Flower colour and phylogeny along an altitudinal gradient in the Himalayas of Nepal



  1. Both the phylogenetic structure and trait composition of flowering plant communities may be expected to change with altitude. In particular, floral colours are thought to vary with altitude because Hymenoptera typically decline in importance as pollinators while Diptera and Lepidoptera become more important at higher elevations. Thus, ecological filtering among elevation zones and competitive processes among co-occurring species within zones could influence the floral chromatic cues present at low and high elevations.
  2. We collected data from 107 species of native flowering plants in the Himalaya Mountains of central Nepal over an elevation range of 900–4100 m, which includes habitat ranging from subtropical to subalpine within a relatively small geographical area. Using a phylogenetic framework, we asked whether and how flower colour diversity differed between species assemblages at lower and higher elevation, between monocots and eudicots, and between our sample from central Nepal and angiosperms from other regions of the world.
  3. There was significant phylogenetic clustering in the communities as a result of monocots, particularly orchids, which were found overwhelmingly at lower elevations. Phylogenetic signal for floral colours indicated that related species had colours that were more disparate than expected under Brownian motion evolution. Floral colours were significantly more diverse in the higher elevation subalpine zone than in the subtropical zone. However, the chromatic cues at both elevations were consistent with the hue discrimination abilities of the trichromatic hymenopteran visual system.
  4. Synthesis. Flower colour is not highly differentiated between subtropical and subalpine vegetation due to differences in the available orders of insect pollinators, or by the rate or direction of colour evolution in the lineages composing the two communities. Differences in colour diversity between zones may reflect differences in the ecologically available morphospace based on pollinator species richness and the constancy of their foraging behaviour. The chromatic signals present in Nepali species are similar to the signals found in insect-pollinated floras of other regions of the world.


Both evolutionary history and functional traits may influence the mix of species that contribute to the diversity and structure of biotic communities (Webb et al. 2002). Among flowering plants, plant–pollinator interactions may be an important contributor to community structure (Sargent & Ackerly 2008), and floral colour is undoubtedly an important functional trait that affects animal pollination (Schemske & Bradshaw 1999; Fenster et al. 2004; Streisfeld & Kohn 2007; Hopkins & Rausher 2012). Is floral colour, therefore, an important determinant of the composition and structure of angiosperm communities? Do floral colours reflect the visual capacities of the pollinator fauna, or the phylogenetic origins of the local flora? How do floral colours change over geographical or environmental gradients? Evidence to address these questions is surprisingly limited, although new insights from vision science combined with advances in phylogenetic comparative methods offer the potential for rapid advances in our understanding of colour structure in angiosperm communities (Shrestha et al. 2013; Stournaras et al. 2013).

Here, we investigate floral colours in two vegetation types along a steep elevation gradient in central Nepal. Elevation gradients are a particularly appealing natural experiment for the role that floral colour plays in community structure because pollinator faunas typically change with altitude. In particular, hymenopterans often decrease in abundance and species richness while dipterans and lepidopterans have been reported to become more frequent with increasing altitude. This pattern has been observed in mountainous areas of New Zealand (Primack 1983), the Snowy Mountains of southeastern Australia (Inouye & Pyke 1988), the Chilean Andes (Arroyo, Primack & Armesto 1982), the European Alps (Müller 1880), the White Mountains of New Hampshire (McCall & Primack 1992), the Colorado Rocky Mountains (Inouye & Pyke 1988; Kearns 1992) and the Wasatch and Uinta Mountains of Utah (Warren, Harper & Booth 1988).

Flies are reported to favour flowers that appear white or yellow to human vision (Lázaro, Hegland & Totland 2008), and a shift to a fly-dominated pollinator fauna at higher altitudes may be expected to favour an enrichment in species with these floral colours. However, human colour categories are unlikely to be ecologically relevant for an assessment of floral colours (Shrestha, Dyer & Burd 2013c), and the apparent prominence of white and yellow flowers in montane and arctic habitats (Kevan 1972; Wardle 1978) must be taken with caution. Arnold, Savolainen & Chittka (2009), for example, found no evidence of altitudinal change in flower colours evaluated with respect to bee and fly visual capacities, over elevations ranging from 700 to 1600 m in alpine areas of Norway.

