Fruit advertisement strategies in two Neotropical plant–seed disperser markets

Abstract

Mutualisms can be seen as biological markets in which participating species exchange resources and services. Advertisements like the colors fleshy fruits are commonly used to attract mutualistic partners such seed dispersers. Although advertisements are common, the strategies employed in partner attraction and shaping the diversity of advertisements such as fruit colors remain largely unknown. Here, we adopt a market perspective on fruit color advertisement in multi-specific ensembles of fleshy-fruited plants and their avian seed dispersers. We develop and test the following non-exclusive hypotheses about fruit advertisement strategies in two Neotropical plant ensembles: (1) some low-rewarding plants offering low-energy fruits have fruit advertisements indistinguishable from those of some highly rewarding ones offering high-energy fruits thus forming possible mimicry pairs; (2) highly rewarding plants advertise their fruits with distinctive colors; and (3) fruit colors indicate the type of nutrient offered. We find support for two of the advertisement strategies. Further, we discuss how constraints on signal diversity may affect the evolution of advertisement strategies and we provide a perspective on which processes could characterize plant advertisement strategies in the biological market of seed-dispersal mutualisms.

Appendix

Possible mimicry pairs hypothesis

The program, which was written to search for possible mimicry pairs among the species in each geographic area, proceeded as follows. First, species were identified as having high-energy content based on whether their energy content was larger or equal to the 80 % quantile of energy content in the particular area (high-energy species). Remaining species were defined as those whose energy content fell below the 80 % quantile. Next, for each of the high-energy species in turn, the program calculated the chromatic distance in JNDs to each of the remaining species (for citation see the methods of the main text) and identified those pairs for which the distance was <4 JNDs. In Cardoso, the program also calculated the achromatic distance between each high-energy fruit and each of the remaining ones. In a next step, the difference between fruit sizes was calculated for each pair, and those pairs were eliminated with more than 20 % difference between their fruit volumes (calculated from fruit length and height). Similarly, only pairs that fruited in the same forest storey (Esmeralda) or which had the same growth form (Cardoso) were retained in the list of possible mimicry model pairs. Lastly, taxonomic information, specifically family and order, was compared, but none of the remaining pairs needed to be eliminated due to belonging to the same family.

Distinctiveness hypothesis

In a first step, we calculated the fruit color density of each species by applying a kernel density function npudens in R, and using the default bandwidth estimate given by the function npudensbw (R, package np). This calculation, as well as all following ones, was carried out for each geographic area separately. Subsequently, we wrote a program in MATLAB to test whether highly rewarding fruits have more distinctive colors than expected by chance alone within each geographic area. Highly rewarding fruits were defined as fruits with energy content higher than or as large as the 90, 85 or 80 % quantile of energy content. All following procedures were repeated for each of the three definitions of highly rewarding species. The distribution of average fruit color density expected by chance alone was obtained by repeatedly sampling (10,000 iterations; sampling without replacement, i.e. one species can be only once in the same sample) as many species from the respective geographic area as there were highly rewarding species in the current definition. For each of the 10,000 random samples, the average color density was calculated and stored, yielding the random distribution of average fruit color density. The average fruit color density of rewarding species was lower than expected by chance if it fell below the 5 % quantile of the random distribution (i.e. at the 5 % significance level). The actual p value was calculated as the percent of random data points smaller than or as large as the observed fruit color density.

Nutrient recognition hypothesis

We wrote a program in MATLAB, which tested whether fruits rich in a particular nutrient type had smaller average distances to each other in color space than expected by chance. All following procedures were carried out in each geographic area separately. First, we applied three definitions of richness in nutrient type. For each type of nutrient, nutrient-rich fruits were defined as those fruits with nutrient content as large as or larger than the 90, 85 or 80 % quantile of the nutrient content of all fruits in the area. To get an estimate of how similar fruit colors of nutrient-rich species are, we calculated the average of all pairwise Euclidean distances in color space between all nutrient-rich species of one nutrient type and definition.

In a next step, the program calculated the distribution of average distance in color space expected by chance alone by sampling repeatedly (10,000 iterations; sampling without replacement) as many color points from the respective geographic area as there were nutrient-rich fruits. In each of the random samples, the average Euclidean distance of all pairs of species was calculated and stored, yielding the random distribution of average distance. The average distance of nutrient-rich species was smaller than expected by chance if it fell below the 5 % quantile of the random distribution (see above). This procedure was repeated for each nutrient type and each definition of nutrient richness.

Subsequently, for species rich in a particular nutrient type that were found to show more similar advertisement than expected by chance, we tested whether they also differed in their advertisement from species rich in other nutrients. To do this, we compared the regions that nutrient-rich fruits of each type occupied in the avian color space. The region in color space that color points occupy is commonly defined using their convex hull (the minimal volume that contains all points and all lines connecting pairs of points; see citations in main text). To compare the occupied regions of nutrient-rich fruits of two types we determined the percent overlap between the convex hulls of the two nutrient-rich fruit samples. In the following, we compared the observed overlap to the overlap expected by chance for samples of this size (details in Stournaras et al. 2013, see main text).