Australian native flowers were collected from Maranoa Gardens, Balwyn. The gardens were chosen due to the diverse collection of labelled species, from all states in Australia. Flowers were collected once a month, from May to January.
A colour photograph of the flower with a scale was taken to confirm the flower ID. A UV photograph was also taken for all flowers, using a digital UV camera [Fuji Finepix Pro S3 UVIR modified CCD for UV imaging] with calibrated UV-vis grey scales. As UV rays are invisible to the human eye (ref), this photo enabled the confirmation of any UV reflectance areas of the flower to be measured by the spectrophotometer.
The spectral reflection functions of flowers were calculated from 300 to 700 nm using a spectrophotometer(S2000) with a PX-2 pulsed xenon light source attached to a PC running SpectraSuite software (Ocean Optics Inc., Dunedin, FL, USA).The white standard was a freshly pressed pellet of dry BaSO4 used to calibrate the spectrophotometer. If the petals were smaller that 10 mm, they were arranged like fish scales. A minimum of three flowers from each plant, were used for each spectral analysis. We evaluated a sample of 113 spectral measurements from Australian flowering plants.
2.1.1.1 Possible confounds
It was difficult to separate plants based on their pollinator(s), as the pollinators are not known for a large amount of Australian flowers. In addition, observations of putative pollinators can be very rare and when examined very brief (e.g. Indsto, Weston et al. 2006). Hence, we collected flowers independently of their pollinator(s), enabling a global look at what visual systems can evaluate Australian flower colouration.
It is possible that flower colour pigments can be constrained by phylogeny (Kalisz and Kramer 2008; (Menzel and Shmida 1993). Therefore, it was important to make sure that our sample did not contain an overrepresentation of one plant family group. Indeed, many plant family groups were contained in our sample, see fig. 1. Furthermore, there was no evidence of a plant family over representing a colour in our sample.
2.1.2. Correlations between spectral reflectance functions of different plant flowers and trichomatic vision of the honeybees
To understand if there is a link between hymenopterans and Australian native flowers, we used the methodology used by Chittka and Menzel (1992). In that study, they looked for correlations between flower spectra sharp steps of different plant flowers and trichomatic vision of the honeybees. Sharp steps are a rapid change as the spectra wavelength shifts from the spectral band dominated by one receptor to the band dominated by the next (see fig. 2 for an example of a sharp step). These steps produce the greatest differences between the signals in the different receptor colour types, producing vivid colours that stand out from the surroundings. Furthermore, a colour signal will be more distinguishable, if the sharp steps match up with the overlap of receptors in a visual system. Therefore, the most important characteristic of a flower spectrum is a sharp step. For this study, I defined a sharp step as a change in greater than 20 % reflectance over more than 50 nm. The midpoint of the slope was determined by eyesight, as the nature of curves varied with each flower. The frequency of midpoint slopes by wavelength were tallied up for each flower and averaged to a maximum of 1. As hybrid plants are artificially selected by humans, hybrid flowers were not included in the analyses.
2.1.3. Generating a Hexagon colour space
To evaluate how flower colours are seen by Trigona carbonaria, we plotted the flower colour positions in a colour hexagon space. A colour space is a numerical representation of an insect's colour perception (Chittka 1992). In a colour space, the distances between locations of a two colour objects link with the insect's capacity to differentiate those colours. To make the colour space, the spectral reflectance of the colour objects were required, as well as the receptor sensitivities of the insect. I then predicted how the brain processed these colour signals by using the average reflectance from each flower, and calculating the photoreceptor excitation (E) values, according to the UV, blue and green receptor sensitivities (using the methods explained by chittka and kevan 2005). The E-values were calculated using the spectral sensitivity curves of honeybee photoreceptors with three areas of peak sensitivity: 345 nm (UV), 440 nm (blue) and 535 nm (green) (Briscoe and Chittka 2001), Chittka and Kevan, 2005). It is reasonable to use the honeybee model sensitivities to represent the Trigona carbonaria visual system as receptors are phylogenetically ancient (Chittka 1996) and therefore similar to the majority of hymenopteran trichomats. The UV, blue and green e-values of flower spectra were used as coordinates and plotted in a colour space (Chittka 1992), Chittka et al 1994, Chittka and Kevan 2005). The colour difference as perceived by a bee was calculated by the Euclidean distance between two objects locations in the colour hexagon space (Chittka 1992), Chittk and Kevan, 2005).
