1 Type of article: Original article 2 3 Pollinators show flower colour preferences but flowers with similar colours do not 4 attract similar pollinators. 5 Sara Reverté1*, Javier Retana1,2, José M. Gómez3,4, Jordi Bosch1 1 6 CREAF, Cerdanyola del Vallès, 08193 Barcelona, Spain 2 7 Univ. Autònoma Barcelona, Cerdanyola del Vallès, 08193 Barcelona, Spain 3 8 CSIC, Estación Experimental de Zonas Áridas, Almería, Spain 4 9 Universidad de Granada, Granada, Spain 10 11 Running title: Flower colour and pollinator composition 12 Author for correspondence: Sara Reverté. Email: s.reverte@creaf.uab.cat 13 14 This is a pre-copyedited, author-produced PDF of an article accepted for publication in Annals of botany (Ed. OUP) following peer review. The version of record Reverté, Sara, et.al. “Pollinators show flower colour preferen-ces but flowers with similar colours do not attract similar pollinators” in Annals of botany, vol. 118, issue 2 (Aug. 2016), p. 249-257 is available online at DOI 10.1093/aob/mcw103 1 1 Abstract 2 3 Background and aims. Colour is one of the main floral traits used by pollinators to locate 4 flowers. Although pollinators show innate colour preferences, the view that the colour of a 5 flower may be considered an important predictor of its main pollinators is highly 6 controversial because flower choice is highly context-dependent, and initial innate preferences 7 may be overridden by subsequent associative learning. Our objective is to establish whether 8 there is a relationship between flower colour and pollinator composition in natural 9 communities. 10 11 Methods. We measured the flower reflectance spectrum and pollinator composition in four 12 plant communities (85 plant species represented by 109 populations, and 32305 plant- 13 pollinator interactions in total). Pollinators were divided into six taxonomic groups: bees, 14 ants, wasps, coleopterans, dipterans and lepidopterans. 15 16 Key results. We found consistent associations between pollinator groups and certain colours. 17 These associations matched innate preferences experimentally established for several 18 pollinators and predictions of the pollination syndrome theory. However, flowers with similar 19 colours did not attract similar pollinator assemblages. 20 21 Conclusions. The explanation for this paradoxical result is that most flower species are 22 pollination generalists. We conclude that although pollinator colour preferences seem to 23 condition plant-pollinator interactions, the selective force behind these preferences has not 24 been strong enough to mediate the appearance and maintenance of tight colour-based plant- 25 pollinator associations. 2 1 2 Key words: colour preferences, floral colour, floral reflectance spectra, generalization, floral 3 traits, phylogenetic signal, plant-pollinator interactions, pollinator assemblage, pollination 4 syndromes. 5 6 INTRODUCTION 7 8 Pollinators use flower traits, such as odour, shape, size and colour, as cues to locate pollen- 9 nectar sources and discriminate between different flower species (Chittka and Raine 2006). 10 Although ultimate flower choice undoubtedly depends on a combination of stimuli, various 11 studies have demonstrated that some pollinators rely strongly on colour to make their foraging 12 decisions (Dafni et al. 1990; Heiling et al. 2003; Ômura and Honda 2005; Dötterl et al. 2014). 13 14 Diurnal pollinators have well-developed colour vision, which in most cases covers a wider 15 range of the spectrum than human vision. The vast majority of pollinators in temperate 16 regions are insects, and different visual models have been developed for different groups 17 (Chittka 1992; Troje 1993). The best studied species are the western honey bee, Apis 18 mellifera, and the bumblebee Bombus terrestris, which share a similar trichromatic system. 19 The visual system of the honey bee, which is usually used as a model for all bees, ranges 20 between 300 and 700 nm, with three types of photoreceptors peaking in the UV, blue and 21 green regions of the spectrum (corresponding to 344, 438 and 560 nm, respectively). Most 22 other insects studied so far also have a trichromatic system, but there are also species with 23 dichromatic (certain flies and coleopterans) and tetrachromatic (mostly butterflies) systems 24 (Briscoe and Chittka 2001). The trichromatic state is supposed to be ancestral, and the loss or 25 gain of photoreceptor types would be secondary (Briscoe and Chittka 2001). Insect colour 3 1 vision appeared long before the emergence of the angiosperms (Chittka 1997). For this 2 reason, it is believed that flowers tuned their visual signals to the sensory system of 3 pollinators, thus becoming as conspicuous and attractive as possible to them. In support of 4 this view, some pollinator species have been found to show innate preferences for certain 5 colours (Lunau and Maier 1995; Raine et al. 2006; Raine and Chittka 2007; Willmer 2011). 6 7 To this extent, colour constitutes one of the main traits used in pollination syndrome theory 8 (Faegri and Van der Pijl 1979). According to this theory, unrelated plant species adapted to 9 the same pollinators should show convergence of floral traits, including colour. For example, 10 bee-visited flowers are expected to be blue or violet, and beetle-visited flowers are expected 11 to be white or cream (Faegri and Van der Pijl 1979). Accordingly, some studies have shown 12 that colour changes among related plant species or between populations within a same species 13 are sometimes accompanied by changes in pollinator composition (Bradshaw and Schemske 14 2003; Rodríguez-Gironés and Santamaría 2004; Wolfe and Sowell 2006). Other studies have 15 shown pollinator colour preferences through experimental manipulations of flower colour 16 (Campbell et al. 2010). Ultimately, flower colour could be considered an important predictor 17 of pollinator group (Fenster et al. 2004). However, other studies do not support this view. 18 Colour preferences predicted by pollination syndromes are not always matched by innate 19 colour preferences (Lunau and Maier 1995), which in many cases are displayed at a finer 20 taxonomic level, such as genus or species, rather than functional group (Raine et al. 2006). 