Integrating DNA-Based Prey Occurrence Probability into Stable Isotope Mixing Models

Abstract

Synopsis The introduction of laboratory methods to animal dietary studies has allowed researchers to obtain results with accuracy and precision, not possible with observational techniques. For example, DNA barcoding, or the identification of prey with taxon-specific DNA sequences, allows researchers to classify digested prey tissues to the species-level, while stable isotope analysis paired with Bayesian mixing models can quantify dietary contributions by comparing a consumer's isotopic values to those derived from their prey. However, DNA-based methods are currently only able to classify, but not quantify, the taxa present in a diet sample, while stable isotope analysis can only quantify dietary taxa that are identified a priori as prey isotopic values are a result of life history traits, not phylogenetic relatedness. Recently, researchers have begun to couple these techniques in dietary studies to capitalize on the reciprocal benefits and drawbacks offered by each approach, with some even integrating DNA-based results directly into Bayesian mixing models as informative priors. As the informative priors used in these models must represent known dietary compositions (e.g., percentages of prey biomasses), researchers have scaled the DNA-based frequency of occurrence of major prey groups so that their normalized frequency of occurrence sums to 100%. Unfortunately, such an approach is problematic as priors stemming from binomial, DNA-based data do not truly reflect quantitative information about the consumer's diet and may skew the posterior distribution of prey quantities as a result. Therefore, we present a novel approach to incorporate DNA-based dietary information into Bayesian stable isotope mixing models that preserves the binomial nature of DNA-based results. This approach uses community-wide frequency of occurrence or logistic regression-based estimates of prey occurrence to dictate the probability that each prey group is included in each mixing model iteration, and, in turn, the probability that each iteration's results are included in the posterior distribution of prey composition possibilities. Here, we demonstrate the utility of this method by using it to quantify the prey composition of nestling Louisiana waterthrush (Parkesia motacilla).