Swallowtail butterflies of the species Papilio aegeus oviposit on the leaves of Rutaceae plants in Australia. They possess receptor types with sensitivity peaks around 390 nm (violet receptor) and 610 nm (red receptor), in addition to the receptor types common in insects with sensitivity peaks at 360 nm (ultraviolet receptor), 440 nm (blue receptor) and 540 nm (green receptor). Multiple-and dual-choice experiments show that females of P. aegeus prefer to oviposit on substrata that look green to humans. A class of simple models is developed to describe this choice behaviour in terms of linear interactions between the different spectral types of photoreceptors. The green receptor has a positive influence, whereas the blue (and possibly the ultraviolet and violet) receptor and the red receptor have negative influences on the choice behaviour. Colour choice for oviposition is thus guided by a single chromatic mechanism. Caterpillars of P. aegeus grow faster on young leaves which, according to the model, should be preferred by females for oviposition. The importance of the red receptor for the discrimination between different green leaves is discussed in ecological and comparative contexts. Finally, in an evolutionary perspective, the possibility is discussed that colour vision systems like those of honeybees might have evolved as a combination of two or more such chromatic mechanisms.

Colour-mediated behaviours in insects are commonly described as either colour vision or wavelength-specific behaviours (Menzel, 1979). Colour vision is defined as the ability to discriminate colours by means of their spectral distribution independent of intensity (Menzel, 1979; De Valois and De Valois, 1997). There are two prerequisites for colour vision: there must be at least two spectral types of photoreceptors, and opponent (chromatic) interactions must occur between them. A receptor interaction is called opponent if different receptor inputs have opposite signs.

Recently, several receptor models have been developed which accurately describe honeybee colour vision. According to Vorobyev and Brandt (1997), the term colour vision is reserved for systems with more than one dimension. In their general colour opponent model, Brandt and Vorobyev (1997) assume two linear opponent interactions between the ultraviolet, blue and green receptor types. These two interactions build the axes of the two-dimensional trichromatic colour space of the honeybee. In humans, by comparison, two opponent interactions and one non-opponent interaction build the axes of the colour space (cf. De Valois and De Valois, 1997; Lee 1998). The latter is the intensity signal that gives the human colour space the achromatic third dimension, brightness.

Wavelength-specific behaviour has not gained as much attention as colour vision. This term is based on measurements of the spectral sensitivity S(λ) of behaviours such as the escape response or the optomotor response. According to Menzel (1979), “if an animal displays markedly different S(λ) for different behavior patterns, the individual behavior pattern will be called wavelength-specific”. This is not very helpful since it does not refer to the underlying physiological mechanisms. The term is often used if colour vision (in the sense defined above) cannot be demonstrated. Wavelength-specific behaviours are based on either an achromatic (non-opponent) mechanism or a single chromatic (opponent) mechanism. I am using these physiologically defined terms throughout this study.

A behaviour mediated by an achromatic mechanism is colour-blind. It can be based on either a single receptor signal or the summed signals of different receptor types. There are many well-studied examples such as the optomotor response in bees and the open-space reaction in butterflies, which depend on just a single spectral type of photoreceptor (for a review, see Menzel, 1979). The luminance mechanism in primates is an example of an achromatic mechanism guided by the sum of two receptor signals (the M and L cones; see Lee, 1998).

Behaviours mediated by single chromatic mechanisms have been described mainly in butterflies. Kolb and Scherer (1982) and Scherer and Kolb (1987a,b) have measured the wavelength-dependence of oviposition and the feeding response in nymphalid, satyrid and pierid butterflies. They found that oviposition is elicited by light within a narrow spectral band around 550 nm and that feeding is elicited by light in two spectral regions around 440 nm and 600 nm. Papilionid butterflies have since been shown to use true colour vision for the discrimination of learned feeder colours (Kinoshita et al., 1999; Kelber and Pfaff, 1999). However, the spontaneous colour choices of naive animals in the context of oviposition and feeding depend on the intensity of the stimuli (Scherer and Kolb, 1987a,b). Swihart (1970) and Menzel and Backhaus (1991) proposed that chromatic receptor interactions might account for these behaviours. To my knowledge, no exact models have hitherto been proposed.

An ommatidium of Papilio spp. consists of nine receptors in three tiers. Two distal receptors (R1 and R2) have maximal sensitivity around 360 nm (ultraviolet receptor, UV), 390 nm (violet receptor, V) or 440 nm (blue receptor, B); two distal receptors (R3 and R4) have sensitivity peaks at 550 nm (green receptors, G); and all proximal receptors (R5-R8) and the basal receptor (R9) are maximally sensitive at 550 nm or 610 nm (red receptor, R; Matic, 1983; Ribi, 1987; Bandai et al., 1992; Arikawa and Uchiyama, 1996). R1, R2 and R9 are long visual fibres ending in the third optic ganglion, the medulla, whereas the other receptor fibres end in the second optic ganglion, the lamina.

The behavioural and ecological relevance of the red receptor have so far not been analysed. It is known that butterflies with a red receptor differ from other insects by choosing colours for oviposition that appear green to the human eye (Ilse, 1937; Kolb and Scherer, 1982). The spectral sensitivity of oviposition closely follows that of the green receptor. Females of other insect species that oviposit on green substrata, prefer yellow to green (Prokopy and Owens, 1983).

