Spectral Sensitivity of Photoreceptors and Colour Vision in the Solitary Bee, Osmia Rufa

Colour vision studies in insects have so far been limited to a few species, and only the honeybee’s colour vision system has been analysed in any detail at receptor, interneurone and behavioural levels (see Menzel, 1979, 1985a). This provokes the question of whether other flower-visiting insect species are equipped with a similar colour vision system or whether species-specific adaptations lead to different colour vision systems. Flower-visiting hymenopterans are of particular interest with regard to this question since they span a large evolutionary scale (from primitive wasps to apoide bees), live in very different habitats (from dense tropical rain-forest to alpine regions and northern tundras), and apply different strategies in searching for food (ranging from oligolectic specialists to polylectic generalists) (Kevan & Baker, 1983). We report in another paper that the stingless bee Melipona quadrifasciata has a trichromatic colour vision quite similar to that of the honeybee but that the ultraviolet and blue receptors are shifted to longer wavelengths. Other social hymenopterans such as bumblebees and wasps are equipped with the same spectral receptor types as the honeybee (Fietz, Hertel & Menzel, 1986; R. Menzel, A. Fietz, D. Peitsch & D. Ventura, unpublished observations) and discriminate colours as well as the honeybee (Beier & Menzel, 1972; R. Menzel & D. Ventura, unpublished observations).

We report here that the solitary bee Osmia rufa discriminates colours even better than the honeybee, and is equipped with a trichromatic colour vision system similar to that of Apis mellifera, four species of Bombus (B. terrestris, B. lapidarius, B. monticola, B. jonellus) and two wasp species (Paravespula germanica, P. vulgaris). Model calculations based on the properties of the receptors are used to define a quantitative measure of the discrimination between a pair of colour signals (Backhaus & Menzel, 1987). This measure - the just noticeable different steps (jnd) - correlates very well with the behavioural results from colour discrimination tests.

Conventional techniques were used to record intracellularly from photoreceptors in the compound eye of Osmia. The spectral sensitivity S(λ) was determined by a voltage-clamp technique, which enabled the establishment of a S(λ)-function within 16 s and a spectral resolution of 4nm (Menzel et al. 1986). A grid monochromator was used to scan the spectrum from 300 to 700 nm (and vice versa) and a circular neutral-density wedge was used to adjust the light flux, with the help of a feedback system, in such a way that the response of the cell was clamped to a preselected receptor potential. The receptor potential was clamped to 3 –6 mV above resting potential (about −40 mV) in most of our recordings. A computer calculated and displayed on-line the S(λ)-function from the settings of the neutral density wedge for 100 wavelengths between 300 and 700nm.

Osmia rufa is a solitary bee, which breeds in holes in wood. Fabre (1925) described a method for housing Osmia and other solitary bees in stems of bamboo, and this is the method that was used here. Osmia builds its nest in such a stem, and provides the brood cells with pollen and nectar by flying in and out for 3–4 weeks in late spring. The bees can be trained to colour marks which surround the entrance to the bamboo stem (Steinmann, 1981). Learning of the colour marks has not yet been studied in detail, but after 1 day of training no improvement of choice behaviour was observed and, therefore, tests were performed after the second day of training. Fig. 1 shows the experimental arrangement. During training the bee had open access to the nest through the centre hole (diameter 10 mm) of the vertically arranged colour mark (diameter 70 mm). The colour mark was a 1mm thick glass filter fixed on top of cardboard painted with silver paint or on top of coloured cardboard. The filters used were 1 mm thick Schott filters (Schott & Gen. Mainz; see Table 1). Such colour stimuli permitted the testing of colour discrimination in the ultraviolet without self-luminant stimuli (Menzel & Lieke, 1983). During the test, two colour marks were placed side by side and equally distant from the former position of the colour mark. Two empty and fresh bamboo stems were inserted in the centre hole, and the stem with the nest was removed. The bees approached the colour marks, and landings were counted as choices; the test was completed after 10 landings had been counted. At the next test the same colour pairs were interchanged in position. Special care was taken to remove odour marks on the filters, and only freshly cleaned filters were used during the test. The surrounding background was the wall of a wooden shelf in front of which many bamboo stems were arranged as nests for Osmia. Since Osmia flies out only on sunny days the bright light conditions were similar for all experiments. The same bee could be used in another series of experiments after it had been trained to the new colour signal for at least 1 day. No differences were found between initial trainings and reversal trainings.

