Palm fruit colours are linked to the broad-scale distribution and diversification of primate colour vision systems

A long-standing hypothesis in ecology and evolution is that trichromatic colour vision (the ability to distinguish red from green) in frugivorous primates has evolved as an adaptation to detect conspicuous (reddish) fruits. This could provide a competitive advantage over dichromatic frugivores which cannot distinguish reddish colours from a background of green foliage. Here, we test whether the origin, distribution and diversity of trichromatic primates is positively associated with the availability of conspicuous palm fruits, i.e. keystone fruit resources for tropical frugivores. We combine global data of colour vision, distribution and phylogenetic data for more than 400 primate species with fruit colour data for more than 1700 palm species, and reveal that species richness of trichromatic primates increases with the proportion of palm species that have conspicuous fruits, especially in subtropical African forests. By contrast, species richness of trichromats in Asia and the Americas is not positively associated with conspicuous palm fruit colours. Macroevolutionary analyses further indicate rapid and synchronous radiations of trichromats and conspicuous palms on the African mainland starting 10 Ma. These results suggest that the distribution and diversification of African trichromatic primates is strongly linked to the relative availability of conspicuous (versus non-conspicuous) palm fruits, and that interactions between primates and palms are related to the coevolutionary dynamics of primate colour vision systems and palm fruit colours.


Primate data
We collected, inferred or interpolated data on functional traits (colour vision, diet and activity level), species distributions and phylogenetic relationships for 411 primate species (100% of total species assessed for the International Union for Conservation of Nature's Red List 2017). For the colour vision data, we classified species as routine trichromatic, polymorphic or nontrichromatic (i.e. monochromatic or dichromatic) based on 15 studies from the primary scientific literature (see Appendix S1). Polymorphic primates have males with dichromatic vision and females with dichromatic or trichromatic vision. Six primate species were excluded because their vision data was missing, and therefore 405 primate species remained. For 90 primate species, vision data was available at the family level, for 295 species at the genus level, and for 17 species at the species level. Moreover, three primate species (Cacajao ayresi, C. hosomi and C. melanocephalus) were interpolated with vision data from a species in their genus.
For diet, we classified frugivores as those primates that include fruits as part of their diet (using the 1-3 ranking in the MammalDIET dataset, the ranking indicates the relative importance of fruits in the diet as compared to other food types) (Kissling et al., 2014, for 95% matching the recent update by Gainsbury et al., 2018). For activity status, we collected data on whether a species is day-active or not (i.e. night-active or crepuscular) from the EltonTraits 1.0 database (317 primate species, Wilman et al., 2014), supplemented with data from the Handbook of the Mammals of the World: Primates (61 species, Mittermeier et al., 2013). For 27 primate species the activity data were interpolated from the genus. These species belong to eight genera, and in all of these genera the activity level is conserved (see Appendix S1). This resulted in n = 158 routine trichromatic primates (n = 126 species of dayactive, frugivores trichromatic primates, i.e. 31% of total primates) and n = 126 polymorphic primates (all polymorphic primates are day-active frugivores). All trait data and references are available from Appendix S1 and Dryad 10.5061/dryad.6hdr7sqwn.

Palm data
From a total of 2557 palm species (following the World Checklist of palms, Govaerts and Dransfield, 2005), ripe fruit colour data were assembled for 1749 species (c. 70% of total) from species descriptions in primary literature, monographs, the e-monocot database and from herbaria (Royal Botanic Gardens Kew Herbarium [K], Aarhus University Herbarium [AAU]) (Kissling et al., 2019). Fruit colours can be classified as 'non-conspicuous' when light reflectance spectra of fruits are similar to leaves, whereas fruits can be classified as 'conspicuous' when spectra differ between fruits and leaves (Regan et al., 2001). Since availability of spectral measurements is limited, we measured fruit reflectance spectra from 54 fresh palm fruits belonging to 18 species and compared them to qualitative colour descriptions from the literature (for details see below). This confirmed that colour descriptions of conspicuous fruits based on human vision match up with measured fruit reflectance data (see Table S3 and Fig. S16). This is also supported by fruit reflectance data from a wider variety of plant species (Sinnott-Armstrong et al., 2018). Following (Dominy et al., 2003), we then classified orange, red, yellow and pink fruits as conspicuous, and brown, black, green, blue, cream, grey, ivory, straw-coloured, white and purple fruits as nonconspicuous. Although purple fruits could be seen as conspicuous, both dichromats and trichromats can distinguish or detect the 'dark' colour against the background, and here they were thus included in the 'non-conspicuous' classification. If a fruit was described as a combination of non-conspicuous and conspicuous colours (e.g. 'green/yellow', 'yellowbrown', 'brown orange') then the non-conspicuous colour was the dominant hue and the fruit colour was classified as non-conspicuous. Colours that were described with a suffix -ish orey were considered to have only a touch of that colour. Primates mostly feed on brown, green, orange, yellow, red and purple fruits (Fleming and Kress, 2013). Therefore, we excluded palm fruits that did not have these colours in their description from the analyses, and only included the species with brown, green, orange, yellow, red or purple fruit colours (n = 1444 palm species remained). All trait data, criteria and references are available from Appendix S2 and Dryad 10.5061/dryad.6hdr7sqwn.