The goals of the present study are therefore threefold: (1) to describe colour choice in the context of oviposition in Papilio aegeus; (2) to develop a linear model describing the chromatic receptor interactions underlying this choice; and (3) to point out the ecological relevance of the P. aegeus red receptor in this context. I will investigate whether the behaviour is mediated by a chromatic rather than an achromatic mechanism. I use generalized linear models (McCullagh and Nelder, 1989) and a least-squares procedure to find the optimal fit of models to behaviour.

Experimental animals and arrangement

I collected caterpillars of the orchard butterfly Papilio aegeus Donovan 1805 (Lepidoptera: Papilionidae) from their larval foodplants (Citrus sp. and Choysia ternata) in gardens in Canberra, Australia, during the southern 1997 summer and reared them in the laboratory. Experiments were performed on adults and their offspring. Adult butterflies (20−30) were kept in an indoor flight cage (3 m×2 m×2 m) illuminated by fluorescent tubes from above under long-day conditions (14 h:10 h light:dark). Petri dishes marked with blue tape and filled with 10 % sugar solution were positioned on high stands and served as feeders. The conditions in the cage were such that animals fed and mated and had a lifespan of up to 2 months.

When the females were known to have mated, they were presented with a set of circular pieces of coloured paper (HKS N series, HKS Warenzeichenverband e.V., Sieglestraße25, D-70469 Stuttgart, Germany), 6 cm in diameter, as oviposition cues. The papers were connected to the stand carrying the feeder by wooden sticks (150 mm long, 2 mm in diameter) and positioned horizontally 20 cm above a background made from white paper (the reflectance of the background paper is shown in Fig. 1B). P. aegeus females do not depend on contact chemical stimuli for oviposition, and no influence of experience on the behaviour was found. Following an approach to a leaf, a female lays only a single egg on the underside. Since approaches are made from above, butterflies cannot see whether eggs are already present on the underside. Results of preliminary experiments showed that, under these experimental conditions, butterflies did not avoid papers already carrying eggs. Social facilitation or inhibition effects can therefore be excluded. The positions of the papers were changed several times during each experiment. Eggs were counted every day and this number is referred to as the number of ‘choices’. Approximately half the females in the cage were observed laying eggs on any day. Experiments were repeated with two different populations to ensure that at least ten females contributed to each data set.

Fig. 1.

Spectral distribution of (A) the cage illumination light intensity I(λ), (B) the reflectance S(λ) of several coloured papers and the background, and (C) the sensitivities R(λ) of the five Papilio aegeus receptor types (UV, V, B, G and R), according to Matic (1983) and Bandai et al. (1992). Receptor sensitivities are normalized by setting the integral to 1. The data are given with the spectral resolution dλ of 20 nm, which was used for modelling.

Fig. 1.

Spectral distribution of (A) the cage illumination light intensity I(λ), (B) the reflectance S(λ) of several coloured papers and the background, and (C) the sensitivities R(λ) of the five Papilio aegeus receptor types (UV, V, B, G and R), according to Matic (1983) and Bandai et al. (1992). Receptor sensitivities are normalized by setting the integral to 1. The data are given with the spectral resolution dλ of 20 nm, which was used for modelling.

Receptor quantum catches

The quantum catch Q of a receptor type i (where i=UV, V, B, G or R receptors) was calculated from the spectral distributions of the illuminating light I(λ), the reflectance spectrum of the coloured paper S(λ) and the spectral sensitivity of the receptor Ri(λ), as:
formula
The spectral resolution (dλ) was limited, by the receptor measurements, to 20 nm, for all values between 300 and 700 nm. The illumination curve (Fig. 1A) was measured with a calibrated spectrophotometer (Ocean Optics). The original reflection curves of the HKS papers (see Fig. 1B for examples) were measured (dλ=1 nm) and kindly provided by Robert Brandt, Martin Giurfa and Randolf Menzel (Freie Universität Berlin, Germany). The receptor sensitivities (Fig. 1C) were obtained from the literature. The curves presented in Fig. 3 of Matic (1983) were used for the red (R), green (G), blue (B) and violet (V) receptor types. Since Matic documented only four receptor types, the sensitivity curve of the related species P. xuthus (Bandai et al., 1992) was used for the UV receptor (UV).
Since receptors adapt to the background illumination (Laughlin, 1981), the quantum catch relative to the background qi was used as input to the models:
formula
where Qi and Qib are quantum catches for the colour and the background, respectively.

Linear modelling procedure

The modelling procedure follows the theory of generalized linear models (McCullagh and Nelder, 1989). The model formula has two parts, the linear predictor η and the link function F. In this particular case, colour attractiveness is assumed to be the result of a linear interaction between different receptor types. This is the linear predictor η:
formula
with xi representing the positive or negative coefficients of the respective receptor quantum catches. For P. aegeus this gives:
formula
where qU, qV, qB, qG and qR are the relative receptor quantum catches of the ultraviolet, violet, blue, green and red receptors, respectively. The choice frequency for one colour, Cc, is a function of the colour attractiveness, which is calculated from the relative quantum catches qi in the following way:
formula
The values of Cc lie between 0 (for a colour that is never chosen) and 1 (for a colour that is always chosen); the error follows a binomial distribution and only depends on the number of registered choices N. A logit function (equation 6) was used as the appropriate sigmoid link. Sigmoid link functions can deal with choice frequencies of 0, but choice frequencies close to 0 or close to 1 have little influence on the result since they are given weak weight. The results would be very similar if another sigmoid link function was chosen (see McCullagh and Nelder, 1989; Francis et al., 1993):
formula
For each possible combination of receptors a model was fitted with coefficients xi optimized using a least-squares algorithm. The scaled deviance S.D. was taken as measure for the goodness of fit of the models. According to McCullagh and Nelder (1989) it is calculated as
formula
where n denotes the number of variables, N is the number of choices, Y is the number of choices for the test colour (Y=CcN), and F is the model value (equation 6). If the full model has n1 variables and a reduced model has n2 variables, then the difference between the scaled variances S.D.(n1)−S.D.(n2) approximately follows a χ2 distribution with n1n2 degrees of freedom. The difference between the scaled deviance of the five-receptor model and the scaled deviance of each of the other models was tested. A second measure is the standard error S.E.M.x of a coefficient x. x/S.E.M.x is tested using a t-test. This measure has to be treated with caution if the variables are correlated. As a rule of thumb, a coefficient x about 3 times larger than its standard error is significant, and a value of x smaller than its standard error is usually insignificant (Francis et al., 1993).

