generalization · Color discrimination · Peak shift J. Martínez‑Harms · N. Márquez · R. Menzel  Institut für Biologie, Neurobiologie, FU Biologie, Freie Universität Berlin, Königin‑Luise Str. 28/30, 14195 Berlin, Germany Present Address: J. Martínez‑Harms (*)  Department of Evolutionary Neuroethology, Max Planck Institute for Chemical Ecology, 07745 Jena, Germany e-mail: jmartinez‑[email protected] M. Vorobyev  Department of Optometry and Vision Science, University of Auckland, Private Bag 92019, Auckland 1142, New Zealand

Introduction Stimulus generalization can be defined as “the behavioral fact that a conditioned response formed to one stimulus may also be elicited by other stimuli which have not been used in the course of conditioning” (Hilgard and Marquis 1940). In studies on generalization, responses to novel stimuli are usually tested after differential conditioning, for example, on subjects previously trained to discriminate a positively reinforced stimulus (S+) from an unrewarded or negatively reinforced stimulus (S−). Although novel stimuli generally elicit weaker responses than familiar stimuli, bias responses have often been observed when novel and familiar stimuli vary along a defined physical or perceptual dimension (Thomas et al. 1991; Wills and Mackintosh 1998; Ghirlanda and Enquist 2003; Lynn et al. 2005). This type of response bias, commonly referred to as “peak shift”, was first described by Köhler (1918/1938) who trained chicks to discriminate between different shades of gray. He reported that when subjects were trained to the lighter of two shades of gray and tested with the former S+ color against a novel, even lighter, shade of gray, animals selected the novel shade in over 70 % of the trials. Köhler suggested that animals learned the relationship between stimuli rather than the absolute shade value of the stimuli. Accordingly, when the absolute value of the stimuli varied but the relationship between them was constant, the animals shifted their preference to the “relationally correct” stimulus. Generalization has traditionally been investigated using physical measures of stimulus difference (Shepard 1987). In the case of color vision, wavelength differences have often been used to describe a continuum in the physical space. However, a continuum in a physical space has limitations when studying generalization of color stimuli.

13



Discriminability and perceptual distance are not simply proportional to wavelength differences, nor do monochromatic lights resemble the spectra of natural objects. A perceptual space provides a more appropriate method of accounting for differences between stimuli and establishing a color continuum. In the present study, we used a receptor noise-based color opponent model to characterize a perceptual space for the honeybee, Apis mellifera (Vorobyev and Brandt 1996; Vorobyev and Osorio 1998; Vorobyev et al. 2001). The receptor noise model is based on the assumption that discriminability of colors is limited by noise originating in photoreceptors and uses a chromaticity diagram to map color spectra (Kelber et al. 2003). This model provides a two-dimensional color space that allows estimation of perceptual distance between stimuli and has been used to describe color discrimination in honeybees (Brandt and Vorobyev 1997; Vorobyev and Brandt 1997; Vorobyev et al. 2001). The honeybee represents a useful model for studying generalization of color stimuli. Due to their remarkable learning and memory capabilities, it is relatively easy to train honeybee foragers using a single or set of colored targets to test responses to stimuli that they have not experienced during the course of training (e.g., Giurfa 1991; Benard and Giurfa 2008; Reisenman and Giurfa 2008). In addition, since the capacity of honeybees to perceive colors was demonstrated almost a century ago (von Frisch 1914), extensive behavioral and physiological studies have revealed the dimensions and quantitative interactions of parameters characterizing their color vision (e.g., Backhaus and Menzel 1987; Backhaus et al. 1987; Brandt and Vorobyev 1997). We used the receptor noise model to describe a continuum of color stimuli to evaluate the occurrence of response biases after differential conditioning in the honeybee. Previous studies have shown that differential conditioning affects the color discrimination performance of honeybees, yielding improved discrimination of perceptually similar colors compared to absolute training (Giurfa 2004). Differential conditioning has also been shown to improve color discrimination in other insect models (Dyer and Chittka 2004; Camlitepe and Aksoy 2010; Kelber 2010). Improved discrimination may result from the fact that under differential conditioning, subjects can learn colors in relational terms (Giurfa 2004). This type of learning induced by differential conditioning may also lead to biases in generalization responses of honeybees. Wright et al. (2009) showed that honeybee foragers differentially trained to discriminate between binary mixtures of odors shifted their maximal response toward a mixture that differed from S− in the same way as S+, but to a greater extent. Despite the longstanding tradition of using honeybees as a model to study visual cognition, only a few studies have tested color generalization after differential conditioning in honeybees,

