Statistical analysis
All analyses were conducted in R v4.0.2 (http://www.R-project.org/). The structure of the final linear models and sample sizes break down by the levels of categorical explanatory variables are summarized in (Table 2). Non-significant interactions were removed from the models. Significance of the final models was evaluated using ANOVA tables using Type II and III sums of squares for models without and with significant interaction, respectively. Differences among groups were analysed using Tukey’s HSD post-hoc test. We have accounted for the ontogenetic effects by using fork length as a proxy of individuals age. Fork length has been used as explanatory variable in all models. When models contained other explanatory variables correlated to the fork length, we used residuals from a linear model between the log transformed variable of interest and fork length to prevent collinearity (see Table 2).
We tested the variability in content of ALA and EPA, and of total lipids in macroinvertebrates across all sampling sites (i.e., category with 6 levels – 3 streams × 2 populations [allopatric, sympatric]) and subsidy types (category with 2 levels – aquatic or terrestrial macroinvertebrates). Since there was no difference lipid quality of aquatic and terrestrial subsidies across the six sampling sites (see results for details), we did not include the stream or sampling site in the models. We collected samples of dorsal and ventral muscle tissue because concentration of adipose cells is often higher in ventral than in dorsal muscles and thus these two tissues can differ in their lipid content (Fauconneau et al. 1995; Ebm et al. 2021). However, the total lipid content and the relative content of n-3 LC-PUFA were repeatable across dorsal and ventral muscle samples (total lipids: Radj = 0.25, 95% CI [0.06, 0.44]; ALA: Radj = 0.59, 95% CI [0.44, 0.72]; EPA: Radj = 0.62, 95% CI [0.48, 0.74]; DHA: Radj = 0.56, 95% CI [0.41, 0.70]. Therefore, for further analysis we have used an average value across tissues for each individual. Total lipids in all models were reported as mass fractions (mg/g of the dry mass), while all models and figures including n-3 LC-PUFA reported the relative fatty acid contents (% of fatty acid methyl esters; with exception of Fig. 1). This is a common approach (e.g., Twining et al. 2019; Ebm et al. 2021) because mass fractions of fatty acids are often closely correlated to total lipid mass fractions, while the relative content of fatty acids is independent of the total lipids and thus more comparable across the sample types and studies, particularly when only a part of an organism (e.g.,muscle tissues) is analysed.
Table 2Summary of the final models as reported in the results. The following categorical variables have been used: sex (male, female), sampling site (6 levels, see Table 1), competition (allopatry, sympatry), prey subsidies (aquatic, terrestrial). Note that model 1 was run with and without sex as an explanatory variable because sex was only determined in the subsample of dissected fish, which substantially reduced the sample size compared to the model without sex. However, the two models yielded similar results.