Comparison of beta diversity patterns between bulk DNA and eDNA datasets of the five primer sets
To compare species community composition of the bulk and ethanol samples across primer sets and locations, we rarified all samples in the five datasets at 25 000 reads. Samples with less than 25 000 reads were removed from the dataset. This threshold was chosen because the plateau phase of the rarefaction curves was reached for all primer sets while a minimum of samples had to be removed (6, 1, 2, 4 and 1 sample were removed for primer sets A, B, C, D and E respectively). The resulting count tables were fourth root transformed, a transformation commonly used for ecological datasets with many zeros and a few large values (Quinn & Keough, 2002). Bray-Curtis dissimilarity index was calculated between samples and used to construct a metric multidimensional scaling (MDS) plot for all primer sets combined and for each primer set separate in the R package Vegan 2.5.6 (Dixon, 2003). Next to this abundance based index, the Jaccard similarity index was calculated for each primer set separate. This index is based on presence/absence data, and determines the percentage of shared species between samples.
Each of the five primer datasets was characterized by a fully crossed design for the factors location (four levels: 120, 330, 840 and ZVL) and DNA source (two levels: eDNA from ethanol and bulk DNA). A two-way PERMANOVA was performed with 9999 permutations to test for significant differences in macrobenthos communi ties between locations, between DNA source and the interaction location X DNA source. The homogeneity of multivariate dispersions for the main factors location and DNA source and their interaction was assessed using the betadisper and the permutest functions (9999 permutations) from the Vegan package. Pairwise posthoc tests were conducted for the factor “Location” when the interaction term was non-significant using the pairwiseAdonis library (Martinez Arbizu, 2019). When the interaction term is significant, pairwise tests should be conducted within each level of the factor DNA source. However, in view of the low level of replication at this level, such analyses were inappropriate. The false discovery rate, the expected proportion of false discoveries amongst the rejected hypotheses was controlled by the BH method (Benjamini & Hochberg, 1995) using 999 permutations. The statistical analyses were conducted on the Bray Curtis and the Jaccard distance matrices for each primer set.