Patters and processes in the metacommunity structuring of benthic diatoms
The morphological analysis recovered 40 genera, in contrast to the 90 genera recovered with metabarcoding. We found 98 species based on morphological identification and 219 species based on metabarcoding. The comparison between morphological and molecular inventories through Venn diagrams showed considerable differences between both methods at genus and species level. We found 10 genera detected only by light microscopy (of which 8 of them had reference sequences in the database), 60 genera detected only by metabarcoding and 30 genera (30%) detected by both methods (Figure 3 a). At species level, we found 55 species detected only by light microscopy (of which 27 of them had reference sequences in the database), 176 species detected only by metabarcoding and 43 species (15.7%) detected by both methods (Figure 3 b).
Stress values in nMDS ordinations for morphological (species–level nMDS =0.22, genus–level nMDS =0.23) and molecular (species–level nMDS =0.24, genus–level nMDS =0.22) data suggested a reasonable fit (Clarke, 1993). The Procrustes rotation analysis using nMDS scores(Figure 4) showed that most of the study sites displayed a relatively low degree of similarity, indicating a poor correspondence in the compositional variation between morphology–and molecular–based metacommunities (here, PROTEST for species–level resolution data,\(m_{1-2}\)=0.29 and p=0.33; and PROTEST for genus–level resolution data, \(m_{1-2}\) =0.30 and a p=0.28). The relatively high procrustean residuals (red arrows in Figure 4 ) re–emphasised this mismatch between the ordinal results of the test datasets at different taxonomic resolutions. Importantly, the direction of the movement and the length of the arrows in the procrustean plots were associated with the distribution of both morphology– and molecular–based assemblages. In this vein, the low correspondence found for both species– and genus–level data was partially caused by e.g. Navicula soehrensis , Cyclostephanos invisitatus , Cymbella subhelvetica , Navicula notha , Tabellaria fenestrata ,Luticola goeppertiana and Amphora indistincta , which were present exclusively in the morphology–based samples. Moreover, molecular–based assemblages included a number of taxa that were absent in morphological identifications such as, Nitzschia fruticosa ,Eunotia arcus , Attheya septentrionalis , Haslea pseudostrearia , Lucanicum sp ., Leptocylindrus sp . andPseudictyota sp.
Species–level data based on morphological identifications showed the strongest environment relationship (p≤0.01) of all different study approaches, and the BIOENV routine selected conductivity, fluorides, total phosphorus (TP) and total suspended solids (TSS) for the best environmental distance matrix. Ammonium and TSS related to species–level data based on DNA metabarcoding of the entire assemblage, whereas fluorides and TSS structured the genus–level data based on morphological and molecular approaches, respectively (Table 3) . According to the non–ranked Mantel tests and partial non–ranked Mantel tests, correlations between dissimilarity and distance matrices showed that only environmental distances were significantly correlated with biological –Jaccard– dissimilarities, whereas geographical distances were never significantly correlated with compositional variation in diatom metacommunities (Table 3) . Distance–based redundancy analysis (db–RDA) showed that half of the environmental variables selected by BIOENV were significantly related to variation in community composition, but that only a rather small amount of compositional variation (in terms of adjusted \(R^{2}\)values; here,\(R^{2}\)<0.3) could be explained by these environmental variables (Table 4) . Of the four dissimilarity matrices subjected to db–RDAs, species–level data based on morphological identifications were best explained by the environmental variables and the genus–level data based on DNA metabarcoding the worst.
We used Mantel correlograms to examine if there was significant spatial autocorrelation at any distance class. In this regard, we detected only very weak, but no significant spatial autocorrelation for morphological and molecular data at different taxonomic resolutions (Figure 5) . Perhaps more importantly, re–running the analyses with the spatial turnover component of the Jaccard index did not alter our main results