Relationship between colouration and body condition in a crab spider that lures pollinators

Even though sit-and-wait predators do not actively search for their food, several strategies have evolved that may increase their probability of capturing prey. Examples include the selection of profitable patches (Janetos, 2004; Metcalfe et al., 1997; Scharf and Ovadia, 2006), displaying cryptic and disruptive colouration to avoid detection by prey (Cott, 1957), and the building of traps (Scharf and Ovadia, 2006; Shear, 2004). In addition to these somewhat passive strategies, many sit-and-wait predators employ tactics that actively attract their prey. Prey attraction evolved in many different taxa and often exploits prey signals used in sexual interactions or when searching for food. For instance, the bolas spider Mastophora dizzydeani releases a chemical that mimics the pheromones of female moths, which attracts male moths of the species Spodoptera frugiperda to its sticky ball trap (Eberhard, 1977; Eberhard, 1980). Exploiting a non-sexual response, the common death adder, Acanthophis antarcticus, captures lizards by waving a conspicuous worm-like caudal lure that incites a predatory response in their prey (Nelson et al., 2010).

Several species of crab spiders (Thomisidae) are sit-and-wait predators that exploit the interaction between plants and insects by sitting on flowers to ambush pollinating insects (Morse, 2007). Their colouration usually resembles the colour of the flowers they are sitting on (Chittka, 2001; Morse, 2007; Thery, 2005; Thery and Casas, 2002). Furthermore, crab spiders, such as the European Misumena vatia and Thomisus onustus, can adjust their body colour to match the flower colour background, which makes them less detectable by prey such as honeybees (Gabritschevsky, 1927; Packard, 1905; Thery and Casas, 2002; Thery, 2005; Thery, 2007).

However, camouflage is not the only strategy used by crab spiders. Some crab spiders seem to be highly conspicuous to their prey, rather than blend into the background. These spiders are UV-reflective (they appear white to humans), which, when viewed through honeybee eyes, creates a strong contrast against the UV-absorbing flower (Heiling et al., 2005). Instead of deterring prey by increasing their visibility, these spiders are attractive to pollinators (Heiling et al., 2003), a similar strategy used by some orb-web spiders (Herberstein and Wignall, 2011). When honeybees had to choose between flowers with and without the contrasting UV-reflective Australian crab spider T. spectabilis, they were more likely to land on flowers occupied by their predators (Heiling et al., 2003). When the same experiment was performed with the non-UV-reflective European M. vatia, the honeybees chose flowers at random irrespective of whether they were occupied by a spider (Herberstein et al., 2009). Moreover, when the level of UV reflected by T. spectabilis was experimentally eliminated, honeybees actively avoided flowers occupied by crab spiders (Heiling et al., 2005).

Despite the apparent benefit of UV reflectance for Australian crab spiders, there is a high level of intra- and inter-individual variation in the amount of UV reflected and hence colour contrast by spiders (Llandres et al., 2011). The overall aim of this study was to explain how some of this variation might be generated. First, we aimed to quantify the variation in colour contrast in field-captured spiders and relate this to their recent foraging success. Based on the use of colour to attract prey, we predicted that individuals that reflect more UV light and, consequently, formed a stronger colour contrast against a UV-absorbing flower background would capture more prey and, as a result, would be in a better condition than individuals that reflected less UV light. Such data may indicate the strength of a relationship between two variables, but they do not indicate causality. Our second aim was to use a manipulative experiment to understand the causality of colour contrast in these crab spiders. Only manipulating the feeding level of spiders can reveal whether it is satiation that causes colour contrast or vice versa. Thomisus spectabilis can attract pollinators based on the strength of its colour contrast (via UV reflectance), and thus we expect food-limited spiders to increase contrast and attract more prey. However, if colour contrast is the result rather than the cause of recent foraging success, only food-satiated spiders should be able to increase contrast. If greater visibility, however, entails costs, such as greater risks of predation, we expect satiated spiders to reduce the contrast. In addition to adjusting their colour in response to their satiation level, we also expect crab spiders to adjust their colouration according to the background they are sitting on, as this will affect their visibility to approaching prey.