Fruit colour experiment
Ideally, fruit colours would be quantified as fruit reflectance and modelled according to trichromatic primate visual phenotypes, instead of following an assignment derived from human vision. However, the paucity of spectral measurements limits our ability to assess broad-scale patterns in fruit colours based on such measurements. Nevertheless, to validate our results using qualitative colour descriptions to assign fruits to the 'conspicuous' and 'nonconspicuous' categories, we compared them to quantitative variables derived from measuring fruit reflectance spectra as red, green and blue (RGB) hues from 54 fresh palm fruits belonging to 18 species. These species are cultivated and were collected in Xishuangbanna Tropical Botanical Garden (Yunnan, China) in August 2017. These species come from different phylogenetic lineages and thus broadly represent the taxonomic diversity of palms, and also represent palms from all geographical realms (see Table S2, 'Natural region'). These species were also selected to represent the range of fruit colours observed in palms (i.e. red, orange, yellow, green, black). Fruit colours were quantified by photographing the fruits (using Nikon's macro lens AF-S VR Micro-Nikkor 105mm f/2.8G IF-ED) under standardized light settings and the angle of illumination and reflection was fixed at 45° to minimize glare. We quantified RGB hues from the photographs in Adobe Photoshop. Results showed that conspicuous fruits can be mostly quantified as those with at least as much red as green in the fruits, and at least as much red as 1.5 times the amount of blue (see Table S3 and Fig. S16). This suggests that the human perception of fruit colour as the basis of our study is both reliable and quantifiable.

Structural equation modelling
We started our modelling process with an a priori SEM that included all hypothesized pathways among all predictor variables (Fig. S1). Thus, we examined direct effects of climate (annual mean temperature, annual precipitation, temperature seasonality and precipitation seasonality) on forest heights and on the proportion of conspicuous palm fruits, respectively. We further examined direct effects of climate, forest height, area, and the proportion of conspicuous palm fruits on trichromat richness. In the next step, we evaluated the model's modification indices, model fits and residual correlations (Grace et al., 2012). We used the chi-square test, the root mean square error of approximation (RMSEA), and the comparative fit index (CFI) to measure the fit of the model. To ensure an adequate fit of SEMs we applied the following criteria: P-values of chi-square tests > 0.05, CFI > 0.90 and confidence intervals of RMSEA < 0.05 (Grace et al., 2012). The path with the least statistical significance was deleted from the a priori SEM and this model was then assessed to identify the next, least significant path. We repeated these steps until our final SEM only consisted of significant pathways (at p < 0.05), for which we extracted the standardized coefficients.
We repeated the global SEM procedure separately for the three biogeographical realms (Africa, Americas, Asia) to evaluate whether these places differ in the relationship of colour vision with conspicuous fruits. The number of included botanical countries varied per continent (for day-active, frugivorous routine trichromats; Africa: 39 botanical countries, Americas: 26 botanical countries, Asia: 24 botanical countries, for day-active, frugivorous polymorphic primates: Africa: 1 botanical country, Americas: 29 botanical countries, Asia = 0 botanical countries). Analyses with n = 1 (e.g. African polymorphic primates only occur on Madagascar) or n = 0 (e.g. no polymorphic primates in Asia) botanical countries were not performed.