13

J Comp Physiol A

Fig. 1  Diagram of the experimental setup. The Y-maze was covered with ultraviolet-transparent plexiglass and illuminated by natural light

and none of those studies tested stimuli over a perceptual continuum in a way that could reveal the occurrence of responses biases in color vision (Reisenman and Giurfa 2008; Dyer and Murphy 2009). Studies using bumblebees reported the occurrence of peak shift in individuals differentially trained to color stimuli (Lynn et al. 2005; Leonard et al. 2011). Given that differences in the learning and color discrimination capacities between honeybees and other hymenopterans have been reported (Moreno et al. 2012; Dyer et al. 2008), peak shift findings in the color vision of bumblebees cannot simply be extrapolated to apply to honeybees. In the present work, we evaluated how honeybees generalize colors along a perceptual continuum after differential conditioning. In particular, the effects of varying the position of a novel stimulus in the color space relative to rewarded (S+) and unrewarded stimuli (S−) on the occurrence of peak shift were studied.

Materials and methods Experimental setup and training procedure A group of free-flying honeybees, A. mellifera L., was trained to collect a 30 % sucrose solution from a feeder located 10 m from the hive. Foragers were individually marked and trained to enter an experimental setup to collect a 50 % sucrose solution from a dispenser at the center of a vertically presented colored targets. Learning this task took bees around five foraging bouts during which time they were progressively induced to fly into a maze using a syringe with the sucrose solution. Only one bee was trained and tested at a time. The experimental setup consisted of a wooden Y-shaped maze with a UV-transparent plexiglass ceiling to ensure daylight conditions within the maze (Fig.  1). A sliding door ensured that only one bee could enter the maze at a time. Bees were trained to fly through

J Comp Physiol A

an entrance hole in the middle of the front panel into a decision chamber. Once inside the decision chamber, bees had visual access to both back walls of the maze. The back walls measured 20 × 20 cm2, with a 0.6-cm diameter central hole through which the sucrose solution could be delivered. The stimuli were presented vertically on the center of the back walls. The position of the stimuli was constantly changed pseudorandomly to ensure that the bees did not associate the reward with any particular side of the maze. The advantage of this experimental setup is that the angular size of the stimuli can be controlled. In these experiments, we used colored targets with a diameter of 8 cm located 15 cm from the decision chamber. From this distance, the targets would subtend angular sizes of 30° when viewed from the decision chamber. Foragers were trained to discriminate a rewarded colored target (S+) from an unrewarded target (S−) presented on a gray background. Each bee experienced only one S+/S− combination during training. Training consisted of allowing bees to enter the decision chamber and approach the training stimuli. When the bee chose the arm containing S+, they were rewarded with sugar water delivered at the stimulus. After it filled its crop, the bee returned to the hive to complete the learning trial. If the bee entered the arm containing S−, it was gently pushed out of the maze and had to enter the maze again. This procedure was repeated until the bee made a correct choice, after which it was rewarded and allowed to return to the hive.

where Ri(λ) is the spectral sensitivity of receptor of type i, S(λ) is the reflectance spectrum, I(λ) is the illumination spectrum and ki is a scaling factor. For trichromatic vision, i = S, M, L (corresponding to short-, medium-, and longwavelength receptors, respectively). We assumed that illumination is standard D65 daylight (Wyszecki and Stiles 1982). ki was set such that quantum catches for the gray background were equal to unity.  ki = 1  I()S b ()Ri ()d, (2)

Defining a color perceptual continuum

1 , A=  2 ωL2 + ωM

The reflectance spectra of the colors used in these experiments were measured with a spectral photometer (model S2000; Ocean Optics, Dunedin, FL, USA) over a range of 300–700 nm. The colors were either cut from HKS-colored cardboard (K+E Stuttgart, Stuttgart-Feuerbach, Germany) or produced using a color printer (HP Color LaserJet 4700, Palo Alto, CA, USA). We used the color opponent receptor noise-limited model (Vorobyev and Osorio 1998; Vorobyev et al. 2001) to estimate similarities between colors. This model is based on the assumption that detection and discrimination of light stimuli is limited by noise generated by photoreceptors. The model assumes that chromatic discrimination is independent of stimulus brightness. The model predictions agree with the results of behavioral experiments in several different species (Vorobyev and Osorio 1998; Vorobyev et al. 2001; Koshitaka et al. 2008). The parameters of the model are the photoreceptor noise levels (Vorobyev et al. 2001). For each color stimuli, we calculated the quantum catch qi of corresponding photoreceptor i,  qi = ki I()S()Ri ()d, (1) 



b

where S (λ) is the spectrum of the gray background. According to the log-linear version of the receptor noiselimited model (Vorobyev et al. 2001), receptor signals are related to receptor quantum catches by

(3)

fi = ln(qi ).