21 Other studies report that changes in flower colour are not always associated with distinct 22 pollinator assemblages (Cooley et al. 2008). 23 24 Even if pollinators show innate preferences for certain colours, actual flower choice in natural 25 communities may not reflect these preferences for two reasons. First, in a natural setting, 4 1 flower choice may be influenced by the presence and abundance of other co-pollinators 2 potentially competing for flower resources (Waser 1983; Lázaro et al. 2009). Second, colours 3 may also be used by pollinators as signals of floral rewards, so that initial innate preferences 4 may be modulated by subsequent associative learning (Gumbert 2000; Goyret et al. 2008). 5 Individual pollinators show remarkable plasticity and are known to switch plants in response 6 to changes in pollen or nectar levels (Heinrich 1979). Thus, the role of colour in determining 7 ultimate flower choice will depend on the interaction between innate and learned preferences. 8 A handful of studies have analysed the influence of floral colour on pollinator partitioning in 9 plant communities, and most of them conclude that floral colour is an important cue (McCall 10 and Primack 1992; Bosch et al. 1997; Hingston and McQuillan 2000; Wolfe and Sowell 2006; 11 Lázaro et al. 2008). However, colours in these studies were categorized as perceived by 12 humans. Waser et al. (1996) characterized colour based on wavelength measures that were 13 incorporated into the visual model of the honeybee to generate colour categories. In their 14 study, the association between floral colour category and pollinator composition narrowly 15 failed significance. Clearly, further research on this topic is needed. 16 17 The aim of this study is to establish whether there is a relationship between flower colour and 18 pollinator visitation in natural communities. To do this, we studied the flower reflectance 19 spectrum and flower visitor assemblages in four plant communities comprising 85 species and 20 109 plant populations. The four communities are located in the same geographic region, and 21 therefore have similar climatic conditions and share the same regional pool of pollinators. 22 Because the association between flower colour and pollinator composition may be affected by 23 phylogeny, our analyses account for phylogenetic relatedness. We ask two questions: 1) Do 24 different pollinator groups show preferences for certain flower colours? 2) Do plant species 25 with similar colours attract similar pollinator assemblages? Although apparently similar, these 5 1 two questions address colour-pollinator relationships from two different and complementary 2 perspectives. The first question addresses plant-pollinator relationships from the pollinators’ 3 point of view, and the second from the plants’ perspective. The existence of colour 4 preferences by different pollinator groups does not necessarily imply that plants with the same 5 colour attract similar pollinator assemblages. For example, dipterans may preferentially visit 6 flowers with high levels of reflectance in the yellow region of the spectrum, but different 7 yellow flowers may be visited by different pollinator assemblages. The distinction is 8 important because pollination syndrome theory relies on the assumption that plants with 9 similar traits have evolved to attract similar pollinators. 10 11 MATERIALS AND METHODS 12 13 Study area 14 15 We conducted our study in four Mediterranean communities near Barcelona (NE Spain), 16 whose geographic coordinates are given in Table 1. Distance between sites ranged from 10 to 17 66 km. The climate is Mediterranean, with a strong sea influence. Summers are dry and most 18 precipitation occurs in spring and autumn. Weather conditions are very similar across the four 19 sites (Table 1). We studied the vast majority of the entomophilous species in each community, 20 without selecting species based on their flower traits or pollinator visitation patterns (Table 1). 21 Therefore our plant sample was unbiased and representative of the flora of the region. Some 22 species were present in more than one community. In these cases, we sampled flower colour 23 and pollinator composition of each population separately (total: 85 species, 109 populations). 24 25 Pollinator assemblages 6 1 2 Pollinator data were obtained throughout the main general flowering period in the study area 3 (February-July). Very few species bloom in August. Each species was surveyed on several 4 days covering its entire flowering period. In two communities (CA and CO), pollinator 5 surveys were carried out by slowly walking along 25- or 50-m-long x 1-m-wide vegetation 6 transects. This was done several times throughout the day, from early morning to evening. In 7 the other two communities (GA and PA), selected individuals of each plant species in bloom 8 were observed during 4- or 5-minute periods every 1-2 h. All insects observed contacting the 9 reproductive organs of flowers were identified by sight and recorded. Pollinators were 10 assigned to one of six taxonomic groups: bees, ants, wasps, coleopterans, dipterans and 11 lepidopterans. From these surveys, we characterized the pollinator assemblage of each plant 12 species in each of the four communities as the proportion of visits from each pollinator group. 