The resting potential (−40 to −50mV), the depolarizing light response (up to 60 mV), and the phasic-tonic time course of the graded light response resembled the typical receptor characteristics of fast insect photoreceptors. Discrete potential fluctuations (quantum bumps) appeared at very low light intensities, but they never exceeded 0·5 mV. The V/logl-function (slope 0·7) was independent of the wavelength of the stimulating light and of the receptor type. The average spectral sensitivity, S (λ), of the three spectral receptor types is given in Fig. 2. S(λ) of the ultraviolet receptor (λ_max = 348 nm) is in accordance with the spectral absorption of the ultraviolet photopigment in other insect species (e.g. Ascalaphus, Hamdorf, Paulsen & Schwemer, 1973). Sensitivity above 430 nm is very low - less than 1 % between 430 and 620 nm and 1-2 % between 620 and 700 nm. The small increase at long wavelengths may be an artefact resulting from the secondary wavelength bands of the grid monochromator at half the wavelength of the main band for which the ultraviolet band is most sensitive. The blue receptors (λ_max = 436nm) follow the theoretical absorption function of a rhodopsin pigment with the corresponding λ_max. The hump around 370 nm probably originates from the β-peak of the photopigment, since a model calculation precisely describes the short wavelength part of the S(λ) if one assumes an addition of 10 % of a β-absorption at 380 nm to the Dartnall-function of the rhodopsin. The λ_max of the β-absorption has been calculated from the green receptors of both Osmia and Apis. In Apis, the broadening of the blue receptor sensitivity at shorter wavelengths can also be modelled accurately if one assumes the contribution of such a β-absorption. The enhanced sensitivity of the blue receptor at longer wavelengths (⩾520 nm) is more difficult to interpret. Since recording from blue receptors is very difficult, we consider this to be an artefact due to electrical coupling to the green receptors. This interpretation is supported by the finding of one blue receptor without any enhanced long-wavelength sensitivity. Since the source of the increased green sensitivity of the blue receptors is unknown, we have used the measured S(λ)-function (not the theoretical function) in the receptor model (see below). The same procedure is followed for the other two receptor types. The green receptors (λ_max = 572nm) are easier to record from. S(λ) follows a theoretical absorption function of a rhodopsin pigment with the corresponding λ_max. The β-sensitivity is very flat with a peak around 380 nm. The main peak spectral sensitivity at 572 nm is at much longer wavelengths than in the green receptors of the honeybee (λ_max = 540nm).

Osmia is equipped with three spectral receptor types - ultraviolet, blue and green. These provide its nervous system with the necessary information required for colour vision. How is this information used? To answer this question we tested colour vision by training single animals at the entrance of their nests. The colour signals used in the training experiments are listed in Table 1. In the following only the running numbers given in Table 1 are used to indicate the filter or filter/cardboard combinations. Table 1 also gives the coordinates of the colour loci of the respective colour signals. The colour loci were calculated by a procedure which follows that of Cornsweet (1970) and Rushton (1972), and which takes the three spectral receptor types with their particular spectral sensitivity (see Fig. 2) as being a special set of primary colours. These define an orthogonal three dimensional space, and the three orthonormal basis vectors of the colour space correspond to the tristimulus values of unit amount. Each colour signal is defined by a vector (Rodiek, 1973). The colour vector is obtained by vector addition from the components, i.e. the relative photon fluxes U, B and G absorbed by each of the receptors. These U, B and G values are given in Table 1 together with a measure of the subjective brightness (H), which is assumed to be equal to the sum of the tristimulus values. The vectors obviously depend on the spectral reflection of the filter/cardboard combination and the spectral distribution of the illuminating natural light. These factors are taken into account in our conclusions. (The procedure for the calculation is described in more detail in Menzel, 1985a, and Backhaus & Menzel, 1987.) Since the length of the vector is not important in describing the perceived chromaticness (hue and saturation) of a colour, the normalized tristimulus values or chromaticity coordinates in a chromaticity diagram, from which only two are linearly independent, define unequivocally the chromaticness of a colour signal. Chromaticity diagrams of this kind have been successfully used to describe quantitatively the colour vision of honeybees at the level of both lower and higher colour metrics (Backhaus & Menzel, 1987). In our model calculation, the tristimulus values U, B and G are normalized in such a way that each value is of unit length for an achromatic object (BaSO₄), which is illuminated by natural daylight (spectral radiation of the international D 65 norm function). Fig. 3 gives the chromaticity diagram together with the loci of the colour signals listed in Table 1.