To plot color stimuli, we used a chromaticity diagram where the Euclidean distance between the points corresponds to the predicted ability to discriminate the stimuli. We used the axes corresponding to the respective L − M and S − [L + M] directions in the chromaticity diagram (Kelber et al. 2003):

X1 = A(fL − fM ), (4)

X2 = B(fS − (afL + bfM )), where:

B=



a=

2 ωM , 2 ωL2 + ωM

b=

ωL2 2 ωL2 + ωM

2 ωL2 + ωM , (ωL2 )(ωM )2 + (ωS )2 (ωL )2 + (ωS )2 (ωM )2

Noise values were set to ωS  = 0.13, ωM  = 0.06 and ωL  = 0.12 (Vorobyev et al. 2001). Distance in the color space can be expressed as:

S 2 = X12 + X22

(5)

Tests In all experiments, bees experienced a single color pair during training. After completing 20 training trials, the bees’ choices were tested under conditions of extinction

13



J Comp Physiol A 1.0

0.8

Reflectance

8

S1− S2 S3 S4− S5 S6 S7− S8 S9 Back.

0.6

X2

6

S9 4

S1 −

S3 S2

S8

S7 −

2

X1 −8

0.4

−6

−4

2

−2 S4 −

4

6

8

−2

S5

0.2

−4

S6

0.0 300

400

500

600

−6

700

Wavelength (nm)

−8

Fig. 2  Spectral reflectance functions of the gray background (Back.) (HKS-92N) and the three sets of colored paper used in the experiments. The color of each curve illustrates the colors as they appear to the human eye. Colored stimuli S1− (HKS-24N), S2 (HKS-26N) and S3 (HKS-29N) were used in experiments 1 and 1’. Colored stimuli S4−, S5 and S6 were produced with a color printer and used in experiments 2 and 2’. Colored stimuli S7− (HKS-8N), S8 (HKS-12N) and S9 (HKS-14N) were used in experiments 3 and 3’

Fig. 3  Loci of color stimuli used in the experiments in the chromaticity diagram of Apis mellifera. The color loci of the three sets of stimuli were selected so that they would lie approximately in line with one of the axes of the chromaticity diagram. The S1−, S2, S3, S7−, S8 and S9 colors are approximately in line with the X1 (L − M) axis, while the S4−, S5 and S6 colors are approximately in line with the X2 [S − (L + M)] axis

(no reward presented) in a dual choice situation. The tests lasted 2 min and consisted of allowing the bees to enter the Y-maze and approach the colored targets. The cumulative number of approaches ending with the bee contacting the targets during the test was recorded as choices. Approaches that did not end up with the bee touching the colored target were not considered, given the behavioral nature of the parameter that does not allow determination of whether the stimuli were discriminated (Giurfa et al. 1999; Hempel de Ibarra and Giurfa 2003). All tests were performed using fresh stimuli to avoid the influence of scent marks. Each pair of colors was tested twice, alternating their positions within the maze. Results from the two tests were pooled to compensate for side asymmetries. Two reinforcement trials between tests of the same color pair and eight trials between tests of different color pairs were performed to avoid extinction of the response towards the stimuli. Two types of unrewarded tests were performed per bee, including a control test (pre-test) in which the level of response for the colors used for training was evaluated and a generalization test (test) in which the level of response for the previously rewarded color (S+) was evaluated when presented against a novel color. Three sets of colors containing three colors each (Fig. 2) were used to

study the effect of varying the position of a novel stimulus in the color space with respect to S+ and S− stimuli on generalization by honeybees. Colors were selected so that they would lie approximately in line with one of the axes of the color space of the honeybee (Fig. 3). Chromatic distances between stimuli and their loci in the color space of the honeybee are shown in Table 1 and Fig. 3, respectively. By maintaining the identity of the fixed S− color and exchanging the identity of the other two colors as novel and S+ stimuli, each set of colors was used for two types of experiments. Type I experiments, corresponding to experiments 1, 2 and 3 (Fig. 4a, c, e), were designed to evaluate generalization responses when the former S+ was presented against a novel color with a higher chromatic contrast to S− compared to S+. For this purpose, the colors whose loci were farther away in the color space from their respective S− color were used as novel stimuli, while the colors whose loci were closer to S− were used as S+ colors (Figs. 2, 3). In experiment 1, individuals were trained to S1− and S2+ and tested using S3 as a novel color (Fig. 2; Table 1). In experiment 2, S4− and S5+ were used as training colors and S6 was the novel stimulus (Fig. 2; Table 1). In experiment 3, bees were trained to S7− and S8+ and tested using S9 as a novel color (Fig. 2; Table 1). Type II experiments,

