This work is based on two river networks, which were monitored as part of INRA-AFB research projects; one in the French overseas department of Mayotte, the other on the French metropolitan territory. These watercourses are all subject to regulatory monitoring under the WFD.

For Mayotte, although the surface area of this French department is modest (376 km²), its rivers present a great diversity of natural and anthropogenic situations and are characterised by various benthic diatom assemblages. A total of 45 stream sites were monitored twice – in years 2014 and 2015 – (Figure 2), the sites representing a marked gradient in water quality, including sites with the lowest possible impact, highly polluted sites and finally intermediate sites belonging to the WFD control network (RCS). Most of the methodological developments of diatom DNA metabarcoding, presented in this article, were performed from this dataset.

Figure 2.

Location of sampled sites on rivers in metropolitan France in 2016 and 2017 (left) and in the overseas department of Mayotte in 2014 and 2015 (right).

For river sites sampled in metropolitan France, they were selected to meet two objectives: (i) complete the DNA barcode reference database for some missing metropolitan important diatom species and (ii) compare ecological status assessments obtained using morphological and molecular approaches on a large scale. In collaboration with the regional environmental services in charge of WFD monitoring (DREAL), 461 sites were chosen to meet these two objectives (Figure 2). For the first purpose (completing the reference database), on the basis of previous morphological inventories, we selected sites known to contain a high abundance of taxa missing in the database (in particular rivers located in the Pyrenees, the Alps, the Massif Central and Brittany). For the second objective, we have selected (i) sites all along the length of the rivers with a marked water quality gradient from upstream to downstream (Seine, Loire, Vienne, Adour, Doubs, Somme, Durance, Allier, Oise, Dordogne and Garonne) and (ii) all the sites of the WFD network in the Rhône catchment area (including the departments of Ain, Jura, Haute-Savoie, Savoie, Rhône and Loire) characterised by a great diversity of environments and pressures (e.g. agricultural, industrial, urban, lowlands and mountainous areas).

Aquatic biofilms (periphyton) were collected during the river monitoring campaigns carried out within the framework of the WFD, in agreement with and with the participation of the actors in charge of monitoring (Water Agencies, DREALs, DEALs, consultancies), corresponding to the 2014 and 2015 campaigns for the Mayotte rivers (45 sites, 80 samples) and the 2016 and 2017 campaigns for the metropolitan rivers (461 sites, 461 samples) (Figure 2).

The aquatic biofilm containing the benthic community of diatoms was sampled according to the current standard (NFT 13946, AFNOR 2014). The recommendations of the technical report, CEN/TR 17245 (CEN 2018a) from the European Committee for Standardisation (CEN) have been followed to ensure that these samples are compatible with the subsequent application of molecular biology techniques. Briefly, the biofilm was brushed from at least five submerged stones located in the stream’s lotic zone and fixed in ethanol at a final concentration of at least 70%, which preserves the DNA. In the end, each sample (50 ml) is homogenised and sub-sampled into two batches, for morphological and molecular analysis, respectively.

The morphological approach is based on the identification by optical microscopy of diatom taxa based on the morphology of their siliceous external skeleton (frustule). The fixed biofilm samples are prepared in such a way as to allow the determination and counting of diatom species by microscopy, according to the standard NFT 90-354 (AFNOR 2016) (Figure 3). The first step consists of using several successive chemical treatments (H₂O₂, HCl) to remove all organic matter and calcium carbonates and preserve only the diatom siliceous skeleton. Then the treated sample is homogenised and a permanent preparation is carried out by fixing an aliquot of the sample between a slide and a glass lamella in a resin with a high refractive index (Naphrax), thus allowing observation under an optical microscope. Finally, at least 400 diatom valves are identified at species level (or genus when impossible), using floristic guides adapted to Europe for metropolitan samples (e.g. Krammer and Lange-Bertalot 1986; Lange-Bertalot et al. 2017) or tropical regions for Mayotte samples (e.g. Metzeltin and Lange-Bertalot 1998; Tudesque et al. 2008), producing a floristic inventory with the relative abundances of diatom taxa which is used to calculate a quality index for the sampled site.