Here we addressed these questions by analysing the relationship between the colouration of the spider T. spectabilis and their condition using field data from two years and by testing in the laboratory how this predator responded to different feeding regimes and background colours. We calculated how spiders are perceived by the visual system of the crab spider’s prey, the honeybee Apis mellifera, and used data from reflectance spectra independent of any visual model.

To measure the reflectance spectra of organisms and objects, we used an optical fibre probe (Ocean Optics Inc., Dunedin, FL, USA) connected to a spectrometer (USB2000, Ocean Optics Inc.) and a light source (PX-2 light source, Ocean Optics Inc.). The probe was positioned at 45 deg above the samples. The reference spectrum was taken using the WS-1 Diffuse Reflectance Standard (Ocean Optics Inc.; >98% reflectance from 250 to 1500 nm). The dark spectrum was taken from the black velvet used as background to the measurements. We took five spectral measurements from each organism and object.

In the origin of the colour space lie colours similar to the background that photoreceptors are assumed to be adapted to. In the case of the honeybee colour space, the background is a leaf. Spectra that evenly stimulate all photoreceptors, such as a flat line between 300 and 700 nm, also lie in the centre of the colour space. Bees use only the green photoreceptor to evaluate the achromatic contrast (Giurfa et al., 1997). Therefore, we also calculated the achromatic contrast of spiders against the flower by subtracting the flower’s E_Green from the spider’s E_Green.

We collected female Thomisus spectabilis Doleschall, 1859 (Thomisidae) spiders sitting on white daisies Bidens alba var. radiata (Asteraceae) at Airlie Beach, Queensland, Australia, in May 2009 (N=42). We also used previously published data (reflectance spectra) from T. spectabilis spiders collected from the same site and flowers in April 2008 (N=67) (Llandres et al., 2011). In the laboratory we weighed the spiders and measured the tibia–patella length of their first leg. Spiders whose tibia–patella length did not exceed 2.00 mm were not considered in the analyses because they are too small for an accurate colour measurement. We collected 13 white daisies in 2009 and used data from eight white daisies collected in 2008 (Llandres et al., 2011) from the same sites where we collected spiders. We measured the light reflectance of the dorsal side of the spider abdomen and the flowers using the methodology described above. We compared the spider colour (mid-point wavelength, colour contrast and achromatic contrast against the flower), mass and leg length, and the flower colour (honeybee photoreceptor excitation values) between years using a t-test. Because there was no difference in flower colour between years (supplementary material Table S1), we pooled the colour data of flowers collected in 2008 and 2009 to generate a flower model against which to calculate chromatic and achromatic contrast between spiders and flowers.

Our aim with these analyses was to understand which component of the overall colour of spiders contributes to colour contrast values and variation within. However, it is unadvisable to separate individual E values (e.g. E_UV and E_Blue) from the overall colour contrast, as the hymenopteran visual system does not process these components separately (Chittka and Wells 2004; Dyer et al., 2011). We can, however, analyse the chromatic contrast (generated by the combination of E_UV, E_Blue and E_Green) separately from the achromatic contrast; these contrasts are indeed processed independently by the hymenopteran visual system.

We used residuals of the linear regression ln(mass) × ln(tibia–patella leg length) as an index of spider condition (Jakob et al., 1996). To test the relationship between spider condition and spider colouration, we generated models that could explain their relationship and selected models that best fitted the data. The year that spiders were collected was also included in models as an explanatory variable. We opted to analyse each colour parameter in separate regressions because of the collinearity of these variables.