13

J Comp Physiol A Table 1  Photoreceptor-specific contrast to the gray background (qS, qM, qL), chromatic coordinates (X1, X2), and chromatic distances to the respective S− stimuli [ΔS(S−) and background (ΔS(Back.)] for the colors used in these experiments Color

qS

qM

qL

X1

X2

ΔS (S−)

ΔS (Back.)

S 1− S2 S3 S 4− S5 S6 S 7− S8

0.95 1.24 1.53 0.75 0.91 1.30 0.91 0.88

0.76 1.08 1.34 0.97 1.52 3.00 0.48 0.48

0.63 0.64 0.70 0.65 1.00 1.97 1.14 0.79

−1.38 −3.82 −4.79 −2.99 −3.10 −3.13 6.50 3.76

1.83 1.74 1.88 −1.29 −3.05 −5.34 3.32 3.60

– 2.44 3.42 – 1.77 4.05 – 2.76

2.29 4.20 5.15 3.26 4.34 6.19 7.30 5.21

S9

0.86

0.46

0.64

2.55

4.01

4.01

4.75

Values were calculated according to the receptor noise-limited model (Vorobyev and Osorio 1998; Vorobyev et al. 2001). Chromatic distances are shown in standard units. Colored stimuli S1− (HKS-24N), S2 (HKS-26N) and S3 (HKS-29N) were used in experiments 1 and 1’. Colored stimuli S4−, S5 and S6 were produced with a color printer and used in experiments 2 and 2’. Colored stimuli S7− (HKS-8N), S8 (HKS-12N) and S9 (HKS-14N) were used in experiments 3 and 3’

corresponding to experiments 1’, 2’ and 3’ (Fig. 4b, d, f), were designed to evaluate generalization responses of bees when the former S+ was presented against a novel color with a lower chromatic contrast to S− compared to S+. For this purpose, colors located farther away in the chromaticity diagram from their respective S− color were used as the S+ color, while the colors whose loci were closer to the S− color were used as novel stimuli (Figs. 2, 3). In experiment 1’, individuals were trained to S3+ and S1− and tested using S2 as a novel color (Fig. 2; Table  1). In experiment 2’, S4− and S6+ were used as training colors and S5 as a novel color (Fig. 2; Table 1). In experiment 3’, bees were trained to S7− and S9+ and tested using S8 as a novel color (Fig. 2; Table 1). Different groups of bees were used for each experiment. Statistics To test whether bees had learned the training stimuli, the proportion of choices for S+ in the pre-test was analyzed for each individual bee using a binomial test (Po > 0.5, α ≤ 0.05) (Sokal and Rohlf 2012). After testing for homogeneity, the proportion of choices for each color pair was pooled for further analyses. These data were used to test the null hypothesis of random choices between colors using a binomial test (Po > 0.5, α  ≤ 0.05). To analyze how the chromatic relationship between stimuli affected choices (dependent variable), ANOVA for repeated measures followed by the Bonferroni post hoc test were performed (Sokal and Rohlf 2012). Data were previously transformed using the arcsine √p transformation to reach normality. The level of significance was set at 0.05 for all analyses. All analyses were performed using R software (R Foundation for Statistical Computing, Vienna, Austria 2009).

Results The experiments tested whether generalization of color stimuli along a perceptual continuum is expressed in a peak shift after training bees in a dual choice situation with a rewarded (S+) color and an unrewarded (S−) color. Type I experiments In experiments 1, 2 and 3, animals were subjected to a pre-test and a generalization test in which the former S+ was tested against a novel color with a higher chromatic contrast to S− than S+ (Fig. 4a, c, e). After training, all bees responded more strongly to S+ in the pre-test (binomial test, P  ≤ 0.04), indicating that they had learned the training stimuli. Analysis performed with the pooled data also showed a stronger response towards S+ in the pre-test (binomial test: experiment 1, n = 105, N = 4, P 

Visual generalization in honeybees: evidence of peak shift in color discrimination.

In the present study, we investigated color generalization in the honeybee Apis mellifera after differential conditioning. In particular, we evaluated...
470KB Sizes 0 Downloads 3 Views