Variation in the mineral and physical properties of quartz particles on food surfaces.

(A) Particle sizes followed a bimodal distribution, with most particles featuring metal inclusions. Nearly half the sample is < 25 μm, the “grittiness threshold” of the human oral cavity (Imai et al., 1995). (B) Circularity is a dimensionless shape factor (range: 0 to 1) based on two-dimensional microscopy and estimates for the projected area and perimeter of a given particle. Some examples are illustrated; overall, it is a convenient but imperfect proxy for sphericity, or deviations from spherical (Grace and Ebneyamini, 2021). Here, circularity varied as a function of mean Feret diameter, suggesting that larger particles hold greater potential for damaging attack angles during particle-enamel contact.

Experimental design and results.

(A) To elicit food-cleaning behaviors, we put sliced cucumbers in trays representing three treatments—food surfaces with low ( = 0.2 ± 0.2 mg mm-2), intermediate ( = 0.9 ± 0.1 mg mm-2), and high ( =1.8 ± 0.9 mg mm-2) concentrations of sand—positioned 1.5 m apart and 15 m from the ocean. (B) Monkeys brushed sandier treatments for longer durations, χ2 (2, n = 575 food-handling bouts) = 185.7, p < 0.0001) with no difference across dominance ranks (Table S3). (C) Monkeys washed sandier treatments for longer durations, χ2 (2, n = 362 food-handling bouts) = 50.5, p < 0.0001), and we detected an interaction effect with dominance rank, χ2 = 6.4, p = 0.04 (Table S4). Raw data for panels B and C are illustrated in Figure S2. (D) Energy intake rates also varied as function dominance (ANOVA; F2,104 = 4.3; p = 0.02). Symbols represent mean values and whiskers ± 1 s.e. Photos by Amanda Tan.

Predicted and observed cleaning times.

a, Predicted time (large filled points) vs. observed times (violin plots) for brushing and washing food (note log scale). b, Predicted cleaning time as a function of cleaning inefficiency c, and handling time h, with predicted values (black points) for brushing and washing based on observed cleaning inefficiencies and handling times. The coloured points (as in panel a) represent observed cleaning times. The tradeoff between longer food handling times and efficient cleaning (oceanside food-washing; Region I) and shorter handling times and inefficient cleaning (immediate food-brushing; Region II) is depicted by the black curve.

Comparison of cleaning effectiveness.

We simulated the brushing and washing behaviours of our study animals using the same three treatments of cucumber slices in our experiment. Brushing was less efficient than washing across all treatments, eliminating an average 76 ± 7% vs. 93 ± 4% of surface sands, respectively.

Individual food-cleaning events.

(A) Monkeys put more time into brushing sandier treatments, X2 (2, n = 575 food-brushing events) = 185.7, p < 0.0001) with no difference across dominance ranks (Table S3). (B) Monkeys put more time into washing sandier treatments, X2 (2, n = 362 food-washing events) = 50.5, p < 0.0001), but we detected an interaction effect with dominance rank, X2 = 6.4, p = 0.04 (Table S4).

delta AIC model results for brushing models - testing different model distributions and potential zero-inflation for brushing model (n = 575 observations)

delta AIC model results for washing models - testing different model distributions and potential zero inflation for washing models (n = 362 observations)

Fixed effects for food-brushing (n = 575 events by animals with known rank), analysis of deviance (Type II Wald Chi Square Tests)

Fixed effects for food-washing (n = 362 events by animals with known rank), analysis of deviance (Type II Wald Chi Square Tests)