Using DNA Metabarcoding To Evaluate the Plant Component of Human Diets: a Proof of Concept

Dietary intake is difficult to measure reliably in humans because approaches typically rely on self-reporting, which can be incomplete and biased. In field studies of animals, DNA sequencing-based approaches such as metabarcoding have been developed to characterize diets, but such approaches have not previously been widely applied to humans. Here, we present data derived from sequencing of a chloroplast DNA marker (the P6 loop of the trnL [UAA] intron) in stool samples collected from 11 individuals consuming both controlled and freely selected diets. The DNA metabarcoding strategy resulted in successful PCR amplification in about 50% of samples, which increased to a 70% success rate in samples from individuals eating a controlled plant-rich diet. Detection of plant taxa among sequenced samples yielded a recall of 0.86 and a precision of 0.55 compared to a written diet record during controlled feeding of plant-based foods. The majority of sequenced plant DNA matched common human food plants, including grains, vegetables, fruits, and herbs prepared both cooked and uncooked. Moreover, DNA metabarcoding data were sufficient to distinguish between baseline and treatment diet arms of the study. Still, the relatively high PCR failure rate and an inability to distinguish some dietary plants at the sequence level using the trnL-P6 marker suggest that future methodological refinements are necessary. Overall, our results suggest that DNA metabarcoding provides a promising new method for tracking human plant intake and that similar approaches could be used to characterize the animal and fungal components of our omnivorous diets. IMPORTANCE Current methods for capturing human dietary patterns typically rely on individual recall and as such are subject to the limitations of human memory. DNA sequencing-based approaches, frequently used for profiling nonhuman diets, do not suffer from the same limitations. Here, we used metabarcoding to broadly characterize the plant portion of human diets for the first time. The majority of sequences corresponded to known human foods, including all but one foodstuff included in an experimental plant-rich diet. Metabarcoding could distinguish between experimental diets and matched individual diet records from controlled settings with high accuracy. Because this method is independent of survey language and timing, it could also be applied to geographically and culturally disparate human populations, as well as in retrospective studies involving banked human stool.