Does an Australian native bee (Trigona carbonaria) have innate colour preferences?
2.2.1 Insect model and holding conditions
Trigona carbonaria is an Australian native stingless bee that lives in colonies of 5000-10000 individuals (Heard 1988). They are common to North Eastern Australia and are an important pollinator for several major commercial crops (ref). The bees were obtained from the CSIRO and were not exposed to flowers in the wild. For this study, a colony was placed in a pine nest box (27.5 x 20 x 31 cm; LWH) and connected to the foraging arena by a 16 cm plexiglass tube, containing individual shutters to control bee movements. All laboratory experiments were conducted in a Controlled Temperature Laboratory (CTL) at Monash University, Clayton, School of Biological Sciences (CTL room G12C dimensions 3 x 5m), during the months of July 2009- January 2010. Relative humidity (RH) was set to 30%, and the temperature was set to 27 °C (SPER-Scientific Hygrometer, Arizona, USA). Illumination (10/14 hr day/night) was provided by four Phillips Master TLS HE slimline 28W/865 UV+ daylight fluorescent tubes (Holland) with specially fitted high frequency (>1200Hz) ATEC Jupiter EGF PMD2x14-35 electronic dimmable ballasts. The flight arena (1.2 x 0.6 x 0.5m; LWH) was made of a coated steel frame with laminated white wooden side panels. The arena floor was painted foliage green, and the arena lid was covered with UV transparent plexiglass. This set up approximately matches natural foraging conditions for insect pollinators (ref). Experiments were conducted from 1pm-3pm to control for time of day, as this is when bees are most active (Heard and Hendrikz 1993).
2.2.2 Innate colour preference testing
2.2.2.1 Pre-training
Bees were habituated to the flight arena for seven days. Naive foragers (i.e. bees that had never encountered real or artificial flowers) were initially pre-trained to forage in the flight arena on three rewarded aluminium sanded disks (25 mm in diameter), with a 10-μl droplet of 15% (w/w) sucrose solution placed in the centre. The disks were placed on vertical plastic cylinders (diameter = 25 mm, height = 100 mm), to raise them above the floor of the flight arena. Pre-training allows bees to become used to visiting artificial flowers for further experiments. The aluminium sanded disks were chosen as "neutral" stimuli because they have an even spectral reflectance curve in the spectral visual range of the bees. The sucrose solution reward on these training disks was refilled using a pipette after it was consumed by foraging bees. The spatial positions of these training disks were regularly randomised so that bees would not learn to associate particular locations with reward. Bees were allowed a minimum of two hours to forage on the pre-training disks before data collection.
2.2.2.2 Experimental procedure
To test the innate colour preferences of naive bees, we performed spontaneous choice experiments with flower-naive bees using artificial flowers that simulated the floral colours of natural flowers. The rewarding disks were replaced by the ten unrewarding, coloured artificial disks in the original flight arena. Artificial flower stimuli were cut in a circle (70 mm diameter) from standardized colour papers of the HKS-N-series (Hostmann-Steinberg K+E Druckfarben, H. Schmincke & Co., Germany). In each experiment the same set of ten test colours (1N - pale yellow, 3N - saturated yellow, 21N - light pink, 32N - pink, 33N - purple, 50N - blue, 68N - green, 82N - brown, 92N - grey, back of 92N - white) were used. These colours were chosen as they have been used in experiments with other hymenopterans (e.g. Giurfa, Núñez et al. 1995; Kelber 1997; Gumbert 2000). The coloured papers were placed on vertical plastic cylinders (diameter = 15 mm; height = 50 mm), to raise them above the floor of the flight arena. The gate was shut in the arena to ensure the bees used in each trial were separated from the next trial. The number of landings and approaches to the stimuli were recorded for one hour. Approximately 200 bees were used for each trial. The spatial positions of the artificial flowers were randomised every 15 minutes for each trial. After each trial, the colour disks were aired and wiped with a paper tissue to remove possible scent marks, which are known to affect experiments with honeybees (ref). We conducted each subsequent trial after removing the used bees from the system, to ensure that the bees used in the next trial were replaced with naive foragers.