Hypotheses and experimental design

The spectral preferences of butterflies for oviposition stimuli, as measured by Saxena and Goyal (1978), Kolb and Scherer (1982) and Scherer and Kolb (1987a,b), follow the spectral sensitivity of the green receptor. The green receptor quantum catch is the most probable achromatic cue controlling the behaviour. In a first series of experiments, I tested whether the choice behaviour in P. aegeus is indeed controlled solely by the green receptor. Seven different sets of three or six colours each were chosen in such a way that all colours of one set provided approximately the same green receptor quantum catch. The receptor quantum catches were calculated for all coloured papers to be used (Table 1, equations 1 and 2). All colours of a given set were presented simultaneously, and the number of eggs laid on the underside of each paper was recorded. If the choices were mediated by the green receptor only, choice distributions should be close to random in all experiments of the series. There are many other achromatic cues, including the other receptor quantum catches and the summed quantum catch in all receptors. An additional multiple choice experiment (Table 1, experiment 8) was designed as a preliminary test to determine whether any of these cues was controlling the behaviour.

Table 1.

Quantum catches relative to the background by the single receptors and summed quantum catches for the papers in all multiple choice experiments

Quantum catches relative to the background by the single receptors and summed quantum catches for the papers in all multiple choice experiments
Quantum catches relative to the background by the single receptors and summed quantum catches for the papers in all multiple choice experiments

The second series of experiments was designed to provide data for modelling. In 15 dual-choice experiments, the animals were presented with a fixed reference colour (HKS paper 60) and a test colour. At least 200 choices were registered in each experiment. The choice frequency for the test colour was used to calculate the cofficients in the models. The predictions made by the models were compared with the choices made by the animals, in the dual-choice experiments and in the multiple-choice experiments.

Measurements of leaf reflectance and caterpillar growth

In Canberra, Choysia ternata is a major food source for P. aegeus. To a human observer, young leaves appear light green, medium-old leaves dark green and old leaves yellowish green, as in many other Rutaceae species. The reflectances of three young (a few weeks old), three medium-old leaves (from the previous vegetation period) and three old leaves (1–2 years old) were measured in springtime with a spectrometer (Ocean Optics; see Fig. 6). The attractiveness of the colours was evaluated using the model developed from the experiments with coloured papers.

Freshly hatched caterpillars were released on young, medium-old and old leaves. Two experiments were performed to determine whether the growth rates of the caterpillars depend on the age of leaves. In the first experiment, the caterpillars were kept on the leaves in closed containers and could not move to leaves of other ages. Fresh leaves were provided daily. I measured the length and noted the larval instar of each caterpillar every day for 21 days. The day at which 50 % of the caterpillars had reached a larval instar was determined. In the second experiment, the caterpillars were placed on small branches carrying both young and old leaves. The distance between the old and young leaves was between 5 and 15 cm. In this experiment, in addition to the length and larval instar, the day of pupation was noted for each caterpillar. The temperature was maintained between 20 °C and 22 °C, and experiments with different groups were started on the same day.

Colour choices are not mediated by the green receptor only

Fig. 2A summarizes the results of the seven multiple-choice experiments with papers matched in intensity for the green receptor. With the exception of experiment 7, the papers used in each experiment have very similar green receptor quantum catches. Choice distributions differed from random (hatched lines) in all experiments (G-tests, P<0.0001). In each experiment, the female butterflies chose one or two of the colours much more frequently than the remaining ones. Table 1 shows that the colours that were avoided had either higher blue or red receptor quantum catches than the preferred ones.

Fig. 2.

(A) Green receptor quantum catches versus choices of stimuli for oviposition in seven multiple-choice experiments. The abscissa gives the green receptor quantum catches relative to the background qgreen, 1 indicating a quantum catch equal to that of the background. The ordinate gives the percentage of choices made within each experiment. The experiment number is given above the graph, and the experiments are separated by pink lines. Expected random-choice frequencies are indicated by dashed lines. The numbers of the papers of the HKS series are given beside the coloured disks surrounding the black squares, which are made as similar as possible to the actual paper colour. (B-G) Results from experiment 8. Choice frequencies are given as functions of six different achromatic cues: all single receptor quantum catches qi (B-F) and the summed quantum catch ∑qi (G).

Fig. 2.

(A) Green receptor quantum catches versus choices of stimuli for oviposition in seven multiple-choice experiments. The abscissa gives the green receptor quantum catches relative to the background qgreen, 1 indicating a quantum catch equal to that of the background. The ordinate gives the percentage of choices made within each experiment. The experiment number is given above the graph, and the experiments are separated by pink lines. Expected random-choice frequencies are indicated by dashed lines. The numbers of the papers of the HKS series are given beside the coloured disks surrounding the black squares, which are made as similar as possible to the actual paper colour. (B-G) Results from experiment 8. Choice frequencies are given as functions of six different achromatic cues: all single receptor quantum catches qi (B-F) and the summed quantum catch ∑qi (G).