Osmia quickly learned to find the entrance hole of its nest with the help of a coloured disc, which surrounded the hole. The training started when an Osmia female flew regularly in and out of a bamboo stem, indicating that this particular stem had been selected for breeding and that brood cells were being built. The first visual mark at the entrance was a small ring painted with silver paint. This ring was exchanged for a larger disc, and after the individual had adapted to this disc, a silver disc was presented with the final diameter (70 mm) of the coloured discs. This silver disc was finally exchanged for one of the coloured discs, the trained colour signal. Tests could be undertaken a few hours later, since the animal learned the colour signals very quickly. In most of the experiments the same bee was trained to a second or third colour signal when the tests with the first trained colour had terminated. However, in this case the bee viewed the silver disc for 2 days before learning the new signal. The choice behaviour of a bee for the trained colour was independent of a previously trained colour or colours.

The pretraining to silver paint may have affected choice behaviour in a colour-selective manner. Since we did not use a different pretraining procedure, we could only test the possibility of a colour bias by comparing symmetrical test situations, i.e. tests where the same pairs of colour signals were tested either after training to one colour or training to the other. For example, if colour signal no. 2 is used for training, and the discrimination between no. 2 and no. 7 was tested, the average choice proportion p was 0·95 (95 % correct choices). If the same pair of colour signals was tested after training to no. 7, p was 0·83 (0·12 less than in the first case). We found the following rank order in symmetrical tests: 2, 7, 24,18,17, 21. Violet (2) and blue (7) colours were chosen more frequently than other training colours. The asymmetry effects were, however, never very strong, and rarely exceeded 15 % differences between symmetrical tests. We conclude, therefore, that colours are chosen with a certain but small bias, which might result from the pretraining to the silver-painted disc (but see Discussion). Compared with the strong learning ability this colour bias is relatively weak.

The results of all the discrimination tests are given in Fig. 4A-L. The training experiments show that Osmia discriminates colours very well, and that colour signals further apart in the chromaticity diagram are discriminated better than closer colour signals. As mentioned above, discrimination is only weakly dependent on the colour signal to which the bees have been trained. Fig. 4A-L indicates with a thick top line the average response value of the symmetrical test series, and with a thin or broken top line the response value of the test series in which only one colour signal was trained (the one indicated for the particular set of columns). Symmetrical tests were not carried out for all pairs of colour signals. The difference between symmetrical and non-symmetrical tests was small (see above), showing that Osmia learns any of the colour signals as a guide mark for the nest entrance.Furthermore, Osmia does not show a strong tendency to use particular colours (or a colour component such as brightness), which cannot be overcome by training.

Since two other hymenopteran species have been tested with very similar experimental arrangements (Apis mellifera and Melipona quadrifasciata), a direct comparison between these species could be made in terms of their colour discrimination for identical pairs of colour stimuli. The differences between the colour discrimination tests of the three species were as follows. (1) In experiments with Apis and Melipona the colour discs appeared on an evenly painted (white or grey), large background screen and were 100cm apart from each other, whereas in Osmia the colour discs were very close and appeared on a structured and darker background. (2) Most of our data for Apis come from training experiments at the feeding place, and this data will be used for the comparison made here. We have shown that Apis also discriminates colour stimuli very well at the hive entrance, and discrimination in the two behavioural contexts (feeding place, hive entrance) is similar, but that more accurate choices are found at the feeding place (Menzel, 1986). (3) Melipona was tested at the hive entrance and at the feeding place. We shall use the data from the hive entrance training experiments here, since the data were collected under the same behavioural conditions (i.e. homing) as the tests with Osmia. (4) The criterion for a choice in the experiments with Osmia and Melipona is a landing on the coloured disc. In the experiments with Apis, an approach towards the disc at the feeding place is counted as a choice.