Figure 3.

Main steps of the microscopy morphological approach for biomonitoring with diatoms.

The molecular approach in DNA metabarcoding relies on the identification of diatom taxa based on a short DNA fragment (DNA barcode). It requires several successive molecular biology steps to acquire sequence data (DNA extraction, PCR amplification, high-throughput sequencing) and then computer processing of these data to acquire diatom inventories (bioinformatics processing, statistical analyses) (Figure 4). Roughly the same process was used for the samples of the Mayotte and metropolitan rivers; however, some methodological choices differed (protocols, technologies, settings) and are detailed in the scientific articles and synthesis reports corresponding to these two projects (Mayotte: Vasselon et al. 2017b ; Metropolitan France: Keck et al. 2018, Rivera et al. 2020). The most recent and optimised methodology is presented in this article.

Figure 4.

Main steps of the DNA metabarcoding approach for biomonitoring with diatoms.

DNA extraction is performed from the biofilm pellet obtained after centrifugation of the sample (30 min to 17,000 g) using the NucleoSpin Soil kit (Macherey-Nagel), according to the methodology described by Vasselon et al. (2017b). A 312 bp part of the rbcL chloroplastic gene, encoding the large RuBisCo subunit, is used as a DNA barcode with both good specificity and polymorphism for diatoms, following recommendations of Kermarrec et al. (2013). For each sample, at least 2 PCR amplifications of this DNA barcode are performed using a mix of 3 forward primers (Diat_rbcL_708F_1, Diat_rbcL_708F_2, Diat_rbcL_708F_3) and a mix of 2 reverse primers (R3_1, R3_2). The PCR mix and the amplification conditions are described in Keck et al. (2018). The pool of PCR products is then sent to a sequencing platform (e.g. GeT-Plage Toulouse (France), PGTB Bordeaux (France)) which carried out: (i) purification of PCR products; (ii) preparation of sequencing libraries by adding a sample-specific tag and sequencing adapters to the PCR products; (iii) preparation of the final “pool” corresponding to the equimolar mixture of the sample libraries; (iv) paired-end sequencing of the library pool performed with the Illumina Miseq technology (Keck et al. 2018).

The first processing steps of the sequence data (“demultiplexing” and “pair contigating”) were performed by the sequencing platform which provides a “fastq” file. All bioinformatics steps were carried out with the Mothur programme (Schloss et al. 2009). The second processing step is dedicated to cleaning sequence data by eliminating sequences of incorrect length, containing errors (e.g. ambiguous bases “N”, homopolymers), as well as chimeric sequences. Next, a “de-replication” step allows the grouping of similar sequences together in order to simplify the dataset and facilitate further processing. Only unique sequences represented by more than 2 reads, over the full dataset, are conserved. These de-replicated DNA sequences are then compared with the Diat.barcode reference base (Rimet et al. 2019; previously called RSyst::diatom, Rimet et al. 2016) using a naïve Bayesian classification method (Wang et al. 2007), to assign a taxonomy to each sequence (confidence level > 85%). Only the sequences assigned to the phylum “Bacillariophyta” (diatoms) are kept for further analysis and grouped into OTUs (Operational Taxonomic Units), according to their genetic similarity (> 95% similarity threshold). Finally, a representative DNA sequence and a taxonomy are assigned to each OTU (Keck et al. 2018), providing an inventory of diatom taxa with their relative abundances (estimated by the relative abundances of reads) which is used to calculate a quality index for each sampled site.

Although the BDI is the French diatom index used for WFD, the Indice de Polluosensibilité Spécifique (IPS) (Cemagref 1982) was chosen in this study because it includes a greater number of diatom species into its calculation and is frequently used in WFD intercalibration exercises (Coste et al. 2009; Kelly et al. 2014). For each site, morphological and molecular IPS values were calculated from morphological and molecular inventories using the Omnidia software (version 6.0.2s) (Lecointe et al. 1993).