In our model selection process we included polynomial models because exploratory analyses suggested a curvilinear relationship between spider condition and spider colour. In total, we fitted 11 linear regression models for each colour parameter, including interactions with the year that spiders were collected (2008 or 2009). In these models, spider condition was entered as the dependent variable and the colour parameter and year were entered as the explanatory variables. For each model we calculated Akaike’s information criterion (AIC) and the Bayesian information criterion (BIC), and for each colour parameter we selected models with the lowest values. Both indices are calculated similarly, but BIC punishes the inclusion of new variables more heavily (Zuur et al., 2009). We also inspected the statistical significance of model coefficients for models that generated the lowest AIC and BIC values. Before fitting models and calculating higher-order polynomial variables, we centred the continuous explanatory variable – i.e. colour parameters – otherwise the slope estimation of lower-order variables becomes unreliable (Schielzeth, 2010). We visually validated models, looking for deviation from normality of residuals, heterogeneity and violation of independence (Zuur et al., 2009). Models were fitted and validated using R (R Development Core Team, 2008). BIC was calculated using the R package nlme (Pinheiro et al., 2008).

To understand how T. spectabilis varied their colour in response to feeding regimes, background colour and the interaction of these factors, we submitted females to two treatments in a factorial design for 30 days. Spiders were randomly placed in containers with two different colour backgrounds: UV-bright, white to human eyes, and UV-dull, yellow to human eyes. The spiders in each of the colour treatments were further subjected to one of two feeding regimes: high prey and low prey. At the beginning of the experiment, treatments had the following sample size: low feeding regime and white UV-bright containers (N=10), high feeding regime and white UV-bright containers (N=10), low feeding regime and yellow UV-dull containers (N=9) and high feeding regime and yellow UV-dull containers (N=9). However, from the second to the third measurement we lost three spiders in the high feeding/white UV-bright containers, two in the high feeding/yellow UV-dull containers and one in the low feeding/yellow UV-dull container (because of death of spiders, exclusion of spiders that laid eggs and moulted and spiders that went missing).

The low feeding regime consisted of one housefly (Musca domestica) 15 days after the beginning of the experiment, whereas the high feeding regime consisted of two houseflies per week during the first 2 weeks and eight houseflies per week during the last 2 weeks. The containers were made of coloured cardboard (11 × 11 × 11 cm). We calculated the E_UV, E_Blue and E_Green that each container colour generated on A. mellifera vision using the methodology described earlier (N=5 for white UV-bright and N=5 for yellow UV-dull containers). The top of the containers was covered with plastic cling wrap (Glad Wrap®), which allows the transmission of all wavelengths between 300 and 700 nm. The reflectance spectra of the colour containers are shown in supplementary material Fig. S1. The experiment was conducted in a glasshouse with controlled temperature (night: 16°C; day: 25°C; 12 h:12 h light:dark cycle) and perspex panels, which allowed the passage of all wavelengths between 300 and 700 nm (Heiling and Herberstein, 2004).

We weighed spiders, measured their first leg tibia–patella length and collected reflectance spectra of their abdomen at the start of the experiment, 15 days into the experiment and at the end of the experiment. As for the field data, we used the residuals of the linear regression log(mass) × log(tibia–patella leg length) as an index of spider condition. We calculated spider mid-point wavelength, photoreceptor excitation values and colour contrast to the origin in the A. mellifera hexagon colour space using the methodology described earlier. We decided to use the colour contrast to the origin of the colour space and use E_Green instead of the chromatic and achromatic contrasts because the chromatic and achromatic would necessarily be different between groups (at the start of the experiment spiders had similar colours but the backgrounds were different). The colour contrast to the origin and E_Green provide a way to estimate the change in the bee colour space that is comparable between spiders in both colour treatment groups.

To test the effectiveness of our feeding treatment, we first tested its effect on spider condition during the experiment. We then tested the effect of feeding treatment, background colouration and the interaction of these variables on the spiders’ mid-point wavelength, E_Green and the colour contrast to the origin of honeybee colour space. Because of the collinearity of these variables we decided to analyse each colour parameter in separate regressions. We analysed data using a linear mixed model to take into account the unbalanced sample size and the repeated measures.

