Statistical analysis

The statistical analysis was performed using the statistical program R version 3.2.4⁴¹. For all parameters, we first compared parameters between maternal Daphnia from the control (ASTM medium, C) and from Treatment I_a (PM) to test whether D. rerio was a useful predator for testing anti-predator defence mechanism in maternal D. magna. Secondly, we analysed the differences between Treatment I_a (PM) and Treatments II – V (PM + different concentration of AgNPs), including Treatment I_b (PM + NM-300K DIS) to analyse the influence of PM in combination with AgNPs and to exclude possible effects of the dispersant agent on test animals (maternal Daphnia). For each treatment, we calculated the life-history parameters reproduction (cumulative mean number of offspring) ± standard deviation (sd), time to first brood (days ± sd), maternal body length (mBL; mm ± sd), offspring body length (oBL; mm ± sd), maternal spine length (mSL; mm ± sd), offspring spine length (oSL; mm ± sd), and checked the data for normal distribution (Shapiro-Wilk test) and for homogeneity of variances (Bartlett´s test). If both requirements met, we performed a one-way analysis of variances (ANOVA), followed by a Dunnett´s post hoc-test for multiple comparisons to test for statistical differences within treatments. Was one requirement not fulfilled, the nonparametric alternative, the Kruskal-Wallis test and afterwards the Dunn’s Test of multiple comparisons using rank sums⁴² was used. Because relative spine length of maternal Daphnia (mRSL) and relative spine length in offspring (oRSL) are bounded²⁷, the data were analysed as dependent variables by using a ‘glmer’ (Generalized Linear Mixed Effect Model) of the package lme4⁴³. As fixed factor, we added treatment as the categorical variable to each model. Relative spine length of maternal Daphnia (mRSL) and relative spine length in offspring (oRSL) were modelled using a Gamma error distribution and a Log link function²⁷. We included the number of moults and identity of test animals as nested random effects to the model. Model assumptions were checked visually. The p-values were adjusted with Bonferroni correction. Significant p-values were marked with asterisks (*P < 0.05, **P < 0.01, ***P < 0.001). All p-values are two tailed.