Fig. 2B–G summarizes the results of multiple-choice experiment 8. The choice frequencies are given as functions of each single (Fig. 2B–F) and the summed (Fig. 2G) receptor quantum catches. No clear pattern can be observed, indicating that colour choice for oviposition is probably not guided by an achromatic mechanism. Despite this hint, I tested whether achromatic models could explain the behaviour in dual-choice experiments. All colours used in these experiments are indicated by bold letters in Table 1.

Models of the behaviour in dual-choice experiments

The choice frequencies for each colour tested against the reference colour (HKS paper 60) were used to model the receptor interactions underlying colour choice. Using generalized linear models and a least-squares procedure, coefficients were calculated for all possible combinations of 1–5 receptor types (see Introduction). Table 2 gives the coefficients xi, standard errors of coefficients (S.E.M.xi) and the scaled deviance (s.D.) of all these models. None of the single-receptor models fits the data. In all other models, opponent interactions are involved. A chromatic mechanism thus mediates the choice behaviour.

Table 2.

Coefficients (xi), and standard errors of xi (S.E.M.xi), for each receptor in models for all combinations of 1-5 receptor types, and scaled deviance (s.D.) for all models

Coefficients (xi), and standard errors of xi (S.E.M.xi), for each receptor in models for all combinations of 1-5 receptor types, and scaled deviance (s.D.) for all models
Coefficients (xi), and standard errors of xi (S.E.M.xi), for each receptor in models for all combinations of 1-5 receptor types, and scaled deviance (s.D.) for all models

According to statistics, the U,V,G,R model seems to provide the best description of the data, the scaled deviance not being significantly different (χ2 test, P>0.1, see Materials and methods for details) from that of the full (UV,V,B,G,R) model. However, other factors have to be taken into account. In the ultraviolet region not much light was available, nor did the papers show large differences in reflectance (Fig. 1A,B). The data are therefore inadequate to distinguish clearly the influence of the ultraviolet and violet receptors. More generally, at least three receptor types contribute to the behaviour: the green receptor with a positive sign, the red receptor with a negative sign and at least one of the three short-wavelength (ultraviolet, violet and blue) receptors with a negative sign.

For reasons of parsimony, I present the best three-receptor model in more detail. Of the three-receptor models, the B,G,R model fits the data best although it differs significantly from the five-receptor model (χ2 test, P<0.001). Fig. 3 shows the choice frequencies made by the butterflies together with the model prediction (red line) as a function of the linear predictor η. η is a measure of the attractiveness of a colour, and results from the linear interaction given in the inset (see equation 3 and Table 2). One important common feature of the model and the butterflies’ choice behaviour is that colours as different as HKS papers 50 and 11 (light-blue and pink for humans) are treated in a very similar way, i.e. they are avoided.

Fig. 4 shows how six different models fit the experimental data. These include the five-receptor model (Fig. 4A), the best four-receptor model (Fig. 4B), the best three-receptor model (Fig. 4C) and three two-receptor models, which result from omitting one of the receptors in the B,G,R model (Fig. 4D–F). The latter obviously do not describe the experimental data well (scaled deviances of 367.6, 530.3 and 965.6). Thus, the most parsimonious solution is the B,G,R model (Figs 3, 4G) in which the scaled deviance was only 85.4.

Fig. 3.

The best-fitting three-receptor model (the B,G,R model). The abscissa gives the outcome of the receptor interaction for a colour (colour attractivity, linear predictor η; see equation 3). The receptor interaction underlying the model is given in the inset; the coefficients are normalized to 1 for the green receptor. Black squares surrounded by coloured disks depict the choice frequencies for a colour in the dual-choice experiment, with HKS paper 60 as the reference colour. The red line is the fitted model. Light-blue (50) and pink (11) colours lie close together, in one-dimensional chromatic space. Note that the linear part of the curve fits very well. Arrows mark the calculated η and predicted attractivity of young (y), medium old (m) and old (o) leaves of the larval food plant. For further explanation, see text.

Fig. 3.

The best-fitting three-receptor model (the B,G,R model). The abscissa gives the outcome of the receptor interaction for a colour (colour attractivity, linear predictor η; see equation 3). The receptor interaction underlying the model is given in the inset; the coefficients are normalized to 1 for the green receptor. Black squares surrounded by coloured disks depict the choice frequencies for a colour in the dual-choice experiment, with HKS paper 60 as the reference colour. The red line is the fitted model. Light-blue (50) and pink (11) colours lie close together, in one-dimensional chromatic space. Note that the linear part of the curve fits very well. Arrows mark the calculated η and predicted attractivity of young (y), medium old (m) and old (o) leaves of the larval food plant. For further explanation, see text.

Fig. 4.

The fit of six generalized linear models to the experimental data. The abscissae give the linear predictor η, which is the receptor interaction indicated in each graph (see also equation 3). The ordinate gives the choice frequencies made by the animals. Results from the dual-choice experiments are indicated by crosses, the model values are shown by the line. Arrows point to the predicted η for young (y), medium-old (m) and old (o) leaves. The five-receptor model (A) and the four-receptor model (B) do not differ significantly (scaled deviances not significantly different; see Table 2, χ2 test, P>0.1). The best three-receptor model (C), presented in more detail in Fig. 3, differs significantly from the five-receptor model (χ2 test, P<0.001). It does not differ significantly from the four-receptor models that include an additional short-wavelength receptor (Table 2, P>0.05). The three two-receptor models (D-F) do not fit the data adequately. U, V, B, G, R are ultraviolet, violet, blue, green and red receptors, respectively.

