Testing the effect of sequencing depth on prey composition
Non-metric multidimensional scaling (NMDS) plots were used to visualize how depth of sequencing and data transformation may influence prey composition resolution. Multiple variants were made to explore beta-diversity using both occurrence, and presence/absence-based dissimilarity metrics. All ordinations were computed from sample-wise dissimilarities from compositional data at zOTU-level. Relative abundances were used to compute Bray-Curtis dissimilarities (Bray & Curtis, 1957), whereas a presence/absence-table enforcing a 0.01% relative abundance threshold was used for Jaccard dissimilarities (Jaccard, 1901). We used Scree-plots to find the appropriate dimensions for a conservative acceptance threshold of ≤0.1 stress. Final NMDS plots were calculated iteratively (trymax = 100) with adequate dimensions (k = 4 or 5) using the metaMDS function of vegan (Oksanen et al., 2019). PERMANOVA (nperm = 10000) analyses were used to test if three explanatory variables (season, station and species) accounted for the observed variance in prey composition. We subsequently tested if the grouped samples had homogenous and comparable dispersions (p ≥ 0.05), or if compositional differences in prey could be due to heterogenous dispersion among groups (p < 0.05, Betadisper).