Sensitivity analyses
To evaluate whether an association between colour vision and fruit colour could result from confounding correlations, we performed five sensitivity analyses, and an additional analysis when also including figs (for details see last section in Supplementary methods). First, we repeated the SEMs but used the diversity of non-frugivorous trichromatic and/or polymorphic primates (instead of all or only day-active, frugivorous trichromatic primates) as the response variable. Under this scenario, we expected that the proportion of conspicuous palm fruits would not (significantly) explain species richness of non-frugivorous (mainly leaf-eating or 'folivorous') trichromats/polymorphs, suggesting that our results are not simply driven by confounding correlations between conspicuous fruits and other variables. Second, we evaluated the effect of the proportion of conspicuous fruits on dichromat species richness. Dichromatic species richness is expected to not be influenced by conspicuous fruits. Third, we evaluated whether differences in total trichromat/polymorphic species richness between biogeographical realms (trichromats/polymorphs, Americas n = 14/113, Africa n = 62/14, Asia n = 82/0) may have (erroneously) influenced our results. To this end, we divided trichromatic and/or polymorphic primate species richness in each botanical country by the total primate species richness in the respective continent, and repeated the global SEM. Fourth, we repeated the SEMs but followed the classification of frugivory from Galán-Acedo et al. (2019) (also available from Appendix S1), which only identifies frugivores as those species that predominantly feed on fruits. In these data, 39 routine trichromatic primate species and 36 polymorphic species which were classified as frugivorous (because fruits represent at least a small part of their diets) were classified as non-frugivorous (e.g. in genera Alouatta, Trachypithecus and Chlorocebus, Cebus, Mico and Saguinus). An additional six species (mainly in genus Macaca) were described as frugivorous whereas in our data they were classified as non-frugivorous. We expected that these differences in diet classification would not significantly affect our results. Fifth, we assessed whether our results were consistent with using conspicuous palm fruit species richness (rather than the proportion of conspicuous palm fruits) to explain primate species richness, and whether we fail to find a relationship between non-conspicuous palm fruit species richness and trichromatic and/or polymorphic primate species richness. The rationale for using conspicuous fruit proportions is that it allows for a direct comparison with non-conspicuous fruits, whereas food plant species richness could, potentially, be associated with primate species richness due to other, unmeasured, variables. Results of these sensitivity analyses are provided in Table S2.

Spatial autocorrelation
Spatial autocorrelation can affect significance tests and coefficient estimates in our SEMs Carl, 2008, Legendre andLegendre, 1998). We assessed the impact of spatial autocorrelation on our results by fitting spatial autoregressive (SAR) models, which allows the inclusion of residual spatial autocorrelation (Kissling and Carl, 2008). First, we fitted non-spatial, ordinary least squares (OLS) regression models with the same set of predictor variables on trichromat richness as recovered from the SEMs. As standardized coefficients from the OLS models are equivalent to the path coefficients of the SEMs, they enable a direct comparison of the spatial autocorrelation between the spatial and non-spatial models . The spatial weights matrix for the SAR models was defined using the minimum distance that linked each occupied grid cell to at least one other occupied cell (i.e. 326 km). To identify the extent of spatial autocorrelation in our data, we used correlograms and visualized changes in Moran's I values in the raw data response variables, the residuals of the non-spatial OLS and the SAR models. The standardized coefficients from the SAR models were very similar to those of the OLS models. We therefore focused on the path coefficients from the SEMs. The OLS model residuals were identified using the R package ncf and the spatial weight matrices of the SARs and the Moran's I values were calculated using the R package 'spdep' (Bivand and Wong, 2018). Spatial autocorrelation did not strongly affect the results. For the global analysis, the standardized coefficient for the effect of conspicuous fruits on polymorphic primate richness obtained from the SAR model (std. coeff. = 0.21) was similar to the coefficient obtained from using the (non-spatial) SEM (Fig. 2a). Similarly, the standardized coefficient for the effect of conspicuous fruits on routine trichromat richness obtained from the SAR model (std. coeff. = 0.43) was similar to the coefficient obtained from using the (non-spatial) SEM (Fig. 2b). However, the positive effect of precipitation seasonality on trichromat richness became nonsignificant in the SAR model (p = 0.54) as compared to the SEM. Spatial autocorrelation did not significantly influence the SEMs performed using botanical countries for the different biogeographical realms (Africa, America, Asia), but did affect the Africa grid-based analysis (when including palms only). In Africa, the (positive) effect of conspicuous palm fruits on trichromat richness remained significant in the SAR model (std. coeff. = 0.091, p < 0.001), although the standardized coefficient was lower than the coefficient obtained from using the (non-spatial) SEM (Fig. 2c). Temperature and precipitation did not significantly affect trichromat richness in the SAR model (p = 0.17 and p = 0.08, respectively).