Fig. 4.

The fit of six generalized linear models to the experimental data. The abscissae give the linear predictor η, which is the receptor interaction indicated in each graph (see also equation 3). The ordinate gives the choice frequencies made by the animals. Results from the dual-choice experiments are indicated by crosses, the model values are shown by the line. Arrows point to the predicted η for young (y), medium-old (m) and old (o) leaves. The five-receptor model (A) and the four-receptor model (B) do not differ significantly (scaled deviances not significantly different; see Table 2, χ2 test, P>0.1). The best three-receptor model (C), presented in more detail in Fig. 3, differs significantly from the five-receptor model (χ2 test, P<0.001). It does not differ significantly from the four-receptor models that include an additional short-wavelength receptor (Table 2, P>0.05). The three two-receptor models (D-F) do not fit the data adequately. U, V, B, G, R are ultraviolet, violet, blue, green and red receptors, respectively.

As a simple test of the model, I calculated the choice frequencies predicted by the B,G,R model for the multiple-choice experiments. In Fig. 5, the predicted choices are compared with those actually made by the butterflies in the experiments. The model correctly predicts the choice behaviour of the animals (r2 of the fitted regression line is 0.933). The full UV,V,B,G,R model describes the multiple choice data only slightly better (r2=0.946). Leaving out any of the three receptors in the B,G,R model, however, results in a much worse fit (r2=0.28 for the B,G model, r2=0.21 for the B,R model and r2=0.01 for the B,R model).

Fig. 5.

Model predictions for the multiple-choice experiments. The choice frequencies made by the animals in multiple-choice tests are given as a function of the choice frequencies predicted by the B,G,R model presented in Fig. 3. The equation for the linear regression line is y=−0.03+1.1x (r2=0.933, P<0.0001). The slope is not significantly different from the expected slope of 1.0 (P>0.4).

Fig. 5.

Model predictions for the multiple-choice experiments. The choice frequencies made by the animals in multiple-choice tests are given as a function of the choice frequencies predicted by the B,G,R model presented in Fig. 3. The equation for the linear regression line is y=−0.03+1.1x (r2=0.933, P<0.0001). The slope is not significantly different from the expected slope of 1.0 (P>0.4).

Leaves and caterpillars

The spectral reflectance curves of three young, three medium-old and three old leaves are presented in Fig. 6. Young leaves have a high but narrow reflectance peak around 540 nm. The reflectance decreases with age, due to increased amounts of light-absorbing chlorophyll. In old leaves, the reflectance at 540 nm has increased again, and there is also increased reflectance at shorter and longer wavelengths, giving a broader peak. All leaves reflect little light at wavelengths shorter than 400 nm and large amounts of light at wavelengths longer than 700 nm (in addition, chlorophyll has a fluorescence at 720 nm).

Fig. 6.

Spectral reflectance curves of young, medium-old and old leaves of Choysia ternata. Almost no light of wavelength shorter than 400 nm is reflected. Colours in this figure do not match the actual colours of the leaves.

Fig. 6.

Spectral reflectance curves of young, medium-old and old leaves of Choysia ternata. Almost no light of wavelength shorter than 400 nm is reflected. Colours in this figure do not match the actual colours of the leaves.

Caterpillars raised on young leaves grew much faster than those on old or medium-old leaves (Fig. 7A). When given the opportunity to move to other leaves, most caterpillars released on old leaves moved to young leaves within the first days. They showed a tendency to be smaller, reached larval instars later than those released on young leaves (Fig. 7B) and pupated later than caterpillars released on young leaves (Fig. 7C).

Fig. 7.

Growth and pupation of caterpillars raised on young, medium-old and old leaves of Choysia ternata. (A) Caterpillar length. Caterpillars raised on young leaves were significantly longer than those raised on old leaves, from day 7 onwards, and longer than those raised on medium-old leaves from day 9 onwards (t-tests, P<0.01). Values are means ± S.D. The day on which 50 % of the larvae reached a particular larval instar is also indicated (numbered larval stage inset into data points). 15 caterpillars were raised on each leaf type. (B) Length of caterpillars released onto young or old leaves of a branch bearing at least three young and three old leaves. The caterpillars could walk freely to other leaves of the branch. New branches were provided when necessary. Caterpillars released on young leaves (N=25) reached the different larval instars (number insets) 2 to 4 days earlier than those released on old leaves (N=50) but the difference in length was not significant (t-test, P>0.1). (C) Pupation of the same caterpillars. Caterpillars released on young leaves pupated earlier. Asterisks mark significant differences between the percentage of pupated animals (G-tests; *P<0.01, **P<0.001).

Fig. 7.

Growth and pupation of caterpillars raised on young, medium-old and old leaves of Choysia ternata. (A) Caterpillar length. Caterpillars raised on young leaves were significantly longer than those raised on old leaves, from day 7 onwards, and longer than those raised on medium-old leaves from day 9 onwards (t-tests, P<0.01). Values are means ± S.D. The day on which 50 % of the larvae reached a particular larval instar is also indicated (numbered larval stage inset into data points). 15 caterpillars were raised on each leaf type. (B) Length of caterpillars released onto young or old leaves of a branch bearing at least three young and three old leaves. The caterpillars could walk freely to other leaves of the branch. New branches were provided when necessary. Caterpillars released on young leaves (N=25) reached the different larval instars (number insets) 2 to 4 days earlier than those released on old leaves (N=50) but the difference in length was not significant (t-test, P>0.1). (C) Pupation of the same caterpillars. Caterpillars released on young leaves pupated earlier. Asterisks mark significant differences between the percentage of pupated animals (G-tests; *P<0.01, **P<0.001).