13 14 Flower colour measurement 15 16 Flowers of each species were collected and transported with a portable cooler to the 17 laboratory, where they were temporarily placed in a cold room at 4ºC. Flower reflectance 18 spectra were measured using an USB4000 spectrometer with a USB-DT deuterium tungsten 19 halogen source (Ocean Optics Inc., Dunedin, FL, USA) connected to a computer running 20 SpectraSuite (Ocean Optics). The light spectrum analysed ranges from 300 to 700 nm divided 21 into 0.22 nm intervals, and the spectrometer sensor was fixed at an angle of 45º from the 22 measuring area. Petals were mounted on an adhesive tape to obtain a flat surface, thus 23 minimizing reflectance variability due to uneven distances between the petals and the sensor. 24 For small flowers, we had to use several petals from different flowers to cover the entire 25 measuring area. Some species had corollas displaying two or more clearly different colours. 7 1 When these different colours were largely separated (for example, some Fabaceae in which 2 the keel is clearly differently coloured than the wings and the standard), we took colour 3 measures of the different parts separately and calculated a weighted mean (according to the 4 surface occupied by each part in the corolla), thus obtaining a single colour value. In corollas 5 showing only small colour markings such as nectar guides or small dots, a single measure was 6 taken. 7 8 Spectrometer readings incorporate a certain degree of noise. We used Avicol (Gomez 2006) 9 to clean this noise, thus smoothening the reflectance spectra. This correction (triangular 10 smoothing) is based on a floating mean with weights and a window size of 15 nm. We 11 measured 4-5 flowers per species except for three species for which we could only obtain one 12 measure. Different measures were averaged to obtain a single value per species and 13 community. 14 15 Colour characterization 16 17 We characterized flower colour using two sets of variables: colour descriptors and “colour 18 composition”. Colour descriptors (brightness, chroma and hue) are physical properties of 19 colours extracted directly from the reflectance spectrum (Endler 1990; Smith 2014). 20 Brightness is a measure of the total intensity of light reflected by a surface; chroma is a 21 measure of purity or saturation of a colour, and it is a function of how rapidly intensity 22 changes with wavelength; hue, represents the usual meaning of colour (such as red, pink, 23 yellow, etc.), and it is a function of the shape of the spectrum. These three colour descriptors 24 were calculated following Smith (2014). To obtain “colour composition”, we divided the 25 reflectance spectrum into 4 broad bands of 100 nm each (300-400, 400-500, 500-600, 600- 8 1 700 nm; Chittka et al. 1994). Breakpoints delimiting bands were chosen assuming that the 2 majority of the inflection points in floral reflectance spectra are located near 400, 500 and 600 3 nm (Chittka and Menzel 1992). Roughly, the first band corresponds to the UV part of the 4 spectrum, the second to the blue-violet, the third to the green-yellow and the fourth to the 5 orange-red (henceforth U, B, Y and R bands). We calculated the proportion of the reflectance 6 spectrum attributable to each band, obtained by dividing the brightness of each band by the 7 total brightness of the sample. By using the proportion, instead of the raw values of brightness 8 of each band, we avoid differences between chromatically equivalent spectra, i.e., spectra 9 with the same shape but different brightness (Endler 1990). It is important to note that flower 10 colours result from the combination of reflectance levels across the spectrum. For example, 11 white flowers reflect from 400 to 700 nm and yellow flowers from 500 to 700 nm. Lilac-pink 12 flowers reflect in the blue and red regions with a variable proportion in the yellow region, and 13 purple flowers reflect in the UV, blue and red regions of the spectrum (see Table S1 14 [Supplementary Information]). 15 16 Data analysis 17 18 Phylogenetic signal of colour variables 19 20 To establish whether colour was phylogenetically constrained, we built a phylogenetic tree of 21 the 85 species using Phylocom (Webb et al. 2008), with family names following the 22 Angiosperm Phylogeny Group classification (The Angiosperm Phylogeny Group 2009). We 23 used the “bladj” function in Phylocom to achieve an ultrametric rooted tree. Polytomies 24 generated by the program were hand-resolved. All distances between families (assessed as 25 millions of years of divergence) and some distances between genera were obtained from the 9 1 database www.timetree.org (Hedges et al. 2006). Additional distances between genera and 2 distances between species were extracted from the literature [Allan and Porter, 2000; Lavin et 3 al., 2005 (Fabaceae); Barres et al., 2013 (Cardueae); Bremer, 2009 (Ericales); Guzmán and 4 Vargas, 2009 (Cistaceae); Mansion et al., 2009 (Boraginaceae); Couvreur et al., 2010 5 (Brassicaceae); Bendiksby et al., 2011; Drew and Sytsma, 2012 (Lamiaceae); Koopman and 6 Baum, 2008 (Malvaceae); Ruiz-Sanchez et al., 2012 (Papaveraceae); Riina et al. 2013 7 (Euphorbiaceae); Banasiak et al. 2013 (Apiaceae)]. The resulting tree is shown in Figure S1 8 [Supplementary Information]. 9 10 We tested for the presence of phylogenetic signal in brightness, chroma and hue with the 11 Blomberg’s K test (Blomberg et al. 2003) using the “phylosig” function of the R package 12 “phytools” (Revell 2012). This was done for each community separately and for the 85 13 species together. We also tested for the presence of phylogenetic signal in flower colour 14 composition. To do this, we divided the colour spectrum into 40 bands of 10 nm each and 15 obtained the mean reflectance value of each band. We used 40 bands instead of the 4 bands 16 used to define colour composition to increase the resolution of this analysis. We then used the 17 generalized K statistic described by Adams (2014), Kmult, specifically developed for high- 18 dimensional multivariate data. This analysis was performed with the function “physignal” of 19 the package “geomorph” for R (Adams and Otarola-Castillo 2013). Again, we conducted this 20 analysis for each community separately and for the 85 species together. 