Ordination and statistics
We prepared NMDS plots of all three datasets (“Pilot”, “Pilot full”
and “Full”, see explanation below) with abundance and presence-absence
based dissimilarity metrics to investigate the importance of depth and
transformation for determining the composition of prey (Fig. 2).
Depending on the dataset being used, ordination required 4 or 5
dimensions to reach a conservative and low stress-level of 0.1. The
pilot prey reads required fewer dimensions (k = 4) than the full dataset
and the dataset consisting of the full subset (k = 5). Regardless of
dataset or metric, prey composition differed significantly between
copepods from different seasons and stations (PERMANOVA, p <
0.001, Fig. 2). The most visually distinct clusters were found when
using the season sampled for profiling prey compositions, and samples
acquired during the pilot (“Pilot”, Fig. 2a and d) formed less
distinct clusters than those from the full sequencing. The same physical
samples subset from the full dataset (“Pilot full”, Fig. 2b and e)
formed more divergent clusters. Ordination of the complete set of
samples (”Full”, Fig. 2c and f) returned a pattern typical of a
seasonal transition, with prey compositions from successive seasons
overlapping, and samples from disparate seasons (e.g. August and
April/May) forming separate clusters. Successive Betadisper tests (Table
S3) indicated however that the clusters observed may be influenced by
heterogenous dispersion (e.g. Fig. 2f). The copepod species sampled was
a less significant predictor of pilot and pilot full diets (Fig. 2a and
b) when using Bray-Curtis as dissimilarity metric (p = 0.003 and p =
0.03, respectively), than with Jaccard. For both datasets with greater
depth (“Pilot full”, and “Full”), Jaccard dissimilarities led to
visually greater separation of clusters.