Ancestral colour vision and fruit colour reconstructions
We reconstructed the origin of trichromacy/polymoprhy and conspicuous fruits in primates and palms, respectively, by using stochastic character mapping (Huelsenbeck et al., 2003). This was done by sampling 500 stochastic character maps from the posterior probability distribution using a Markov chain Monte Carlo under the unequal transition rate model (i.e. the evolutionary transition rate of losing trichromacy, and losing conspicuous fruit colours, is different from the transition rate gaining trichromacy, or gaining conspicuous fruit colours, respectively). This resulted in a sample of unambiguous histories for trichromatic/polymorphic colour vision and conspicuous fruits over the primate and palm phylogenies, respectively. We summarized these by showing the posterior probability that the edges and nodes of the tree are inferred to have trichromatic colour version (as compared to non-trichromatic vision) or conspicuous fruits (as compared to non-conspicuous fruits) (see Fig. S11).

Analyses with figs
We used palms as a proxy for primate food plants, because comprehensive global trait, distribution and phylogenetic data for other keystone plant clades of primates are not available. However, we evaluated whether our results for Africa based on palms were consistent with another keystone food plant clade, namely figs (Ficus, Moraceae) (Shanahan et al., 2001) for which good functional trait and distribution data are available for mainland Africa (Kissling et al., 2007).
Fruit colour data were assembled for all 86 African fig species (100% of figs on African mainland, thus excluding Madagascar) from the FigWeb of the Iziko Museums of Cape Town (http://www.figweb.org/Ficus/Species_index/afrotropical_species.htm). Species distribution maps of all 86 African figs were based on expert-based range maps provided by the Iziko Museums of Cape Town (for details see Kissling et al., 2007). For the phylogenetic data, we pruned the fossil-calibrated Moraceae MCC phylogenetic tree from Zhang et al. (2018) to include only African figs (n = 34 species, i.e. 52 African fig species with trait data were missing from the phylogeny).
We repeated the African SEM as well as the ancestral diversity reconstructions on the African fig phylogenetic tree. Results were consistent across palms and figs, however, the SEMs including figs showed weaker effects of the proportion of conspicuous fruits on trichromatic primate species richness as compared to palms only (Table S2). This effect became non-significant when correcting for spatial autocorrelation in model residuals (p = 0.1). Nevertheless, including figs provided a pattern similar to palms only regarding the standardized coefficient (obtained from SEMs) of the proportion of conspicuous food plant species on trichromat richness from arid to tropical climates, peaking in subtropical regions (compare Fig. S9 to Fig. 3). Furthermore, African figs with conspicuous fruits (as compared to those with non-conspicuous fruits) show rapid diversity increases from c. 10 Ma (Fig.  S13), similar to the result obtained for palms (Fig. 4).