Previous work has shown that insect-pollinated flowers in Israel, Australia and New Zealand often have rapid changes in spectral reflectance near 400 and 500 nm wavelengths (Chittka & Menzel 1992; Dyer et al. 2012; Bischoff et al. 2013). These are the wavelengths of the most sensitive colour discrimination by the trichromatic visual system of hymenopterans (von Helversen 1972; Peitsch et al. 1992), and rapid reflectance changes at these wavelengths would render a floral colour especially distinctive to bees (Chittka & Menzel 1992). Given the abundance of bees such as Bombus in Nepal (Williams et al. 2010) and their likely importance as selective agents on floral traits, we anticipate that the Nepali flora will also show the spectral signatures that are typical of other flowers with hymenopteran pollination.

Within plant communities at a given elevation, floral colours can be affected by the phylogenetic history of the coexisting species and by competition or facilitation among those species for pollinator attention. Of five subalpine meadows between 1500 and 1700 m elevation studied by McEwen & Vamosi (2010) in Alberta, Canada, one had a statistically significant degree of phylogenetic clustering and one showed a significant degree of floral colour diversity, but the patterns in other communities were not significantly different from random expectations. This study was exemplary in combining phylogenetic comparative analyses with spectrophotometric measurement of floral reflectance, although ‘colour’ was quantified by principal component analysis of the reflectance spectra without reference to any pollinator visual system. Although such a procedure is objective, it runs the risk that assessments of colour similarity and difference might be irrelevant to the way specific pollinators perceive colour.

While floral colours in montane and alpine plant communities have received some attention, hardly any assessments have been made of floral colour structure in tropical or subtropical communities. One such study is the work of Altshuler (2003), who compared four tropical forest sites in Costa Rica, Panama, Brazil and Peru. Floral colour, quantified by the wavelength that equally divided the area under a reflectance spectrum, was strongly related to bird or bee pollination, but there was no significant effect of closed vs. open habitat, despite the different ambient light environments that would potentially affect colour perception. Our investigation of floral colour in the Himalaya Mountains includes communities at both low latitudes and high elevations. The Himalayas began to rise in the Palaeocene following collision of the Indian subcontinent with Asia (Valdiya 2002). Fusion of the plates was completed 55–50 Ma (Palzett et al. 1996) and uplift of the Himalaya completed ca. 40–50 Ma (Molnar 1986). As a consequence of its location at the juncture of two land masses and its topographical diversity, Nepal is rich in terms of global floral diversity (Chaudhary 1998) and ideally placed to evaluate altitudinal effects on floral coloration, especially because there is rapid change in altitude over relatively short distances.

In the current study, we address the following questions: (i) Is there a phylogenetic structure, either more clustered or more dispersed than expected at random, to the floral communities in subtropical vegetation at lower elevations and subalpine vegetation at higher elevations in central Nepal? (ii) Is there a phylogenetic signal in the array of colours present among the species we sampled? (iii) Are there significant differences in flower coloration or colour diversity between the species in each elevation group? Is there evidence of different evolutionary patterns of differentiation underlying colour signals at different altitudes? (iv) Do the flowers of species in central Nepal have colour cues in their reflectance spectra similar to the cues found in angiosperms from other regions of the world?

Materials and methods

Sites and species

We collected native flowering plants from four locations in the Himalayas of central Nepal, over an elevation range of 900–4100 m oriented largely southwest to northeast (Fig. 1). Our sites encompassed two broad vegetation types: subtropical at elevations below 2000 m, and subalpine, principally above 3000 m (Dobremez 1972; Chaudhary 1998). Both vegetation types experience monsoonal climates (see Appendix S1 in Supporting Information for details). Flowers were collected opportunistically as they were encountered in the field between April and October, 2011. The 107 species came from 78 genera in 35 families of monocots and eudicots; 46 species were sampled in the subtropical zone and 61 species in the subalpine zone.

Figure 1.

Elevation profile of a transect through central Nepal showing sampling areas. (a) Subtropical region, 900–2000 m and (b) Subalpine region, 3000–4100 m. Inset map of Nepal shows location of the transect as a heavy line.