With regard to molecular inventories, it has been shown that the copy number of the rbcL gene is directly correlated to the biovolume of diatom species (Vasselon et al. 2018). Compared to the morphological approach, the molecular approach, therefore, tends to overestimate species with a high biovolume. To limit this bias, a correction factor (CF) was proposed for each species on the basis of its biovolume and applied to correct molecular inventories abundancies in samples from Mayotte and metropolitan France. These corrected taxonomic inventories were then used to calculate a corrected molecular IPS value for each sample.

In the end, three IPS scores were produced, based on the morphological inventory, the molecular inventory, corrected and not corrected by the CF, for each sample.

The use of diatom DNA metabarcoding as a tool for assessing the ecological status of watercourses has been explored over the past 10 years (beginning in France with the PhD thesis of L. Kermarrec (2012)). During this period, complementary studies have made it possible to remove certain methodological obstacles, such as: the validation of the suitability of the rbcL DNA barcode for diatoms identification (Kermarrec et al. 2013, 2014), the definition of good practices for the creation of a barcode reference database (Zimmermann et al. 2014) or the implementation of the first complete methodological framework from the biofilm sample to the assessment of the ecological status of the river site from where it was sampled (Kermarrec 2012; Visco et al. 2015; Zimmermann et al. 2015). However, some methodological biases were still limiting the production of reliable molecular taxonomic inventories through DNA metabarcoding, particularly in terms of taxon quantification, thereby producing weak ecological status assessments. As a result, projects were carried out on the French rivers of Mayotte and of the France metropolitan territory in order to consolidate the DNA approach by optimising certain crucial parameters to increase bioassessment efficiency, namely: the coverage of the barcode reference database, the DNA extraction lab step and the quantification of diatom taxa from rbcL HTS data.

Development and completion of the barcode reference database “Diat.barcode”

The ability of DNA metabarcoding to accurately identify diatom taxa in an environmental sample is directly related to the quality of the DNA barcode reference database used to assign a taxonomy to OTUs. This relies largely on its diversity coverage, with one quality criterion being its completion (ideal objective of hosting at least one reference DNA barcode per diatom species, Weigand et al. 2019) and another quality criterion being the level of information provided for each barcode (DNA sequences, taxonomic and autecological information, related metadata etc.). Although there are several international reference databases gathering genetic information obtained by the scientific community, such as the National Center for Biotechnology Information (NCBI) (Federhen 2012) or the Barcode Of Life Data system (WFD) (Ratnasingham and Hebert 2007), the information they contain is generally very heterogeneous in terms of quality (e.g. origin of DNA sequences, wet-lab protocols) and reliability (e.g. traceability of sequences, taxonomic identification).