It is known that perception of colour can be influenced by background colour (Lunau, Wacht et al. 1996). Therefore, I also tested colour choices on other background colours of grey and black. The results are qualitatively similar but are not shown. I decided to only show data from the green background, as it is more biologically relevant to green foliage. I predicted that bees should prefer certain colours over others similar to the bees in Europe (Giurfra ref)
2.2.3 Analysis of colour stimuli
As bees see colours differently to humans, I quantified stimuli according to three parameters: wavelength, brightness and purity (saturation). Dominant wavelength was calculated by tracing a line from the centre of the colour hexagon through the stimulus location to the corresponding spectrum locus wavelength (see Wyszecki and Stiles 1982 for an explanation). Brightness was measured as the sum of excitation values of the UV, blue and green receptors (see ). Spectral purity of the stimulus was calculated by the percentage distance of the stimulus in relation to the end of the spectrum locus (see ref for explanation).
Does a food deceptive orchid (Caladenia carnea) exploit the innate colour preferences of Trigona carbonaria?
Caladenia carnea is a widespread species, common to eastern Australia. The orchid is highly variable in colour, ranging from pink to white. It is pollinated by Australian native bees of the Trigona species (Adams and Lawson 1993).With bright colours and fragrance, this orchid achieves pollination by food mimicry (Adams and Lawson 1993)..
2.3.1 Experiment 1.Can Trigona carbonaria perceive a difference between pink and white flowers of a food deceptive orchid (Caladenia carnea)?
Colorimetric analysis of the pink and white Caladenia carnea flowers were used to investigate whether different colours of the orchid were perceived as similar or different, to a bees visual system. Caladenia carnea flowers were supplied by private growers from the Australasian Native Orchid Society. Four measurements of the orchid by a spectrophotomer (see above for procedure) were used for each flower colour. The actual measurements used in the analysis were an average of each colour. To predict the probability with which insect pollinators would discriminate between different flowers, these spectra were plotted as loci in a hexagon colour space (Chittka 1992) (see above methods). I predicted that bees should perceive the colours as different-why
2.3.2. Choice experiments
I conducted trials testing the preferences of bees when offered the choice between different Caladenia carnea flowers. Each trial took place inside a flight arena (see above). Each flower was matched for size, placed into a plastic container (diameter= 5 cm, height=5 cm) and placed in the arena with a distance of 10 cm between flower centres. Each container was covered with Glad WrapTM (The Clorox Company, Oaklands, CA, USA) to remove olfactory cues as olfactory cues are known to influence the choice behaviour of honey-bees (e.g. Pelz, Gerber et al. 1997)(Laska, Galizia et al. 1999). The ï¬rst contact with the Glad WrapTM within a distance of 4 cm, was considered the choice of a flower. We allowed up to five minutes for the bees to make a choice. After each trial, the Glad WrapTM was changed to prevent scent marks. In addition, individual flowers and spatial positions were randomised. Bees used in each trial were removed from the system with a vacuum after testing to avoid pseudo replication.
. Experiment 2. Does the UV signal affect the attraction of bees to the orchid flowers?
Experiment 1 confirmed that Caladenia carnea did have a UV reflecting dorsal sepal. To investigate whether the UV reflectance of the dorsal sepal affected the response by bees, we offered bees the choice between two white orchids, one with a UV signal and the other without. The UV signal was removed by applying a thin layer of sunscreen (Hamilton SPF 30+, Adelaide, SA, Australia) over the dorsal sepal. Spectral reflectance measurements were taken to ensure that the sunscreen prevented any reflection of UV light (below 395 nm) from the sepals and did not change the reflectance properties of the orchid. In addition, spectral measurements of orchid petals under Glad WrapTM confirmed that the foil was permeable to all wavelengths of light above 300 nm and did not obscure the reflectance of flowers. see procedures for choice testing.