21 22 Association between pollinator groups and regions of the colour spectrum 23 24 We explored possible preferences of the different pollinators groups for certain regions of the 25 colour spectrum by means of Canonical Correspondence Analyses (CCA) including the 10 1 pollinator assemblage of each population and relative brightness of the four above-mentioned 2 colour bands (U, B, Y and R). This was again done for the four communities separately and 3 for all species together. CCAs were performed using the function “CCorA” in the R package 4 “vegan” (Oksanen et al. 2015). 5 6 Relationship between flower colour and pollinator assemblages 7 8 To test whether flowers with similar colours had similar pollinator assemblages, we 9 conducted a partial Mantel test between distance matrices of colour composition (combination 10 of the proportion of the four colour bands) and pollinator assemblages, including the 11 phylogenetic distance matrix as covariable. We then followed the same approach to test the 12 association between pollinator assemblages and the other colour variables (brightness, 13 chroma, hue). We used Bray-Curtis distances between pairs of species for pollinator 14 composition and colour composition, and Euclidean distances between species for brightness, 15 chroma and hue. These analyses were done for each community separately and for the four 16 communities lumped together. In the latter case, because the same species may attract 17 different pollinators in different communities, we maintained all 109 populations. For plant 18 species present in more than one community, we assigned a very low value of divergence 19 (1000 years) between conspecific populations. These analyses were performed with the 20 function “mantel.partial” in the R package “vegan”. 21 22 RESULTS 23 24 We recorded 32,315 plant-pollinator interactions. The number of interactions recorded in each 25 community ranged from 3505 (CA) to 13673 (CO) (Table S2 [Supplementary 11 1 Information]). Mean number of interactions per population was 206 in CA (range: 43 to 2 1454); 297 in CO (range: 42 to 1911); 364 in GA (range: 26 to 1730) and 307 in PA (range: 3 31 to 1359). Overall, 88.5 % of the populations surveyed had more than 50 recorded 4 interactions. Most of the pollinators recorded were bees, accounting for 48.9% of the flower 5 visits. The second most frequent group was coleopterans (21.5% of the interactions), followed 6 by ants (14.7%), dipterans (8.8%), lepidopterans (3.5%) and wasps (2.5%). Bees and 7 coleopterans were the two most abundant groups in all four communities, except CA, which 8 was largely dominated by ants (Table S2 [Supplementary Information]). 9 10 The four communities showed a high degree of similarity in flower colours. The most 11 common floral colour in the four communities was lilac-pink (30-50% of the species), 12 followed by white (16-29%) (Table S1 [Supplementary Information]). UV-yellow flowers 13 were also well represented (14-24%), although they were lacking in GA. Yellow (12- 15%), 14 purple (4- 9%) and green (4- 6%) flowers were less frequent. The association between colour 15 categories and pollinator composition is shown in Table S3 [Supplementary Information]. 16 17 Phylogenetic signal of colour variables 18 19 All colour descriptors considered showed significant phylogenetic signal when the four 20 communities were pooled together (Table 2). In most cases, however, significance was lost 21 when the communities were analysed separately, possibly due to small sample sizes. In all 22 cases, K and Kmult values were lower than 1, indicating that related species were less similar 23 than expected under the Brownian motion evolution model. 24 25 Association between pollinator groups and regions of the colour spectrum 12 1 2 CCAs revealed clear associations between certain pollinator groups and certain colours (Figs. 3 1 and 2). Visual inspection of the resulting biplots revealed that some of these patterns were 4 relatively consistent across the four communities (Fig. 1, Table 4). 5 6 Overall, bees were associated with purple flowers, and ants with UV-Yellow and green 7 flowers. Wasps and dipterans were mostly associated with UV-yellow flowers. Coleopterans 8 were associated with white and yellow flowers, and lepidopterans with pink flowers (Table 4, 9 Fig. 2). 10 11 Relationship between flower colour and pollinator assemblages 12 13 Results from the partial Mantel test showed no significant association between flower colour 14 and pollinator assemblages (Table 3). Plants with similar colour descriptors, including colour 15 composition, did not attract similar pollinator assemblages in any of the communities, and 16 similar results were obtained when data from the four communities were lumped together. 17 18 DISCUSSION 19 20 We traced the phylogenetic signal of floral colour under a neutral evolution model (Brownian 21 motion). In this model, changes occur slowly and gradually along phylogenetic branches, and 22 character similarity between species is related to phylogenetic proximity (Blomberg et al. 23 2003). Phylogenetic signal for flower colour was mostly lacking when we analysed the four 24 communities separately, but when all species were lumped together all colour properties 25 measured showed significant phylogenetic signal (related species were more similar in colour 13 1 than expected from a random association between species and colours). Notwithstanding this 2 influence of phylogeny on flower colour, the fact that all K values were lower than 1 indicates 3 that similarity among related species tends to be lower than expected under Brownian motion. 