Skip to main content

Juggling options: manipulation ease determines primate optimal fruit size choice

Cite this dataset

Dias-Silva, Renann H. P. et al. (2020). Juggling options: manipulation ease determines primate optimal fruit size choice [Dataset]. Dryad.


Optimal foraging theory predicts that animals will seek simultaneously to minimize food processing time and maximize energetic gain. To test this hypothesis, we evaluated whether a specialist seed-predator primate forages optimally when choosing among variable-sized thick-husked fruits. Our objects of study were the golden-backed-uacari (Cacajao ouakary, Pitheciidae) and single seeded pods of the macucu tree (Aldina latifolia, Fabaceae). We predicted that golden-backed-uacari will consume fruits of the size class that requires the least time to obtain, handle, and ingest. We used scan-sampling, ad libitum to record feeding observations, and measured fruits, their penetrability and the size of taxidermised C. ouakary hands. To test if uacaris selected for optimal characteristics, we compared 8 metrics from 75 eaten and 105 uneaten seeds/fruits collected. Uacaris selected fruits of medium size and weight disproportionately to their abundance. Processing large fruits took six times longer than did medium-sized fruits, but seeds were only four times as large, that is, for energetic yield per unit time, thus choosing medium-sized pods was optimal. Disproportionate selection by C. ouakary of fruits of medium size and mass in relation to their abundance suggests active sub-sampling of the available weight-size continuum. This selectivity probably maximizes trade-offs between the energy derived from a seed, and time and energy expended in processing fruit to access this, so following optimal foraging theory predictions. The greater time spent processing large pods is attributed to difficulties manipulating objects five to seven times the size of the animal’s palm and one-sixth its own body weight.


To test if the ucaris were selecting seeds of larger-sized fruit, we compared seven metrics recorded from eaten and uneaten seeds and fruits. To describe allometry patterns, we regressed seed length (mm) against fruit length (mm) (n = 57), fruit length (mm) against fruit weight (g) (n = 88), maximum husk thickness (mm) against fruit length (mm) (n =79), and maximum husk thickness (mm) against fruit weight (g) (n = 58). The number of seeds/fruits measured varied as a result of the uacaris de-husking behavior, so that, for example, on occassion, only the length or the weight of a given fruit could be recovered reliably. For husk allometries we also fitted an asymptotic model to account for possible disproportionalities in husk thickness during fruit ontogeny. We then compared model fit (linear and asymptotic) using the Akaike Information Criterion (AIC). The AIC is widely used to measure the goodness of fit of a particular model, relative to other models, when the data are the same (Akaike 1974). We used the difference between models (Delta AIC >2), to select the candidate model.

For the subset of observations for which fruit processing times were available (n= 21), we individually regressed fruit length (mm) against time spent: (i) retrieving fruit from the water, (ii) de-husking, and (iii) eating the seed.

To test for possible fruit selection by uacaris, we performed an F test to compare variances of total length, width, maximum and minimum husk thickness between eaten and uneaten A. latifolia fruits. The null hypothesis for this test was that the ratio of the variances of the eaten and uneaten fruits would be equal to 1. To control for possible bias within an unbalanced sampling design (eaten = 56, uneaten fruits = 105) we used a bootstrap procedure (permutation with replacement). In each run the eaten and uneaten fruits measures were randomized and F test variance computed. We then compared the statistics of the 999 permutations with the observed value to calculate the probability that the observed value was larger than random. Descriptive statistics and frequency of hand measurements was performed to obtain minimum, maximum and average sizes. All analyses were made in R (R Core Team 2016).