For this reason, an expert reference database has been developed for diatoms: R-Syst::diatom (Rimet et al. 2016), for which the traditional completion strategy is based on the isolation and culture of environmental cells, their taxonomic identification and the sequencing of the gene of interest from the DNA extracted from the monospecific culture. This diatom database incorporates information about the genes most frequently used as DNA barcodes for diatom metabarcoding: 18S (ribosomal RNA 18S) and rbcL. Data from algae culture collections (e.g. Thonon Culture Collection, www6.inra.fr/carrtel-collection_eng/), the NCBI database and scientific articles are used to annually update this database. Before being integrated into the database, each new DNA sequence is submitted to several curation steps in order to keep only quality sequences with correct taxonomic identification (Rimet et al. 2016). The data from strains that were isolated, identified and sequenced during the French projects have thus contributed to the development of this database, in particular with the addition of 112 rbcL sequences, corresponding to 29 Mayotte diatoms species, some of them tropical or endemic. Despite these efforts, the completeness of this database remains partial as stated by Weigand et al. (2019) in their large gap analysis. For example, of the 100 taxa most frequently identified in morphological inventories during the last monitoring campaigns (1992 to 2014) of French metropolitan rivers, 47 were still absent from the database in 2016 (version 6). As these taxa are very abundant, their weight in the calculation of diatom indices is important and not detecting them is detrimental to the efficiency of molecular indices. To overcome the traditional completion strategy, which is time-consuming and has a moderate success rate, we recently proposed an alternative strategy to recover barcodes from environmental DNA sequences (Rimet et al. 2018). Based on morphological inventories, we focused on low-diversity samples, characterised by diatom assemblages dominated by one taxon absent from the reference database. It was therefore highly likely that a large majority of the metabarcoding DNA sequence reads that were unidentified correspond to this missing taxon. Based on several validation criteria (morphological, genetic, phylogenetic), we associated this environmental DNA sequence with the taxonomy of the species observed in morphology and integrated it into the reference database (Rimet et al. 2018). This operational approach allowed us to easily and quickly fill crucial gaps in the reference database by targeting abundant taxa, important for index calculation. Thus, from the French metropolitan river samples, we could identify 61 “environmental” rbcL barcodes corresponding to 21 species of diatoms, 18 of which were amongst those 100 most frequently identified species in France. Moreover, since 2018, this database has been enriched with collections from other countries (England, Russia) and is managed by a collective of International diatom experts (Rimet et al. 2019), which gives it significant visibility (Pawlowski et al. 2018). This international evolution has led to a change in the name of the reference database now called Diat.barcode, which is hosted and available in open-access on www6.inra.fr/carrtel-collection/Barcoding-database.

DNA extraction from benthic diatom communities

The first laboratory step for diatoms metabarcoding is to extract DNA from aquatic biofilm samples. A wide variety of extraction methods and protocols have been developed, depending on the nature of the sample (e.g. water, soil, biofilm, organic tissue) and the targeted biological group (e.g. bacteria, fungi, diatoms) (Dhaliwal 2013). Despite this wide variety of methods, it is sometimes difficult to obtain DNA of good quality and in sufficient quantity to allow the subsequent lab steps (PCR, sequencing) to be carried out. Two main limitations are observed for aquatic biofilms: (i) it is common to co-extract, together with DNA, organic compounds (humic acids), minerals (ions, metals) or enzymes responsible for DNA degradation or PCR inhibition (Schrader et al. 2012); (ii) diatom cells are protected by a silica frustule which may be difficult to break since the strength of the frustule varies from species to species (Hamm et al. 2003; Moreno et al. 2015). It is therefore important to use a method that allows DNA extraction from all diatom species, in order to ensure a good representation of the community.

Five DNA extraction methods frequently used for diatoms were tested on biofilm samples from the Mayotte project (Vasselon et al. 2017a). These methods were compared according to their ability to: (i) extract DNA, regarding both quality and quantity, (ii) obtain molecular inventories using DNA metabarcoding with the rbcL barcode that correlates well with taxonomic inventories obtained by microscopy and (iii) provide valid ecological status assessments from diatom indices. Although some disparities could be observed amongst the five methods in terms of quality (presence of PCR inhibitors) and quantity of DNA obtained, none of them prevented the subsequent metabarcoding. All molecular taxonomic inventories provided were identical in terms of specific composition (same diatom taxa detected), but significant variations in the relative abundance of some taxa (e.g. belonging to the genera Nitzschia, Amphora, Encyonema and Gomphonema) were observed in two of the five methods. However, these variations had no significant impact on the assessment of the ecological status of the rivers tested, so all the methods tested can be safely used for biomonitoring using diatom metabarcoding (Vasselon et al. 2017a).