4 Most previous studies have found lack of phylogenetic signal in flower colour (Smith et al. 5 2008; Arnold et al. 2009; McEwen and Vamosi 2010). However, one recent study on four 6 Himalayan plant communities (Shrestha et al. 2014) obtained results similar to ours 7 (significant phylogenetic signal with K values < 1). A corollary of our phylogenetic analysis 8 is that the different colour categories considered are widely spread across the phylogenetic 9 tree. This outcome is in agreement with reports showing important changes in flower colour 10 as a result of relatively simple genetic changes (Rausher 2008). 11 12 We analysed the colour preferences of the pollinator groups present in our communities. 13 Results of the CCAs showed notable coincidences among communities and provided 14 compelling evidence for the association between pollinator groups and certain colours. These 15 associations were less clear in the communities with lower number of plant species, but were 16 very apparent when the results of all 109 populations were lumped together (Fig. 2). 17 Importantly, these associations are notably congruent with innate colour preferences found for 18 specific pollinators (Lunau and Maier 1995; Willmer 2011). For example, inexperienced 19 Eristalis tenax hover flies were found to land only on artificial flowers with yellow stimuli in 20 the range of 520 to 600 nm (Lunau and Maier 1995). Naive honey bees showed preference for 21 stimuli reflecting between 410 and 530 nm (Giurfa et al. 1995), and bumblebees preferred 22 artificial flowers reflecting in one or two of the three colour bands corresponding to their 23 photoreceptor types (ultraviolet, blue, green) (Lunau 1992). Papilio demoleus butterflies 24 selected blue, green, and red colours while neglecting yellow (Ilse and Vaidya 1956) and 25 Pieris brassicae preferred artificial flowers reflecting at 450 and 600 nm (Scherer and Kolb 14 1 1987). The hawkmoth Macroglossum stellatarum was shown to have a strong preference for 2 wavelengths around 440 nm and a weaker preference for wavelengths of 540 nm (Kelber 3 1997; Telles et al. 2014). 4 5 Importantly, the above-mentioned pollinator-colour associations found in our study are 6 consistent with colour preferences described in pollination syndromes, according to which 7 bees favour blue, flies yellow and white, lepidopterans pink and red, coleopterans white and 8 cream and wasps favour brown and yellow flowers (Faegri and Van der Pijl 1979; Willmer 9 2011). Waser et al. (1996) explored the association between pollinator groups and bands of 10 the colour spectrum in a natural community in Germany. Their results narrowly failed 11 significance, but the observed trends were also consistent with associations predicted by 12 pollination syndromes. Other community studies using more subjective (human-perceived) 13 colour measures also found associations that were more or less consistent with pollination 14 syndromes (McCall and Primack 1992; Wolfe and Sowell 2006; Lázaro et al. 2008). 15 16 However, our results cannot be interpreted as supporting pollination syndrome theory because 17 flowers with similar colours did not attract similar pollinators (Table 3). That is, pollinator 18 preferences are a necessary but not sufficient condition for the establishment of colour-based 19 pollination syndromes. The lack of relationship between flower colour and pollinator 20 composition can be explained by the influence of other traits, or trait combinations, on flower 21 choices. Some studies have found phenology (Herrera 1988), pollen/nectar rewards (Bosch et 22 al. 1997), or corolla depth (Stang et al. 2007) to be the main drivers of pollinator partitioning 23 in flower communities. A second factor contributing to the lack of relationship between 24 flower colour and pollinator composition is pollinator generalisation. As in most temperate 25 systems (Herrera 1996; Waser et al. 1996), the majority of plant species in our study are 15 1 pollinator generalists (with two or more pollinator groups each accounting for more than 10% 2 of the observed visits), and therefore cannot be readily assigned to a given pollination 3 syndrome. Even then, we could have found a correspondence between certain flower colours 4 and pollinator assemblages (rather than single pollinator groups), but this was clearly not the 5 case. 6 7 Our results are in agreement with the findings of Rezende et al. (2007), who found that plant- 8 pollinator networks are more strongly associated to animal than to plant phylogeny. That is, 9 related animal species are more likely to share host plants than related plant species are to 10 share pollinator visitors. This result is attributed to pollinator mobility (Bascompte and 11 Jordano 2007), which allows pollinators to play a more active role in the definition of 12 interaction identity. 13 14 Recently, Renoult et al. (2014) have used a novel approach to analyze the association between 15 flower colour and generalization in pollinators. They argue that pollinators should 16 preferentially visit flowers that are most conspicuous to them, and that colour perception 17 should be analyzed as a measure of colour conspicuousness between emitters (flowers) and 18 receivers (insects). They measure flower and background colour, and use evidence from 19 various sources to assign one of eleven described visual systems to each pollinator species. 20 Then they calculate ‘colour matching’ for each plant-pollinator pair as a measure of how a 21 floral colour stands out from the environment for each visual system. They find that colour 22 matching influences generalization in one of three regions studied. However, they also find 23 that other factors (spatiotemporal overlap and co-abundance between plants and pollinators) 24 play a greater role as determinants of plant generalization. The results of Renoult et al. (2014) 25 in general, and the strong influence of these other factors in particular, support our conclusion 16 1 that even though pollinator colour preferences seem to be conditioning plant-pollinator 2 interactions, the selective force behind these preferences may not been strong enough to 3 mediate the appearance and maintenance of tight colour-based plant-pollinator associations. 