Dependent variables of the models were spider condition, mid-point wavelength, colour contrast to the origin and E_Green. Colour contrast and E_Green were Box–Cox transformed before model fitting. We first fitted, using the restricted maximum likelihood (REML), a full model with all fixed factors, interactions and a random intercept and slope for each spider during the measurements (Zuur et al., 2009). Full models started with the following set of fixed effects: (1) value of the dependent variable before the experiment, (2) day of dependent variables measurement (categorical with two values: 15 days after the beginning of the experiment and at the end of the experiment), (3) feeding treatment (categorical: low and high), (4) colour treatment (categorical: white UV-bright and yellow UV-dull) and (5) the interaction between feeding and colour treatments. In the random effects we included: (1) spider individuals (random intercept) and (2) day of dependent variables measurement (random slope). To select the model with the best random term, we compared the full model with a model with the same fixed effects but with only the intercept in the random effects. We selected the best model based on the likelihood ratio. We then proceeded to find the optimal fixed effect structure. We fitted a series of nested models by removing, one by one, model fixed effects, but fitting models using maximum likelihood (ML) instead of REML. We then compared models using AIC and BIC and inspected t-statistics of model’s coefficients. We validated models looking for deviation from normality of residuals, heterogeneity and violation of independence (Zuur et al., 2009). Models were fitted and validated using R (R Development Core Team, 2008). Linear mixed models were fitted using the R package nlme (Pinheiro et al., 2008).

The reflectance spectrum of field-collected T. spectabilis was different between years. In 2008, the spiders reflected more light between 350 and 450 nm than in 2009 (Fig. 1). As a consequence, the mid-point wavelength of spiders and spider colouration as perceived by the honeybee was also different between years (Table 1). The mid-point wavelength of spiders in 2008 was on average 13 nm lower than in 2009. The colour contrast of the spiders against the flower background was 0.09 units higher in 2008 than 2009 (Table 1). Similarly, E_UV was 0.10 units higher in 2008 than in 2009. In contrast, E_Blue and E_Green had very similar values in both years, although the average difference of only 0.02 units in E_Blue was statistically significant at the 5% level (Table 1). E_Blue and E_Green of spiders and flowers had very similar values; however, E_UV were higher for spiders than for flowers (Table 1, supplementary material Table S1). Even though spiders in 2009 were slightly heavier and had longer leg lengths than in 2008, these differences were not statistically significant (Table 1).

Variation in leg length explained more than 85% of the variation in mass (F_1,107=672.950, P<0.001). Residuals of this regression, used as an index of spider condition, suggested that the condition of the spiders was more variable in 2008 than in 2009, but the difference was not statistically significant (Levene’s test: F_1,107=1.87, P=0.17).

Model selection for the relationship between spider condition and colouration depended on the colour parameter evaluated (Table 2, Fig. 2). For colour contrast, AIC and BIC identified a second-order polynomial as the best model (Table 2). All coefficients in this model were statistically significant (supplementary material Table S2). High values of colour contrast yielded the highest condition index values. Simpler models with only first-order colour contrast and year terms also produced significant coefficients and indicated a positive relationship between condition and colour contrast (e.g. condition index ∼ colour contrast + year; colour contrast coefficient, mean ± s.e.m.=1.495±0.438, t₁=3.414, P<0.001). For mid-point wavelength, the lowest value of AIC was also a second-order polynomial (Fig. 2). The shape of this curve was similar to that of colour contrast (Fig. 2), but mirrored, because in this system colour contrast and mid-point wavelength are negatively correlated. The model predicts that small values of mid-point wavelength produce the highest values of condition. For BIC, however, there were two models with the lowest values. One was the same model selected using AIC. The second was a linear relationship with no interaction term and a negative coefficient for mid-point wavelength. Both models have in common that they point to the smallest values of mid-point wavelength generating the highest values of condition index. For the achromatic contrast, AIC and BIC pointed to different models (Table 2): a third-order polynomial had the lowest AIC value, but a model with just year as an explanatory variable presented the lowest BIC values. Therefore, the polynomial regression should be treated with care, especially because in this case the third-order polynomial coefficient presented a P-value very close to the established statistical significance level (supplementary material Table S2). Models for all colour parameters evaluated pointed to year as an important variable (2009 having a higher mean spider condition index value than 2008; Table 2).

