Harnessing the power of metabarcoding in the ecological interpretation of plant-pollinator DNA data: strategies and consequences of reads filtering

Abstract

DNA metabarcoding approaches to analyse complex mixtures of pollen has become the standard in pollination biology, especially in the light of recent threats affecting pollination. In spite of the increasing adoption of High Throughput Sequencing (HTS) approaches, these studies generate huge numbers of raw reads, some of which might be associated to false positives or infrequently recorded species with potentially little biological information. If these reads are not discarded (i.e. pruned), they can lead to changes in the ecological findings and lead to biased conclusions. In this study we reviewed 42 papers in the recent pollen DNA metabarcoding literature and focused on the type of pruning applied. We also tested whether the different types of those cut off threshold may leave a mark on the DNA metabarcoding data. To do so, we compared for the first time community composition, species richness and networks of species interactions (i.e. Connectace, Modularity, Connectivity and Shannon entropy) associated with the most relevant ways of treating HTS outputs: no cut (no reads filtering), or cutting levels obtained as proportional 1% of sample total reads, or as fixed amount of 100 reads, or from ROC (Receiver operator characteristic). Results clearly indicated that pruning type shapes species composition and that to apply or not a threshold dramatically impacts ecological indices, potentially increasing the risk of misinterpreting DNA metabarcoding data under an ecological point of view. Given the high methodological heterogeneity from the revised literature, we discuss in what conditions filtering types may be more appropriate, and also recommend to biologically justify the pruning threshold when analysing DNA metabarcoding raw reads, and to develop shared approaches to make future studies more comparable.