For comparison with the experimental data, the receptor quantum catches for the leaves were calculated in the same way and in the same conditions as for the coloured papers. The arrows in Figs 3 and 4 show the attractiveness of the different leaves, as predicted by the different models. In to the B,G,R model (Figs 3, 4C), young leaves are more attractive than any of the coloured papers (mean ± S.D. predicted choice frequency versus HKS paper 60 is 81±4 %) and more attractive than medium-old (predicted choice frequency 60±9 %) or old leaves (predicted choice frequency 63±4 %). Thus, the B,G,R model predicts that a female butterfly will choose those leaves for oviposition which guarantee rapid development of caterpillars. Omission of any one of the three receptors in the B,G,R model results in much poorer discrimination, especially between young and old leaves (Fig. 4D–F). The BG model, which is the best two-receptor model, predicts choice frequencies of 71±3 % for young leaves and 65±4 % for old leaves (Fig. 4D).

Colour choices made by butterfly females can be described accurately using generalized linear models that assume a single linear opponent (chromatic) interaction of three spectral types of photoreceptors: a colour is attractive to an ovipositing female if it gives rise to a high green receptor quantum catch and low red and blue (and/or ultraviolet/violet) receptor catches. The model predicts that a female should lay her eggs on young leaves, which are indeed the best food for a young caterpillar.

The models presented here are based on multiple regressions. How well is this statistical method suited for receptor data? As can be seen from the receptor sensitivity curves (Fig. 1C), all receptor types have large areas of overlap. The largest areas of overlap occur between the three short-wavelength receptors: UV, V and B. This results in a high colinearity of the receptor quantum catches by these receptors, and thus large variances in the estimated partial regression coefficients (Freund and Minton, 1979). The high standard errors of the coefficients for the three short-wavelength receptors in the five-receptor model (Table 2) suggest over-determination of the system.

Multiple regression is a useful statistical tool which measures the significance of the influence of each of the tested variables. However, colours have to be chosen in a way to present high and independent variations in all receptor quantum catches. This is difficult in the short-wavelength range since all three short-wavelength receptors of Papilio spp. have large areas of overlap.

Which receptors in the ommatidium of Papilio spp. are involved?

At least one of the short-wavelength receptors contributes to the observed colour choice behaviour. In an ommatidium of Papilio spp., the proximal receptors R1 and R2 (blue-, violet-or ultraviolet-sensitive) and the basal R9 (red-or green-sensitive) are long visual fibres and have been assumed to serve as colour-coding systems (see Arikawa and Uchiyama, 1996). My model predicts that R1 and/or R2 are involved in a chromatic mechanism. It remains to be seen which of the green and red receptors are involved. R9 is sensitive to the same polarization direction of light as R1 and R2 (dorso-ventral or vertical; Ribi, 1987; Bandai et al., 1992). If R9 was the only green/red receptor involved, no interference of polarization and colour cues would be expected. However, recent work suggests that colour choice behaviour is sensitive to the polarization direction of light (A. Kelber, unpublished data). Thus, R9 is unlikely to be the only red-or green-sensitive receptor involved in chromatic interactions.

A red receptor to look at green leaves

Green leaves are among the most common objects in nature. The reflectance spectra of all photosynthesizing leaves are similar: little reflection occurs below 500 nm, there is a peak around 540 nm, a shoulder towards longer wavelengths, decreased reflection around 650 nm, and fluorescence and high reflection in the infrared (Fig. 6). Surprisingly, little is known about the way in which green leaves are coded by different colour vision systems. Prokopy and Owens (1983) proposed that to discriminate green leaves from any yellowish or reddish substratum, a red-sensitive receptor is needed. Similarly, Mollon (1989) and Osorio and Vorobyev (1996) proposed that the evolution of the primate red-green receptor system has been driven by the need to discriminate yellow and red fruits from green leaves.

Lythgoe (1979) suggested that green leaves differ mainly in the red region of the spectrum and do not reflect much light at wavelengths shorter than 500 nm. He concluded that a useful strategy would be to compare red and green receptor signals to derive information about the ratio of chlorophyll and red pigments in a leaf, and to use a blue receptor as a reference for the overall amount of pigments present (Lythgoe, 1979). More recently, Lythgoe and Partridge (1989) and Chittka (1996) have measured the spectral reflectances of a large number of leaves and investigated the optimal set of receptors to discriminate as many of them as possible. For two-receptor sets, Lythgoe and Partridge (1989) found optimal discrimination with one pigment sensitive in the blue (420−450 nm) and another sensitive in the green (510−520 nm) region. Chittka (1996) assumed that the ultraviolet, blue and green receptor set of honeybees and other insects evolved to discriminate green leaves optimally. Osorio and Vorobyev (1996) also found that primate trichromacy did not enhance discrimination between different leaves.

