Statistical analyses
All statistical analyses were performed using R version 4.0.4 (R Core Team, 2021). The decay rates of Japanese jack mackerel eDNA were estimated using the time-series changes in their eDNA concentrations from each sampling tank. Previous studies estimated eDNA decay rates by fitting a monophasic exponential decay model (Strickler et al., 2015; Jo et al., 2020b) as follows:
\begin{equation} C_{t}=C_{0}e^{-kt}\nonumber \\ \end{equation}
where \(C_{t}\) is the eDNA concentration (copies) at time \(t\)[hour], \(C_{0}\) is the eDNA concentration at time 0, and \(k\) is the decay rate constant (/hour). We used a linear model to compare the decay rates of each type of eDNA between BAC treatments, where log-transformed eDNA concentration was included as the dependent variable and sampling time point (hour), BAC treatment, and their interaction were included as explanatory variables.
Alternatively, following Eichmiller et al. (2016), we used a biphasic exponential decay model if the fitness of a monophasic decay model was poor and there was an obvious breakpoint between two distinct phases of eDNA degradation as follows:
\begin{equation} C_{t}=C_{0}e^{-k_{1}t^{{}^{\prime}}}e^{-k_{2}(t-t^{{}^{\prime}})}\nonumber \\ \end{equation}
where \(k_{1}\) and \(k_{2}\) are the eDNA decay rate constants at the initial rapid and following slower phases, respectively, and \(t^{{}^{\prime}}\)is the time of breakpoint between different degradation phases (hour). We estimated eDNA decay rates with 95% confidence intervals (CIs) and breakpoints using the package ‘segmented’ (Muggeo, 2017). We compared the fitness of monophasic and biphasic decay models between BAC treatments by calculating Akaike’s information criterion (AIC). All eDNA samples with concentrations below one copy per reaction were excluded.
Furthermore, we compared time-series changes in fish species composition inferred by eDNA metabarcoding between BAC treatments. For each time point, the number of fish species detected by eDNA metabarcoding was compared between BAC treatments using the exact McNemer test in the package ‘exact2×2’ (Fay, 2010). We then visualized the community compositions based on Jaccard dissimilarities using a two-dimensional non-metric multidimensional scaling (nMDS) with 10000 permutations byvegdist and metaMDS functions in the package ‘vegan’ (Oksanen et al., 2019). In addition, we performed a permutational multivariate analysis of variance (PERMANOVA) with 10000 permutations using adonis function to examine whether the community compositions were different among BAC treatments and/or time points.