To conduct phylogenetically informed analyses in the absence of sequence data for these Nepali species, we assembled a phylogenetic tree from existing phylogenetic estimations. Much of the evolutionary diversity in our sample occurred at the family level, and because deep phylogenetic relationships among angiosperms are now fairly robustly elucidated, we used the Soltis et al. (2011) family-level tree topology as a scaffold and grafted subfamilial topologies for taxa in our sample to this family tree (Appendix S1). The resulting tree is shown in Fig. 2.

Figure 2.

Phylogenetic relationships of the species in the sample. Square symbols indicate elevation (blue = 900–2000 m, subtropical vegetation; red = 3000–4100 m, subalpine vegetation). Circles indicate locus in the colour hexagon. The colours in the left circle correspond to the hexagon sector in which the species occurs (see inset hexagon for the colour map and Fig. 3 for exact loci of the sample species). The colours of these symbols are meant only to be distinguishable and to indicate an approximate radial angle. They are neither the apparent colour to human vision nor a representation of bee colour vision. The right circles use a grey scale to indicate relative distance from the centre of the hexagon (white) to the periphery (black).

Floral reflectance properties

We measured the reflectance of light from 300 to 700 nm wavelength on three fresh flowers of each species and characterized species by the wavelengths at which inflections occur in the reflectance spectra, as previously done by Shrestha et al. (2013) (see Appendix S1 for details and Fig. S1 for examples). Inflections represent the location of large changes in reflectance, which are especially discriminable by colour-opponent visual processing. Once inflections had been identified, we quantified the overall fit of each species' inflections to the hue discrimination optima of bee vision at 400 and 500 nm using the metrics of Shrestha et al. (2013). The metrics were the mean absolute deviation (MAD) of the inflections from either 400 or 500 nm, whichever was nearer, and the minimum absolute deviation from 400 nm (minAD400) and from 500 nm (minAD500) by any single inflection in a reflectance profile. These latter metrics allow that a single inflection may have more relevance for colour discrimination by a bee than the ensemble of inflections.

Colorimetric model

Because the data appeared to show a high frequency of chromatic cues oriented towards hymenopteran vision (see Results), we translated the floral reflectance spectrum of each species to a locus in a colour hexagon, a two-dimensional representation of the colour perception of hymenopteran vision (Chittka 1992). Mathematical details of this translation are given in Chittka (1992) and in Appendix S1. Loci in the colour hexagon can be represented as Cartesian or polar coordinates. In polar notation, the vector magnitude r (distance from the hexagon centre) is related to colour saturation, while the vector angle θ corresponds to hue. Another use of the colour hexagon is to represent relative differences in colour. Euclidian distance between two loci in the hexagon corresponds to difference in perceived colour (Chittka 1992), an interpretation that has been verified through behavioural tests involving Apis and Bombus (Dyer & Chittka 2004; Dyer 2006; Dyer, Spaethe & Prack 2008).

Data analysis

1 Phylogenetic structure within altitudinal samples

To investigate the phylogenetic organization of our altitudinal samples, we tested whether the species occurring in subtropical vegetation at lower altitudes and in subalpine vegetation at higher altitudes had more phylogenetic structure (either closer relatedness or greater dispersion) than expected by chance. We used the mean pair-wise phylogenetic distance (MPD) between species in each vegetation group as the metric of structure (Webb et al. 2002; McEwen & Vamosi 2010). The significance of this metric was tested by comparison to a null distribution derived from 10 000 random permutations among the tips of the phylogenetic tree followed by calculation of MPD. Calculations were carried out with the ses.mpd function of the R package Picante (Kembel et al. 2010).

2 Phylogenetic signal for floral colour

We examined phylogenetic signal for hue and saturation to determine whether colour similarity between species was a function of their evolutionary relatedness. Tests for signal in each element may have heuristic value in suggesting the lability of colour evolution, and previous work indicates that colour saturation is potentially an important cue for pollinators (Lunau 1990; Rhode, Papiorek & Lunau 2013). Phylogenetic signal was measured by Pagel's (1999) λ and by the K statistic of Blomberg, Garland & Ives (2003). Pagel's λ was calculated in the R package GEIGER (Harmon et al. 2008). Significance of λ was tested against null hypotheses of λ = 0 and λ = 1 using procedures outlined in Nunn (2009). K was calculated, and its significance tested against a null distribution generated by 1000 random permutations of the tips of the phylogeny using the Picante package (Kembel et al. 2010).

