2. Statistical analyses
All statistical analyses were performed in R version 3.6.1 (R Core Team, 2019). We first performed a generalized linear model (GLM) with Gaussian distribution to assess the relationship between eDNA persistence, eDNA state, and environmental conditions. The eDNA decay rate constants (per hour) were treated as the dependent variable, and the filter pore size (µm), DNA fragment size (bp), target gene (mitochondrial or nuclear), water temperature (°C), water source (artificial, freshwater, or seawater), and their primary interactions were included as the explanatory variables. We first confirmed that the multi-collinearity among the variables was negligible (1.028 to 1.096), by calculating the generalized variance inflation factors (GVIF). We then selected models based on Akaike’s Information Criterion (AIC), using the dredgefunction in the ‘MuMIn’ package in R (Bartoń, 2019). We adopted the model with the smallest AIC value, and all models with ⊿AIC (i.e. difference in the AIC value) less than two were selected as the supported models (Burnham & Anderson, 2002).
We performed an additional meta-analysis to examine the relationship between the DNA fragment size and eDNA decay rate constant. Most eDNA studies conducted to date have targeted short DNA fragments (<200 bp), and only three papers have reported eDNA decay rates targeting longer DNA fragments (>200 bp); however, they yielded inconsistent conclusions (Tables 1 & S1). Taking this into consideration and targeting eDNA decay rate constants derived from <200 bp DNA fragments, we performed a linear regression to assess the effect of DNA fragment size on eDNA degradation.