2.3.2.2. Experiment 3. Does Trigona carbonaria display preferences when choosing among pink vs. white flowers of a food deceptive orchid (Caladenia carnea)?
To assess whether bees show a preferences for pink or white variants of the orchid Caladenia carnea, we offered bees (N=16) a simultaneous choice between a pink or white flower. For methods see procedures for choice testing. I predicted that bees should prefer white flowers given the results from part 2.
Experiment 4. & Experiment 5. Do bees habituate to non-rewarding flowers? Do habituation rates differ with different flower colours? Does a novel flower colour affect learning to avoid an unrewarding flower decoy?
We conducted a two stage experiment to investigate if bees could learn to habituate to a non-rewarding flower colour over time and whether bees adjusted their subsequent flower choice depending on the flower colour encountered previously. At stage 1 of the experiment, native bees were presented with one flower, either a white or pink flower. Flowers were placed in a container with Glad WrapTM. The flower was left in the arena for 35 minutes. We then scored the landings to the flower for 5 minutes at the start and end of the trial. At stage 2, the flower from stage 1 was swapped with either the same or new flower colour and the number landings were scored for 5 minutes. Flowers were randomised and Glad WrapTM changed to prevent scent marks after each trial. I predicted that hymenopterans will be capable of learning unrewarding flower decoys based on previous studies (Wong and Schiestl 2002) and this may influence floral colour (ref).
2.3 Statistical analyses
The impact of wavelength on number of landings by Trigona carbonaria was investigated using a single factor analysis of variance (ANOVA) and a post hoc Tukeys HSD test (α=0.05) (Quinn and Keough 2002) using the number of landings as the dependent variable and wavelength of stimuli as the independent variable.
For experiment 2 , 3 & 4, number of landings by naive bees to flower pairs were compared using a paired t-tests. All tests were two-tailed. Experiment 5 was analysed with a two factor ANOVA. The dependent variable was the landings and the two independent variables were previous flower colour and new flower colour.
Statistical analyses were conducted using R statistical and graphical environment (R Development Core Team, 2007)(R and Team 2007). Statistical significance was set to P≤0.05.
3. Results
Is there a link between hymenopteran vision and Australian floral colouration?
The floral reflection curves were analysed to see if there were any correlations between the inflection curves of different plant flowers and trichomatic vision of hymenopterans. The analysis of 113 spectral reflection curves of Australian flowers reveals that sharp steps occur at those wavelengths where hymenoterans are most sensitive to spectral differences (see fig. 3b). The results are in good agreement with the study by Chittka and Menzel (1992), see fig. 3a. There are three clear peaks in sharp steps (fig. 3b). We know that hymenopteran trichomats are all sensitive to spectral differences at approximately 400 and 500 nm (Menzel and Backhaus 1991)(Peitsch, Fietz et al. 1992). Hence, the peaks at 400 and 500 nm can be discriminated well by hymenopteran trichomats, as illustrated by the inverse Δ λ/λ function of the honeybee (von Helversen 1972, solid curve shown in fig. 3a), which is an empirically determined threshold function which shows the region of the electromagnetic function that a bees visual system discriminates colours best. In summary, the receptors of trichomatic hymenopterans are well placed in the spectrum to allow for optimal evaluation of steps in the floral spectra of Australian flowers.
Does an Australian native bee (Trigona carbonaria) have innate colour preferences?
3.2.1 Effect of wavelength on colour choices
Stimuli colours are plotted in fig 4a, as they appear to a human viewer to enable readers to understand the correlation between colour choices. However, all statistical analyses were conducted with stimuli plotted as wavelength, as bees see colours very differently to humans. There is a significant effect of wavelength on the number of landings by Trigona carbonaria (Single factor ANOVA, F = 5.60, df= 9, 110, P <0.001), see figure 4b. Tukey's post hoc test revealed that the wavelength of 437 nm (white) had significantly higher landings than other wavelengths of 528 nm (brown) (P<0.01), 432 nm (grey) (P <0.01), 431 nm (light pink) (P<0.01), 420 nm (purple) (P<0.01), 455 nm (blue) (P=0.0196) and 535 nm (green) (P=0.0266). In addition, the number of landings to wavelengths of 530 nm (pale yellow) (P=0.0321) and 422 nm (pink) (P=0.0318) disks were significantly higher than that of 432 nm (grey), see fig 4b.