In order to examine phylogenetic signal for the full coordinates of colour loci, we used Euclidian distance between points to quantify colour differences. We tested the correlation between a distance matrix for the 107 colour loci and a matrix of phylogenetic distances (the sum of branch lengths separating any two taxa on the phylogenetic tree in Fig. 1) using a Mantel test.

3 Comparison of chromatic cues between elevation zones

Because pollinator-mediated selection on floral colour may differ with elevation, we compared the altitudinal samples for floral spectral features and the pattern of inferred colour changes on the phylogenetic tree. For the 89 species that had inflections in their floral reflectance spectra, we compared the values of MAD, minAD400 and minAD500 between the subtropical and subalpine groups using phylogenetic anova (Garland et al. 1993) implemented in GEIGER (Harmon et al. 2008). Because we found that monocots contributed significantly to the phylogenetic structure of the elevation groups (see 'Results'), we wished to compare the MAD and minAD metrics between monocots and eudicots within the subtropical zone, where 21 of the 22 monocots we sampled occurred. For this comparison, we used phylogenetic anova with monocots vs. eudicots as the categorical factor.

We also investigated differences between elevation groups in the diversity of colours present. We quantified colour variation within each vegetation type by the mean pair-wise Euclidian distance between loci in the colour hexagon and compared these values to null distributions based on random sampling from the entire pool of colours occurring in our sample. As with the test of phylogenetic structure within elevation zones in (a) above, the calculations were carried out with the ses.mpd function of Picante using the taxa.labels randomization method with 10 000 randomizations per test. We also used this procedure to compare colour diversity between monocots and eudicots within the subtropical zone.

To visualize how evolutionary history underlies colour variation in our sample, we superimposed the phylogeny of Fig. 2 on the species loci in hexagonal colour space. This procedure creates a phylomorphospace (Sidlauskas 2008). We compared the two altitudinal samples to assess how differences in colour diversity derive from differences in the rate and direction of colour evolution. Because the subtropical and subalpine samples share basal branches of the phylogenetic tree, we analysed only those branches that lead exclusively to tips in one group or the other.

4 Comparison between Nepali species and other floras

We compared the chromatic cues of the Nepali species in our sample to the cues in a sample of 180 mostly insect-pollinated species from the Israeli flora (Chittka & Menzel 1992) and 147 insect-pollinated Australian species characterized in Shrestha et al. (2013). We used the frequency of loci in radial segments of the colour hexagon (roughly, the frequency of different hues) to compare qualitatively among the floras. To make a quantitative comparison, we conducted phylogenetic anova for the Nepali and Australian species with country of origin as the categorical factor and MAD, minAD400 and minAD500 as dependent variables. For the phylogenetic basis of these tests, we pruned the bird-visited species from the phylogenetic tree used in Shrestha et al. (2013) and then merged the resulting tree with the tree of Nepali species in Fig. 2 to provide a unified phylogeny for the tests.


Phylogenetic structure within altitudinal samples

Both subtropical and subalpine samples were phylogenetically less diverse than expected by chance, given the phylogeny of the entire pool of species. Specifically, each sample had a lower MPD than any of the 10 000 randomly permuted groups (Table 1). Inspection of Fig. 2 shows that much of the phylogenetic homogeneity is due to the monocots, all but one of which occurred in the lower elevation community. This impression was corroborated by removing monocots from the sample and repeating the analysis for eudicots only. There was no significant difference between the MPD of observed and random communities of eudicots for either altitudinal group (Table 1).

Table 1.  Phylogenetic dispersion of species in the two elevation zones. = number of species in each group. MPDobs = observed mean pair-wise phylogenetic distance between all species pairs. The mean and standard deviation of MPD in the null distribution were obtained by randomization of species in the phylogenetic distance matrix (taxa.labels method in Picante). = (MPDobs – mean MPD of the null distribution)/SD of the null distribution. Negative values of Z indicate greater phylogenetic homogeneity (clustering) while positive values indicate greater dispersion among members of a group than expected by chance. P is the probability of drawing an MPD from the null distribution at least as extreme as MPDobs, based on 10 000 randomizations
 Null distribution
N MPDobsMean ± SD Z P
Subtropical (all species)46216.2236.5 ± 2.6−7.720.0001
Subalpine (all species)59222.3236.6 ± 2.0−7.140.0001
Subtropical (eudicots)22216.4219.8 ± 5.8−0.590.2507
Subalpine (eudicots)58220.1219.9 ±