Metabarcoding and quantification: impact of the variation in gene copy number

As diatom indices rely on the equation from Zelinka and Marvan (1961), they take into account the relative abundance of taxa. Since the proportions of DNA read sequences obtained in metabarcoding are not clearly correlated to the proportions of cells in microscopy, the ecological status assessments obtained by the two approaches may differ. Amongst the various methodological and biological biases that may potentially influence the relative abundance of DNA read sequences (Pawlowski et al. 2018), the bias related to the variation in the gene copy number appears to have the strongest impact (Figure 1). This bias is dependent on the marker gene used and the targeted biological group, as demonstrated for macroinvertebrates (Elbrecht et al. 2017), fish and amphibians (Evans et al. 2016), oligochaetes (Vivien et al. 2016), foraminifera (Weber and Pawlowski 2013) or microbial communities (Angly et al. 2014). Indeed, Godhe et al. (2008) were able to demonstrate, using quantitative PCR, a correlation between the number of copies of the 18S gene per cell and the cell biovolume of various algae, including diatoms.

We carried out an experiment to verify whether such a correlation existed for the rbcL gene and whether it could be used to correct the relative abundances of diatom species in molecular inventories to reconcile them with those obtained by microscopy (Vasselon et al. 2018). Based on tests on monoclonal cultures from 8 diatom species, the results showed that the copy number of the rbcL gene varies from one species to another and that this relationship is correlated to cell biovolumes (Figure 5). The equation of the correlation curve could thus be used to make the link between cell biovolume and gene copy number per cell, allowing the calculation of correction factors (CFs) specific to each diatom species. These correction factors were calculated and their efficiency was tested on molecular taxonomic inventories obtained from five mock-communities, made from mixtures of the DNA from the 8 tested species in various controlled proportions. The application of the CFs has made it possible to correct the sequence proportions of each of the 8 species in the molecular inventories, making them comparable to those obtained using the morphological approach.

Figure 5.

Correlation between the number of copies of the rbcL gene per diatom cell and the cell biovolume. Adapted from Vasselon et al. (2018).

These CFs were then successfully applied to correct the molecular inventories obtained for the Mayotte rivers WFD network (Vasselon et al. 2018). The calculation method was subsequently refined and new CFs were applied to efficiently correct the molecular inventories from the French metropolitan rivers network (Rivera et al. 2020). The implementation of these quantification corrections is discussed in the following sections of this article.

The DNA metabarcoding optimisations presented above were all integrated at the scale of WFD river networks in Mayotte, France (80 samples) and in metropolitan France (447 samples). These tests were amongst the first to be applied at such a network scale with the objectives of assessing: (i) the large-scale applicability of the molecular approach for diatoms, (ii) its ability to produce taxonomic inventories similar to those obtained with microscopy and (iii) its ability to provide a reliable assessment of the ecological status of rivers.

Potential of diatom DNA metabarcoding to characterise river communities

Molecular inventories have identified 66 diatom species in Mayotte rivers (Vasselon et al. 2017b) and 288 species in those of metropolitan France (Rivera et al. 2020). Morphological taxonomic inventories have identified 204 and 783 species of diatoms in these same rivers, respectively, which is about three times higher. The structures of diatom communities (richness and relative abundances of species), obtained with the two approaches, show a significant but partial correlation (Mantel test; Mayotte: r = 0.42, Metropolitan France: r = 0.63). There are several reasons for the discrepancies observed between morphological and molecular taxonomic inventories:

Gaps in the reference database: despite the efforts made to complete the DNA barcode reference database, it remains incomplete, avoiding the full assignment of DNA read sequences to diatom species. Unassigned sequences represent 12.4% of the dataset in metropolitan France and 40.7% in Mayotte. This percentage is higher in Mayotte due to the poor knowledge of diatom diversity in this tropical island, which includes endemic and tropical species that are not yet morphologically described (Vasselon et al. 2017b).