4 5 ACKNOWLEDGEMENTS 6 7 Many people (including A. Bonet, A. Rodrigo, S. Osorio, H. Barril-Graells, C. Primante, A. 8 Martín González and V. Flo) helped during pollinator surveys. We are also grateful to C- 9 Hernández-Castellano, M. Escolà, A. Torné-Noguera, R. Molowny-Horas, G. Bagaria and M. 10 Galbany-Casals for their assistance during different phases of the study. We also thank two 11 anonymous reviewers for their valuable comments. This study was supported by the Spanish 12 MICINN, projects CICYT CGL2005-00491, CICYT CGL2009-12646, MICINN CGL2012- 13 34736, Junta de Andalucía P11-RNM-7676 and CONSOLIDER CSD2008-0040. 14 15 17 1 LITERATURE CITED 2 3 Adams DC. 2014. A generalized K statistic for estimating phylogenetic signal from shape 4 and other high-dimentional multivariate data. Systematic Biology 63: 685–697. 5 Adams DC, Otarola-Castillo E. 2013. geomorph: an R package for the collection and 6 analysis of geometric morphometric shape data. Methods in Ecology and Evolution 4: 393– 7 399. 8 Allan GJ, Porter JM. 2000. Tribal delimitation and phylogenetic relationships of Loteae and 9 Coronilleae (Faboideae: Fabaceae) with special reference to Lotus: Evidence from nuclear 10 ribosomal ITS sequences. American Journal of Botany 87: 1871–1881. 11 Arnold SEJ, Savolainen V, Chittka L. 2009. Flower colours along an alpine altitude 12 gradient, seen through the eyes of fly and bee pollinators. Arthropod-Plant Interactions 3: 27– 13 43. 14 Banasiak Ł, Piwczyński M, Uliński T, Downie SR, Watson MF, Shakya B, Spalik K. 15 2013. Dispersal patterns in space and time: a case study of Apiaceae subfamily Apioideae (M 16 Carine, Ed.). Journal of Biogeography 40: 1324–1335. 17 Barres L, Sanmartín I, Anderson CL, Susanna A, Buerki S, Galbany-Casals M, 18 Vilatersana R. 2013. Reconstructing the evolution and biogeographic history of tribe 19 Cardueae (compositae). American Journal of Botany 100: 867–882. 20 Bascompte J, Jordano P. 2007. Plant-Animal Mutualistic Networks: The Architecture of 21 Biodiversity. Annual Review of Ecology, Evolution, and Systematics 38: 567–593. 22 Beardsley PM, Yen A, Olmstead RG. 2003. AFLP phylogeny of Mimulus section 23 Erythranthe and the evolution of hummingbird pollination. Evolution 57: 1397–1410. 24 Bendiksby M, Thorbek L, Scheen AC, Lindqvist C, Ryding O. 2011. An updated 25 phylogeny and classification of Lamiaceae subfamily Lamioideae. Taxon 60: 471–484. 26 Blomberg SP, Garland T, Ives AR. 2003. Testing for phylogenetic signal in comparative 27 data: behavioral traits are more labile. Evolution 57: 717–745. 28 Bosch J, Retana J, Cerdá X. 1997. Flowering phenology, floral traits and pollinator 29 composition in a herbaceous Mediterranean plant community. Oecologia 109: 583–591. 30 Bradshaw HD, Schemske DW. 2003. Allele substitution at a flower colour locus produces a 31 pollinator shift in monkeyflowers. Nature 426: 176–178. 32 Bremer B. 2009. Asterids. The Timetree of Life: 177–187. 33 Briscoe AD, Chittka L. 2001. The evolution of color vision in insects. Annual Review of 34 Entomology 46: 471–510. 35 Campbell DR, Bischoff M, Lord JM, Robertson AW. 2010. Flower color influences insect 36 visitation in alpine New Zealand. Ecology 91: 2638–2649. 37 Chittka L. 1992. The colour hexagon: a chromaticity diagram based on photoreceptor 38 excitations as a generalized representation of colour opponency. Journal of Comparative 39 Physiology A 170: 533–543. 40 Chittka L. 1997. Bee color vision is optimal for coding flower color, but flower colors are 41 not optimal for being coded - why? Israel Jornal of Plant Sciences 45: 115–127. 42 Chittka L, Menzel R. 1992. The evolutionary adaptation of flower colours and insect 18 1 pollinators’ colour vision. Journal of Comparative Physiology A 171: 171–181. 2 Chittka L, Raine NE. 2006. Recognition of flowers by pollinators. Current Opinion in Plant 3 Biology 9: 428–435. 4 Chittka L, Shmida A, Troje N, Menzel R. 1994. Ultraviolet as a component of flower 5 reflections, and the colour perception of hymenoptera. Vision Research 34: 1489–1508. 6 Cooley AM, Carvallo G, Willis JH. 2008. Is floral diversification associated with pollinator 7 divergence? Flower shape, flower colour and pollinator preference in Chilean Mimulus. 8 Annals of Botany 101: 641–650. 9 Couvreur TLP, Franzke A, Al-Shehbaz I, Bakker FT, Koch M, Mummenhoff K. 2010. 10 Molecular phylogenetics, temporal diversification, and principles of evolution in the mustard 11 family (Brassicaceae). Molecular Biology and Evolution 27: 55–71. 12 Dafni A, Bernhardt P, Shmida A, Ivri Y, Greenbaum S, O’Toole C, Losito L. 1990. Red 13 bowl-shaped flowers: convergence for beetle pollination in the Mediterranean region. Israel 14 Journal of Botany 39: 81–92. 15 Dötterl S, Glück U, Jürgens A, Woodring J, Aas G. 2014. Floral reward, advertisement and 16 attractiveness to honey bees in dioecious Salix caprea. PloS ONE 9: e93421. 17 Drew BT, Sytsma KJ. 2012. Phylogenetics, biogeography, and staminal evolution in the 18 tribe Mentheae (Lamiaceae). American Journal of Botany 99: 933–953. 19 Endler JA. 1990. On the measurement and classification of colour in studies of animal colour 20 patterns. Biological Journal of the Linnean Society 41: 315–352. 21 Faegri K, Van der Pijl L. 1979. The principles of pollination ecology. Oxford; New York: 22 Pergamon Press. 23 Fenster CB, Armbruster WS, Wilson P, Dudash MR, Thomson JD. 2004. Pollination 24 syndromes and floral specialization. Annual Review of Ecology, Evolution, and Systematics 25 35: 375–403. 26 Giurfa M, Núñez J, Chittka L, Menzel R. 1995. Colour preferences of flower-naive 27 honeybees. Journal of Comparative Physiology A 177: 247–259. 