3.2.2. Effect of brightness on colour choices
To analyse whether brightness of the stimuli, accounted for choice frequency by bees, the Spearman's rank correlation test was used (rs: Spearman rank correlation coefficient). There was no significant correlation between choice frequency and brightness of stimuli (rs= 0.333, n=10, p= 0.347; see fig. 5a). Note, when the stimulus which bees preferred on the basis of wavelength was removed from the brightness analysis, R squared =0.0173, confirming there was no correlation of innate preferences with brightness.
3.2.3. Effect of spectral purity on colour choices
There was no significant correlation between choice frequency and spectral purity (rs = 0.224, n=10, p= 0.533; see fig. 6b). Note, when the stimulus which bees preferred on the basis of wavelength was removed from the spectral purity analysis, R squared =0.0235, confirming there was no correlation of innate preferences with spectral purity.
3.3 Does a food deceptive orchid (Caladenia carnea) exploit the innate colour preferences of Trigona carbonaria?
Experiment 1. Can Trigona carbonaria perceive a difference between pink and white flowers of a food deceptive orchid (Caladenia carnea)?
Ultraviolet photographs and reflectance measurements revealed that lateral sepals were different from the dorsal sepals (see fig. 6). The spectra of the pink and white lateral sepals indicated no UV reflection. In contrast, the spectra of the dorsal sepals show reflection in the UV region (320 nm, Fig. 3). Figure 7 shows the loci of the respective flower spectra in a hexagon colour space. Dyer and Chittka double check this ((Dyer and Chittka 2004)) showed that with increasing colour distance between flowers and distractor flowers, less errors were made by foraging bees. Colour distance between the white and pink flowers is measured in hexagon units (Euclidean colour metric); see Table 1. The lateral sepals (UV-) of pink and white flowers are separated by only 0.082 colour hexagon units, while pink and white dorsal sepals (UV+) are separated by 0.039 hexagon units. Thus, pink and white lateral sepals are distinguishable to a bee. In contrast, pink and white dorsal sepals (UV+) are perceptually similar to a bee. Therefore, most of the white vs. pink flowers of Caladenia carnea can thus most likely be discriminated with between 70-90% accuracy (see fig. 8). This means that these white/pink flower colours can be discriminated by the bees, with occasional pollinator perceptual errors (1-3 errors/10 visits). This is thus potentially a very interesting system to consider for testing flower mimicry hypotheses.
3.3.2. Experiment 2. Does the UV signal affect the attraction of bees to the orchid flowers?
When test subjects were presented with a choice between two white orchid flowers, one with a UV signal and one without, there was a significant preference for the flower with the UV reflectance (paired t-test: t= 6.949, df= 15, p<0.001, n=16; Fig. 9).
3.3.3 Experiment 3. Does Trigona carbonaria display preferences when choosing among pink vs. white flowers of a food deceptive orchid (Caladenia carnea)?
When test subjects were presented with a choice between two flower colours, pink and white, there was a significant preference for the white flower (paired t-test: t= 3.484 (was negative -3.484), df= 15, p= 0.00333389 or p<0.005, n=16; see fig. 10).
3.3.4 Experiment 4. Do bees habituate to non-rewarding flowers? Do habituation rates differ with different flower colours?
Bees were found to habituate to non-rewarding flowers, as the mean number of landings by Trigona carbonaria to the flower at the first time stage (T1) were found to be significantly different from the second time stage (T2) for white (paired t-test: t= 8.34, df= 15, p<0.001) and pink flowers (paired t-test: t= 8.11, df= 15, p<0.001) (see fig. 11). Habituation rates were found to differ with different flower colours, as the mean number of landings by Trigona carbonaria to the white flower were found to be significantly higher from that of the pink flower (paired t-test: t=3.59, df=15, p=0.002659862, see fig. 11). I also looked at delta, which is calculated as the rate of change between landings at the first and second time stage for pink and white flowers separately, which was found to be significantly different (paired t-test: t=3.94, df=15, p=0.001316129). In summary, bees were able to habituate to pink and white flower colours. Interestingly, it takes longer for bees to habituate to white flowers in comparison to pink flowers.