In conclusion, the results including figs indicate that fig fruits probably played a less important role for trichromat distribution and diversification in mainland Africa than palm fruits (also see Dominy et al., 2003).  Table S2: Summary overview of sensitivity analyses. The base structural equations model (SEM) ( Figure S1) was adjusted in terms of the response variable (i.e. different measures of primate species richness) and/or the explanatory variable (i.e. different measures of fruits), to assess how the standardized coefficient of fruits on primate richness changed as compared to the empirical results ( Figures S2 and S7). Sensitivity SEMs were performed for the global dataset (with botanical countries) as well as for the mainland African grid dataset (110 × 110 km), and in most cases the analyses were performed for all primates vs. only day-active frugivorous primates. Results were also explored for trichromatic, polymorphic or trichromatic + polymorphic primates combined. 'Type' refers to the type of sensitivity analysis (A-G). (A) effect of proportion of conspicuous palm fruits on non-frugivorous primates as the response variable (no effect expected); (B) effect of proportion of conspicuous palm fruits on dichromatic primate species richness (no effect expected); (C) effect of conspicuous palm fruits on primate species richness, corrected for the continental differences in primate species richness (effect expected to be similar to empirical results, see Figure S2); (D) effect of conspicuous fruits on day-active frugivorous primate species richness in which the classification of frugivorous primates followed Galán-Acedo et al. (2019); (E1) effect of species richness (instead of proportion) of palms with conspicuous fruits on primate species richness (effect expected to be similar to empirical results, see Figure S2); (E2) effect of species richness of palms with non-conspicuous fruits on primate species richness (no or a negative effect expected); (F) effect of conspicuous fruits on primate species richness when also including figs in the model (effect expected to be similar to empirical results, see Figure  S7). The last column indicates if the result supported our expectation. In case it did not, it applies to polymorphic primates for which we indeed did not find strong support for an association with conspicuous fruits (see main text), or when including figs, suggesting that these are not as closely associated with primate colour vision as expected (see main text). Last, although the effect of conspicuous fruits species richness on routine trichromatic primate species richness at the global scale was positive but not statistically supported, the effect was positive when focusing on Africa only, supporting our primary result (see main text). The complete results from the sensitivity analysis of SEMs (i.e. also including effects of climate and forest structure) are available on request from the first author. Tri = trichromatic primate species richness, Poly = polymorphic primate species richness, Tri+Poly = trichromatic and polymorphic species richness combined, PSR = primate species richness, All = all primates were included in the SEM, frugi = only day-active, frugivorous primates were included in the SEM, non-frugi = only non-frugivorous primates were included in the SEM, Consp. = proportion of conspicuous (palms, unless figs is indicated) fruits, Std. coeff. = standardized coefficient of the explanatory on the response variable as obtained with the SEM, None = no significant effect of the explanatory variable on the response variable was detected in the SEM, n.a. = not applicable.  Table S3: Fruit colour experiment. Measurements of red (R), green (G) and blue (B) in the reflectance spectra of palm fruits. The percentages R, G and B were calculated for 18 species (3 fruits per species), and matched against descriptions of fruit colours from the literature which were used to assign species to the non-conspicuous/conspicuous categories in this study. If fruits had more R than G in their reflectance, and more red than 1.5 times the amount of B, they were quantified as 'conspicuous'. If not, they were mostly 'non-conspicuous'. In the few cases in which the literature did not match the quantified colour (last column 'no'), the fruits were mostly immature.  . In addition, we show the standardized coefficient (from SEMs) of proportion of conspicuous fruits on primate richness, the coefficient in the spatial autoregressive (SAR) model if spatial autocorrelation affected the results, whether the effect of proportion of conspicuous fruits on primate richness was higher than any of the other direct effects (i.e., forest structure, area, temperature, precipitation, temperature seasonality and precipitation seasonality, see Fig. S1), which supplementary figure shows the result, and whether results were robust to simulations (i.e., 1000 randomisations of primates across spatial units and repeating the SEM -if the observed effect lied outside the 95% confidence interval of simulated effects, it was considered robust with respect to simulation). Tri = trichromatic primate species richness, Poly = polymorphic primate species richness, Tri+Poly = trichromatic and polymorphic species richness combined, All = all primates were included in the SEM, frugi = only day-active, frugivorous primates were included in the SEM, n.a. = not applicable.  illustrates the direct and indirect effects of forest height, climate, area, and the proportion of conspicuous palm fruits on trichromat richness. Arrows indicate the direction of the effect. The red arrows highlight the relation between palm fruit colour and trichromat richness. Prec. = annual precipitation, Temp. = annual mean temperature, T Seas. = temperature seasonality, P Seas. = precipitation seasonality, Conspicuous fruits = proportion of conspicuous palm fruits, Trichromats = day-active, frugivorous, trichromatic primate species richness.

Figure S2: Determinants of trichromatic and polymorphic primate species richness.