28 Gomez D. 2006. AVICOL, a program to analyse spectrometric data. Last update october 29 2011. Free executable available at http://sites.google.com/site/avicolprogram/ or from the 30 author at dodogomez@yahoo.fr. 31 Goyret J, Pfaff M, Raguso RA, Kelber A. 2008. Why do Manduca sexta feed from white 32 flowers? Innate and learnt colour preferences in a hawkmoth. Naturwissenschaften 95: 569– 33 576. 34 Gumbert A. 2000. Color choices by bumble bees (Bombus terrestris ): innate preferences and 35 generalization after learning. Behavioral Ecology Sociobiology 48: 36–43. 36 Guzmán B, Vargas P. 2009. Historical biogeography and character evolution of Cistaceae 37 (Malvales) based on analysis of plastid rbcL and trnL-trnF sequences. Organisms Diversity 38 and Evolution 9: 83–99. 39 Hedges SB, Dudley J, Kumar S. 2006. TimeTree: A public knowledge-base of divergence 40 times among organisms. Bioinformatics 22: 2971–2972. 41 Heiling AM, Herberstein ME, Chittka L. 2003. Pollinator attraction: Crab-spiders 42 manipulate flower signals. Nature 421: 334. 43 Heinrich B. 1979. Majoring and minoring by foraging bumblebees, Bombus vagans: an 44 experimental analysis. Ecology 60: 245–255. 19 1 Herrera J. 1988. Pollination relationships in southern Spanish Mediterranean shrublands. 2 Journal of Ecology 76: 274–287. 3 Herrera CM. 1996. Floral traits and plant adaptation to insect pollinators: a devil’s advocate 4 approach. In: Lloyd DG,, In: Barrett SCH, eds. Floral biology: Studies on floral evolution in 5 animal-pollinated plants. New York: Chapman & Hall, 65–87. 6 Hingston AB, McQuillan PB. 2000. Are pollination syndromes useful predictors of floral 7 visitors in Tasmania ? Austral Ecology 25: 600–609. 8 Ilse D, Vaidya V. 1956. Spontaneous feeding response to colours in Papilio demoleus L. 9 Proceedings of the Indian Academy of Sciences, section B 43: 23–31. 10 Kelber A. 1997. Innate preferences for flower features in the hawkmoth Macroglossum 11 stellatarum. Journal of Experimental Biology 200: 827–836. 12 Koopman MM, Baum DA. 2008. Phylogeny and biogeography of tribe Hibisceae 13 (Malvaceae) on Madagascar. Systematic Botany 33: 364–374. 14 Lavin M, Herendeen PS, Wojciechowski MF. 2005. Evolutionary rates analysis of 15 Leguminosae implicates a rapid diversification of lineages during the tertiary. Systematic 16 biology 54: 575–594. 17 Lázaro A, Hegland SJ, Totland Ø. 2008. The relationships between floral traits and 18 specificity of pollination systems in three Scandinavian plant communities. Oecologia 157: 19 249–257. 20 Lázaro A, Lundgren R, Totland Ø. 2009. Co-flowering neighbors influence the diversity 21 and identity of pollinator groups visiting plant species. Oikos 118: 691–702. 22 Lunau K. 1992. Innate recognition of flowers by bumble bees: orientation of antennae to 23 visual stamen signals. Canadian Journal of Zoology 70: 2139–2144. 24 Lunau K, Maier EJ. 1995. Innate colour preferences of flower visitors. Journal of 25 Comparative Physiology A 177: 1–19. 26 Mansion G, Selvi F, Guggisberg A, Conti E. 2009. Origin of Mediterranean insular 27 endemics in the Boraginales: Integrative evidence from molecular dating and ancestral area 28 reconstruction. Journal of Biogeography 36: 1282–1296. 29 McCall C, Primack RB. 1992. Influence of flower characteristics, weather, time of day, and 30 season on insect visitation rates in three plant communities. American Journal of Botany 79: 31 434–442. 32 McEwen JR, Vamosi JC. 2010. Floral colour versus phylogeny in structuring subalpine 33 flowering communities. Proceedings of the Royal Society B 277: 2957–2965. 34 Oksanen J, Blanchet FG, Kindt R, Legendre P, Minchin PR, O’Hara RB, Simpson GL, 35 Solymos P, Stevens HMH, Wagner H. 2015. vegan: Community Ecology Package. R 36 package version 2.3-0. http://CRAN.R-project.org/package=vegan. 37 Ômura H, Honda K. 2005. Priority of color over scent during flower visitation by adult 38 Vanessa indica butterflies. Oecologia 142: 588–596. 39 Raine NE, Chittka L. 2007. The adaptive significance of sensory bias in a foraging context: 40 Floral colour preferences in the bumblebee Bombus terrestris. PLoS ONE 2: 1–8. 41 Raine NE, Ings TC, Dornhaus A, Saleh N, Chittka L. 2006. Adaptation, Genetic Drift, 42 Pleiotropy, and History in the Evolution of Bee Foraging Behavior. Advances in the Study of 43 Behavior 36: 305–354. 44 Rausher MD. 2008. Evolutionary Transitions in Floral Color. International Journal of Plant 20 1 Sciences 169: 7–21. 2 Revell LJ. 2012. phytools: An R package for phylogenetic comparative biology (and other 3 things). Methods in Ecology and Evolution 3: 217–223. 4 Rezende EL, Lavabre JE, Guimarães PR, Jordano P, Bascompte J. 2007. Non-random 5 coextinctions in phylogenetically structured mutualistic networks. Nature 448: 925–929. 6 Riina R, Peirson J a., Geltman D V., Molero J, Frajman B, Pahlevani A, Barres L, 7 Morawetz JJ, Salmaki Y, Zarre S, Kryukov A, Bruyns P V., Berry PE. 2013. A 8 worldwide molecular phylogeny and classification of the leafy spurges, Euphorbia subgenus 9 Esula (Euphorbiaceae). Taxon 62: 316–342. 10 Rodríguez-Gironés M a., Santamaría L. 2004. Why are so many bird flowers red? PLoS 11 Biology 2. 12 Ruiz-Sanchez E, Rodriguez-Gomez F, Sosa V. 2012. Refugia and geographic barriers of 13 populations of the desert poppy, Hunnemannia fumariifolia (Papaveraceae). Organisms 14 Diversity & Evolution 12: 133–143. 15 Scherer C, Kolb G. 1987. Behavioral experiments on the visual processing of color stimuli 16 in Pieris brassicae L. (Lepidoptera). Journal of Comparative Physiology A 160: 645–656. 17 Shrestha M, Dyer AG, Bhattarai P, Burd M. 2014. Flower colour and phylogeny along an 18 altitudinal gradient in the Himalayas of Nepal. Journal of Ecology 102: 126–135. 19 Smith SD. 2014. Quantifying Color Variation: Improved Formulas for Calculating Hue with 20 Segment Classification. Applications in Plant Sciences 2: 1300088. 