The best random effect configuration for all models included only the parameter ‘spider identity’ (intercept). The best-fit model for the effect of treatments on the condition index of spiders did not include background colour or the interaction between background colour and feeding treatment. Only the feeding treatment was included in the model (Table 3, supplementary material Table S3). The mean (±s.d.) condition index of spiders in the low feeding treatment changed from –0.01±0.14 to –0.06±0.10 during the 30 day treatment whereas the condition index of spiders in the high feeding treatment changed from –0.03±0.12 to 0.14±0.14 at the end of the experiment. Coefficient statistics show that the effect of feeding treatment was highly significant (supplementary material Table S3). White UV-bright containers generated an E_UV of 0.91±0.00, an E_Blue of 0.90±0.00 and an E_Green of 0.86±0.00, whereas yellow UV-dull containers generated an E_UV of 0.47±0.13, an E_Blue of 0.39±0.08 and an E_Green of 0.78±0.01.

Overall, during the experiment all spiders increased their UV reflectance (supplementary material Fig. S1), significantly increased E_Green and decreased their mid-point wavelength and chromatic contrast in relation to the origin of the bee colour space, regardless of the treatment applied (Table 4, Fig. 3). There was evidence supporting a small effect of background colouration on the spiders’ colouration. This variable was present in the best model indicated by both AIC and BIC for the colour contrast to the origin (Table 3). Moreover, background colouration coefficient had a P-value smaller than the established statistical significance level (Table 4). Models indicated that spiders in yellow UV-dull containers produced a slightly higher colour contrast to the origin than spiders in white UV-bright containers (Table 4, Fig. 3).

There was either a very weak effect or no evidence for an effect of feeding treatment on spider colouration, because feeding treatment was not present in any of the models indicated by BIC (Table 3). Although some of the lowest AIC models did include feeding treatment, their AIC values were very similar to those of models that did not include feeding treatment (Table 3, difference <2), indicating that the inclusion of this variable did not substantially improve models. Moreover, feeding treatment coefficient had a P-value higher than the established statistical significance level (Table 4).

Similar to the feeding treatment, there was no evidence of an interaction between colour background and feeding treatment on spider colouration. The lowest AIC model for mid-point wavelength did include the interaction term, but the AIC value was very similar to that of models that did not include interaction (Table 3). In addition, the t-test indicated that the interaction term could be excluded from the model (Table 4). When this was done, both feeding regime and background colouration became non-significant at the 5% level (feeding treatment: t₁=0.945, P=0.351; colour treatment: t₁=1.387, P=0.175).

We found temporal and individual differences in spider reflectance, especially in the UV region of the spectrum (Fig. 1), with the average mid-point wavelength shifted to the right from 2008 to 2009 (Table 1). As a consequence, when modelled into the honeybee vision, the spiders varied in the overall colour contrast against the UV-dull flowers, probably because of differences in the excitation of the honeybee UV photoreceptor. On average, the spiders were more UV-reflective and created a greater colour contrast in 2008 than in 2009. Colour contrast can be seen as a gradient of difficulty in the discrimination between two colours. The lower the value, the more difficult the task. In behavioural experiments, honeybees have been shown to discriminate targets that differ by as little as 0.06 units in their colour space (Dyer and Chittka, 2004). Furthermore, the lower the colour contrast, the longer honeybees take to learn to discriminate between two targets and the longer the bees take to make a choice between targets (Dyer and Chittka, 2004). Moreover, the ability to discriminate between two different colours is likely to be lower under natural foraging conditions, where honeybees are subjected to distracting factors (Spaethe et al., 2006). In 2009, 10 spiders fell below 0.06 units of colour contrast, but none did so in 2008. In addition, 17 spiders had a colour contrast above 0.25 in 2008, but none did in 2009. Thus, overall our results suggest that from a honeybee perspective, in 2009 more spiders were adopting a strategy of low conspicuousness, whereas in 2008 more spiders were adopting a strategy of higher colour contrast and thus greater visibility (Table 1).