Phylogenetic signal for floral colour

The two measures of phylogenetic signal, λ and K, have different interpretations, but the results are compatible and suggest that floral colours in related species are less similar than would be expected under Brownian motion evolution. Pagel's λ for floral hue was not significantly different from zero (λ = 0.137, = 0.36), indicating that the covariance among species was compatible with independent evolution of hue among all lineages (effectively, evolution along a star phylogeny). The value of K for floral hue was significantly different from random expectations and was less than unity (= 0.30, = 0.023), indicating less similarity among species than would be expected from their phylogenetic relationships under a Brownian motion model of hue evolution. The value of λ for colour saturation, λ = 0.36, was significantly different from both zero (= 0.007) and unity (P ≪ 0.001). Covariance in saturation was therefore intermediate between complete independence among lineages and Brownian motion evolution. K for colour saturation was significant and less than unity (= 0.35, = 0.001). Mantel tests showed a statistically significant but weak correlation between phylogenetic distance and distance in the hexagonal colour space (= 0.045, = 0.018). Overall, the phylogenetic signal for elements of floral colour within our sample is compatible with considerable lability in colour evolution, and, in particular, with stronger than expected divergence among related species.

Comparison between elevation zones

As judged by human vision, flower colours in our sample included whites (24%), pinks (20%), reds (6%), yellows (22%), blues (11%) and purples (17%). Despite this variety, the distribution of inflection points shows pronounced clustering near 400 and 500 nm in both elevation zones (Fig. S2). Similar clustering of inflections near these wavelengths of maximum hue discrimination by hymenoptera is a feature of other insect-pollinated species (Chittka & Menzel 1992; Dyer et al. 2012). Thus, the chromatic cues possessed by the flowers of both elevation zones should allow them to be distinguished easily by the hymenopteran visual system.

The metrics of fit between inflection points and hymenopteran vision, MAD, minAD400, or minAD500 showed no significant differences between the two elevation zones in phylogenetic anova (Table 2). Similarly, there were no significant differences in the MAD and minAD metrics between monocots and eudicots within the subtropical zone (Table 2). That is, neither elevation zone had flowers with chromatic cues that would make them more discriminable by hymenopteran pollinators than flowers from the other zone.

Table 2.  Results of phylogenetic anova for MAD and minAD metrics (fit between chromatic cues and hymenopteran hue discrimination). Comparisons were made between the two elevation groups of the Nepali species, and between the Nepali species and the insect-pollinated Australian flowers analysed in Shrestha et al. (2013). Conventional P values are based on an F distribution with d.f. = 1, 88 for the altitudinal comparison, d.f. = 1, 39 for the monocot/eudicot comparison and d.f. = 1, 235 for the Nepal–Australia comparison. Phylogenetic P values are based on 10 000 simulations of Brownian motion evolution of the dependent variable along either the phylogenetic tree in Fig. 1 or a combined phylogenetic tree of Nepali and Australian species
FactorDependent variable F Conventional PPhylogenetic P
Subtropical vs. subalpineMAD0.980.330.39
Monocots vs. eudicotsMAD0.710.400.32
Nepal vs. AustraliaMAD4.000.0470.31

The ensemble of floral colours in the sample species would be perceived by the bee visual system to be highly diverse, as indicated by the scatter of loci in the colour hexagon (Fig. 3). Colour diversity contrasted markedly between the lower and higher elevation samples. The minimum convex polygon containing the 46 species in subtropical vegetation occupied 0.39 squared units of the colour hexagon in Fig. 3, while the 61 subalpine species occupied 0.73 squared units. In particular, subalpine species had the greatest extent towards the bee-UV, bee-UV-blue and bee-green vertices of the colour hexagon (Fig. 3). On average, the subalpine taxa occupied 40% more area of the colour hexagon per species than did subtropical taxa. These differences represent significantly greater colour diversity for subalpine species and significantly less diversity for subtropical species than expected by chance (Table 3).