Even with an optimal receptor set, the choice between different plant species cannot be made using colour vision alone. There are simply too many similar plants. The choice between plant species is mainly made by means of chemical cues (for a review, see Renwick and Chew, 1994). It is unlikey that insect eyes have been optimized to discriminate as many different species of plants as possible by means of leaf colour. However, as the present study shows, the choice of a suitable leaf within a plant can be made using colour vision. This allows ovipositing females to avoid landings on unsuitable leaves. As proposed by Lythgoe (1979), this discrimination is possible using one receptor sensitive to those wavelengths predominantly reflected by the leaf (green), and two others sensitive at longer (red) and shorter (blue) wavelengths. Given the similarity of the reflectance curves, the B,G,R model predicts surprisingly good discrimination between leaves of different age (see Fig. 6, arrows in Figs 3 and 4). Field experiments have yet to reveal to what extent the light environment, the colours of natural backgrounds and mirror reflections by shiny leaves influence discrimination between similarly green leaves. A recent study on Macaca fascicularis suggests that in primates, the red receptor is also involved in the discrimination of palatable leaves within a plant (Lucas et al., 1998).

Red receptors seem to be the rule rather than the exception in butterflies (Bernard, 1979; Kinoshita et al., 1997). Bernard and Remington (1991) argue that in butterfly females of the genus Lycaena, a red-sensitive pigment (absorption maximum at 568 nm) is needed for the discrimination of red larval food plants. Most Hymenoptera (bees, wasps and ants) do not have red receptors. They are not interested in green leaves as food for themselves or their offspring. For the discrimination of flower colours, the ultraviolet, blue and green receptor set has been demonstrated to be very powerful (Menzel and Shmida, 1993; Chittka, 1996), a red receptor not being required for this task. One group of hymenopteran insects, sawflies of the genus Tenthredo, oviposit on green leaves, and they are indeed known to have a red receptor (Peitsch et al., 1992).

An evolutionary perspective

It will be rewarding to test whether the modelling procedure developed here can be applied to other data sets and therefore be of wider use for the description of chromatic mechanisms in insects. This might lead to an interesting perspective. Spontaneous behaviours are often guided by chromatic mechanisms. These work as filters matched to the specific ecological needs of a species, such as discrimination between sky and earth, between the solar and antisolar halves of the sky, between flowers and green leaves, between male and female conspecifics, etc (for a review, see Menzel, 1979). Colour vision may have evolved from the combination of primarily independent chromatic mechanisms like the one described in the present study. Thus, as Menzel (1979) speculated, chromatic mechanisms “may give us clues about the evolution of the sensory and interneuron mechanisms of color vision in invertebrates”. It will be interesting to study the mechanisms underlying spontaneous behaviours in more primitive arthropod species in order to test this hypothesis.

I am extremely grateful to Gert Stange who started it all and contributed many ideas throughout the project. I would also like to thank Shao-Wu Zhang and Mandyam Srinivasan for the invitation to Canberra, Andrew James for his most helpful introduction to generalized linear models, Gerlinde Lenz for help with the figures, Michael Pfaff for patient help breeding the animals and Misha Vorobyev for inspiring discussions. The suggestions of Misha Vorobyev, Kentaro Arikawa, Eric Warrant, Ron Rutowski and two referees greatly improved the manuscript. The project was jointly funded by the RSBS and the Alexander von Humboldt-Foundation, which deserves special thanks for the unique and always excellent support. The paper was completed at the Zoological Institute of Lund University with the help of a Postdoctoral grant from the Swedish Natural Sciences Research Council, which is gratefully acknowledged.

