Evaluation of fish biodiversity in estuaries using environmental DNA metabarcoding

the river and shore from the river mouth, using Google Earth Pro (http://support.google.com/ earth/answer/21995?hl=ja). Statistical analysis To examine the effect of salinity or water temperature on the ratio of freshwater, brackish, or seawater species, we used general linear models (GLMs) with a negative binominal distribution and a log link function. To this end, we applied the glm.nb function in the MASS package. The number of freshwater, brackish, or seawater species (classified by Nelson [36]) in each sample was used as a response variable; salinity or water temperature were explanatory variables; and the total number of fish species represented an offset term. To verify the accuracy of the six models, the areas under the Receiver Operating Characteristic curves (AUCs) were calculated, using the roc function in the pROC package [37]. Accuracy was defined as low (AUC < 0.7), moderate (0.7 � AUC < 0.9), and high (AUC � 0.9) (Table 2). To examine the human impact on the number of fish species, we again applied the above GLMs using the glm.nb function in the MASS package. The number of species in each river was used as a response variable. We used data about TN, TP, and revetment rates as indicators of human impact. However, both TN and TP had a high variance inflation factor (VIF), which indicated high multi- collinearity among these variables (VIF = 26.1 and 15.6 for TN and TP, respectively, VIF = 7.3 for revetment rate). After removal of TP, there was no multicollinearity between TN and revetment

rate (VIF = 7.0), so we used TN and revetment rates as explanatory variables for our GLM analyses. These VIF values were calculated using the vif function in the car package [38]. The number of sam- ples was used as an offset variable. For model selection among GLMs, we used the dredge function in the MuMIn package [39]. The best model was selected using Akaike’s information criterion (AIC), which stipulates that the best model for any candidate set applied to a given data set is that with the lowest AIC value. Following Burnham and Anderson [40], models with ΔAIC < 2 were assumed to be reasonable alternatives to the best model and thus were retained (Table 3). All statistical tests were carried out using R software ver. 3.5.2 [41]. Results Species occurrence A total of 182 species from 67 families were detected in the present eDNA survey (S2 Table). Most species (94) occurred in the Sendai River and fewest (25) in the Tama River; whereas the Table 2. Summary of models† used to assess the effect of each environmental factor on the rate of freshwater, brackish, or marine fish. Variable Freshwater species Brackish water species Seawater species Effect of salinity (Intercept) 2.615��� 2.884��� 1.688��� Salinity -0.046�� -0.008 0.042��� AUC 0.839 0.825 0.838 Effect of water temperature (Intercept) 4.377�� 2.114��� -1.525� Water temperature -0.098 0.024 0.152��� AUC 0.856 0.841 0.961 †Based on comparison of null and full models in general linear model results; β coefficients of predictor variables are shown.

Abbreviations: AUC, area under the Receiver Operating Characteristic curve �p < 0.05 �� p < 0.01, and ���p < 0.001 in a Ward test https://doi.org/10.1371/journal.pone.0231127.t002 PLOS ONE Biodiversity of five estuaries PLOS ONE | https://doi.org/10.1371/journal.pone.0231127 October 6, 2020 6 / 15