The 22 taste categories and 46 therapeutic uses studied here. Each taste and use is represented by an icon that is used throughout the manuscript.

Strong-tasting drugs with relatively few perceived unique tastes predict therapeutic versatility a) Botanical drugs with more tastes are significantly negatively associated with the number of therapeutic uses, px = 0.008. b) Botanical drugs with strong tastes are significantly positively associated with the number of therapeutic uses, px = <0.001. For both relationships, we plot the mean regression line as estimated from the parameter estimates of our phylogenetic Bayesian regression analyses (dark line) along with 100 random samples of the posterior distribution (faded lines).

Diversity of use predicted by taste. The balance is a pictorial representation of how the strength of specific tastes contributes to significant associations with fewer or more uses.

The relationship between individual therapeutic uses and taste intensity/complexity. Uses predicted by intensity and complexity of taste. An arrow is shown indicating the direction of any significant association (px < 0.05) with either taste intensity (left) or taste complexity (right). An arrow points upward if a drug is more likely to be used for a given purpose with increasing taste intensity or complexity. Arrows are shaded by the strength of the relationship (magnitude of estimated parameter, see scale).

The magnitude of effect for 22 different specific tastes across 46 different therapeutic uses. a) A heat map showing the strength of the association between specific tastes (columns) and therapeutic uses (rows). Where we find a significant association between the strength of taste and whether a drug is used for a particular therapeutic purpose in our phylogenetic binary (probit) response model, the corresponding cell is coloured according to its estimated parameter value; see colour scale in panel (b). All other variables are zeroed and coloured white for visualization. b) An overall ‘magnitude of effect’ is calculated for each taste by summing the significant betas (px < 0.05, i.e. those displayed in the heat-map above) across all different uses and plotted as bars coloured by their magnitude; see colour scale alongside axis. The sum of all betas regardless of significance are shown as grey bars. Note that the large parameter estimate for “smoky” tastes is owing to the fact that no botanical drug with any smoky taste (scored by any participant) is ever used for androgynous therapeutic purposes and thus is poorly estimated in that model (see supporting information).