Table 3.  Colour diversity in elevation zones and between monocots and eudicots in the subtropical zone. Notation is the same as in Table 1, except that colour distance in the colour hexagon rather than phylogenetic distance was the variable analysed
 Null distribution
NMPDobs(Mean ± SD) Z P
Subtropical (all species)460.3440.422 ± 0.026−2.940.0015
Subalpine (all species)590.4730.422 ± 0.0212.490.0025
Subtropical (eudicots)220.4040.464 ± 0.045−1.300.0921
Subalpine (eudicots)580.4750.464 ± 0.0170.680.2697
Monocots (Subtropical)240.2720.344 ± 0.026−2.810.0041
Eudicots (Subtropical)220.4050.345 ± 0.0282.140.0105
Figure 3.

(a) Floral colour loci of 107 Nepali angiosperm species in a hexagonal space that represents the visual system of hymenopteran trichromats. The species occurred in two vegetation types (blue: subtropical, 900–2000 m; red: subalpine 3000–4100 m). The curve inside the hexagon is the spectral locus representing complete colour saturation (the perception of monochromatic light of different wavelengths). (b) Frequency distributions based on the occurrence of flower colour loci in 10° sections of the hexagonal space (insert i). This histogram can be compared with comparable data on flower colour in samples from the Middle East (dotted lines, Chittka & Menzel 1992) and Australia (solid lines, Dyer et al. 2012) (insert ii).

Monocots, which were largely restricted to the subtropical zone and created the significant phylogenetic structure in the altitudinal communities, had significantly more homogeneous floral colours than expected by chance (Table 3). Eudicots were neither more clustered nor more dispersed in colour space than expected by chance (Table 3).

Further phylogenetic basis for the differences in colour diversity can be seen in the phylomorphospace diagram (Fig. S3), equivalent to the colour hexagon in Fig. 3 but with branches connecting species to the inferred colour loci of ancestral nodes. The direction of evolutionary change in colour space appears equally diverse in the two altitudinal communities, but several branches leading to subalpine species are noticeably long, ostensibly indicating that the subalpine community comprises more lineages that have undergone large evolutionary change. This is true in a sense, in that the average branch length in colour hexagon units among subalpine species is 30% longer than among subtropical species (0.234 units vs. 0.179 units). However, the subtropical community tends to have shorter phylogenetic branch lengths due to the preponderance of orchids at lower elevations. When colour space branch lengths are divided by phylogenetic branch lengths (a heuristic estimate of the rate of colour evolution), the two communities are nearly identical, on average: 0.0099 hexagon units Ma−1 in the subalpine zone vs. 0.0094 hexagon units Ma−1 in the subtropical zone. However, 7 of the 10 branches with the highest rate of colour evolution lead to subalpine taxa. A visualization of the rates and directions of the phylomorphospace branches in the two communities is given in Fig. S4.

Comparison between Nepali species and other floras

The hue distributions of Israeli and Australian species are shown in the inset of Fig. 3, along with the corresponding distribution for the Nepali sample. There were many species with bee-blue–green flowers (60°) in all samples, and a secondary peak in frequency at bee-green (120°) among the Nepali and Israeli but not the Australian species. There was another secondary peak in abundance of bee-blue flowers among the subalpine Nepali species, but not among the Israeli and Australian samples, which came from much lower elevations. There was also a somewhat higher frequency of flowers with a pronounced bee-ultraviolet colour (240°) among the Nepali species than among the other floras.

Although the samples from different continents seem to show some differences in the frequency of colours, there were no significant differences between the Nepali and Australian species in the quantitative match between chromatic cues and hymenopteran vision measured by MAD, minAD400 and minAD500 (Table 2). It is worth noting that a non-phylogenetic anova would have incorrectly indicated that significant differences exist between these samples (conventional probability values, Table 2).


In common with many plant communities around the world (Emmerson & Gillespie 2008; Vamosi et al. 2009), we found a degree of phylogenetic clustering among the flowering plants in subtropical and subalpine habitats in central Nepal. This pattern was due to the abundance of monocots, particularly orchids, at lower elevations; eudicots showed neither phylogenetic clustering nor overdispersion. Thus, the phylogenetic clustering that we found occurred at the level of anciently diverged lineages, a scale dependence that seems to be typical of other plant communities (Emmerson & Gillespie 2008). For example, Hardy & Senterre (2007) found phylogenetic clustering among trees over a 300–1300 m range in Monte Alén National Park in Equatorial Guinea due to the restriction of magnoliids to lower elevations than eudicots. As in the sample of Hardy & Senterre (2007), however, our taxon sampling would not be sufficiently dense to allow detection of phylogenetic dispersion patterns at fine taxonomic scales.

