raits, broadly speaking, are measurable attributes or characteristics of organisms. Traits related to function (for example, leaf size, body mass, tooth size or growth form) are often used to understand how organisms interact with their environment and other species via key vital rates such as survival, development and reproduction 1-5. Trait-based approaches have long been used in systematics and macroevolution to delineate taxa and reconstruct ancestral morphology and function 6-8 and to link candidate genes to phentoypes 9-11. The broad appeal of the trait concept is its ability to facilitate quantitative comparisons of biological form and function. Traits also allow us to mechanistically link organismal responses to abiotic and biotic factors with measurements that are, in principle, relatively easy to capture across large numbers of individuals. For example, appropriately chosen and defined traits can help identify lineages that share similar life-history strategies for a given environmental regime 12,13. Documenting and understanding the diversity and composition of traits in ecosystems directly contributes to our understanding of organismal and ecosystem processes, functionality, productivity and resilience in the face of environmental change 14-19. In light of the multiple applications of trait data to address challenges of global significance (Box 1), a central question remains: How can we most effectively advance the synthesis of trait data within and across disciplines? In recent decades, the collection, compilation and availability of trait data for a variety of organisms has accelerated rapidly. Substantial trait databases now exist for plants 20-23 , reptiles 24,25 , invertebrates 23,26-29 , fish 30,31 , corals 32 , birds 23,33,34 , amphibians 35 , mammals 23,36-38 and fungi 23,39 , and parallel efforts are no doubt underway for other taxa. Though considerable effort has been made to quantify traits for some groups (for example, Fig. 1), substantial work remains. To develop and test theory in biodiversity science, much greater effort is needed to fill in trait data across the Tree of Life by combining and integrating data and trait collection efforts.

Summary Dietary inferences are a key foundation for paleoecological, ecomorphological and macroevolutionary studies because they inform us about the direct relationships between the components of an ecosystem. However, we need to consider the range of dietary variation we want to investigate and characterize before choosing a proxy. The goal of the present work is to evaluate the differences in dietary discrimination power between our new method, the multidimensional multi‐proxy dental morphology analysis (MPDMA) and unidimensional dental morphology proxies such as orientation patch count (OPCR), relief index (RI) or slope. In order to do that, we three‐dimensionally scanned the dentitions of 134 extant mammals including 28 marsupials (order Diprotodontia) and 106 placentals (orders Carnivora, Primates and Rodentia) and classified their diets using a new classification scheme that emphasizes the primary resource in a given diet. Diet categories included herbivory, carnivory, frugivory, granivory, insectivory, fungivory, gumivory and generalist. Unidimensional proxies significantly discriminate (P < 0·05) between one or two diet categories on the one hand and the rest on the other. For example, OPCR discriminates well between carnivorous and non‐carnivorous species. However, none of the individual proxies discriminate all eight dietary categories. Multi‐proxy dental morphology analysis demonstrates significant morphological differences across diets (MANOVA, d.f. = 7; F = 7·56; P < 0·05) and correctly discriminates diet for 67–82% of the specimens in the data set including and excluding rodents respectively. Combining different morphological variables makes it possible to draw better dietary inferences and fully represent the multidimensional nature of dental morphology and dietary specializations. Our results have important applications in ecological, paleoecological and evolutionary research.

Understanding the feeding behaviour of the species that make up any ecosystem is essential for designing further research. Mammals have been studied intensively, but the criteria used for classifying their diets are far from being standardized. We built a database summarizing the dietary preferences of terrestrial mammals using published data regarding their stomach contents. We performed multivariate analyses in order to set up a standardized classification scheme. Ideally, food consumption percentages should be used instead of qualitative classifications. However, when highly detailed information is not available we propose classifying animals based on their main feeding resources. They should be classified as generalists when none of the feeding resources constitute over 50% of the diet. The term ‘omnivore’ should be avoided because it does not communicate all the complexity inherent to food choice. Moreover, the so-called omnivore diets actually involve several distinctive adaptations. Our dataset shows that terrestrial mammals are generally highly specialized and that some degree of food mixing may even be required for most species.