Taxonomic differences: the presence of morphologically-related species and the constant evolution of diatom taxonomy (Mann et al. 2016) make it difficult to obtain reliable taxonomic identification with microscopy (Figure 1b). In addition, morphological inventories were produced by several taxonomic experts, which may have led to different identifications of the same morphological entity (species). This phenomenon tends to artificially increase the number of species detected through the morphological approach and thus create variability in ecological status assessments. Therefore, intercalibration exercises between diatom experts are required in order to harmonise assessments (Kahlert et al. 2009).

Detection capacity and limit: Morphological and molecular approaches do not give the same insight into diatom communities and, therefore, do not have the same detection capacity for species (Figure 1a). The relative abundance of small species tends to be overestimated by microscopy (more easily observed in microscopy because they are very abundant), while the molecular approach tends to overestimate the relative abundance of species with a high biovolume (more copies of the rbcL gene). Some taxa could therefore not be observed by microscopy, but detected by metabarcoding and vice versa.

False positives: whether through the presence of dead frustules recorded in morphological inventories (Stevenson and Peterson 1991) or the presence of free DNA in the environment from dead individuals (Deiner et al. 2017), different kinds of false positives can be detected by both approaches (Figure 1c–e). Although this artificially increases the discrepancy between methods and the number of species detected, these species generally have a low relative abundance and therefore a moderate impact on final index values.

The taxonomic differences between molecular and morphological inventories mainly affect species present in low abundance. Overall, dominant species, when present in the reference database, are properly detected by both approaches. The application of CFs to molecular inventories has thus made it possible to obtain DNA sequence proportions closer to cell proportions obtained by microscopy. This correction was most effective on molecular inventories from Mayotte samples. This was probably due to the fact that they were strongly dominated by genera with high biovolume (Eunotia: 31.9% of sequences, Ulnaria: 11.7% of sequences) that were poorly detected in morphological inventories (Vasselon et al. 2017b). The application of CF strongly reduced their importance in molecular inventories, allowing proportions closer to those expected in microscopy to be obtained (reduction in the difference of 99% for Eunotia and of 83% for Ulnaria).

Potential of diatom DNA metabarcoding to monitor the ecological status of rivers

Despite the compositional differences described previously, the IPS values, calculated from both inventories, were highly correlated (Figure 6). This is because the most dominant species 1) are detected by both approaches and 2) have the greatest impact on the final index value (Bigler et al. 2009). Other studies on river biofilm samples have also shown this interesting correlation (Visco et al. 2015; Apotheloz-Perret-Gentil et al. 2017; Bailet et al. 2019; Mortágua et al. 2019). In addition, the application of CFs has greatly contributed to reducing the differences in scores between morphological and molecular IPS, improving both the percentage and slope of correlations (Figure 6) (Vasselon et al. 2017b ; Rivera et al. 2020) and has been applied recently on the WFD network of northern Portugal (Mortágua et al. 2019).

Figure 6.

Correlation between morphological and molecular IPS values for samples from the WFD network of Mayotte (left) and of metropolitan France (right). Correction factors (CFs) were (below) or were not (above) applied on molecular taxonomic inventories. The limits of the WFD ecological status classes defined for the IPS are indicated in colour (red = bad, orange = poor, yellow = moderate, green = good, blue = very good), except for Mayotte where these classes are not yet defined. IPS: Indice de Polluosensibilité Spécifique.

Although the IPS index was not developed for the biomonitoring of Mayotte rivers, the corrected morphological and molecular IPS values were both in agreement with the expected quality of the rivers, allowing for a quality gradient ranging from highly impacted to low impacted situations (Figure 6). However, the corrected molecular IPS values were on average 1.9 points higher than the morphological IPS values (Vasselon et al. 2018). This overestimation is partly explained by the lack of some poor quality indicator taxa in molecular inventories. Parts of them were not referenced in the database. Moreover, some of them could still not be identified in metabarcoding, despite the presence of a reference barcode in the database. This is particularly the case of Nitzschia inconspicua, which was a paraphyletic species whose taxonomy was poorly defined until recently (Rovira et al. 2015). This taxonomic instability limited its identification in metabarcoding at the genus level. As this species was abundant in some samples (up to 33.5% of the valves observed in microscopy) and is an indicator of rather degraded environments, we failed to correctly identify it, thereby raising the IPS scores. These taxonomic problems were solved in the latest version (version 7) of the Diat.barcode database (Rimet et al. 2019) and this species is now correctly identifiable.