When we analysed the relationship between colour and body condition, we found that condition had a quadratic relationship with overall colour contrast (Table 2, Fig. 2). In addition, the models suggested either a quadratic or linear relationship between mid-point wavelength and condition of spiders (Table 2, Fig. 2). These results, together with those of previous experiments (Bhaskara et al., 2009; Heiling et al., 2005; Heiling et al., 2003; Herberstein et al., 2009), suggest that greater conspicuousness, most likely achieved by a higher E_UV contrast, is advantageous for these spiders. This advantage is likely to be the result of the fact that honeybees are attracted to UV-reflective spiders (Heiling et al., 2005; Heiling et al., 2003; Herberstein et al., 2009) and, therefore, the foraging success of highly UV-reflective spiders would be greater than that of UV-dull spiders. Moreover, the fact that colour contrast and condition have a quadratic relationship suggests that the benefit comes after a certain degree of colour contrast has been achieved (Fig. 2). Contrary to the other variables evaluated, achromatic contrast did not show a clear relationship with spider condition. Model selection predicts that the inclusion of achromatic contrast does not substantially increase its quality (Table 2), but the possibility of an actual effect cannot be ruled out. Our data show that T. spectabilis collected from the field show little variability in E_Blue and E_Green compared with E_UV (see s.d. in Table 1). In a hypothetical scenario where the achromatic contrast was more variable – creating a higher contrast against the background – it is likely that this would influence spider conspicuousness and hence their foraging success.

If there were a clear benefit to adopt a high-UV-reflective, highly contrasting strategy, why would some crab spiders adopt the potentially less efficient foraging strategy of low conspicuousness? Prey often change their foraging behaviour when predators are present, usually with the consequence of reducing their food intake. Common anti-predatory responses include a reduction in foraging time and the selection of less risky foraging patches (Lima, 1998). However, predator pressure can also directly affect prey colouration. For instance, the fiddler crab Uca vomeris can change its carapace from a bright to a dull colour over the course of a few minutes (Hemmi et al., 2006). Colonies of fiddler crabs that are highly exposed to bird predators have, on average, a duller colouration than less exposed colonies and there is evidence that individuals reduce their conspicuousness if the danger of predation is experimentally increased (Hemmi et al., 2006). Similar to fiddler crabs, crab spiders can change their body colour over several days (Gabritschevsky, 1927; Schmalhofer, 2000; Thery, 2007). Therefore, the difference in spider colouration between years quantified in our study may indicate that spiders are adjusting their body colouration in response to variation in predation pressure. Insects, such as wasps and birds, the most likely predators of spiders, are able to perceive UV light (Briscoe and Chittka, 2001; Foelix, 1996; Hart, 2001), and thus highly contrasting UV-bright spiders may suffer a higher risk of predation because of an increased conspicuousness, whereas less contrasting UV-dull spiders may reduce the probability of detection of predators by matching the UV-dull flower. Nonetheless, information on the actual predation rates of crab spiders is very limited (Morse, 2007) and further studies are needed to test those ideas. It is also noteworthy that the overall condition in 2009 was greater than that in 2008 (Table 3, Fig. 2). This suggests that, in terms of foraging success, 2009 was a more productive year than 2008. Therefore, 2009 spiders could be avoiding a riskier strategy of high conspicuousness because a more conservative low colour contrast strategy was yielding similar foraging success compared with highly contrasting spiders in 2008. As an example, the model for the relationship between condition and colour contrasts predicts that a 0.10 colour contrast in 2009 produces a condition similar to a 0.30 colour contrast in 2008 (Fig. 2).