Experiment 5. Does a novel flower colour affect learning to avoid an unrewarding flower decoy?
The number of landings to a flower were found to be significantly affected by the interaction between the previous flower colour and new flower colour, (two factor ANOVA, F3,28=6.846, p=0.001374, see fig. 12). When the second flower colour presented was the same colour as the previous flower, landings were not significantly different to the second flower (F1,14=4.332 p=0.0562). In contrast, when the second flower colour was different to the previous colour, landings were found to be significantly different to the second flower (F1,14=9.168 p=0.00904), see fig. 12. In addition, preferences depended on the previous colour bees were exposed to. When the previous flower was white, landings to the second pink or white flower were not found to be significantly different (F1,14=5.332,p=0.2301018). In contrast, when the previous flower colour was pink, landings were found to be significantly higher to second white flower than to new pink flower (F1,14=8.395, p=0.0117, see fig. 12). Bees, in this regard, were adjusting their choices to the second flower depending on their previous flower experience.
Figure 3. A) Taken from Chittka and Menzel (1992). The bars show the frequency of slope midpoints of flower spectral reflections (180 flowers) from Israel against their wavelength. There are three clear peaks in sharp steps. We know that hymenopteran trichomats are all sensitive to spectral differences at approximately 400 and 500 nm(Menzel and Backhaus, 1991; Peitsch et al. 1992). Hence, the peaks at 400 and 500 nm can be discriminated well by hymenopteran trichomats, as illustrated by the inverse Δ λ/λ function of the honeybee (von Helversen 1972, solid line). The columns are normalised to a maximum. The highest column corresponds to a total of 36 slopes. B) The frequency of slope midpoints from the flower spectral reflections of 113 Australian native flowers. It can be seen that results are similar to Figure A. The highest column corresponds to a total of 19 slopes
Figure 1. The range of plant families encountered in a sample of 113 Australian native plants from Maranoa Gardens, Balwyn. Flowers were collected from May 2009 to January 2010.
Figure 2. Typical floral spectral reflections of two flowers. The arrows show the slope midpoints of the sharp steps in the curve. The wavelength positions were determined by eyesight for 113 flowers.
Figure 4. A) Mean (±SE) percentage of choices by Trigona carbonaria for the ten HKS paper colour disks (pale yellow, saturated yellow, light pink, pink, purple, blue, green, brown, grey, white) in one hour. Note: colours are plotted as they appear to a human viewer. B) Mean (±SE) percentage choices by Trigona carbonaria plotted against wavelength of stimuli. Choices were observed on a standard green background (n=12). In total, 2946 choices were made by 2400 bees. Symbols above bars show significant differences between means (Tukey HSD) at α = 0.05. C) Results in B are remarkably similar to the innate preferences of flower naive honeybees in Europe (Giurfa, Núñez et al. 1995).
Figure 5. Mean (±SE) percentage choices by Trigona carbonaria plotted against A) wavelength of stimuli B) brightness of stimuli and C) spectral purity of stimuli. Choices were observed on a standard green background (n=12). In total, 2946 choices were made by 2400 bees.
Figure 6. Photograph showing the positions of the dorsal and lateral sepals of the orchid Caladenia carnea. A) photographs in the visible spectrum and B) in the UV spectrum. Note the UV reflectance of the dorsal sepal.
Figure 7: Flower spectra (see supplied excel file for full details), and plots in a hexagon colour space [p=Pink (+/-UV), w=White (+/-UV) forms of Caladenia carnea flowers; loci map shown to the lower LHS of plots]. The visual system is assumed to be adapted to a half maximum response from green foliage which is plotted as a '+' in the centre of colour space. The corners of the colour space show maximum excitation (E) of the bee UV, Blue and Green photoreceptors. See Table 1 for model predictions.
Figure 8. Bee flower colour discrimination ability follows a sigmoidal type function depending upon perceptual similarity (Dyer and Chittka 2004).
Table 1. Predicted colour difference (Hexagon Units) between pink and white flowers.
