The precautionary principle and dietary DNA metabarcoding: Commonly used abundance thresholds change ecological interpretation

Abstract

Dietary DNA metabarcoding enables researchers to identify and characterize trophic interactions with a high degree of taxonomic precision. It is also sensitive to sources of bias and contamination in the field and laboratory. One of the earliest and most common strategies for dealing with such sensitivities has been to remove all low-abundance sequences and conduct ecological analyses based on the presence or absence of food taxa. Although this step is now often perceived to be necessary, evidence of its sufficiency is lacking and more attention to the risk of introducing other errors is needed. Using computer simulations, we demonstrate that common strategies to remove low-abundance sequences can erroneously eliminate true dietary sequences in ways that impact downstream inferences. Using real data from well-studied wildlife populations in Yellowstone National Park, we further show how these strategies can markedly alter the composition of dietary profiles in ways that scale-up to obscure ecological interpretations about dietary generalism, specialism, and composition. Although the practice of removing low-abundance sequences may continue to be a useful strategy to address research questions that focus on a subset of relatively abundant foods, its continued widespread use risks generating misleading perceptions about the structure of trophic networks. Researchers working with dietary DNA metabarcoding data-or similar data such as environmental DNA, microbiomes, or pathobiomes-should be aware of drawbacks and consider alternative bioinformatic, experimental, and statistical solutions.

Keywords: Hill numbers; bighorn sheep; bison; frequency of occurrence; grazer-browser continuum; herbivore; microhistology; relative read abundance.

FIGURE 1

Different impacts of RRA thresholds on simulated specialist and generalist diets. The dietary profiles of a (a) specialist, (b) intermediate, and (c) generalist feeder as simulated using Pareto distributions. When the shape parameter (α) and total number of food taxa (Taxa_n) are low, there is a large skew in the rank‐abundance distribution (i.e., few food taxa with high relative abundance); increasing these values increases the richness and evenness of the dietary profile (i.e., many food taxa, each with lower relative abundance). In each stacked barplot, the color of each segment represents the relative abundance of each simulated taxon in the diet profile. Increasing the threshold from 0% to 5% for each diet profile resulted in differential impacts on the (d) inferred dietary richness and (e) % loss of initial dietary richness from each sample. A 2.8% threshold (grey dashed lines in d and e) results in similar levels of inferred dietary richness and % losses of taxa across samples

FIGURE 2

Using RRA thresholds differentially impacted samples that varied in evenness, altering the inferred rank‐order of dietary richness. Stacked barplots show four representative dietary DNA metabarcoding profiles from (a, b) bighorn sheep and (c, d) bison in summer. The color of each segment represents the relative abundance of a taxon prior to abundance‐based filtering. Increasing the minimal RRA threshold from 0% to 5% for each dietary profile resulted in different impacts on (e) the inferred level and rank‐order of dietary richness across samples as well as (f) the % loss of dietary richness. A 4% threshold (grey dashed lines in e and f) resulted in similar levels of inferred richness and % losses of taxa across samples

FIGURE 3

Thresholds altered ecological patterns in dietary DNA metabarcoding data. We compared (a) mean dietary richness and (b) total population‐level dietary richness of bighorn sheep and bison in summer and winter. Total population‐level dietary richness was estimated for winter and summer based on extrapolation to double the minimum seasonal sample size for each species (N = 8 bighorn sheep; N = 20 bison). Error bars represent (a) standard deviations and (b) 95% upper and lower confidence intervals. In all plots, lines connect mean dietary richness for winter and summer at each relative threshold

FIGURE 4

Hill numbers applied to both simulated and real dietary DNA metabarcoding data. All curves show a decline in apparent dietary diversity with increasing q due to the increasing emphasis on abundant taxa. Curves show how sensitive each set of diet profiles is to increasing q based on (a) simulated dietary profiles using different Pareto distributions (Figure 1), (b) a set of representative samples from Yellowstone (Figure 2), (c) the average population‐level values from Yellowstone (Figure 3a), and (d) the total population‐level estimated values from Yellowstone (Figure 3b). In contrast to the effect of applying RRA thresholds to the same data, these curves convey more information about the relative abundance of both common and rare taxa while retaining clearer rank‐order of samples

FIGURE 5

Seasonal changes in dietary diversity based on DNA metabarcoding and microhistology. For both bighorn sheep and bison, we compare log‐transformed (a) dietary richness (q ⁰), (b) the number of “typical” (q ¹), and (c) the number of dominant (q ²) plant taxa identified in DNA metabarcoding data (dark solid lines) and microhistology data (light dashed lines). Lines connect the mean values with error bars that represent standard deviations. For microhistological analysis, when multiple samples were collected per herd per season, these samples were pooled into a composite scan of fecal material