Phylogenetic clustering, as a result of the restricted altitudinal range of monocots in our sample, is consistent with environmental filtering based on conservative traits associated with niche tolerance, although there is no indication that floral colour and pollination interactions are the relevant traits. Orchids are prominent among the monocots we sampled. Although orchids are not absent from high elevations, declining species richness of orchids with increasing altitude has been found in other parts of the Himalaya (Singh, Rawat & Jalal 2009) and elsewhere (Jacquemyn et al. 2005). In addition to the various abiotic environmental factors associated with elevation, interactions with mycorrhizal fungi and, for epidendroid orchids, with host trees may limit the altitudinal range of orchids and contribute to the phylogenetic structure within our sample.

The phylogenetic signal for floral hue and saturation among our species indicated that related species were more independent (λ) or more disparate (K) than expected under a Brownian motion model of trait evolution. These two interpretations of phylogenetic signal are compatible with each other, since a larger disparity than expected between related species implies less effect of recent ancestry. It may be that changes in colour saturation are more readily evolved than changes in hue because they require only quantitative variation in the expression of existing pathways of pigment synthesis. However, any differences in the evolution of these two elements of colour are probably relevant only for recently derived lineages (e.g. Hopkins & Rausher 2012). Floral colour is often highly labile among closely related species (Rausher 2008; Smith, Ane & Baum 2008; Eaton et al. 2012), so large colour disparity within the wide but not dense taxon sampling of this study is unsurprising. The observed pattern of phylogenetic signal could result from competition for pollinator services between sympatric-related species, or from adaptation to different suites of pollinators in related species that occur in different communities. Greater density of taxon sampling of the Nepali flora would be needed to explore processes that have produced the phylogenetic distribution of floral colours.

Flowers in our sample had spectral inflections aggregated near 400 and 500 nm wavelength, the regions of best colour discrimination by hymenopteran trichromatic pollinators (Fig. S2 and Table 2). Similar patterns are seen in insect-pollinated flowers in other parts of the world (Chittka & Menzel 1992; Dyer et al. 2012; Shrestha et al. 2013). Interestingly, when birds are the major pollinators, there is a significant shift in floral spectral cues towards longer wavelengths consistent with the tetrachromatic colour visual system of birds (Shrestha et al. 2013).

The predominance of ‘bee signals’ in both our montane communities (Fig. S2) is consistent with ubiquitous importance of Hymenoptera as pollinators in Nepal throughout our sampling range up to 4100 m, even though they would be absent at these elevations at higher latitudes. For example, Williams et al. (2010) found that Bombus species richness increases with increasing elevation up to about 4000 m in Nepal and declines only above that elevation. Eleven Bombus species occur between 1000 and 1999 m, while 27 Bombus species occur between 3000 and 3999 m. Similarly, data in Thapa (2000) show that species richness of Apidae in Nepal rises from 5 species at 500 m elevation to an approximate plateau around 20 species between 2300 and 3800 m, with a rapid decline at higher elevation. In contrast, flower-visiting dipterans in Nepal reach their peak species richness at about 1900 m and decline in richness to a single species at 4000 m and above (Thapa 2000). Although species numbers need not correspond to population abundance or efficacy as pollinators, it seems likely that the relative importance of hymenopterans and dipterans has a very different pattern in Nepal to that in montane regions at higher latitudes. The presence of hymenopteran pollinators throughout the elevation range of this study provides a simple explanation for the ubiquity of ‘bee signals’ that we found in the floral reflectance spectra.

