Statistical analyses
For minnow sampling, two different kind of gear was used (i.e. gillnets with one mesh size in the lake locations and electrofishing in the stream locations). To test if total length of minnows differed between habitats (i.e. lake and streams), thus to rule out a potential size-selective effect of sampling gear on my interpretations, I conducted an ANOVA with total length as dependent variable and location nested within habitat as independent variable.
Variation in morphology between the habitats (i.e. lake and stream), and locations was examined using MorphoJ v.1.06d (40). I checked for outliers using the “Find outliers” function. To correct the shape data for body size, I used a regression of the shape scores (Procrustes coordinates) on size (centroid size) for each location separately and the residuals of this regression was used for all further analyses (41). A Discriminant Function analysis (DFA) and a Canonical Variate analysis (CVA) were used to assess significance of shape differences between habitats. A second CVA was conducted for pairwise comparison between the six locations. The shape analysis was restricted to a maximum of 30 individuals of each location.
As minnows crush their food using pharyngeal teeth many individuals solely had unidentified items and mucus in their guts (37.5 % of all minnows caught) and these individuals were excluded from the analyses. Proportional data was arcsine-square root transformed prior to statistical analyses. Ordination of multivariate diet composition was based on Bray-Curtis similarities and analyzed using a PERMANOVA with location nested within habitat, setting location as a random factor and habitat as a fixed factor. The significance of the model was tested with unrestricted permutations (999 permutations) with type III sums of squares. To test whether the contribution of the three diet categories (i.e. zooplankton, benthic invertebrates, and terrestrial insects) differed between the individuals caught in the lake versus streams, I applied Mann-Whitney U-tests. I further conducted Kruskal-Wallis tests with Bonferroni –adjusted Dunn´s pairwise comparisons to analyze if the contribution of the three diet categories differed between the six different locations.
To test the relationship between resource use and minnow body shape, I used Spearman´s rank correlation on the individual proportions of zooplankton in gut content and the first axis of the CVA (CV 1).
PRIMER v 7.0.13 with the PERMANOVA add-on (Primer E Ltd. Plymouth, United Kingdom) was used to analyze the multivariate dataset, whereas univariate analyses were conducted using IBM SPSS v.25 (IBM Corp., Armonk, NY, USA).