Alternatively, but not mutually exclusive to the previous hypotheses, spiders could be adjusting their colouration in accordance with the behaviour of the most common prey. Different species of pollinators respond differently to the presence of crab spiders (Brechbühl et al., 2009; Brechbühl et al., 2010) and some prey species are attracted to the high UV contrast (Heiling and Herberstein, 2004; Heiling et al., 2003; Llandres et al., 2011). In this scenario, one would expect that in 2008 the most common prey were those attracted to conspicuous UV-bright spiders, but the most common prey in 2009 were not. For example, honeybees are attracted and land more frequently on flowers harbouring UV-bright spiders (Herberstein et al., 2009), but Australian native bees are less likely to land on such flowers (Heiling and Herberstein, 2004; Llandres et al., 2011). If, in 2008, the most common prey were honeybees, the best strategy would be to adopt a high UV-reflectance strategy, whereas if in 2009 the most common prey were native bees, the most efficient strategy would be to reduce conspicuousness.

In our experiment on the effect of background colouration and food intake on spider colouration, we found that the spiders have the ability to change colour independently of food intake. This is supported by two pieces of evidence. First, spiders in all treatments decreased their colour contrast to the origin throughout the experiment (Fig. 3). Second, there was no statistically significant effect of feeding treatment on spider colouration. At the end of the experiment, the spiders in both feeding treatments had remarkably different body conditions (supplementary material Table S3) but almost the same colour contrast and mid-point wavelength (Table 4), suggesting that it is not body condition that causes the difference in colouration, but that higher colour contrast results in greater foraging success and hence greater body condition. Second, the lack of feeding treatment effect, particularly with the yellow background, suggests that satiated spiders do not opt for a potentially less risky strategy of low conspicuousness. However, in our experiment, spiders were isolated from predators; therefore, the overall change in colour contrast, regardless of the feeding treatment, could have been at least partly influenced by the absence of predator cues. Without predator cues, spider may have adopted a high contrast strategy in order to increase prey capture.

Although the effect was small, spiders did respond differently to background colouration (Table 4, Fig. 3). Spiders on the white UV-bright backgrounds tended to produce a lower colour contrast to the origin than spiders on the yellow UV-dull background. Despite this difference, spiders on the yellow UV-dull background did not match the background, i.e. they did not change their reflectance to match the background reflectance (supplementary material Fig. S1). We do not know, however, whether the housefly diet we fed spiders and the artificial backgrounds affected their ability to change colours. The ommochrome pigments related to the yellow colouration in crab spiders are the same class of pigment found in fly eyes (Insausti and Casas, 2008). Indeed, crab spiders fed with red-eyed Drosophila melanogaster (ommochrome rich) changed to a slightly brighter yellow colour than crab spiders fed white-eyed flies (Thery, 2007). Nonetheless, the houseflies in our experiment were red-eyed, and other species of crab spiders have changed colour even against artificial backgrounds (Gabritschevsky, 1927; Packard, 1905; Thery, 2007). In contrast to our study, crab spiders Misumena vatia did change their colouration to match the colouration of white and yellow natural and artificial backgrounds (Gabritschevsky, 1927; Packard, 1905; Thery, 2007). However, M. vatia is apparently not UV-reflective and does not lure pollinators. Thus one can predict that the best strategy for this species is to always reduce conspicuousness, whereas for T. spectabilis it could be advantageous to increase conspicuousness in certain circumstances to lure prey via UV-reflectance.

In conclusion, our results show that the colour contrast created by UV-reflectance in Australian crab spiders is beneficial for spiders most likely because of an increase in their foraging success. It also reveals the nature of the relationship between spider body condition and body colouration, showing that there is a positive relationship between colour contrast and condition. Also, contrary to other crab spiders, UV-reflective Thomisus spectabilis did not change their colour in order to match the background. Their response to background colouration and satiation may additionally depend on the composition of the prey population.

We would like to thank Sarah van Broekhoven for field assistance and anonymous reviewers for valuable comments on the manuscript.