Statistics
Differences in the ESVs/OTUs/morphospecies richness and correlation analysis between the three treatments (HTS 10, HTS 0.5 and microscopy) were tested using Kruskal-Wallis Analysis of Variance (ANOVA) by Ranks and Spearman Rank Order Correlations in STATISTICA 7 (StatSoft 2004). Permutational Analysis of Variance (PERMANOVA, with 9999 permutations) and Principal Co-ordinate Analysis (PCoA) of Bray-Curtis similarity matrices of log transformed data were performed in PRIMER v6 (Clarke & Gorley 2006). The values of the PCoA first axis were selected as a response variable for Random Forest analysis to identify most important environmental variables for diatom community assembly. The latter was performed in R (R-Core-Team 2015), using the package ‘randomForest’ (Liaw & Wiener 2002). The variables included for Random Forest analyses were maximum depth, conductivity, and concentration of zinc (Zn), calcium (Ca), silica (Si), chromium (Cr), nitrate (NO3), copper (Cu), manganese (Mn), iron (Fe), strontium (Sr), vanadium (V) (Table S1). Variables, selected by Random Forest model per treatment, were used in marginal test analysis (with 9999 permutations), in PRIMER. Only variables with P-values < 0.05 were displayed as vectors in the PCoA ordination plots.
Bray-Curtis similarity matrices of Hellinger-transformed data per treatment were compared with Mantel tests (with 9999 permutations, method=”spear”) to assess the correlations between sample similarities as implemented in the ‘vegan’ package (Oksanen et al. 2015). In addition, Procrustes analyses were used to compare the similarity of PCoA ordinations between different treatments (in ‘vegan’). To examine the presence of diatom OTUs that were consistently detected either by the HTS 10 or the HTS 0.5 treatment, an indicator species analysis (with 9999 permutations) was performed using the ‘indicspecies’ package (De Caceres, Jansen & De Caceres 2016).
The analyses for comparing metabarcoding treatments (HTS 10 and HTS 0.5) were performed for 20 corresponding samples. Because of smaller sample size for the microscopy data, analyses for HTS 10 vs . microscopy and HTS 0.5 vs . microscopy were performed with 14 and 11 samples, respectively (see Table S1).