With regard to rivers in metropolitan France, the ecological status obtained using both approaches are congruent (Figure 6). Some 66.2% of the samples belong to the same quality class, 31.8% of the samples have a deviation of one class and only 2% of the samples have a deviation of two classes. In addition, the average difference between the morphological and corrected molecular IPS values is only 1.5. These results are similar to what could be obtained by comparing morphological IPS scores produced by two different diatom experts, with differences of three points already observed for the same sample, corresponding to a difference of one quality class (Kahlert et al. 2012). These results provide additional evidence for the use of DNA metabarcoding as an operational biomonitoring tool to assess the ecological status of watercourses.

Given all the results from the studies presented here and from other recent tests (e.g. Bailet et al. 2019; Chonova et al. 2019; Mortágua et al. 2019), the use of diatom metabarcoding for ecological assessment seems to be an increasingly promising perspective (Keck et al. 2017; Pawlowski et al. 2018). Recently, England has even abruptly and totally replaced the diatom morphological approach with the molecular one for river monitoring in 2017 (Kelly et al. 2018; Kelly 2019). This still recent and fast developing approach has prompted researchers and biomonitoring stakeholders in Europe to create the COST DNAqua-Net network (Leese et al. 2016, 2018). The objective of this network is to accelerate the process of optimising the molecular approach, harmonise existing methods and protocols between laboratories and define the conditions for implementing molecular approaches within the framework of the WFD. Although the molecular approach gets closer and closer to being technically operational for diatoms, its implementation should guarantee reliable and improved biomonitoring in favour of aquatic ecosystems. This leaves many issues to be addressed with a view to its routine regulatory implementation.

Molecular and morphological approaches are complementary tools

While some recent studies have shown the potential of taxonomy-free approaches (Apotheloz-Perret-Gentil et al. 2017) and machine learning (Tapolczai et al. 2019) to assess water quality using diatom’s metabarcoding, one of the questions is to determine the optimum role of molecular tools. These tools will generally come after the morphology-based ones used in the WFD that benefit from a long time series of valuable biomonitoring. In the case of Mayotte rivers, the implementation of the WFD and the development of a diatom index are currently underway (Tapolczai et al. 2017). As the monitoring history in Mayotte is still short, it seems possible to consider using either approach directly, without losing too much history on the evolution of the ecological status of the rivers. In the case of rivers in metropolitan France, monitoring networks have relied on morphological tools for decades (mainly since the 1990s). In such a case, in addition to the political, economic and social consequences that this decision could have (Kelly et al. 2019), it seems difficult to replace it abruptly with the molecular approach without losing valuable information, especially on long-term changes. The loss of time-series continuity could have serious impacts on the actions undertaken over the past several years to recover water quality. Accordingly, it seems preferable, as a first step, to perform both approaches in parallel when long-time series are available.

DNA metabarcoding has the ability to quickly and cost-effectively evaluate a large number of samples. This high-throughput potential could be used, for example, to monitor large numbers of sites as a first step. Then for sites showing higher risk of not achieving good ecological status or for sites subject to management actions to restore good status, it is more crucial to reliably assess their temporal trajectory. In this case, monitoring with the classical approach will ensure continuity in action and in ecological understanding (Kelly et al. 2015), while tiling with the molecular approach will increase the molecular data log and its promises for the future.