The choice proportions p for identical pairs of colour stimuli are plotted for Osmia and Apis in Fig. 5. The values scatter around a symmetry line, and this indicates that both species have similar colour discrimination. In contrast, Melipona differs considerably from Osmia in its colour discrimination (Fig. 6). Most data lie below the symmetry line indicating a better colour discrimination in Osmia. Closer inspection of the data shows that Osmia is also much better at discriminating between colour signals in the ultraviolet, violet and blue regions, whereas Melipona is at least equally as good as Osmia at discriminating bluish-green colour signals. We have shown elsewhere that the lesser ability of Melipona to discriminate colour signals in the ultraviolet, violet and blue depends on the behavioural context. For example, Melipona discriminates ultraviolet, violet and blue signals better at the feeding place than at the hive entrance. It is not surprising, therefore, that Osmia (as well as Apis) discriminates in the ultraviolet, violet and blue better than Melipona, if Melipona’s performance at the hive entrance is used as a basis for comparison.

We have recently derived a model calculation in which the perceptual distance of colour signals can be determined from the spectral properties and the intrinsic noise of the photoreceptors (Backhaus & Menzel, 1987). The model assumes that the just noticeable difference steps at the receptor level (rjnd) constitute a measure of the perceptual just noticeable difference steps (pjnd) and, therefore, the perceptual distance between two colour signals. These assumptions are valid for the honeybee Apis mellifera (Backhaus & Menzel, 1987) and for the stingless bee Melipona quadrifasciata. As in the model calculations with Apis and Melipona we can determine the rjnd-steps in the colour space or in the colour plane from the spectral sensitivity functions (Fig. 2), if we assume that the photoreceptors in the eye of Osmia resemble the intrinsic noise components of Apis (0·4 % of the maximum voltage response). The number of pjnd-steps derived from the colour space is the minimal number of jnd-steps at the receptor level between two colour stimuli and is referred to as the spatial distance sD(jnd). Ignoring the components that result from intensity differences of the two colour signals, and referring only to pjnd-steps at the colour plane, we call the distance value pD(jnd). Both sD(jnd) and pD(jnd) are calculated by a procedure which is described in detail in Backhaus & Menzel (1987). Essentially, a computer program is used to calculate the smallest number of pjnd-steps between two colour loci (in the colour space or in the colour plane). A pjnd is defined by the criterion that a variance in the effective quantal flux has to be significantly larger than the intrinsic noise in at least one of the three spectral receptor types.

An example of the discrimination/distance functions for pD(jnd) is given in Fig. 7A. The choice proportions p are linearized by probability transformation into z-values, z(p) (see Backhaus, Menzel & Kreissl, 1987, p. 297 formula 2); z(p) = 0 corresponds to a proportion of 0-75 correct choices. There is a high correlation between the distance value pD(jnd) and the choice behaviour. The correlation coefficients r for different series of experiments are given in Table 2. The distances in the colour plane, pD(jnd), correlate much better with the choice behaviour than those in the colour space, sD(jnd). This means that subjective brightness differences between the colour stimuli are less important for colour discrimination and may be completely ignored (compare H values in Table 1). An analysis of the correlation between the distance values and the choice behaviour reveals that large distances correlate less well with the choice behaviour than shorter distances. Distances larger than 55pD(jnd)-steps or 90sD(jnd) are, therefore, not included in Fig. 7 and Table 2.