21 Smith SD, Ané C, Baum D a. 2008. The role of pollinator shifts in the floral diversification 22 of Iochroma (Solanaceae). Evolution 62: 793–806. 23 Stang M, Klinkhamer PGL, van der Meijden E. 2007. Asymmetric specialization and 24 extinction risk in plant–flower visitor webs: a matter of morphology or abundance? Oecologia 25 151: 442–453. 26 Telles FJ, Lind O, Henze JM, Rodríguez-Gironés MA, Goyret J, Kelber A. 2014. Out of 27 the blue: the spectral sensitivity of hummingbird hawkmoths. Journal of Comparative 28 Physiology A 200: 537–546. 29 The Angiosperm Phylogeny Group. 2009. An update of the Angiosperm Phylogeny Group 30 classification for the orders and families of flowering plants: APG III. Botanical Journal of 31 the Linnean Society 161: 105–121. 32 Troje N. 1993. Spectral Categories in the learning behaviour of blowflies. Zeitschrift für 33 Naturforschung 48: 96–104. 34 Waser NM. 1983. Competition for pollination and floral character differences among 35 sympatric plant species: a review of evidence. In: C. E. Jones,, In: Little RJ, eds. Handbook 36 of experimental pollination biology. New York: Van Nostrand Reinhold New York, 277–293. 37 Waser NM, Chittka L, Price M V., Williams NM, Ollerton J. 1996. Generalization in 38 pollination systems, and why it matters. Ecology 77: 1043–1060. 39 Webb CO, Ackerly DD, Kembel SW. 2008. Phylocom: software for the analysis of 40 phylogenetic community structure and trait evolution. Bioinformatics 24: 2098–2100. 41 Willmer P. 2011. Pollination and floral ecology. New Jersey: Princeton University Press. 42 Wolfe LM, Sowell DR. 2006. Do pollination syndromes partition the pollinator community? 43 A test using four sympatric morning glory species. International Journal of Plant Sciences 44 167: 1169–1175. 21 1 2 22 1 Figure 1. CCA biplots of pollinator groups and bands of the colour spectrum (colour squares) 2 corresponding to UV, blue, yellow, and red, respectively, in each of the four communities 3 (CA, CO, GA, PA). Each dot represents a plant population and dot colour corresponds to 4 flower colour categories as shown in the legend (see Table S1 [Supplementary Information] 5 for example spectra of each category). Bee = bees, Ant = ants, Was = wasps, Dip = dipterans, 6 Col = coleopterans, Lep = lepidopterans. 7 23 1 Figure 2. CCA biplot of pollinator groups and bands of the colour spectrum (colour squares) 2 corresponding to UV, blue, yellow, and red, respectively (data from the four communities 3 lumped together). Each dot represents a plant population and dot colour corresponds to the 4 flower colour categories as shown in the legend (see Table S1 [Supplementary Information] 5 for example spectra of each category). Bee = bees, Ant = ants, Was = wasps, Dip = dipterans, 6 Col = coleopterans, Lep = lepidopterans. 24 1 Table 1. Descriptors of the four communities studied, including vegetation type, dominant species, location, elevation, mean annual 2 precipitation, mean annual temperature, and number of species sampled. 3 Community Vegetation Dominant species Location Elevation Mean annual Mean annual Plant (m.a.s.l.) precipitation temperature species (mm) (ºC) studied Hyparrhenia hirta, Canet de Mar CA Grassland 50 590 16.1 17 Brachypodium retusum 41º 35’ N, 2º 34’ E Hyparrhenia hirta, Open Collserola CO Foeniculum vulgare, 280 630 14.8 46 scrubland 41º 24’ N, 2º 6’ E Cistus monspeliensis Quercus coccifera, Garraf GA Scrubland Rosmarinus officinalis, 330 580 15.7 25 41º 16’ N, 1º 55’ E Pistacia lentiscus Quercus coccifera, El Papiol PA Scrubland Ulex parviflorus, 150 628 15.4 21 41º 27’ N, 2º 0’ E Pistacia lentiscus 4 25 1 Table 2. Analyses of phylogenetic signal for colour descriptors brightness, chroma, hue 2 (Blomberg’s K values) and colour composition (Kmult values) in the four study communities 3 separately and lumped together. Significant results (p < 0.05) in bold. 4 Community n Brightness Chroma Hue Colour species composition CA 17 0.73 0.78 0.85 0.48 CO 46 0.35 0.57 0.71 0.33 GA 25 0.26 0.61 0.44 0.25 PA 21 0.34 0.52 0.76 0.26 CA+CO+GA+PA 85 0.57 0.60 0.71 0.38 5 26 1 Table 3. R values of phylogenetically-controlled partial Mantel tests between colour 2 descriptors and pollinator composition in the four communities and overall (data of the four 3 communities lumped together). All results are non-significant. Community Brightness Chroma Hue Colour composition CA -0.089 0.096 0.199 0.158 CO -0.002 -0.031 0.015 -0.080 GA -0.025 0.115 -0.082 -0.038 PA -0.103 -0.063 -0.035 0.051 Overall 0.021 -0.023 0.019 0.006 4 5 27 1 Table 4. Relationships between pollinator groups and the different floral colours in the four 2 communities and overall (data of the four communities lumped together) estimated visually 3 from the CCA biplots of Fig. 1 for the four communities and Fig. 2 for the overall tendencies. 4 Community Bees Ants Wasps Coleopterans Dipterans Lepidopterans White Purple Purple White White CA UV-yellow Yellow UV-yellow UV-yellow UV-yellow Pink Pink Yellow White Purple Yellow CO Purple Yellow Green Yellow UV-yellow UV-yellow White Purple Yellow White Yellow GA Purple Pink Pink Green Pink Green Yellow Purple White PA Green White Pink Pink Pink Pink White UV-yellow White Overall Purple UV-yellow UV-yellow Pink Green Yellow 5 6 28 1 Suplementary Information 2 Figure S1. Phylogenetic tree of the 85 species surveyed. Colour dots represent the colour 3 category of each plant species (see colour legend in Fig. 1). 4 29 1 Table S1. Percentage of species in each of the four communities (CA, CO, GA, PA) 2 corresponding to the flower colour categories defined by Chittka et al. (1994) with an 3 example species of each category. 4 30 1 Table S2. Number (and percentage) of pollinators of each pollinator group surveyed in each 2 of the four communities (CA, CO, GA, PA). 3 31 1 Table S3. Mean percentage of visits of each pollinator group in each colour category (data of 2 the four communities lumped together). 3 4 5 32