Indeed, the molecular approach opens up new perspectives in terms of environmental monitoring. For the time being, the WFD monitoring network for rivers in metropolitan France includes 1,673 sites monitored once a year, for a total river length of 623,464 km, which remains a poor coverage. The molecular approach, with its ability to deal with a large number of samples, offers the perspective to increase: (i) the number of sites monitored annually to provide more spatially complete monitoring and (ii) the number of monitoring operations throughout the year and thus allow the temporal variability of ecosystems to be explored. The latter could be particularly interesting for rivers submitted to large seasonal variations in terms of pressures (e.g. temporary rivers, Stubbington et al. 2019), that could remain undetected with low frequency monitoring, however damaging they may be.

WFD compatibility of molecular approaches for diatoms

In addition to the need to continue the methodological development of the molecular approach, it is also essential to propose strategies to make the ecological status assessments it produces more reliable:

Validate molecular inventories: as with the minimum number of valves required in microscopic counts (AFNOR 2016), it is necessary to define minimum requirements for molecular taxonomic inventories. This can be thresholds, such as a minimum sequencing depth required per sample, taxon detection limits or the maximum proportion of unassigned sequences per sample. The development of such thresholds will increase the reliability and the comparability of molecular inventories.

Control the metabarcoding workflow: a wide variety of methods, technologies and protocols can be used throughout the workflow of diatom DNA metabarcoding. To validate and compare results, controls should be integrated at each step. For example, it has been proposed to systematically include in each pool of samples, prior to sequencing, a controlled mock-community built from cultures (Kozich et al. 2013). This enables to check whether the whole process has been carried out correctly. Such controls could also make it possible to verify that changes in the metabarcoding process, such as changes in sequencing technology, do not affect the final results.

Intercalibrate and standardise protocols: as for morphological inventories, inter-calibration exercises should be implemented in order to validate methodologies and results from different laboratories and to limit the level of uncertainty in data quality. This should lead to the development of standards which are crucial for the implementation of the molecular approach. For diatom metabarcoding, two CEN technical reports TR 17245 (CEN 2018a) and TR 17244, (CEN 2018b) are already available for the field sampling and the management of reference databases, respectively. The recently launched EcoAlpsWater (INTERREG Alpine Space) project aims at pursuing these efforts at the scale of Alpine region by involving scientists and environmental agencies from 6 countries (France, Italy, Switzerland, Slovenia, Germany and Austria). Further work is required to propose European standards at each step of the workflow, to ensure proper sample preservation, DNA extraction and sequencing, data analysis etc.

Transfer knowledge: it is crucial to set up training courses to enable the transfer of technical and theoretical knowledge to environment managers and stakeholders. The definition and transfer of good practices for implementing the molecular approach will be essential for their effective use in river monitoring and future implementation in environmental regulations.

Deploying the molecular approach in parallel with the morphological during the next WFD monitoring cycle on a large set of WFD river sites will require the definition of a common strategy for: (i) sample collection and storage, (ii) laboratory molecular biology steps (DNA extraction, PCR, sequencing), (iii) bioinformatics processing of DNA sequencing data and (iv) calculation of diatom indices and evaluation of water quality status. Considering the high throughput provided by the molecular approach, it is necessary to meet new needs, such as the sustainability of DNA barcode reference databases, the storage of samples and sequence data and the maintenance and improvement of the molecular approach. To provide a solution for these needs, it is necessary to bring together the different actors (e.g. decision-makers, researchers, managers, monitoring bodies, standardisation bodies, private companies, biotechnology companies), in order to define their respective skills and roles and, thus, to implement this methodology in a concerted and routine manner. This has been initiated at European level within the framework of DNAqua-Net (Hering et al. 2018; Leese et al. 2018) and at a French level thanks to the Mayotte and metropolitan France projects presented in this article. To go further on the implementation of these new monitoring tools, round tables and consultation workshops between actors held in 2019 should provide valuable implementation scenarios (SYNAQUA France-Swiss INTERREG project; Lefrançois et al. 2018). Such efforts should be enlarged to smooth the path towards next generation biomonitoring in Europe.