Structural equation models (SEMs) illustrate how the proportion of conspicuous palm fruits (red arrows) determines (a) trichromatic and polymorphic primate species richness, (b) dayactive, frugivorous trichromatic and polymorphic primate species richness, (c) routine trichromatic primate species richness, (d) day-active, frugivorous routine trichromatic species richness, (e) polymorphic primate species richness and (f) day-active, frugivorous polymorphic primate species richness. SEMs were performed at the global scale (botanical country resolution). Effect sizes represent standardized coefficients and arrows indicate the direction of the effect, with arrow thickness being proportional to effect strength. Only statistically significant effects (at p < 0.05) are illustrated. Prec. = annual precipitation, Temp. = annual mean temperature, T Seas. = temperature seasonality, P Seas. = precipitation seasonality, Conspicuous fruits = proportion of conspicuous palm fruits, Tri+Poly = Trichromats + polymorphic primate species richness (combined), Trichromats = routine trichromatic primate species richness, Polymorphs = polymorphic primate species richness. Figure S3: Simulations of trichromatic and polymorphic primates globally. We compared the observed standardized coefficient of proportion of conspicuous fruits on primate species richness as obtained from structural equation models (SEMs) to 1000 simulated effects. Simulations were performed by randomly assigning primate species (and their colour vision) to botanical countries, keeping the total species richness for each botanical country the same. The observed effect of conspicuous fruits on routine trichromat richness (c, d) is much higher than the effect expected under a random distribution, whereas the effect on polymorphic primate species richness (e, f) or routine trichromats and polymorphic primates combined (a, b) does not deviate from a random expectation.

Figure S4: Structural equation models for Africa (botanical countries, including Madagascar).
(a) all polymorphic and trichromatic primates, (b) day-active, frugivorous polymorphic and trichromatic primates, (c) all trichromatic primates, (d) day-active, frugivorous trichromatic primates. Please note that this is using botanical countries as spatial units whereas Fig. 2c (main text) illustrates the model using the fine resolution grid cells. SEMs for polymorphic primates only were not possible due to low sample size (n = 1, i.e. Madagascar). SEMs illustrate how the proportion of conspicuous palm fruits (red arrows) determines primate species richness once direct and indirect effects of forest height, climate, and area are accounted for. Effect sizes represent standardized coefficients and arrows indicate the direction of the effect, with arrow thickness being proportional to effect strength.

Figure S5: Structural equation models for the Americas.
(a) all polymorphic and trichromatic primates, (b) day-active, frugivorous polymorphic and trichromatic primates, (c) all trichromatic primates, (d) day-active, frugivorous trichromatic primates, (e) all polymorphic primates, (f) day-active, frugivorous polymorphic primates. SEMs illustrate how the proportion of conspicuous palm fruits (red arrows) determines primate species richness once direct and indirect effects of forest height, climate, and area are accounted for. Effect sizes represent standardized coefficients and arrows indicate the direction of the effect, with arrow thickness being proportional to effect strength. Only statistically significant effects (at p < 0.05) are illustrated. Prec. = annual precipitation, Temp. = annual mean temperature, T Seas. = temperature seasonality, P Seas. = precipitation seasonality, Conspicuous fruits = proportion of conspicuous palm fruits, Poly+Trichromats = polymorphic and trichromatic primate species richness, Trichromats = trichromatic primate species richness, Polymorphs = polymorphic primate species richness. Figure S6: Structural equation models for Asia. (a) all trichromatic primates, (b) dayactive, frugivorous trichromatic primates. Polymorphic primates are absent in Asia. SEMs illustrate how the proportion of conspicuous palm fruits (red arrows) determines primate species richness once direct and indirect effects of forest height, climate, and area are accounted for. Effect sizes represent standardized coefficients and arrows indicate the direction of the effect, with arrow thickness being proportional to effect strength. Only statistically significant effects (at p < 0.05) are illustrated. Prec. = annual precipitation, Temp. = annual mean temperature, T Seas. = temperature seasonality, P Seas. = precipitation seasonality, Conspicuous fruits = proportion of conspicuous palm fruits, Trichromats = trichromatic primate species richness. Figure S7: Determinants of trichromatic primate species richness in mainland Africa using grid cells. Structural equation models (SEMs) illustrate how the proportion of conspicuous palm fruits (red arrows) determines (a) trichromatic and (b) day-active, frugivorous trichromatic primate species richness across mainland Africa (where no polymorphic primates occur) (resolution of 110 × 110 km grid cells) once direct and indirect effects of forest height, climate, and area are accounted for. Effect sizes represent standardized coefficients and arrows indicate the direction of the effect, with arrow thickness being proportional to effect strength. Only statistically significant effects (at p < 0.05) are illustrated. Prec. = annual precipitation, Temp. = annual mean temperature, T Seas. = temperature seasonality, P Seas. = precipitation seasonality, Conspicuous fruits = proportion of conspicuous palm fruits, Trichromats = routine trichromatic primate species richness. Figure S8: Simulations of trichromatic primates in mainland Africa. We compared the observed standardized coefficient of proportion of conspicuous fruits on trichromatic primate richness as obtained from structural equation models (SEMs) to 1000 simulated effects. Simulations were performed by randomly assigning primate species (and their colour vision) to African grid cells (n = 794 grid cells), keeping the total species richness for each grid cell the same. The observed effect of conspicuous fruits on routine trichromat richness is much higher than the effect expected under a random distribution.    The pie charts reflect ancestral nodes in the phylogeny of African figs and their probability to have conspicuous fruits (red colour in pie-charts, obtained from ancestral state reconstructions). Similar to palms (Fig. 4f), figs with conspicuous fruits show rapid diversity increases at ca. 10 million years ago as compared to figs with non-conspicuous fruits (grey colour in pie-charts). This happens in parallel with the radiation of trichromats in Africa (Fig. 4e).
