Using biotracer techniques to uncover consumer diets: A comparison of stable isotopes, fatty acids, and amino acids

Abstract

AbstractBiotracers are commonly used in food‐web studies to estimate consumer diets. Increasingly, multiple biotracer types are combined to provide more nuanced pictures of consumer resource use, unravel more complex diet mixtures, and improve the discriminatory power of mixing models. However, few studies compare different biotracer types, including the impact of tracer selection choices, and few methods exist for combining multiple types into a single analysis. We processed three biotracer types (stable isotopes, fatty acids, and amino acids) from the same samples to determine how common galaxias (Galaxias maculatus) and flathead gudgeon (Philypnodon grandiceps) utilize different basal resources (algae, macrophytes, and detritus) using Bayesian mixing models. We then combined subsets of different biotracer types by either combining them directly into one model using generalist priors (all possible two and three biotracer type combinations) or informative priors based on other biotracer types. We compared our mixing model results with independent food webs constructed from gut contents to assess accuracy. We found fatty acids gave the most accurate diet estimates, amino acid models were the least accurate, and stable isotope models did not converge. Model precision tended to increase with the number of individual tracers included (with exceptions). Adding biotracer types that produced similar diet estimates in their respective models tightened the posterior probability distributions of combined models, while adding ones that disagreed expanded those distributions, as expected. However, tracer selection and the combination of tracer sets must be carefully considered, as they can dramatically affect model accuracy. We also demonstrate the utility of a complementary method that combines information from multiple biotracers using informative priors based on other biotracer types. Overall, our work shows that choices made when constructing mixing models can greatly influence resultant diet estimates, and we provide heuristics for making those choices, so they are most appropriate for an individual study's objectives and available data.