One might expect that choice behaviour would simply saturate for large perceptual distances. This is indeed the case for many tests as Fig. 4A-L shows. But there are quite a few exceptions, which are more difficult to understand (e.g. Fig. 4G, no. 8; H, no. 4; I, no. 5; K, nos 2, 8). We have not yet analysed the divergence for large distances, but it is interesting to note that violet and blue colours are more frequently involved in these cases particularly when bluish-green or yellow colours are used for training. One might suspect, therefore, the influence of pretraining in the same sense as has been discussed above (colour bias as a result of pretraining) or an innate bias towards violet or blue colours.

All correlations between discrimination and distance are incorporated in Fig. 7B for pD(jnd) values smaller than 55 jnd and choice proportions of z(p) ⩾ ±1 ·3 (⩾95 % correct choices). The data are best fitted by a regression Une (r = 0 ·851) with a slope of 0 ·09. This means that a change of z(p) = ±0 ·5 around z(p) = 0 (response change between 65 and 85 %) is reached by an average distance of 11 pD(jnd).

One of the most informative methods of assessing a colour vision system is the function which describes the wavelength-dependency of spectral discrimination. The critical test for such a function requires narrowband or monochromatic light of equal subjective brightness (see von Helversen, 1972). We carried out our experiments under natural daylight with reflecting colour signals. However, Fig. 3 clearly shows that several of the colour signals were of such pure chromatic light, that the corresponding colour loci are positioned very close to the corresponding monochromatic lights (e.g. nos 2, 3, 4, 5, 6, 9, 18, 19, 20, 21, 22, 23, 24). The calculated brightness value H (see Table 1) demonstrates that these colour signals are not of equal subjective brightness, but since differences in brightness contribute very little or nothing to the discrimination of these colours (see above), the distances between those two pairs of spectrally pure signals are calculated for those discrimination tests that were carried out with spectrally pure colour signals (three tests for colour 2, one test for colour 4, one test for colour 18, one test for colour 19, three tests for colour 22, two tests for colour 21; choice proportions p & 0 ·95 correct choices were not included). Regression lines are calculated for the three series of correlations (colours 2, 21, 22) between discrimination [z(p)-value] and Δ λ, the wavelength distance between the corresponding monochromatic lights (dominant wavelength) (see inset of Fig. 8; compare number of colour stimulus with Fig. 3). Such functions yield the wavelength distance (Δλ) which produces a discrimination of 75 % correct choices [z(p) = 0]. The other results for which only one test exists can also be used by applying the closest response - Δλ-function. The results of these calculations are given in Fig. 8 together with the theoretical function which was calculated using the same procedure as described above for the perceptual distance pD(jnd) in the plane of equally bright spectral lights.

Osmia is able to discriminate best in the violet part of the light spectrum, and to a slightly lesser degree light in the bluish-green region. The five values of spectral discrimination sensitivity are close to the theoretical function. We can conclude, therefore, that our model predicts well Osmia’s sensitivity to spectral differences.

Like the social honeybee Apis mellifera, the solitary bee Osmia rufa has a trichromatic input system for colour vision. The spectral sensitivity functions of the ultraviolet and blue receptors are similar for the two species (Apis, λ_max 336 nm and 423 nm; Osmia, λ_max 348 nm and 436nm). The green receptor, however, is more sensitive at longer wavelengths in Osmia (Apis, λ_max 532nm; Osmia, λ_max 572nm). This should have consequences for colour discrimination. A receptor model for colour vision in Apis has been successfully tested in various colour discrimination experiments (Backhaus & Menzel, 1987; Backhaus et al. 1987), and thus may also be applied to the colour discrimination data in Osmia. Such a model provides us with predictions concerning the degree of discrimination of two colour signals, since the calculation of jnd-steps gives us a measure of the subjective difference between pairs of colour signals. Furthermore, the receptor-based model allows the comparison of colour vision among insect species solely on the basis of spectral measurements of single photoreceptors. This is particularly valuable since electrophysiological data on photoreceptors are relatively easy to collect whereas behavioural discrimination tests are often impossible to perform.