Figure S14: Diversification rates for polymorphic and trichromatic primates and palms with conspicuous fruits. We used Multiple State Speciation and Extinction (MuSSE) and Binary State Speciation and Extinction (BiSSE) models to model speciation, extinction, and transition rates of trichromatic/polymorphic colour vision vs. non-trichromatic/polymorphic colour vision (a) and of conspicuous fruits vs. non-conspicuous fruits (b). The MuSSE and BiSSE models jointly estimate speciation, extinction and transition rates of a trait with multiple trait states by using dated phylogenetic trees, and trait states assigned to the species at the tips of the trees. We fitted eight MuSSE or BiSSE models with decreasing complexity (parameters) and selected the best-fitting models based on likelihood-ratio tests under a Chisquare distribution and the Akaike Information Criterion (AIC). These models included constraints on speciation, extinction and transition rates between trait states. Maximum likelihood was used to optimize full and constrained models, and a Bayesian MCMC was run for the best-fitting models (constrained extinction and transition rates for primates and constrained speciation rates for palms) for 1000 generations on the maximum clade credibility trees. We show the posterior distribution of the diversification rates (speciation -extinction rates), indicating that trichromatic primates have the fastest diversification rates followed by polymorphic primates and non-trichromatic/non-polymorphic primates (a). Palm lineages with conspicuous fruits have faster diversification rates than palms with non-conspicuous fruits (b). Significance is indicated by the fact that the 95% Bayesian credibility intervals between parameter states do not overlap. The analyses were performed using the 'make.bisse' and 'make.musse' functions in the 'diversitree' R package (FitzJohn, 2012). Non-conspicuous fruit colour Conspicuous fruit colour Figure S15: Diversity changes of folivorous and frugivorous primates in mainland Africa through geological time. The pie charts reflect ancestral nodes in the phylogenies and their probability to have the focal trait (red colour, obtained from ancestral state reconstructions): (a) folivory or (b) frugivory. In mainland Africa (i.e., excluding Madagascar), rapid diversity increases of frugivores happened ca. 10 million years ago (Ma) onward, compared to more gradual diversification of non-frugivorous African primates. In contrast, both folivorous and non-folivorous primates diversified rapidly from ca. 10 Ma onward, and this increase is thus unlikely to be related to this trait (i.e., folivory).
Figure S16: Fruit colour experiment. Visualization of measurements of red (R), green (G) and blue (B) and corresponding colours in the reflectance spectra of palm fruits (n = 18 species). Conspicuous and non-conspicuous refers to the classification of fruits following literature descriptions.  Figure S17: Global species richness of palms with conspicuous or non-conspicuous fruits. Maps depict the global distribution and variation of species richness of palm species with (a) conspicuous fruit colours and (b) non-conspicuous fruit colours. Only palm species with fruit colours known to be eaten by primates are included (brown, green, orange, yellow, red and purple fruits). Values are quantified at the spatial resolution of botanical countries. Grey colours indicate countries where no primates and / or < 2 palm species occur.