Floral colours were more diverse at subalpine altitudes than at subtropical altitudes (Fig. 3). This pattern did not result from different tendencies in the direction of colour evolution in the two communities, but rates and absolute amounts of colour evolution were greater in several lineages leading to higher altitude taxa (Fig. S4). This pattern is consistent with different patterns of ecological filtering of plant assemblages at different altitudes. Abiotic factors such as UV and temperature may play a role in the assembly of floral colour communities, but pollinator-mediated filtering of coexisting colours is also possible. In particular, the greater species richness of Hymenoptera at subalpine elevations in Nepal may allow coexistence of a suite of plant species drawn from lineages that produce greater floral colour diversity than occurs at lower elevations. It would be worthwhile to investigate whether floral colour diversity is correlated with pollinator richness or diversity in communities in other geographical regions.

Although floral colour diversity was greater in the higher elevation habitat, it did not seem to represent a shift to a different order of pollinating insects, and not a shift towards substantial fly pollination, in particular. We note, however, that 15% of our species lacked inflection points and were therefore relatively achromatic. These species were included in all analyses based on the colour hexagon, but they may represent a special class of flowers that deserves further attention. Fly-mediated pollination may favour the presence of such floral colours within a community. Yet colour discrimination in flies seems to be very rudimentary and based on simple categorical processing mechanisms (Troje 1993). Some Diptera, such as the hoverflies Eristalis tenax and Episyrphus balteatus, show evidence of a fixed innate response to colours appearing yellow to human vision (Lunau & Wacht 1994; Wacht, Lunau & Hansen 1996), possibly as an adaptation to the feeding behaviour of these insects on yellow pollen (Lunau 2000). However, such innate colour preferences for yellow pollen seem unlikely to affect flower colour evolution to the extent observed with our current data set. It is likely that volatile compounds provide the major cue that flies use to discriminate flowers via the olfactory system (Goulson & Wright 1998; Shuttleworth & Johnson 2010).

An important contemporary question is the extent to which future climate change may influence geographical distributions, local abundances, and phenology of plants and their pollinators, and whether such changes would disrupt their current interactions by creating spatial or temporal mismatches (Parmesan 2006; Hegland et al. 2009). Interest in potential mismatches in pollination interactions has focused on flowering times and pollinator emergence (Visser & Both 2005), while mismatches based on chromatic cues and colour vision are rarely explored and their demographic consequences are largely unknown. By directly measuring spectral signals at different altitudes, we were able to get a snapshot of how pollination may be affected by shifts in the elevation range of plant species or pollinators brought about by climate disruption. Over the attitudinal range of our sampling, flower colours are very consistent in possessing spectral characteristics that readily allow discrimination by hymenopteran trichromats. Thus, it seems that plant–pollinator interactions in the Nepal Himalayas – or at least the component of these interactions based on floral colours – could be largely resilient to climate-induced shifts in elevation bands and disruption of current community assemblages, at least throughout the large range of current subtropical and subalpine vegetation. There has also been some evidence from simulation modelling of elevation shifts in Argentinian communities that plant–pollinator interactions may be reasonably resilient to the potential effects of climate change (Devoto, Zimmerman & Medan 2007). However, we also found some evidence that flower colour signals at higher altitudes were less clustered in colour space, suggesting possible effects of pollinator diversity on ecological filtering of coexisting species. Future work could thus consider specific plant–pollinator interactions to test for potential influences of environmental changes of flower pollination and colour signalling.


We thank the Department of National Parks and Wildlife Conservation (DNPWC), Nepal, Lantang National Park, Langtang, and Shivapuri National Park, Kathmandu, Nepal, for providing permission to work inside the parks. We are grateful to Dr. Lokesh Shakya, Suresh Shakya and Prakash Tiwari for assistance in Nepal and to Dr Alan Dorin for discussion of this topic. We thank the anonymous referees and the associate editor for their comments on earlier versions of the manuscript. M.S. was supported by a Monash Graduate Scholarship, Monash International Postgraduate Research Scholarship and Faculty of Science Postgraduate Publication Award. A.G.D. was supported by Australian Research Council DP0878968/DP0987989/DP130100015 and the Alexander von Humboldt Foundation. M.B. was supported by a Sabbatical Scholar fellowship from the National Evolutionary Synthesis Center (NESCent), NSF #EF-0905606.

Data accessibility

The phylogenetic tree in Fig. 2 is available as a Nexus file, and all colour data used in the analyses are available at the Dryad data repository (datadryad.org): http://dx.doi.org/10.5061/ dryad.2p8v2 (Shrestha et al. 2013).