The geometrical distance between loci of colour stimuli in the receptor diagram is not correlated with the dissimilarity of colours. Helmholtz (1896) and Schrodinger (1920) pointed out that a line element has to be established which defines the shortest way between two colour loci. We have developed a model calculation which defines the perceptual line element from the noise components of the photoreceptors (Backhaus & Menzel, 1987). Here we present two independent sets of experiments which support the predictive capability of the jnd-measure for the subjective difference of colour signals. First, we found a high correlation between pD(jnd), the jnd-steps in the colour plane, and the proportion of correct choices as expressed in the z(p)-values (probability-transformed choice proportion) (Fig. 7A,B). It turns out that Osmia is less concerned with the brightness differences of colour pairs than differences in chromaticness (hue, saturation).

This is concluded from the better correlation of the behavioural data with the jnd-steps in the colour plane than in the colour space (Table 2). Second, we calculated the AA(A)-function for five colour signals and found a good agreement with the corresponding sensitivity values predicted by the receptor model (Fig. 8).

Such a function with many more data points has been established for Apis (von Helversen, 1972), and we have recently shown in this species that the line elements along the spectral line of the chromaticity diagram follow exactly the behaviourally measured function (Backhaus & Menzel, 1987). In Apis, spectral discrimination is optimal around 400 and 490nm, 490nm being somewhat better than 400nm. Osmia differs in that its discrimination is best at 400 nm, and somewhat less in the blue-green (Fig. 8). A comparison between the species should be made with some caution since the test arrangements were quite different. The test lights were further apart for Apis than Osmia, and the tests were carried out under dim light conditions and without the ultraviolet-reflecting filters that were used in tests with Osmia under natural bright light. Nevertheless, the line elements of the receptor model predict the spectral discrimination of both species to a good fit, indicating that spectral discrimination of Osmia and Apis is relatively independent of the colourless surround and ambient light conditions.

A comparison between species can also be performed without reference to a colour vision model, and without any assumptions of the line elements. This has been carried out (Figs 5,6) for the three bees Osmia, Apis and Melipona. It should be noted that the colour signals used in the experiments with Osmia and Apis (nos 7 and 17, see Fig. 3) are in the chromatic range for which one would not expect differences between the two species on the basis of their Δλ/ λ-function, and this is in fact the case (Fig. 5). In the comparison of Osmia and Melipona, colour signals in the violet (no. 3) and bluish-green (no. 18) are used. The results clearly show that Osmia is considerably better than Melipona at discriminating violet colours, and that Melipona is better in the blue-green region of the spectrum. Little or no difference was found for the other test colours (nos. 7, 17). Since Melipona ‘s colour vision system is very similar to that of Apis we conclude that there are additional factors which reduce the discrimination of Melipona in the violet.

Colours are important signals for animals. For an individual, the significance of colour is the result of its own experience and the result of its evolutionary history (Menzel, 1985b). One observation - the asymmetry effect - provides additional evidence for the unique perception of colour. When the same pair of colour signals is tested after training an animal either to one or the other colour signal, small but statistically significant (x²-test, P between 0·03 and 0·07) differences in the choice behaviour may be found. An analysis of the bias for certain colour signals reveals the following rank order: violet, blue, yellow, bluish-green. What might be the reason for such an order of preference? The animals had to be pretrained to a disc painted silver before training to a particular colour signal started. A generalization process might transfer the pretraining effect selectively to the colour signals. However, the rank order does not depend on any other pretrainings, e.g. a different colour signal. Thus, it is unlikely that the pretraining causes the colour bias. Another possibility is that Osmia has an innate preference for colours which is not completely overcome by learning. It is interesting to note that the same rank order of colours has been observed in colour learning of honeybees when the speed of acquisition is taken as a parameter (Menzel, 1967). In the honeybee this rank order is independent of pretraining and several other parameters, e.g. colour of the alternative signal in the dual-choice tests, age of the bee, time of year, weather conditions etc. The existence of a colour bias, as in the case of Apis and Osmia, is a strong indication of the perceptual uniqueness of colours.

We thank Mr M. Whitfield for his help with the English and the typing of the manuscript. The comments of the anonymous referees are greatly appreciated. The study was supported by DFG grant Me 365/12 and a visiting grant for J. de Souza as part of the binational agreement between Brazil and the Federal Republic of Germany.