Arikawa
,
K.
and
Uchiyama
,
H.
(
1996
).
Red receptors dominate the proximal tier of the retina in the butterfly Papilio xuthus
.
J. Comp. Physiol. A
178
,
55
61
.
Bandai
,
K.
,
Arikawa
,
K.
and
Eguchi
,
E.
(
1992
).
Localization of spectral receptors in the ommatidium of butterfly compound eye determined by polarization sensitivity
.
J. Comp. Physiol. A
171
,
289
297
.
Bernard
,
G. D.
(
1979
).
Red-absorbing visual pigment of butterflies
.
Science
203
,
1125
1127
.
Bernard
,
G. D.
and
Remington
,
C. L.
(
1991
).
Color vision in Lycaena butterflies: spectral tuning of receptor arrays in relation to behavioural ecology
.
Proc. Natl. Acad. Sci. USA
88
,
2783
2787
.
Brandt
,
R.
and
Vorobyev
,
M.
(
1997
).
Metric analysis of increment threshold sensitivity in ther honeybee
.
Vision Res.
37
,
425
439
.
Chittka
,
L.
(
1996
).
Optimal sets of receptors and color opponent systems for coding of natural objects in insect vision
.
J. Theor. Biol.
181
,
179
196
.
De Valois
,
R. L.
and
De Valois
,
K. K.
(
1997
).
Neural coding of color
. In
Readings on Color, Vol. 2: The Science of Color
(ed.
A.
Byrne
and
D. R.
Hilbert
), pp.
93
140
.
Cambridge, London
:
Bradford
.
Francis
,
B.
,
Green
,
M.
and
Payne
,
C.
(
1993
).
GLIM4. The Glim System: Generalized Linear Interactive Modelling.
Oxford
:
Oxford University Press
.
Freund
,
R. J.
and
Minton
,
P. D.
(
1979
).
Regression Methods. A Tool for Data Analysis.
New York
:
Marcel Dekker Inc
.
Ilse
,
D.
(
1937
).
New observations on responses to colours in egglaying butterflies
.
Nature
140
,
544
545
.
Kelber
,
A.
and
Pfaff
,
M.
(
1999
).
True colour vision in the Orchard butterfly, Papilio aegeus
.
Naturwissenschaften
86
,
221
224
.
Kinoshita
,
M.
,
Sato
,
M.
and
Arikawa
,
K.
(
1997
).
Spectral receptors of nymphalid butterflies
.
Naturwissenschaften
84
,
199
201
.
Kinoshita
,
M.
,
Shimada
,
N.
and
Arikawa
,
K.
(
1999
).
Colour vision of the foraging swallowtail butterfly Papilio xuthus
.
J. Exp. Biol.
202
,
95
102
.
Kolb
,
G.
and
Scherer
,
C.
(
1982
).
Experiments on wavelength specific behavior of Pieris brassicae L. during drumming and egg-laying
.
J. Comp. Physiol.
149
,
325
332
.
Laughlin
,
S. B.
(
1981
).
Neural principles in the peripheral visual system on invertebrates
. In
Handbook of Sensory Physiology
, vol.
VII/6B
(ed.
H.
Autrum
), pp.
133
280
.
Berlin, Heidelberg
,
New York
:
Springer
.
Lee
,
B. B.
(
1998
).
Receptors, channels and color in primate retina
. In
Color Vision. Perspectives from Different Disciplines
(ed.
W.
Backhaus
,
R.
Kliegl
and
J. S.
Werner
), pp.
79
88
.
Berlin, New York
:
Walter de Gruyter
.
Lucas
,
P. W.
,
Darvell
,
B. W.
,
Lee
,
P. K. D.
,
Yuen
,
T. D. B.
and
Choong
,
M. F.
(
1998
).
Colour cues for leaf food selection by long-tailed macaques (Macaca fascicularis) with a new suggestion for the evolution of trichromatic colour vision
.
Folia Primatol.
69
,
139
152
.
Lythgoe
,
J. N.
(
1979
).
The Ecology of Vision
.
Oxford
:
Clarendon Press
.
Lythgoe
,
J. N.
and
Partridge
,
J. C.
(
1989
).
Visual pigments and the acquisition of visual information
.
J. Exp. Biol.
146
,
1
20
.
Matic
,
T.
(
1983
).
Electrical inhibition in the retina of the butterfly Papilio. I. Four spectral types of photoreceptors
.
J. Comp. Physiol.
152
,
169
182
.
McCullagh
,
P.
and
Nelder
,
J. A.
(
1989
).
Generalized Linear Models. Second edition. London
,
New York
:
Chapman and Hall
.
Menzel
,
R.
(
1979
).
Spectral sensitivity and color vision in invertebrates
. In
Handbook of Sensory Physiology
, vol.
VII/6A
(ed.
H.
Autrum
), pp.
503
580
.
Berlin, Heidelberg, New York
:
Springer
.
Menzel
,
R.
and
Backhaus
,
W.
(
1991
).
Colour vision in insects
. In
The Perception of Colour
(ed.
P.
Gouras
), pp.
262
293
.
London
:
Macmillan
.
Menzel
,
R.
and
Shmida
,
A.
(
1993
).
The ecology of flower colours and the natural colour vision of insect pollinators: The Israeli flora as a study case
.
Biol. Rev.
68
,
81
120
.
Mollon
,
J. D.
(
1989
).
“Tho’ she kneel’d in that place where they grew…”: The uses and origins of primate colour vision
.
J. Exp. Biol.
146
,
21
38
.
Osorio
,
D.
and
Vorobyev
,
M.
(
1996
).
Colour vision as an adaptation to frugivory in primates
.
Proc. R. Soc. Lond. B
263
,
593
599
.
Peitsch
,
D.
,
Fietz
,
A.
,
Hertel
,
H.
,
de Souza
,
J.
,
Ventura
,
D. F.
and
Menzel
,
R.
(
1992
).
The spectral input systems of hymenopteran insects and their receptor-based colour vision
.
J. Comp. Physiol. A
170
,
23
40
.
Prokopy
,
R. J.
and
Owens
,
E. D.
(
1983
).
Visual detection of plants by herbivorous insects
.
Annu. Rev. Entomol.
28
,
337
364
.
Renwick
,
J. A. A.
and
Chew
,
F. S.
(
1994
).
Oviposition behaviour in Lepidoptera
.
Annu. Rev. Entomol.
39
,
377
400
.
Ribi
,
W.
(
1987
).
Anatomical identification of spectral receptor types in the retina of the Australian Orchard butterfly, Papilio aegeus D
.
Cell Tissue Res.
247
,
393
407
.
Saxena
,
K. N.
and
Goyal
,
S.
(
1978
).
Host-plant relations of the citrus butterfly Papilio demoleus L.: Orientational ovipositional responses
.
Ent. Exp. Appl
.
24
,
1
10
.
Scherer
,
C.
and
Kolb
,
G.
(
1987a
).
Behavioural experiments on the visual processing of color stimuli in Pieris brassicae L
.
(Lepidoptera). J. Comp. Physiol. A
160
,
645
656
.
Scherer
,
C.
and
Kolb
,
G.
(
1987b
).
The influence of color stimuli on visually controlled behavior in Aglais urticae L. and Pararge aegeria L
.
(Lepidoptera). J. Comp. Physiol. A
161
,
891
898
.
Swihart
,
S. L.
(
1970
).
The neural basis of colour vision in the butterfly, Papilio troilus
.
J. Insect Physiol.
16
,
1623
1636
.
Vorobyev
,
M.
and
Brandt
,
R.
(
1997
).
How do insect pollinators discriminate colours?
Israel J. Plant Sci.
45
,
103
113
.